diff --git a/tensorflow/compiler/tests/cholesky_op_test.py b/tensorflow/compiler/tests/cholesky_op_test.py
index 0a739df1163..41a72fda402 100644
--- a/tensorflow/compiler/tests/cholesky_op_test.py
+++ b/tensorflow/compiler/tests/cholesky_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.compiler.tests import xla_test
 from tensorflow.python.framework import constant_op
@@ -79,7 +78,7 @@ class CholeskyOpTest(xla_test.XLATestCase):
 
       # Generate random positive-definite matrices.
       matrices = np.random.rand(10, 5, 5).astype(dtype)
-      for i in xrange(10):
+      for i in range(10):
         matrices[i] = np.dot(matrices[i].T, matrices[i])
       self._verifyCholesky(matrices, atol=1e-4)
 
diff --git a/tensorflow/compiler/tests/clustering_test.py b/tensorflow/compiler/tests/clustering_test.py
index 64d587268c1..ab7ed89d1e2 100644
--- a/tensorflow/compiler/tests/clustering_test.py
+++ b/tensorflow/compiler/tests/clustering_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.compiler.tests import xla_test
 from tensorflow.python.framework import constant_op
@@ -56,7 +55,7 @@ class ClusteringTest(xla_test.XLATestCase):
         input2 = constant_op.constant(val2, name="const2")
       with self.test_scope():
         output = math_ops.add(input1, input2)
-      for _ in xrange(10):
+      for _ in range(10):
         result = self.evaluate(output)
         self.assertAllClose(result, expected, rtol=1e-3)
 
diff --git a/tensorflow/compiler/tests/conv3d_test.py b/tensorflow/compiler/tests/conv3d_test.py
index 155b513e0d6..f0086c88614 100644
--- a/tensorflow/compiler/tests/conv3d_test.py
+++ b/tensorflow/compiler/tests/conv3d_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.compiler.tests import xla_test
 from tensorflow.python.framework import constant_op
@@ -100,11 +99,11 @@ class Conv3DTransposeTest(xla_test.XLATestCase):
       #   kernel_depth * ceil(kernel_height/2) * kernel_width or
       #   kernel_depth * kernel_height * ceil(kernel_width/2)
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[3]):
-          for w in xrange(y_shape[3]):
-            for h in xrange(y_shape[2]):
-              for d in xrange(y_shape[1]):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[3]):
+          for w in range(y_shape[3]):
+            for h in range(y_shape[2]):
+              for d in range(y_shape[1]):
                 d_in = d > 0 and d < y_shape[1] - 1
                 h_in = h > 0 and h < y_shape[2] - 1
                 w_in = w > 0 and w < y_shape[3] - 1
@@ -137,11 +136,11 @@ class Conv3DTransposeTest(xla_test.XLATestCase):
           x, f, y_shape, strides=strides, padding="SAME")
       value = self.evaluate(output)
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[3]):
-          for w in xrange(y_shape[3]):
-            for h in xrange(y_shape[2]):
-              for d in xrange(y_shape[1]):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[3]):
+          for w in range(y_shape[3]):
+            for h in range(y_shape[2]):
+              for d in range(y_shape[1]):
                 # We add a case for locations divisible by the stride.
                 d_in = d % strides[1] == 0 and 0 < d < y_shape[1] - 1
                 h_in = h % strides[2] == 0 and 0 < h < y_shape[2] - 1
@@ -180,11 +179,11 @@ class Conv3DTransposeTest(xla_test.XLATestCase):
       # The amount of padding added
       pad = 1
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[3]):
-          for w in xrange(y_shape[3]):
-            for h in xrange(y_shape[2]):
-              for d in xrange(y_shape[1]):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[3]):
+          for w in range(y_shape[3]):
+            for h in range(y_shape[2]):
+              for d in range(y_shape[1]):
                 # We add a case for locations divisible by the stride.
                 d_in = d % strides[1] == 0 and pad < d < y_shape[1] - 1 - pad
                 h_in = h % strides[2] == 0 and pad < h < y_shape[2] - 1 - pad
diff --git a/tensorflow/compiler/tests/depthwise_conv_op_test.py b/tensorflow/compiler/tests/depthwise_conv_op_test.py
index a49985f0446..61dcc919512 100644
--- a/tensorflow/compiler/tests/depthwise_conv_op_test.py
+++ b/tensorflow/compiler/tests/depthwise_conv_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.compiler.tests import xla_test
 from tensorflow.python.framework import constant_op
@@ -39,7 +38,7 @@ def ReferenceDepthwiseConv2D(input_tensor, filter_tensor, strides, padding,
   convs = []
   in_channels = filter_tensor.shape[2]
   # Use a custom implementation of depthwise conv2d using slicing.
-  for channel in xrange(in_channels):
+  for channel in range(in_channels):
     # Slice the input along channel
     if data_format == "NCHW":
       input_slice = input_tensor[:, channel:channel+1, :, :]
diff --git a/tensorflow/compiler/tests/fifo_queue_test.py b/tensorflow/compiler/tests/fifo_queue_test.py
index ba80fa0a0b2..f2ff9db6d51 100644
--- a/tensorflow/compiler/tests/fifo_queue_test.py
+++ b/tensorflow/compiler/tests/fifo_queue_test.py
@@ -20,7 +20,6 @@ from __future__ import print_function
 
 import time
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.compiler.tests import xla_test
 from tensorflow.python.framework import dtypes as dtypes_lib
@@ -90,7 +89,7 @@ class FIFOQueueTest(xla_test.XLATestCase):
 
       # Dequeue every element using a single thread.
       results = []
-      for _ in xrange(len(elems)):
+      for _ in range(len(elems)):
         results.append(dequeued_t.eval())
       self.assertItemsEqual(elems, results)
 
@@ -128,7 +127,7 @@ class FIFOQueueTest(xla_test.XLATestCase):
       for enqueue_op in enqueue_ops:
         enqueue_op.run()
 
-      for i in xrange(len(elems)):
+      for i in range(len(elems)):
         vals = self.evaluate(dequeued_t)
         self.assertEqual([elems[i]], vals)
 
@@ -149,7 +148,7 @@ class FIFOQueueTest(xla_test.XLATestCase):
       results = []
 
       def dequeue():
-        for _ in xrange(len(elems)):
+        for _ in range(len(elems)):
           results.append(sess.run(dequeued_t))
 
       enqueue_thread = self.checkedThread(target=enqueue)
@@ -172,7 +171,7 @@ class FIFOQueueTest(xla_test.XLATestCase):
       for enqueue_op in enqueue_ops:
         enqueue_op.run()
 
-      for i in xrange(len(elems)):
+      for i in range(len(elems)):
         x_val, y_val = sess.run(dequeued_t)
         x, y = elems[i]
         self.assertEqual([x], x_val)
diff --git a/tensorflow/compiler/tests/image_ops_test.py b/tensorflow/compiler/tests/image_ops_test.py
index 5889a011296..2225cc021dd 100644
--- a/tensorflow/compiler/tests/image_ops_test.py
+++ b/tensorflow/compiler/tests/image_ops_test.py
@@ -24,7 +24,6 @@ import math
 from absl.testing import parameterized
 import numpy as np
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.compiler.tests import xla_test
 from tensorflow.python.framework import dtypes
@@ -230,7 +229,7 @@ class AdjustHueTest(xla_test.XLATestCase):
     x_v = x_np.reshape([-1, 3])
     y_v = np.ndarray(x_v.shape, dtype=x_v.dtype)
     channel_count = x_v.shape[0]
-    for i in xrange(channel_count):
+    for i in range(channel_count):
       r = x_v[i][0]
       g = x_v[i][1]
       b = x_v[i][2]
@@ -350,7 +349,7 @@ class AdjustSaturationTest(xla_test.XLATestCase):
     x_v = x_np.reshape([-1, 3])
     y_v = np.ndarray(x_v.shape, dtype=x_v.dtype)
     channel_count = x_v.shape[0]
-    for i in xrange(channel_count):
+    for i in range(channel_count):
       r = x_v[i][0]
       g = x_v[i][1]
       b = x_v[i][2]
diff --git a/tensorflow/compiler/tests/listdiff_op_test.py b/tensorflow/compiler/tests/listdiff_op_test.py
index b061ac3c817..a6f512b9588 100644
--- a/tensorflow/compiler/tests/listdiff_op_test.py
+++ b/tensorflow/compiler/tests/listdiff_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.compiler.tests import xla_test
 from tensorflow.python.framework import dtypes
@@ -65,7 +64,7 @@ class ListDiffTest(xla_test.XLATestCase):
     int_low = -7
     int_high = 8
     max_size = 50
-    for _ in xrange(num_random_tests):
+    for _ in range(num_random_tests):
       x_size = np.random.randint(max_size + 1)
       x = np.random.randint(int_low, int_high, size=x_size)
       y_size = np.random.randint(max_size + 1)
diff --git a/tensorflow/compiler/tests/test_utils.py b/tensorflow/compiler/tests/test_utils.py
index 0e77dbf1a79..799a677502e 100644
--- a/tensorflow/compiler/tests/test_utils.py
+++ b/tensorflow/compiler/tests/test_utils.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 
 def ConvertBetweenDataFormats(x, data_format_src, data_format_dst):
@@ -69,7 +68,7 @@ _JIT_WARMUP_ITERATIONS = 10
 
 def RunWithWarmup(sess, op_to_run, feed_dict, options=None, run_metadata=None):
   """Runs a graph a few times to ensure that its clusters are compiled."""
-  for _ in xrange(0, _JIT_WARMUP_ITERATIONS):
+  for _ in range(0, _JIT_WARMUP_ITERATIONS):
     sess.run(op_to_run, feed_dict, options=options)
   return sess.run(
       op_to_run, feed_dict, options=options, run_metadata=run_metadata)
diff --git a/tensorflow/compiler/tests/unary_ops_test.py b/tensorflow/compiler/tests/unary_ops_test.py
index 349dabbb393..ea3f70e35a5 100644
--- a/tensorflow/compiler/tests/unary_ops_test.py
+++ b/tensorflow/compiler/tests/unary_ops_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import unittest
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.compiler.tests import xla_test
 from tensorflow.python.framework import dtypes
@@ -81,7 +80,7 @@ class UnaryOpsTest(xla_test.XLATestCase):
   def ListsAreClose(self, result, expected, rtol, atol):
     """Tests closeness of two lists of floats."""
     self.assertEqual(len(result), len(expected))
-    for i in xrange(len(result)):
+    for i in range(len(result)):
       self.assertAllClose(result[i], expected[i], rtol, atol)
 
   def testAllTypeOps(self):
diff --git a/tensorflow/compiler/xla/cpu_function_runtime.cc b/tensorflow/compiler/xla/cpu_function_runtime.cc
index 517b30a8251..43ca4806711 100644
--- a/tensorflow/compiler/xla/cpu_function_runtime.cc
+++ b/tensorflow/compiler/xla/cpu_function_runtime.cc
@@ -16,6 +16,7 @@ limitations under the License.
 #include "tensorflow/compiler/xla/cpu_function_runtime.h"
 
 #include "tensorflow/core/platform/dynamic_annotations.h"
+#include <cstdint>
 
 namespace xla {
 namespace {
diff --git a/tensorflow/compiler/xla/python_api/xla_shape.py b/tensorflow/compiler/xla/python_api/xla_shape.py
index bdcd4abd6cc..5b59c0086eb 100644
--- a/tensorflow/compiler/xla/python_api/xla_shape.py
+++ b/tensorflow/compiler/xla/python_api/xla_shape.py
@@ -20,7 +20,6 @@ from __future__ import print_function
 
 import numpy as _np  # Avoids becoming a part of public Tensorflow API.
 
-from six.moves import xrange
 
 from tensorflow.compiler.xla import xla_data_pb2
 from tensorflow.compiler.xla.python_api import types
@@ -121,7 +120,7 @@ def _CreateShapeFromNumpy(ndarray):  # pylint: disable=invalid-name
   else:
     # Row-major layout. This corresponds to a "dimension order is
     # major-to-minor" layout int XLA.
-    layout = list(reversed(xrange(ndarray.ndim)))
+    layout = list(reversed(range(ndarray.ndim)))
 
   return Shape(element_type, dimensions, layout)
 
diff --git a/tensorflow/contrib/constrained_optimization/python/candidates.py b/tensorflow/contrib/constrained_optimization/python/candidates.py
index 66d7ebed74d..c1bcb30fae7 100644
--- a/tensorflow/contrib/constrained_optimization/python/candidates.py
+++ b/tensorflow/contrib/constrained_optimization/python/candidates.py
@@ -48,7 +48,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 
 def _find_best_candidate_distribution_helper(objective_vector,
@@ -294,7 +293,7 @@ def find_best_candidate_index(objective_vector,
     maximum_ranks = scipy.stats.rankdata(objective_vector, method="min")
   else:
     maximum_ranks = np.zeros(nn, dtype=np.int64)
-  for ii in xrange(mm):
+  for ii in range(mm):
     # Take the maximum of the constraint functions with zero, since we want to
     # rank the magnitude of constraint *violations*. If the constraint is
     # satisfied, then we don't care how much it's satisfied by (as a result, we
@@ -306,7 +305,7 @@ def find_best_candidate_index(objective_vector,
   best_index = None
   best_rank = float("Inf")
   best_objective = float("Inf")
-  for ii in xrange(nn):
+  for ii in range(nn):
     if maximum_ranks[ii] < best_rank:
       best_index = ii
       best_rank = maximum_ranks[ii]
diff --git a/tensorflow/contrib/cudnn_rnn/python/kernel_tests/cudnn_rnn_ops_benchmark.py b/tensorflow/contrib/cudnn_rnn/python/kernel_tests/cudnn_rnn_ops_benchmark.py
index 933df6d71dd..72a07bd0c83 100644
--- a/tensorflow/contrib/cudnn_rnn/python/kernel_tests/cudnn_rnn_ops_benchmark.py
+++ b/tensorflow/contrib/cudnn_rnn/python/kernel_tests/cudnn_rnn_ops_benchmark.py
@@ -20,7 +20,6 @@ from __future__ import print_function
 
 import time
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 from tensorflow.contrib import rnn as contrib_rnn
 from tensorflow.contrib.cudnn_rnn.python.ops import cudnn_rnn_ops
 from tensorflow.contrib.rnn.python.ops import lstm_ops
@@ -74,7 +73,7 @@ class CudnnRNNBenchmark(test.Benchmark):
     benchmark_steps = 20
     with session.Session() as sess:
       sess.run(variables.global_variables_initializer())
-      for i in xrange(burn_in_steps + benchmark_steps):
+      for i in range(burn_in_steps + benchmark_steps):
         if i == burn_in_steps:
           start_time = time.time()
         sess.run(op)
diff --git a/tensorflow/contrib/eager/python/examples/densenet/densenet_test.py b/tensorflow/contrib/eager/python/examples/densenet/densenet_test.py
index a9fb0035d29..16744faf549 100644
--- a/tensorflow/contrib/eager/python/examples/densenet/densenet_test.py
+++ b/tensorflow/contrib/eager/python/examples/densenet/densenet_test.py
@@ -235,13 +235,13 @@ class DensenetBenchmark(tf.test.Benchmark):
       num_iters = 30
       with tf.device(device):
         images, _ = random_batch(batch_size, data_format)
-        for _ in xrange(num_burn):
+        for _ in range(num_burn):
           model(images, training=False).cpu()
         if execution_mode:
           tfe.async_wait()
         gc.collect()
         start = time.time()
-        for _ in xrange(num_iters):
+        for _ in range(num_iters):
           model(images, training=False).cpu()
         if execution_mode:
           tfe.async_wait()
@@ -286,7 +286,7 @@ class DensenetBenchmark(tf.test.Benchmark):
         num_iters = 10
         with tf.device(device):
           iterator = make_iterator((images, labels))
-          for _ in xrange(num_burn):
+          for _ in range(num_burn):
             (images, labels) = iterator.next()
             apply_grads(model, optimizer,
                         compute_gradients(model, images, labels))
@@ -296,7 +296,7 @@ class DensenetBenchmark(tf.test.Benchmark):
           gc.collect()
 
           start = time.time()
-          for _ in xrange(num_iters):
+          for _ in range(num_iters):
             (images, labels) = iterator.next()
             apply_grads(model, optimizer,
                         compute_gradients(model, images, labels))
diff --git a/tensorflow/contrib/eager/python/examples/resnet50/resnet50_test.py b/tensorflow/contrib/eager/python/examples/resnet50/resnet50_test.py
index fb81979d7bd..941bfd68200 100644
--- a/tensorflow/contrib/eager/python/examples/resnet50/resnet50_test.py
+++ b/tensorflow/contrib/eager/python/examples/resnet50/resnet50_test.py
@@ -22,7 +22,6 @@ import gc
 import tempfile
 import time
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 
 import tensorflow.contrib.eager as tfe
@@ -227,13 +226,13 @@ class ResNet50Benchmarks(tf.test.Benchmark):
       num_iters = 30
       with tf.device(device):
         images, _ = random_batch(batch_size, data_format)
-        for _ in xrange(num_burn):
+        for _ in range(num_burn):
           model(images, training=False).cpu()
         if execution_mode:
           tfe.async_wait()
         gc.collect()
         start = time.time()
-        for _ in xrange(num_iters):
+        for _ in range(num_iters):
           model(images, training=False).cpu()
         if execution_mode:
           tfe.async_wait()
@@ -273,7 +272,7 @@ class ResNet50Benchmarks(tf.test.Benchmark):
         num_iters = 10
         with tf.device(device):
           iterator = make_iterator((images, labels))
-          for _ in xrange(num_burn):
+          for _ in range(num_burn):
             (images, labels) = iterator.next()
             apply_grads(model, optimizer,
                         compute_gradients(model, images, labels))
@@ -283,7 +282,7 @@ class ResNet50Benchmarks(tf.test.Benchmark):
           gc.collect()
 
           start = time.time()
-          for _ in xrange(num_iters):
+          for _ in range(num_iters):
             (images, labels) = iterator.next()
             apply_grads(model, optimizer,
                         compute_gradients(model, images, labels))
diff --git a/tensorflow/contrib/eager/python/examples/revnet/cifar_tfrecords.py b/tensorflow/contrib/eager/python/examples/revnet/cifar_tfrecords.py
index 377844ad8fb..ec1c9323498 100644
--- a/tensorflow/contrib/eager/python/examples/revnet/cifar_tfrecords.py
+++ b/tensorflow/contrib/eager/python/examples/revnet/cifar_tfrecords.py
@@ -28,7 +28,7 @@ import sys
 import tarfile
 
 from absl import flags
-from six.moves import cPickle as pickle
+import pickle as pickle
 from six.moves import urllib
 import tensorflow as tf
 
diff --git a/tensorflow/contrib/eager/python/examples/spinn/spinn_test.py b/tensorflow/contrib/eager/python/examples/spinn/spinn_test.py
index 6e47cc57051..0ee1f945bf2 100644
--- a/tensorflow/contrib/eager/python/examples/spinn/spinn_test.py
+++ b/tensorflow/contrib/eager/python/examples/spinn/spinn_test.py
@@ -26,7 +26,6 @@ import tempfile
 import time
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 
 # pylint: disable=g-bad-import-order
@@ -462,7 +461,7 @@ class EagerSpinnSNLIClassifierBenchmark(test.Benchmark):
 
       gc.collect()
       start_time = time.time()
-      for _ in xrange(benchmark_iterations):
+      for _ in range(benchmark_iterations):
         trainer.train_batch(labels, prem, prem_trans, hypo, hypo_trans)
       wall_time = time.time() - start_time
       # Named "examples"_per_sec to conform with other benchmarks.
diff --git a/tensorflow/contrib/factorization/examples/mnist.py b/tensorflow/contrib/factorization/examples/mnist.py
index 06a62db0049..ce25a85f384 100644
--- a/tensorflow/contrib/factorization/examples/mnist.py
+++ b/tensorflow/contrib/factorization/examples/mnist.py
@@ -33,7 +33,6 @@ import sys
 import tempfile
 import time
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 
 from tensorflow.examples.tutorials.mnist import input_data
@@ -106,7 +105,7 @@ def do_eval(sess,
   batch_size = min(FLAGS.batch_size, data_set.num_examples)
   steps_per_epoch = data_set.num_examples // batch_size
   num_examples = steps_per_epoch * batch_size
-  for _ in xrange(steps_per_epoch):
+  for _ in range(steps_per_epoch):
     feed_dict = fill_feed_dict(data_set,
                                images_placeholder,
                                labels_placeholder,
@@ -227,7 +226,7 @@ def run_training():
 
     # Start the training loop.
     max_test_prec = 0
-    for step in xrange(FLAGS.max_steps):
+    for step in range(FLAGS.max_steps):
       start_time = time.time()
 
       # Fill a feed dictionary with the actual set of images and labels
diff --git a/tensorflow/contrib/factorization/python/ops/factorization_ops.py b/tensorflow/contrib/factorization/python/ops/factorization_ops.py
index e04de0579b1..d5407b70830 100644
--- a/tensorflow/contrib/factorization/python/ops/factorization_ops.py
+++ b/tensorflow/contrib/factorization/python/ops/factorization_ops.py
@@ -20,7 +20,6 @@ from __future__ import print_function
 
 import numbers
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.factorization.python.ops import gen_factorization_ops
 from tensorflow.contrib.util import loader
@@ -512,7 +511,7 @@ class WALSModel(object):
       assert var
       cache = [
           WALSModel._transient_var(name="%s_shard_%d" % (name, i))
-          for i in xrange(len(var))
+          for i in range(len(var))
       ]
       reset_ops = []
       for i, c in enumerate(cache):
@@ -667,7 +666,7 @@ class WALSModel(object):
       sharded_values = data_flow_ops.dynamic_partition(values, assignments,
                                                        num_shards)
       updates = []
-      for i in xrange(num_shards):
+      for i in range(num_shards):
         updates.append(
             state_ops.scatter_update(factor[i], sharded_ids[i], sharded_values[
                 i]))
diff --git a/tensorflow/contrib/factorization/python/ops/factorization_ops_test.py b/tensorflow/contrib/factorization/python/ops/factorization_ops_test.py
index 6aa62fb82e8..c2f4bd9be2b 100644
--- a/tensorflow/contrib/factorization/python/ops/factorization_ops_test.py
+++ b/tensorflow/contrib/factorization/python/ops/factorization_ops_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.factorization.python.ops import factorization_ops
 from tensorflow.contrib.factorization.python.ops import factorization_ops_test_utils
@@ -659,7 +658,7 @@ class WalsModelTest(test.TestCase):
 
     model.initialize_op.run()
     model.worker_init.run()
-    for _ in xrange(num_iterations):
+    for _ in range(num_iterations):
       model.row_update_prep_gramian_op.run()
       model.initialize_row_update_op.run()
       row_update_op.run()
@@ -676,7 +675,7 @@ class WalsModelTest(test.TestCase):
     with ops.Graph().as_default(), self.cached_session():
       data = np.dot(np.random.rand(rows, 3), np.random.rand(
           3, cols)).astype(np.float32) / 3.0
-      indices = [[i, j] for i in xrange(rows) for j in xrange(cols)]
+      indices = [[i, j] for i in range(rows) for j in range(cols)]
       values = data.reshape(-1)
       inp = sparse_tensor.SparseTensor(indices, values, [rows, cols])
       model = factorization_ops.WALSModel(
@@ -706,7 +705,7 @@ class WalsModelTest(test.TestCase):
     with ops.Graph().as_default(), self.cached_session():
       data = np.dot(np.random.rand(rows, 3), np.random.rand(
           3, cols)).astype(np.float32) / 3.0
-      indices = [[i, j] for i in xrange(rows) for j in xrange(cols)]
+      indices = [[i, j] for i in range(rows) for j in range(cols)]
       values = data.reshape(-1)
       inp = sparse_tensor.SparseTensor(indices, values, [rows, cols])
       model = factorization_ops.WALSModel(
@@ -744,7 +743,7 @@ class WalsModelTest(test.TestCase):
       indices = np.array(
           list(
               filter(keep_index,
-                     [[i, j] for i in xrange(rows) for j in xrange(cols)])))
+                     [[i, j] for i in range(rows) for j in range(cols)])))
       values = data[indices[:, 0], indices[:, 1]]
       inp = sparse_tensor.SparseTensor(indices, values, [rows, cols])
       model = factorization_ops.WALSModel(
@@ -760,8 +759,8 @@ class WalsModelTest(test.TestCase):
       row_factor = model.row_factors[0].eval()
       col_factor = model.col_factors[0].eval()
       out = np.dot(row_factor, np.transpose(col_factor))
-      for i in xrange(rows):
-        for j in xrange(cols):
+      for i in range(rows):
+        for j in range(cols):
           if keep_index([i, j]):
             self.assertNear(
                 data[i][j], out[i][j], err=0.4, msg="%d, %d" % (i, j))
diff --git a/tensorflow/contrib/factorization/python/ops/gmm_ops_test.py b/tensorflow/contrib/factorization/python/ops/gmm_ops_test.py
index 112e4d289b0..d208c9c5751 100644
--- a/tensorflow/contrib/factorization/python/ops/gmm_ops_test.py
+++ b/tensorflow/contrib/factorization/python/ops/gmm_ops_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import time
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.factorization.python.ops import gmm_ops
 from tensorflow.python.framework import constant_op
@@ -59,7 +58,7 @@ class GmmOpsTest(test.TestCase):
     """
     vectors = []
     classes = []
-    for _ in xrange(num_vectors):
+    for _ in range(num_vectors):
       if np.random.random() > 0.5:
         vectors.append([np.random.normal(2.0, 0.6), np.random.normal(2.0, 0.9)])
         classes.append(0)
@@ -82,7 +81,7 @@ class GmmOpsTest(test.TestCase):
     """
     vectors = []
     classes = []
-    for _ in xrange(num_vectors):
+    for _ in range(num_vectors):
       current_class = np.random.random_integers(0, len(centers) - 1)
       vectors.append([
           np.random.normal(centers[current_class][0],
@@ -128,7 +127,7 @@ class GmmOpsTest(test.TestCase):
         variables.global_variables_initializer().run()
         sess.run(init_op)
         first_loss = sess.run(loss_op)
-        for _ in xrange(self.iterations):
+        for _ in range(self.iterations):
           sess.run(training_op)
         assignments = sess.run(assignments)
         end_loss = sess.run(loss_op)
@@ -152,7 +151,7 @@ class GmmOpsTest(test.TestCase):
       training_ops = gmm_tool.training_ops()
       variables.global_variables_initializer().run()
       sess.run(gmm_tool.init_ops())
-      for _ in xrange(self.iterations):
+      for _ in range(self.iterations):
         sess.run(training_ops)
 
       # Only the probability to each class is updated.
@@ -170,7 +169,7 @@ class GmmOpsTest(test.TestCase):
       training_ops = gmm_tool.training_ops()
       variables.global_variables_initializer().run()
       sess.run(gmm_tool.init_ops())
-      for _ in xrange(self.iterations):
+      for _ in range(self.iterations):
         sess.run(training_ops)
       alphas = sess.run(gmm_tool.alphas())
       self.assertAlmostEqual(alphas[0], alphas[1])
@@ -189,7 +188,7 @@ class GmmOpsTest(test.TestCase):
       training_ops = gmm_tool.training_ops()
       variables.global_variables_initializer().run()
       sess.run(gmm_tool.init_ops())
-      for _ in xrange(self.iterations):
+      for _ in range(self.iterations):
         sess.run(training_ops)
       alphas = sess.run(gmm_tool.alphas())
       self.assertAlmostEqual(alphas[0], alphas[1])
diff --git a/tensorflow/contrib/factorization/python/ops/kmeans.py b/tensorflow/contrib/factorization/python/ops/kmeans.py
index 260ea5bf127..f4ec267a386 100644
--- a/tensorflow/contrib/factorization/python/ops/kmeans.py
+++ b/tensorflow/contrib/factorization/python/ops/kmeans.py
@@ -265,7 +265,7 @@ class KMeansClustering(estimator.Estimator):
   # train
   num_iterations = 10
   previous_centers = None
-  for _ in xrange(num_iterations):
+  for _ in range(num_iterations):
     kmeans.train(input_fn)
     cluster_centers = kmeans.cluster_centers()
     if previous_centers is not None:
diff --git a/tensorflow/contrib/image/python/kernel_tests/distort_image_ops_test.py b/tensorflow/contrib/image/python/kernel_tests/distort_image_ops_test.py
index ac8573445ca..03be762c9c2 100644
--- a/tensorflow/contrib/image/python/kernel_tests/distort_image_ops_test.py
+++ b/tensorflow/contrib/image/python/kernel_tests/distort_image_ops_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import time
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.image.python.ops import distort_image_ops
 from tensorflow.core.protobuf import config_pb2
@@ -271,7 +270,7 @@ class AdjustHueInYiqBenchmark(test.Benchmark):
         outputs = distort_image_ops.adjust_hsv_in_yiq(inputs, delta, 1, 1)
         run_op = control_flow_ops.group(outputs)
         sess.run(variables.global_variables_initializer())
-        for i in xrange(warmup_rounds + benchmark_rounds):
+        for i in range(warmup_rounds + benchmark_rounds):
           if i == warmup_rounds:
             start = time.time()
           sess.run(run_op)
@@ -315,10 +314,10 @@ class AdjustSaturationInYiqBenchmark(test.Benchmark):
         outputs = distort_image_ops.adjust_hsv_in_yiq(inputs, 0, scale, 1)
         run_op = control_flow_ops.group(outputs)
         sess.run(variables.global_variables_initializer())
-        for _ in xrange(warmup_rounds):
+        for _ in range(warmup_rounds):
           sess.run(run_op)
         start = time.time()
-        for _ in xrange(benchmark_rounds):
+        for _ in range(benchmark_rounds):
           sess.run(run_op)
     end = time.time()
     step_time = (end - start) / benchmark_rounds
diff --git a/tensorflow/contrib/kernel_methods/python/mappers/random_fourier_features_test.py b/tensorflow/contrib/kernel_methods/python/mappers/random_fourier_features_test.py
index bad0a596a78..12ff66982e7 100644
--- a/tensorflow/contrib/kernel_methods/python/mappers/random_fourier_features_test.py
+++ b/tensorflow/contrib/kernel_methods/python/mappers/random_fourier_features_test.py
@@ -18,7 +18,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.kernel_methods.python.mappers import dense_kernel_mapper
 from tensorflow.contrib.kernel_methods.python.mappers.random_fourier_features import RandomFourierFeatureMapper
@@ -134,7 +133,7 @@ class RandomFourierFeatureMapperTest(TensorFlowTestCase):
     points_shape = [1, input_dim]
     points = [
         random_ops.random_uniform(shape=points_shape, maxval=1.0)
-        for _ in xrange(num_points)
+        for _ in range(num_points)
     ]
 
     normalized_points = [nn.l2_normalize(point, dim=1) for point in points]
diff --git a/tensorflow/contrib/labeled_tensor/python/ops/core.py b/tensorflow/contrib/labeled_tensor/python/ops/core.py
index 394254cbd90..ef48a6de017 100644
--- a/tensorflow/contrib/labeled_tensor/python/ops/core.py
+++ b/tensorflow/contrib/labeled_tensor/python/ops/core.py
@@ -33,7 +33,6 @@ import numpy as np
 from six import binary_type
 from six import string_types
 from six import text_type
-from six.moves import range  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.labeled_tensor.python.ops import _typecheck as tc
 from tensorflow.python.framework import dtypes
diff --git a/tensorflow/contrib/labeled_tensor/python/ops/core_test.py b/tensorflow/contrib/labeled_tensor/python/ops/core_test.py
index e378db56afb..738c45523a4 100644
--- a/tensorflow/contrib/labeled_tensor/python/ops/core_test.py
+++ b/tensorflow/contrib/labeled_tensor/python/ops/core_test.py
@@ -22,7 +22,6 @@ import re
 import textwrap
 
 import numpy as np
-from six.moves import range  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.labeled_tensor.python.ops import _typecheck as tc
 from tensorflow.contrib.labeled_tensor.python.ops import core
diff --git a/tensorflow/contrib/labeled_tensor/python/ops/ops_test.py b/tensorflow/contrib/labeled_tensor/python/ops/ops_test.py
index 9a402d888cf..68289bca64b 100644
--- a/tensorflow/contrib/labeled_tensor/python/ops/ops_test.py
+++ b/tensorflow/contrib/labeled_tensor/python/ops/ops_test.py
@@ -18,7 +18,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import range  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.labeled_tensor.python.ops import core
 from tensorflow.contrib.labeled_tensor.python.ops import ops
diff --git a/tensorflow/contrib/labeled_tensor/python/ops/sugar_test.py b/tensorflow/contrib/labeled_tensor/python/ops/sugar_test.py
index 2797e7d5251..6c71918449f 100644
--- a/tensorflow/contrib/labeled_tensor/python/ops/sugar_test.py
+++ b/tensorflow/contrib/labeled_tensor/python/ops/sugar_test.py
@@ -17,7 +17,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import range  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.labeled_tensor.python.ops import core
 from tensorflow.contrib.labeled_tensor.python.ops import ops
diff --git a/tensorflow/contrib/layers/python/layers/embedding_ops.py b/tensorflow/contrib/layers/python/layers/embedding_ops.py
index 14bbe5f9b30..2fcce32799a 100644
--- a/tensorflow/contrib/layers/python/layers/embedding_ops.py
+++ b/tensorflow/contrib/layers/python/layers/embedding_ops.py
@@ -17,7 +17,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.framework.python.framework import tensor_util as contrib_tensor_util
 from tensorflow.contrib.layers.python.ops import sparse_feature_cross_op
@@ -799,13 +798,13 @@ def _embedding_lookup_with_distributed_aggregation(params,
         # partitions based on a constant number of ids per partition. Optimize
         # if we already know the full shape statically.
         dim_0_size = params[0].get_shape().dims[0]
-        for p in xrange(1, np):
+        for p in range(1, np):
           dim_0_size += params[p].get_shape().dims[0]
         if dim_0_size.value:
           num_total_ids = constant_op.constant(dim_0_size, flat_ids.dtype)
         else:
           dim_0_sizes = []
-          for p in xrange(np):
+          for p in range(np):
             if params[p].get_shape().dims[0].value is not None:
               dim_0_sizes.append(params[p].get_shape().dims[0].value)
             else:
@@ -842,7 +841,7 @@ def _embedding_lookup_with_distributed_aggregation(params,
                                                  p_assignments, np)
       # Do np separate lookups, finding embeddings for plist[p] in params[p]
       partitioned_result = []
-      for p in xrange(np):
+      for p in range(np):
         with ops.colocate_with(params[p]):
           partitioned_result.append(_do_gather(params[p], gather_ids[p]))
 
@@ -850,14 +849,14 @@ def _embedding_lookup_with_distributed_aggregation(params,
       if not ignore_weights:
         # Partition weights according to pindices.
         partitioned_weight = []
-        for p in xrange(np):
+        for p in range(np):
           partitioned_weight.append(array_ops.gather(weights, pindices[p]))
       # Reshape each partition result.
       element_shape = params[0].get_shape()[1:]
       for p in params[1:]:
         element_shape = element_shape.merge_with(p.get_shape()[1:])
       if element_shape.is_fully_defined():
-        for p in xrange(np):
+        for p in range(np):
           with ops.colocate_with(params[p]):
             partitioned_result[p] = array_ops.reshape(
                 partitioned_result[p],
@@ -866,7 +865,7 @@ def _embedding_lookup_with_distributed_aggregation(params,
       else:
         with ops.colocate_with(params[0]):
           params_shape = array_ops.shape(params[0])
-        for p in xrange(np):
+        for p in range(np):
           with ops.colocate_with(params[p]):
             partitioned_result[p] = array_ops.reshape(
                 partitioned_result[p],
@@ -875,12 +874,12 @@ def _embedding_lookup_with_distributed_aggregation(params,
                     array_ops.slice(params_shape, [1], [-1])
                 ], 0))
       # Normalize each partition result.
-      for p in xrange(np):
+      for p in range(np):
         with ops.colocate_with(params[p]):
           partitioned_result[p] = maybe_normalize(partitioned_result[p])
       if not ignore_weights:
         # Multiply each partition result with partition weights.
-        for p in xrange(np):
+        for p in range(np):
           with ops.colocate_with(params[p]):
             if partitioned_weight[p].dtype != partitioned_result[p].dtype:
               partitioned_weight[p] = math_ops.cast(partitioned_weight[p],
@@ -902,7 +901,7 @@ def _embedding_lookup_with_distributed_aggregation(params,
                   ]))
             partitioned_result[p] *= partitioned_weight[p]
       partitioned_segment_ids = []
-      for p in xrange(np):
+      for p in range(np):
         if not ignore_weights:
           # Partition segment_ids according to pindices.
           p_segment_ids = array_ops.gather(segment_ids, pindices[p])
diff --git a/tensorflow/contrib/layers/python/layers/rev_block_lib.py b/tensorflow/contrib/layers/python/layers/rev_block_lib.py
index 06da32072f3..e16a1380c4b 100644
--- a/tensorflow/contrib/layers/python/layers/rev_block_lib.py
+++ b/tensorflow/contrib/layers/python/layers/rev_block_lib.py
@@ -31,7 +31,6 @@ import re
 
 import numpy as np
 import six
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.framework.python import ops as contrib_framework_ops
 from tensorflow.python.eager import backprop
@@ -145,7 +144,7 @@ def _rev_block_forward(x1,
                        gate_outputs=False):
   """Forward for a series of reversible layers."""
   out = (x1, x2)
-  for i in xrange(num_layers):
+  for i in range(num_layers):
     out = _rev_layer_forward(
         out, f[i], g[i], f_side_input, g_side_input, gate_outputs=gate_outputs)
 
@@ -294,7 +293,7 @@ class RevBlock(base.Layer):
       g.reverse()
 
       with variable_scope.variable_scope(self.scope_name, reuse=True):
-        for i in xrange(self.num_layers):
+        for i in range(self.num_layers):
           ys, grad_ys, f_ret, g_ret = _rev_layer_backward(
               ys, grad_ys, f[i], g[i], f_vars[i], self.f_side_input, g_vars[i],
               self.g_side_input)
@@ -369,7 +368,7 @@ class RevBlock(base.Layer):
     f.reverse()
     g.reverse()
 
-    for i in xrange(self.num_layers):
+    for i in range(self.num_layers):
       gy1 = g[i](y1, self.g_side_input) if self.g_side_input else g[i](y1)
       x2 = y2 - gy1
       fx2 = f[i](x2, self.f_side_input) if self.f_side_input else f[i](x2)
diff --git a/tensorflow/contrib/learn/python/learn/datasets/mnist.py b/tensorflow/contrib/learn/python/learn/datasets/mnist.py
index abbb44c2f5b..d30e81f8ba5 100644
--- a/tensorflow/contrib/learn/python/learn/datasets/mnist.py
+++ b/tensorflow/contrib/learn/python/learn/datasets/mnist.py
@@ -26,7 +26,6 @@ from __future__ import print_function
 import gzip
 
 import numpy
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.learn.python.learn.datasets import base
 from tensorflow.python.framework import dtypes
@@ -188,8 +187,8 @@ class DataSet(object):
         fake_label = [1] + [0] * 9
       else:
         fake_label = 0
-      return [fake_image for _ in xrange(batch_size)], [
-          fake_label for _ in xrange(batch_size)
+      return [fake_image for _ in range(batch_size)], [
+          fake_label for _ in range(batch_size)
       ]
     start = self._index_in_epoch
     # Shuffle for the first epoch
diff --git a/tensorflow/contrib/learn/python/learn/estimators/estimator_test.py b/tensorflow/contrib/learn/python/learn/estimators/estimator_test.py
index 153d4867961..85e1807c117 100644
--- a/tensorflow/contrib/learn/python/learn/estimators/estimator_test.py
+++ b/tensorflow/contrib/learn/python/learn/estimators/estimator_test.py
@@ -26,7 +26,6 @@ import tempfile
 
 import numpy as np
 import six
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from google.protobuf import text_format
 
@@ -1284,7 +1283,7 @@ class InferRealValuedColumnsTest(test.TestCase):
     with self.assertRaisesRegexp(
         ValueError, 'on integer or non floating types are not supported'):
       estimator.infer_real_valued_columns_from_input(
-          np.array([[False for _ in xrange(8)] for _ in xrange(7)]))
+          np.array([[False for _ in range(8)] for _ in range(7)]))
 
   def testBoolInputFn(self):
     with self.assertRaisesRegexp(
@@ -1299,7 +1298,7 @@ class InferRealValuedColumnsTest(test.TestCase):
         ValueError, 'on integer or non floating types are not supported'):
       # pylint: disable=g-long-lambda
       estimator.infer_real_valued_columns_from_input(
-          np.array([['%d.0' % i for i in xrange(8)] for _ in xrange(7)]))
+          np.array([['%d.0' % i for i in range(8)] for _ in range(7)]))
 
   def testStringInputFn(self):
     with self.assertRaisesRegexp(
@@ -1308,8 +1307,8 @@ class InferRealValuedColumnsTest(test.TestCase):
       estimator.infer_real_valued_columns_from_input_fn(
           lambda: (
               constant_op.constant([['%d.0' % i
-                                     for i in xrange(8)]
-                                    for _ in xrange(7)]),
+                                     for i in range(8)]
+                                    for _ in range(7)]),
               None))
 
   def testBostonInputFn(self):
diff --git a/tensorflow/contrib/learn/python/learn/learn_io/data_feeder.py b/tensorflow/contrib/learn/python/learn/learn_io/data_feeder.py
index b7dba0f2775..901ce6062e0 100644
--- a/tensorflow/contrib/learn/python/learn/learn_io/data_feeder.py
+++ b/tensorflow/contrib/learn/python/learn/learn_io/data_feeder.py
@@ -30,7 +30,6 @@ import math
 
 import numpy as np
 import six
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import tensor_util
@@ -223,7 +222,7 @@ def setup_predict_data_feeder(x, batch_size=None):
     if batch_size <= 0:
       raise ValueError('Invalid batch_size %d.' % batch_size)
     n_batches = int(math.ceil(float(len(x)) / batch_size))
-    return [x[i * batch_size:(i + 1) * batch_size] for i in xrange(n_batches)]
+    return [x[i * batch_size:(i + 1) * batch_size] for i in range(n_batches)]
   return [x]
 
 
@@ -502,7 +501,7 @@ class DataFeeder(object):
     def assign_label(data, shape, dtype, n_classes, indices):
       shape[0] = indices.shape[0]
       out = np.zeros(shape, dtype=dtype)
-      for i in xrange(out.shape[0]):
+      for i in range(out.shape[0]):
         sample = indices[i]
         # self.n_classes is None means we're passing in raw target indices
         if n_classes is None:
@@ -752,7 +751,7 @@ class StreamingDataFeeder(DataFeeder):
       inp = init_array(self.input_shape, self._input_dtype)
       out = init_array(self.output_shape, self._output_dtype)
 
-      for i in xrange(self._batch_size):
+      for i in range(self._batch_size):
         # Add handling when queue ends.
         try:
           next_inp = six.next(self._x)
diff --git a/tensorflow/contrib/learn/python/learn/learn_io/data_feeder_test.py b/tensorflow/contrib/learn/python/learn/learn_io/data_feeder_test.py
index 284a4f45f63..9098550a630 100644
--- a/tensorflow/contrib/learn/python/learn/learn_io/data_feeder_test.py
+++ b/tensorflow/contrib/learn/python/learn/learn_io/data_feeder_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import os.path
 import numpy as np
 import six
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 # pylint: disable=wildcard-import
 from tensorflow.contrib.learn.python.learn.learn_io import *
@@ -122,12 +121,12 @@ class DataFeederTest(test.TestCase):
     self._assert_dtype(np.float64, dtypes.float64, self._wrap_dict(data))
 
   def test_input_bool(self):
-    data = np.array([[False for _ in xrange(2)] for _ in xrange(2)])
+    data = np.array([[False for _ in range(2)] for _ in range(2)])
     self._assert_dtype(np.bool, dtypes.bool, data)
     self._assert_dtype(np.bool, dtypes.bool, self._wrap_dict(data))
 
   def test_input_string(self):
-    input_data = np.array([['str%d' % i for i in xrange(2)] for _ in xrange(2)])
+    input_data = np.array([['str%d' % i for i in range(2)] for _ in range(2)])
     self._assert_dtype(input_data.dtype, dtypes.string, input_data)
     self._assert_dtype(input_data.dtype, dtypes.string,
                        self._wrap_dict(input_data))
diff --git a/tensorflow/contrib/learn/python/learn/learn_io/graph_io_test.py b/tensorflow/contrib/learn/python/learn/learn_io/graph_io_test.py
index 8e68a17e478..be026dcfeec 100644
--- a/tensorflow/contrib/learn/python/learn/learn_io/graph_io_test.py
+++ b/tensorflow/contrib/learn/python/learn/learn_io/graph_io_test.py
@@ -23,7 +23,6 @@ import os
 import random
 import tempfile
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.learn.python.learn.learn_io import graph_io
 from tensorflow.python.client import session as session_lib
@@ -365,7 +364,7 @@ class GraphIOTest(test.TestCase):
     num_records = 49999
     lines = [
         "".join([sequence_prefix, str(l)]).encode("ascii")
-        for l in xrange(num_records)
+        for l in range(num_records)
     ]
     filename = self._create_file_from_list_of_features(lines)
     batch_size = 10000
@@ -421,7 +420,7 @@ class GraphIOTest(test.TestCase):
     """
     gfile.Glob = self._orig_glob
     num_records = 1000
-    lines = ["".join(str(l)).encode("ascii") for l in xrange(num_records)]
+    lines = ["".join(str(l)).encode("ascii") for l in range(num_records)]
     filename = self._create_file_from_list_of_features(lines)
     batch_size = 10
     queue_capacity = 1000
diff --git a/tensorflow/contrib/learn/python/learn/monitors_test.py b/tensorflow/contrib/learn/python/learn/monitors_test.py
index d4a7169bb63..b0abc8f3a1d 100644
--- a/tensorflow/contrib/learn/python/learn/monitors_test.py
+++ b/tensorflow/contrib/learn/python/learn/monitors_test.py
@@ -23,7 +23,6 @@ import shutil
 import tempfile
 import time
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib import testing
 from tensorflow.contrib.framework.python.framework import checkpoint_utils
@@ -109,7 +108,7 @@ class MonitorsTest(test.TestCase):
     else:
       max_steps = None
     monitor.begin(max_steps=max_steps)
-    for epoch in xrange(num_epochs):
+    for epoch in range(num_epochs):
       monitor.epoch_begin(epoch)
       should_stop = False
       step = epoch * num_steps_per_epoch
@@ -538,7 +537,7 @@ class MonitorsTest(test.TestCase):
               counter_var.name: step + 1.0,
               assign_add.name: step + 1.0,
           }
-          for step in xrange(30)
+          for step in range(30)
       }, monitor0.data)
 
       self._run_monitor(monitor1, num_epochs=3, num_steps_per_epoch=10)
@@ -548,10 +547,10 @@ class MonitorsTest(test.TestCase):
               counter_var.name: step + 31.0,
               assign_add.name: step + 31.0,
           }
-          for step in xrange(30)
+          for step in range(30)
       }, monitor1.data)
 
-      for step in xrange(30):
+      for step in range(30):
         matched, non_matched = monitor1.compare(monitor0, step=step)
         self.assertEqual([const_var.name], matched)
         self.assertEqual({
diff --git a/tensorflow/contrib/learn/python/learn/utils/gc.py b/tensorflow/contrib/learn/python/learn/utils/gc.py
index f69a4dd6ad5..74109a757e9 100644
--- a/tensorflow/contrib/learn/python/learn/utils/gc.py
+++ b/tensorflow/contrib/learn/python/learn/utils/gc.py
@@ -37,7 +37,7 @@ Note that functions should always return a sorted list.
 For example,
   base_dir = "/tmp"
   # Create the directories.
-  for e in xrange(10):
+  for e in range(10):
     os.mkdir("%s/%d" % (base_dir, e), 0o755)
 
   # Create a simple parser that pulls the export_version from the directory.
diff --git a/tensorflow/contrib/learn/python/learn/utils/gc_test.py b/tensorflow/contrib/learn/python/learn/utils/gc_test.py
index af93e517f51..2de8d73d599 100644
--- a/tensorflow/contrib/learn/python/learn/utils/gc_test.py
+++ b/tensorflow/contrib/learn/python/learn/utils/gc_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import os
 import re
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.learn.python.learn.utils import gc
 from tensorflow.python.framework import test_util
@@ -103,7 +102,7 @@ class GcTest(test_util.TensorFlowTestCase):
 
   def testUnion(self):
     paths = []
-    for i in xrange(10):
+    for i in range(10):
       paths.append(gc.Path("/foo", i))
     f = gc.union(gc.largest_export_versions(3), gc.mod_export_version(3))
     self.assertEqual(
@@ -131,7 +130,7 @@ class GcTest(test_util.TensorFlowTestCase):
   def testPathsWithParse(self):
     base_dir = os.path.join(test.get_temp_dir(), "paths_parse")
     self.assertFalse(gfile.Exists(base_dir))
-    for p in xrange(3):
+    for p in range(3):
       gfile.MakeDirs(os.path.join(base_dir, "%d" % p))
     # add a base_directory to ignore
     gfile.MakeDirs(os.path.join(base_dir, "ignore"))
diff --git a/tensorflow/contrib/legacy_seq2seq/python/ops/seq2seq.py b/tensorflow/contrib/legacy_seq2seq/python/ops/seq2seq.py
index 9555bb2fe8b..bbc3ab74fe3 100644
--- a/tensorflow/contrib/legacy_seq2seq/python/ops/seq2seq.py
+++ b/tensorflow/contrib/legacy_seq2seq/python/ops/seq2seq.py
@@ -59,8 +59,6 @@ from __future__ import print_function
 import copy
 
 # We disable pylint because we need python3 compatibility.
-from six.moves import xrange  # pylint: disable=redefined-builtin
-from six.moves import zip  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.rnn.python.ops import core_rnn_cell
 from tensorflow.python.framework import dtypes
@@ -622,7 +620,7 @@ def attention_decoder(decoder_inputs,
     hidden_features = []
     v = []
     attention_vec_size = attn_size  # Size of query vectors for attention.
-    for a in xrange(num_heads):
+    for a in range(num_heads):
       k = variable_scope.get_variable(
           "AttnW_%d" % a, [1, 1, attn_size, attention_vec_size], dtype=dtype)
       hidden_features.append(nn_ops.conv2d(hidden, k, [1, 1, 1, 1], "SAME"))
@@ -642,7 +640,7 @@ def attention_decoder(decoder_inputs,
           if ndims:
             assert ndims == 2
         query = array_ops.concat(query_list, 1)
-      for a in xrange(num_heads):
+      for a in range(num_heads):
         with variable_scope.variable_scope("Attention_%d" % a):
           y = Linear(query, attention_vec_size, True)(query)
           y = array_ops.reshape(y, [-1, 1, 1, attention_vec_size])
@@ -662,7 +660,7 @@ def attention_decoder(decoder_inputs,
     prev = None
     batch_attn_size = array_ops.stack([batch_size, attn_size])
     attns = [
-        array_ops.zeros(batch_attn_size, dtype=dtype) for _ in xrange(num_heads)
+        array_ops.zeros(batch_attn_size, dtype=dtype) for _ in range(num_heads)
     ]
     for a in attns:  # Ensure the second shape of attention vectors is set.
       a.set_shape([None, attn_size])
diff --git a/tensorflow/contrib/linear_optimizer/python/ops/sdca_ops.py b/tensorflow/contrib/linear_optimizer/python/ops/sdca_ops.py
index 950840c6b77..6b6d932752a 100644
--- a/tensorflow/contrib/linear_optimizer/python/ops/sdca_ops.py
+++ b/tensorflow/contrib/linear_optimizer/python/ops/sdca_ops.py
@@ -26,8 +26,6 @@ from __future__ import print_function
 
 import collections
 
-from six.moves import range
-
 from tensorflow.contrib.linear_optimizer.python.ops.sharded_mutable_dense_hashtable import ShardedMutableDenseHashTable
 from tensorflow.python.compat import compat
 from tensorflow.python.framework import constant_op
diff --git a/tensorflow/contrib/linear_optimizer/python/ops/sharded_mutable_dense_hashtable.py b/tensorflow/contrib/linear_optimizer/python/ops/sharded_mutable_dense_hashtable.py
index 8fda828e994..796dec348be 100644
--- a/tensorflow/contrib/linear_optimizer/python/ops/sharded_mutable_dense_hashtable.py
+++ b/tensorflow/contrib/linear_optimizer/python/ops/sharded_mutable_dense_hashtable.py
@@ -23,8 +23,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import range
-
 from tensorflow.contrib import lookup
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
diff --git a/tensorflow/contrib/metrics/python/ops/metric_ops_large_test.py b/tensorflow/contrib/metrics/python/ops/metric_ops_large_test.py
index 5777e64c294..0a16710bf4e 100644
--- a/tensorflow/contrib/metrics/python/ops/metric_ops_large_test.py
+++ b/tensorflow/contrib/metrics/python/ops/metric_ops_large_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 from tensorflow.contrib.metrics.python.ops import metric_ops
 from tensorflow.python.framework import dtypes as dtypes_lib
 from tensorflow.python.framework import ops
@@ -49,7 +48,7 @@ class StreamingPrecisionRecallAtEqualThresholdsLargeTest(test.TestCase):
 
     with self.cached_session() as sess:
       sess.run(variables.local_variables_initializer())
-      for _ in xrange(num_updates):
+      for _ in range(num_updates):
         sess.run(update_op)
 
       prdata = sess.run(result)
diff --git a/tensorflow/contrib/metrics/python/ops/metric_ops_test.py b/tensorflow/contrib/metrics/python/ops/metric_ops_test.py
index 906bebe3b82..b7ca4539de9 100644
--- a/tensorflow/contrib/metrics/python/ops/metric_ops_test.py
+++ b/tensorflow/contrib/metrics/python/ops/metric_ops_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import math
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 from tensorflow.contrib import metrics as metrics_lib
 from tensorflow.contrib.metrics.python.ops import metric_ops
 from tensorflow.python.framework import constant_op
@@ -562,7 +561,7 @@ class StreamingAccuracyTest(test.TestCase):
       accuracy, update_op = metrics.streaming_accuracy(predictions, labels)
 
       sess.run(variables.local_variables_initializer())
-      for _ in xrange(3):
+      for _ in range(3):
         sess.run(update_op)
       self.assertEqual(0.5, sess.run(update_op))
       self.assertEqual(0.5, accuracy.eval())
@@ -649,7 +648,7 @@ class StreamingAccuracyTest(test.TestCase):
                                                        weights)
 
       sess.run(variables.local_variables_initializer())
-      for _ in xrange(3):
+      for _ in range(3):
         sess.run(update_op)
       self.assertEqual(1.0, sess.run(update_op))
       self.assertEqual(1.0, accuracy.eval())
@@ -1971,11 +1970,11 @@ class StreamingDynamicAUCTest(test.TestCase):
     with self.cached_session() as sess:
       sess.run(variables.local_variables_initializer())
       # Run several updates.
-      for _ in xrange(10):
+      for _ in range(10):
         sess.run(update_op)
       # Then verify idempotency.
       initial_auc = auc.eval()
-      for _ in xrange(10):
+      for _ in range(10):
         self.assertAlmostEqual(initial_auc, auc.eval(), 5)
 
   def testAllLabelsOnes(self):
@@ -2082,7 +2081,7 @@ class StreamingDynamicAUCTest(test.TestCase):
     auc, update_op = metrics.streaming_dynamic_auc(tf_labels, tf_predictions)
     with self.cached_session() as sess:
       sess.run(variables.local_variables_initializer())
-      for _ in xrange(num_batches):
+      for _ in range(num_batches):
         new_labels = np.random.randint(0, 2, size=batch_size)
         noise = np.random.normal(0.0, scale=0.2, size=batch_size)
         new_predictions = 0.4 + 0.2 * new_labels + noise
@@ -2152,7 +2151,7 @@ class StreamingDynamicAUCTest(test.TestCase):
                                                    weights=tf_weights)
     with self.cached_session() as sess:
       sess.run(variables.local_variables_initializer())
-      for _ in xrange(num_batches):
+      for _ in range(num_batches):
         new_labels = np.random.randint(0, 2, size=batch_size)
         noise = np.random.uniform(-0.2, 0.2, size=batch_size)
         new_predictions = 0.4 + 0.2 * new_labels + noise
@@ -2325,7 +2324,7 @@ class AucWithConfidenceIntervalsTest(test.TestCase):
                                                            tf_predictions)
     with self.cached_session() as sess:
       sess.run(variables.local_variables_initializer())
-      for _ in xrange(num_batches):
+      for _ in range(num_batches):
         new_labels = np.random.randint(0, 2, size=batch_size)
         noise = np.random.normal(0.0, scale=0.2, size=batch_size)
         new_predictions = 0.4 + 0.2 * new_labels + noise
@@ -4161,7 +4160,7 @@ class StreamingSparsePrecisionTest(test.TestCase):
     precision_ex1 = (0.0 / 1, 1.0 / 2, 1.0 / 3, 2.0 / 4)
     avg_precision_ex1 = (0.0 / 1, precision_ex1[1] / 2, precision_ex1[1] / 3,
                          (precision_ex1[1] + precision_ex1[3]) / 4)
-    for i in xrange(4):
+    for i in range(4):
       k = i + 1
       self._test_streaming_sparse_precision_at_k(
           predictions, labels, k, expected=precision_ex1[i])
@@ -4181,7 +4180,7 @@ class StreamingSparsePrecisionTest(test.TestCase):
     precision_ex2 = (0.0 / 1, 0.0 / 2, 1.0 / 3, 2.0 / 4)
     avg_precision_ex2 = (0.0 / 1, 0.0 / 2, precision_ex2[2] / 3,
                          (precision_ex2[2] + precision_ex2[3]) / 4)
-    for i in xrange(4):
+    for i in range(4):
       k = i + 1
       self._test_streaming_sparse_precision_at_k(
           predictions, labels, k, expected=precision_ex2[i])
@@ -4203,7 +4202,7 @@ class StreamingSparsePrecisionTest(test.TestCase):
         (ex1 + ex2) / 2
         for ex1, ex2 in zip(avg_precision_ex1, avg_precision_ex2)
     ]
-    for i in xrange(4):
+    for i in range(4):
       k = i + 1
       self._test_streaming_sparse_precision_at_k(
           predictions, labels, k, expected=streaming_precision[i])
@@ -4222,7 +4221,7 @@ class StreamingSparsePrecisionTest(test.TestCase):
         (weights[0] * ex1 + weights[1] * ex2) / (weights[0] + weights[1])
         for ex1, ex2 in zip(avg_precision_ex1, avg_precision_ex2)
     ]
-    for i in xrange(4):
+    for i in range(4):
       k = i + 1
       self._test_streaming_sparse_average_precision_at_k(
           predictions,
@@ -4246,7 +4245,7 @@ class StreamingSparsePrecisionTest(test.TestCase):
     precision_ex1 = (0.0 / 1, 1.0 / 2, 1.0 / 3, 2.0 / 4)
     avg_precision_ex1 = (0.0 / 1, precision_ex1[1] / 2, precision_ex1[1] / 3,
                          (precision_ex1[1] + precision_ex1[3]) / 4)
-    for i in xrange(4):
+    for i in range(4):
       k = i + 1
       self._test_streaming_sparse_precision_at_k(
           predictions, labels, k, expected=precision_ex1[i])
@@ -4526,7 +4525,7 @@ class StreamingSparsePrecisionTest(test.TestCase):
         [[[0, 0, 1, 0, 0, 0, 0, 1, 1, 0], [0, 1, 1, 0, 0, 1, 0, 0, 0, 0]],
          [[0, 1, 1, 0, 0, 1, 0, 1, 0, 0], [0, 0, 1, 0, 0, 0, 0, 0, 1, 0]]])
 
-    for class_id in xrange(10):
+    for class_id in range(10):
       self._test_streaming_sparse_precision_at_k(
           predictions,
           labels,
@@ -5231,7 +5230,7 @@ class StreamingSparseRecallTest(test.TestCase):
         [[[0, 0, 1, 0, 0, 0, 0, 1, 1, 0], [0, 1, 1, 0, 0, 1, 0, 0, 0, 0]],
          [[0, 1, 1, 0, 0, 1, 0, 1, 0, 0], [0, 0, 1, 0, 0, 0, 0, 0, 1, 0]]])
 
-    for class_id in xrange(10):
+    for class_id in range(10):
       self._test_streaming_sparse_recall_at_k(
           predictions,
           labels,
diff --git a/tensorflow/contrib/model_pruning/examples/cifar10/cifar10_input.py b/tensorflow/contrib/model_pruning/examples/cifar10/cifar10_input.py
index 6a3b535eb44..4cb59dbfc10 100644
--- a/tensorflow/contrib/model_pruning/examples/cifar10/cifar10_input.py
+++ b/tensorflow/contrib/model_pruning/examples/cifar10/cifar10_input.py
@@ -20,7 +20,6 @@ from __future__ import print_function
 
 import os
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 
 # Process images of this size. Note that this differs from the original CIFAR
@@ -149,7 +148,7 @@ def distorted_inputs(data_dir, batch_size):
     labels: Labels. 1D tensor of [batch_size] size.
   """
   filenames = [
-      os.path.join(data_dir, 'data_batch_%d.bin' % i) for i in xrange(1, 6)
+      os.path.join(data_dir, 'data_batch_%d.bin' % i) for i in range(1, 6)
   ]
   for f in filenames:
     if not tf.gfile.Exists(f):
@@ -217,7 +216,7 @@ def inputs(eval_data, data_dir, batch_size):
   """
   if not eval_data:
     filenames = [
-        os.path.join(data_dir, 'data_batch_%d.bin' % i) for i in xrange(1, 6)
+        os.path.join(data_dir, 'data_batch_%d.bin' % i) for i in range(1, 6)
     ]
     num_examples_per_epoch = NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN
   else:
diff --git a/tensorflow/contrib/optimizer_v2/momentum_test.py b/tensorflow/contrib/optimizer_v2/momentum_test.py
index e69f12839e9..c1a9ee4ba16 100644
--- a/tensorflow/contrib/optimizer_v2/momentum_test.py
+++ b/tensorflow/contrib/optimizer_v2/momentum_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.optimizer_v2 import momentum as momentum_lib
 from tensorflow.python.eager import context
@@ -443,7 +442,7 @@ class MomentumOptimizerTest(test.TestCase):
       mom_opt = momentum_lib.MomentumOptimizer(learning_rate=0.1, momentum=0.1)
       mom_update = mom_opt.apply_gradients(zip([grads0], [var0]))
       variables.global_variables_initializer().run()
-      for i in xrange(num_samples):
+      for i in range(num_samples):
         mom_update.run(feed_dict={grads0: db_grad[i]})
         self.assertAllClose(np.array(db_out[i]), var0.eval())
 
diff --git a/tensorflow/contrib/recurrent/python/kernel_tests/recurrent_test.py b/tensorflow/contrib/recurrent/python/kernel_tests/recurrent_test.py
index aea80a5256e..a945c414722 100644
--- a/tensorflow/contrib/recurrent/python/kernel_tests/recurrent_test.py
+++ b/tensorflow/contrib/recurrent/python/kernel_tests/recurrent_test.py
@@ -20,7 +20,6 @@ from __future__ import print_function
 
 import collections
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.recurrent.python.ops import recurrent
 from tensorflow.python.framework import dtypes
@@ -155,7 +154,7 @@ class RecurrentTest(test_util.TensorFlowTestCase):
       # Statically unrolled.
       s = state0
       out = []
-      for i in xrange(seqlen):
+      for i in range(seqlen):
         inp = _ElmanInputs(x=inputs.x[i, :])
         s, _ = RecurrentTest.Elman(theta, s, inp)
         out += [s.h]
diff --git a/tensorflow/contrib/resampler/python/ops/resampler_ops_test.py b/tensorflow/contrib/resampler/python/ops/resampler_ops_test.py
index e30e7255fac..acaa978e258 100644
--- a/tensorflow/contrib/resampler/python/ops/resampler_ops_test.py
+++ b/tensorflow/contrib/resampler/python/ops/resampler_ops_test.py
@@ -20,7 +20,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib import resampler
 from tensorflow.python.framework import constant_op
@@ -147,8 +146,8 @@ class ResamplerTest(test.TestCase):
 
     # Generate reference output via bilinear interpolation in numpy
     reference_output = np.zeros_like(out)
-    for batch in xrange(batch_size):
-      for c in xrange(data_channels):
+    for batch in range(batch_size):
+      for c in range(data_channels):
         reference_output[batch, :, :, c] = _bilinearly_interpolate(
             data[batch, :, :, c],
             warp[batch, :, :, 0],
diff --git a/tensorflow/contrib/session_bundle/gc.py b/tensorflow/contrib/session_bundle/gc.py
index 4a9adcf4838..5d6c8e0123d 100644
--- a/tensorflow/contrib/session_bundle/gc.py
+++ b/tensorflow/contrib/session_bundle/gc.py
@@ -33,7 +33,7 @@ Note that functions should always return a sorted list.
 For example,
   base_dir = "/tmp"
   # create the directories
-  for e in xrange(10):
+  for e in range(10):
     os.mkdir("%s/%d" % (base_dir, e), 0o755)
 
   # create a simple parser that pulls the export_version from the directory
diff --git a/tensorflow/contrib/session_bundle/gc_test.py b/tensorflow/contrib/session_bundle/gc_test.py
index 02725bb1cbb..f225f48cf21 100644
--- a/tensorflow/contrib/session_bundle/gc_test.py
+++ b/tensorflow/contrib/session_bundle/gc_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import os
 import re
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.contrib.session_bundle import gc
 from tensorflow.python.framework import test_util
@@ -75,7 +74,7 @@ class GcTest(test_util.TensorFlowTestCase):
 
   def testUnion(self):
     paths = []
-    for i in xrange(10):
+    for i in range(10):
       paths.append(gc.Path("/foo", i))
     f = gc.union(gc.largest_export_versions(3), gc.mod_export_version(3))
     self.assertEquals(
@@ -97,7 +96,7 @@ class GcTest(test_util.TensorFlowTestCase):
   def testPathsWithParse(self):
     base_dir = os.path.join(test.get_temp_dir(), "paths_parse")
     self.assertFalse(gfile.Exists(base_dir))
-    for p in xrange(3):
+    for p in range(3):
       gfile.MakeDirs(os.path.join(base_dir, "%d" % p))
     # add a base_directory to ignore
     gfile.MakeDirs(os.path.join(base_dir, "ignore"))
diff --git a/tensorflow/core/kernels/BUILD b/tensorflow/core/kernels/BUILD
index 22d2facf5ca..de3d5d34760 100644
--- a/tensorflow/core/kernels/BUILD
+++ b/tensorflow/core/kernels/BUILD
@@ -1339,8 +1339,6 @@ tf_kernel_library(
     ],
     deps = if_cuda_or_rocm([
         ":cuda_solvers",
-    ]) + if_cuda([
-        "@cub_archive//:cub",
     ]) + if_rocm([
         "@rocprim_archive//:rocprim",
     ]) + ARRAY_DEPS,
@@ -2502,7 +2500,7 @@ tf_kernel_library(
     deps = DYNAMIC_DEPS + [
         ":fill_functor",
         ":gather_functor",
-    ] + if_cuda(["@cub_archive//:cub"]) + if_rocm([
+    ] + if_rocm([
         "@rocprim_archive//:rocprim",
     ]),
 )
@@ -2905,7 +2903,7 @@ tf_kernel_library(
 tf_kernel_library(
     name = "non_max_suppression_op",
     prefix = "non_max_suppression_op",
-    deps = IMAGE_DEPS + if_cuda(["@cub_archive//:cub"]),
+    deps = IMAGE_DEPS,
 )
 
 tf_kernel_library(
@@ -3823,9 +3821,7 @@ tf_kernel_library(
     name = "reduction_ops",
     gpu_srcs = ["reduction_gpu_kernels.cu.h"],
     prefix = "reduction_ops",
-    deps = MATH_DEPS + [":transpose_functor"] + if_cuda([
-        "@cub_archive//:cub",
-    ]) + if_rocm([
+    deps = MATH_DEPS + [":transpose_functor"] + if_rocm([
         "@rocprim_archive//:rocprim",
     ]),
 )
@@ -3850,9 +3846,7 @@ tf_kernel_library(
         "scan_ops_gpu_half.cu.cc",
         "scan_ops_gpu_int.cu.cc",
     ],
-    deps = MATH_DEPS + if_cuda([
-        "@cub_archive//:cub",
-    ]) + if_rocm([
+    deps = MATH_DEPS + if_rocm([
         "@rocprim_archive//:rocprim",
     ]),
 )
@@ -4284,7 +4278,6 @@ tf_kernel_library(
         "//tensorflow/core:framework",
         "//tensorflow/core:lib",
     ] + if_cuda([
-        "@cub_archive//:cub",
         "@local_config_cuda//cuda:cudnn_header",
     ]) + if_rocm([
         "@rocprim_archive//:rocprim",
@@ -4373,7 +4366,6 @@ tf_kernel_library(
     ] + if_cuda_or_rocm([
         ":reduction_ops",
     ]) + if_cuda([
-        "@cub_archive//:cub",
         "//tensorflow/core:stream_executor",
         "//tensorflow/stream_executor/cuda:cuda_stream",
     ]) + if_rocm([
@@ -4415,8 +4407,6 @@ tf_kernel_library(
     prefix = "softmax_op",
     deps = NN_DEPS + if_cuda_or_rocm([
         ":reduction_ops",
-    ]) + if_cuda([
-        "@cub_archive//:cub",
     ]) + if_rocm([
         "@rocprim_archive//:rocprim",
     ]),
@@ -4451,9 +4441,7 @@ tf_kernel_library(
         "topk_op_gpu_int8.cu.cc",
         "topk_op_gpu_uint8.cu.cc",
     ],
-    deps = NN_DEPS + if_cuda([
-        "@cub_archive//:cub",
-    ]) + if_rocm([
+    deps = NN_DEPS + if_rocm([
         "@rocprim_archive//:rocprim",
     ]),
 )
@@ -4478,7 +4466,7 @@ tf_kernel_library(
         "//tensorflow/core:lib",
         "//tensorflow/core:lib_internal",
         "//third_party/eigen3",
-    ] + if_cuda(["@cub_archive//:cub"]) + if_rocm([
+    ] + if_rocm([
         "@rocprim_archive//:rocprim",
     ]),
 )
@@ -4491,7 +4479,7 @@ tf_kernel_library(
         "//tensorflow/core:lib",
         "//tensorflow/core:lib_internal",
         "//third_party/eigen3",
-    ] + if_cuda(["@cub_archive//:cub"]) + if_rocm([
+    ] + if_rocm([
         "@rocprim_archive//:rocprim",
     ]),
 )
@@ -4506,7 +4494,7 @@ tf_kernel_library(
         "//tensorflow/core:lib",
         "//tensorflow/core:lib_internal",
         "//tensorflow/core:nn_grad",
-    ] + if_cuda(["@cub_archive//:cub"]) + if_rocm([
+    ] + if_rocm([
         "@rocprim_archive//:rocprim",
     ]),
 )
@@ -5084,8 +5072,6 @@ tf_kernel_library(
         "//third_party/eigen3",
     ] + if_cuda_or_rocm([
         ":reduction_ops",
-    ]) + if_cuda([
-        "@cub_archive//:cub",
     ]) + if_rocm([
         "@rocprim_archive//:rocprim",
     ]),
@@ -5655,8 +5641,6 @@ tf_kernel_library(
         "//tensorflow/core:lib_internal",
     ] + if_cuda_or_rocm([
         ":reduction_ops",
-    ]) + if_cuda([
-        "@cub_archive//:cub",
     ]) + if_rocm([
         "@rocprim_archive//:rocprim",
     ]),
diff --git a/tensorflow/core/kernels/bincount_op_gpu.cu.cc b/tensorflow/core/kernels/bincount_op_gpu.cu.cc
index a2ad9cacf12..303056387b3 100644
--- a/tensorflow/core/kernels/bincount_op_gpu.cu.cc
+++ b/tensorflow/core/kernels/bincount_op_gpu.cu.cc
@@ -18,7 +18,7 @@ limitations under the License.
 #define EIGEN_USE_GPU
 
 #if GOOGLE_CUDA
-#include "third_party/cub/device/device_histogram.cuh"
+#include "third_party/gpus/cuda/include/cub/cub.cuh"
 #elif TENSORFLOW_USE_ROCM
 #include "external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp"
 #endif
diff --git a/tensorflow/core/kernels/cuda_sparse.cc b/tensorflow/core/kernels/cuda_sparse.cc
index dff78bc9c7e..04d9e7294f0 100644
--- a/tensorflow/core/kernels/cuda_sparse.cc
+++ b/tensorflow/core/kernels/cuda_sparse.cc
@@ -209,7 +209,7 @@ static inline Status GtsvImpl(SparseFn op, cusparseHandle_t cusparse_handle,
                     dl, d, du, B, ldb);                                      \
   }
 
-TF_CALL_LAPACK_TYPES(GTSV_INSTANCE);
+//TF_CALL_LAPACK_TYPES(GTSV_INSTANCE);
 
 #define GTSV_NO_PIVOT_INSTANCE(Scalar, sparse_prefix)                          \
   template <>                                                                  \
@@ -221,7 +221,7 @@ TF_CALL_LAPACK_TYPES(GTSV_INSTANCE);
                     m, n, dl, d, du, B, ldb);                                  \
   }
 
-TF_CALL_LAPACK_TYPES(GTSV_NO_PIVOT_INSTANCE);
+//TF_CALL_LAPACK_TYPES(GTSV_NO_PIVOT_INSTANCE);
 
 template <typename Scalar, typename SparseFn>
 static inline Status GtsvStridedBatchImpl(SparseFn op,
@@ -247,7 +247,7 @@ static inline Status GtsvStridedBatchImpl(SparseFn op,
                                 batchCount, batchStride);                    \
   }
 
-TF_CALL_LAPACK_TYPES(GTSV_STRIDED_BATCH_INSTANCE);
+//TF_CALL_LAPACK_TYPES(GTSV_STRIDED_BATCH_INSTANCE);
 
 template <typename Scalar, typename SparseFn>
 static inline Status Gtsv2Impl(SparseFn op, cusparseHandle_t cusparse_handle,
diff --git a/tensorflow/core/kernels/dynamic_partition_op_gpu.cu.cc b/tensorflow/core/kernels/dynamic_partition_op_gpu.cu.cc
index 7fe519b063c..49f31718396 100644
--- a/tensorflow/core/kernels/dynamic_partition_op_gpu.cu.cc
+++ b/tensorflow/core/kernels/dynamic_partition_op_gpu.cu.cc
@@ -36,10 +36,7 @@ limitations under the License.
 #define EIGEN_USE_GPU
 
 #if GOOGLE_CUDA
-#include "third_party/cub/device/device_radix_sort.cuh"
-#include "third_party/cub/device/device_reduce.cuh"
-#include "third_party/cub/iterator/constant_input_iterator.cuh"
-#include "third_party/cub/thread/thread_operators.cuh"
+#include "third_party/gpus/cuda/include/cub/cub.cuh"
 #elif TENSORFLOW_USE_ROCM
 #include "external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp"
 #endif
diff --git a/tensorflow/core/kernels/histogram_op_gpu.cu.cc b/tensorflow/core/kernels/histogram_op_gpu.cu.cc
index 5f35adccba3..f82d8d0efbd 100644
--- a/tensorflow/core/kernels/histogram_op_gpu.cu.cc
+++ b/tensorflow/core/kernels/histogram_op_gpu.cu.cc
@@ -19,7 +19,7 @@ limitations under the License.
 
 #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
 #if GOOGLE_CUDA
-#include "third_party/cub/device/device_histogram.cuh"
+#include "third_party/gpus/cuda/include/cub/cub.cuh"
 #elif TENSORFLOW_USE_ROCM
 #include "external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp"
 #endif
diff --git a/tensorflow/core/kernels/non_max_suppression_op.cu.cc b/tensorflow/core/kernels/non_max_suppression_op.cu.cc
index 1749a66579b..da6f0174119 100644
--- a/tensorflow/core/kernels/non_max_suppression_op.cu.cc
+++ b/tensorflow/core/kernels/non_max_suppression_op.cu.cc
@@ -19,9 +19,7 @@ limitations under the License.
 
 #include "absl/strings/str_cat.h"
 #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
-#include "third_party/cub/device/device_radix_sort.cuh"
-#include "third_party/cub/device/device_segmented_radix_sort.cuh"
-#include "third_party/cub/device/device_select.cuh"
+#include "third_party/gpus/cuda/include/cub/cub.cuh"
 #include "tensorflow/core/framework/numeric_types.h"
 #include "tensorflow/core/framework/op_kernel.h"
 #include "tensorflow/core/framework/tensor_types.h"
diff --git a/tensorflow/core/kernels/reduction_gpu_kernels.cu.h b/tensorflow/core/kernels/reduction_gpu_kernels.cu.h
index 7024d62a53e..62a9faa45d7 100644
--- a/tensorflow/core/kernels/reduction_gpu_kernels.cu.h
+++ b/tensorflow/core/kernels/reduction_gpu_kernels.cu.h
@@ -24,12 +24,7 @@ limitations under the License.
 
 #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
 #if GOOGLE_CUDA
-#include "third_party/cub/device/device_reduce.cuh"
-#include "third_party/cub/device/device_segmented_reduce.cuh"
-#include "third_party/cub/iterator/counting_input_iterator.cuh"
-#include "third_party/cub/iterator/transform_input_iterator.cuh"
-#include "third_party/cub/warp/warp_reduce.cuh"
-#include "third_party/gpus/cuda/include/cuComplex.h"
+#include "third_party/gpus/cuda/include/cub/cub.cuh"
 #elif TENSORFLOW_USE_ROCM
 #include "external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp"
 #endif
diff --git a/tensorflow/core/kernels/scan_ops_gpu.h b/tensorflow/core/kernels/scan_ops_gpu.h
index eaa9360a5b7..1665d998ca6 100644
--- a/tensorflow/core/kernels/scan_ops_gpu.h
+++ b/tensorflow/core/kernels/scan_ops_gpu.h
@@ -25,11 +25,7 @@ limitations under the License.
 #endif  // CUDA_VERSION >= 9000
 
 #if GOOGLE_CUDA
-#include "third_party/cub/block/block_load.cuh"
-#include "third_party/cub/block/block_scan.cuh"
-#include "third_party/cub/block/block_store.cuh"
-#include "third_party/cub/iterator/counting_input_iterator.cuh"
-#include "third_party/cub/iterator/transform_input_iterator.cuh"
+#include "third_party/gpus/cuda/include/cub/cub.cuh"
 #include "third_party/gpus/cuda/include/cuComplex.h"
 #elif TENSORFLOW_USE_ROCM
 #include "external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp"
diff --git a/tensorflow/core/kernels/topk_op_gpu.h b/tensorflow/core/kernels/topk_op_gpu.h
index 12717ca11fe..a48a86ca89d 100644
--- a/tensorflow/core/kernels/topk_op_gpu.h
+++ b/tensorflow/core/kernels/topk_op_gpu.h
@@ -23,9 +23,7 @@ limitations under the License.
 #include <vector>
 
 #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
-#include "third_party/cub/device/device_segmented_radix_sort.cuh"
-#include "third_party/cub/iterator/counting_input_iterator.cuh"
-#include "third_party/cub/iterator/transform_input_iterator.cuh"
+#include "third_party/gpus/cuda/include/cub/cub.cuh"
 #include "tensorflow/core/framework/op_kernel.h"
 #include "tensorflow/core/framework/register_types.h"
 #include "tensorflow/core/framework/tensor.h"
diff --git a/tensorflow/core/kernels/where_op_gpu.cu.h b/tensorflow/core/kernels/where_op_gpu.cu.h
index 8c228d60ebb..e49de52fcfd 100644
--- a/tensorflow/core/kernels/where_op_gpu.cu.h
+++ b/tensorflow/core/kernels/where_op_gpu.cu.h
@@ -22,10 +22,7 @@ limitations under the License.
 
 #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
 #if GOOGLE_CUDA
-#include "third_party/cub/device/device_reduce.cuh"
-#include "third_party/cub/device/device_select.cuh"
-#include "third_party/cub/iterator/counting_input_iterator.cuh"
-#include "third_party/cub/iterator/transform_input_iterator.cuh"
+#include "third_party/gpus/cuda/include/cub/cub.cuh"
 #elif TENSORFLOW_USE_ROCM
 #include "external/rocprim_archive/hipcub/include/hipcub/hipcub.hpp"
 #endif
@@ -246,6 +243,8 @@ class WhereOutputIterator {
     return *(ptr_ + (valid ? (NDIM * n) : 0));
   }
 
+ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE operator int64*(){ return ptr_; }
+
  private:
   int64* ptr_;
   const Eigen::DenseIndex max_row_;
diff --git a/tensorflow/core/lib/bfloat16/bfloat16.h b/tensorflow/core/lib/bfloat16/bfloat16.h
index a133f7e0f17..c2c5681417d 100644
--- a/tensorflow/core/lib/bfloat16/bfloat16.h
+++ b/tensorflow/core/lib/bfloat16/bfloat16.h
@@ -18,6 +18,7 @@ limitations under the License.
 
 #include <cmath>
 #include <complex>
+#include <cstdint>
 
 #include "tensorflow/core/platform/byte_order.h"
 
diff --git a/tensorflow/core/lib/core/coding.h b/tensorflow/core/lib/core/coding.h
index bfab80dd007..33deb95faf7 100644
--- a/tensorflow/core/lib/core/coding.h
+++ b/tensorflow/core/lib/core/coding.h
@@ -24,6 +24,7 @@ limitations under the License.
 #include "tensorflow/core/lib/core/raw_coding.h"
 #include "tensorflow/core/lib/core/stringpiece.h"
 #include "tensorflow/core/platform/types.h"
+#include <cstdint>
 
 namespace tensorflow {
 namespace core {
diff --git a/tensorflow/core/lib/gif/gif_io.cc b/tensorflow/core/lib/gif/gif_io.cc
index dc5406920a4..ae2f0b373d4 100644
--- a/tensorflow/core/lib/gif/gif_io.cc
+++ b/tensorflow/core/lib/gif/gif_io.cc
@@ -17,6 +17,8 @@ limitations under the License.
 
 #include "tensorflow/core/lib/gif/gif_io.h"
 #include <algorithm>
+#include <cstdint>
+#include <cstring>
 #include "tensorflow/core/lib/gtl/cleanup.h"
 #include "tensorflow/core/lib/strings/strcat.h"
 #include "tensorflow/core/platform/gif.h"
diff --git a/tensorflow/core/platform/posix/port.cc b/tensorflow/core/platform/posix/port.cc
index 47f4abae3bb..f70d5ea5e24 100644
--- a/tensorflow/core/platform/posix/port.cc
+++ b/tensorflow/core/platform/posix/port.cc
@@ -37,6 +37,7 @@ limitations under the License.
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
+#include <cstdint>
 #ifdef TF_USE_SNAPPY
 #include "snappy.h"
 #endif
diff --git a/tensorflow/examples/speech_commands/input_data.py b/tensorflow/examples/speech_commands/input_data.py
index 2bb48e01b79..4ad31f64fc7 100644
--- a/tensorflow/examples/speech_commands/input_data.py
+++ b/tensorflow/examples/speech_commands/input_data.py
@@ -29,7 +29,6 @@ import tarfile
 
 import numpy as np
 from six.moves import urllib
-from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 
 from tensorflow.contrib.framework.python.ops import audio_ops as contrib_audio
@@ -545,7 +544,7 @@ class AudioProcessor(object):
     pick_deterministically = (mode != 'training')
     # Use the processing graph we created earlier to repeatedly to generate the
     # final output sample data we'll use in training.
-    for i in xrange(offset, offset + sample_count):
+    for i in range(offset, offset + sample_count):
       # Pick which audio sample to use.
       if how_many == -1 or pick_deterministically:
         sample_index = i
diff --git a/tensorflow/examples/speech_commands/train.py b/tensorflow/examples/speech_commands/train.py
index 446e351cb81..54b01e9f2a7 100644
--- a/tensorflow/examples/speech_commands/train.py
+++ b/tensorflow/examples/speech_commands/train.py
@@ -75,7 +75,6 @@ import os.path
 import sys
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 
 import input_data
@@ -206,7 +205,7 @@ def main(_):
 
   # Training loop.
   training_steps_max = np.sum(training_steps_list)
-  for training_step in xrange(start_step, training_steps_max + 1):
+  for training_step in range(start_step, training_steps_max + 1):
     # Figure out what the current learning rate is.
     training_steps_sum = 0
     for i in range(len(training_steps_list)):
@@ -243,7 +242,7 @@ def main(_):
       set_size = audio_processor.set_size('validation')
       total_accuracy = 0
       total_conf_matrix = None
-      for i in xrange(0, set_size, FLAGS.batch_size):
+      for i in range(0, set_size, FLAGS.batch_size):
         validation_fingerprints, validation_ground_truth = (
             audio_processor.get_data(FLAGS.batch_size, i, model_settings, 0.0,
                                      0.0, 0, 'validation', sess))
@@ -280,7 +279,7 @@ def main(_):
   tf.compat.v1.logging.info('set_size=%d', set_size)
   total_accuracy = 0
   total_conf_matrix = None
-  for i in xrange(0, set_size, FLAGS.batch_size):
+  for i in range(0, set_size, FLAGS.batch_size):
     test_fingerprints, test_ground_truth = audio_processor.get_data(
         FLAGS.batch_size, i, model_settings, 0.0, 0.0, 0, 'testing', sess)
     test_accuracy, conf_matrix = sess.run(
diff --git a/tensorflow/examples/tutorials/mnist/fully_connected_feed.py b/tensorflow/examples/tutorials/mnist/fully_connected_feed.py
index 8eb57100058..455312644e5 100644
--- a/tensorflow/examples/tutorials/mnist/fully_connected_feed.py
+++ b/tensorflow/examples/tutorials/mnist/fully_connected_feed.py
@@ -24,7 +24,6 @@ import os
 import sys
 import time
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 
 from tensorflow.examples.tutorials.mnist import input_data
@@ -103,7 +102,7 @@ def do_eval(sess,
   true_count = 0  # Counts the number of correct predictions.
   steps_per_epoch = data_set.num_examples // FLAGS.batch_size
   num_examples = steps_per_epoch * FLAGS.batch_size
-  for step in xrange(steps_per_epoch):
+  for step in range(steps_per_epoch):
     feed_dict = fill_feed_dict(data_set,
                                images_placeholder,
                                labels_placeholder)
@@ -160,7 +159,7 @@ def run_training():
     sess.run(init)
 
     # Start the training loop.
-    for step in xrange(FLAGS.max_steps):
+    for step in range(FLAGS.max_steps):
       start_time = time.time()
 
       # Fill a feed dictionary with the actual set of images and labels
diff --git a/tensorflow/examples/tutorials/mnist/input_data.py b/tensorflow/examples/tutorials/mnist/input_data.py
index fa148ae3e6f..5f3e57cdca3 100644
--- a/tensorflow/examples/tutorials/mnist/input_data.py
+++ b/tensorflow/examples/tutorials/mnist/input_data.py
@@ -25,7 +25,6 @@ import tempfile
 
 import numpy
 from six.moves import urllib
-from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 from tensorflow.contrib.learn.python.learn.datasets.mnist import read_data_sets
 # pylint: enable=unused-import
diff --git a/tensorflow/examples/tutorials/word2vec/word2vec_basic.py b/tensorflow/examples/tutorials/word2vec/word2vec_basic.py
index 5eef590c1dd..af05a1e6992 100644
--- a/tensorflow/examples/tutorials/word2vec/word2vec_basic.py
+++ b/tensorflow/examples/tutorials/word2vec/word2vec_basic.py
@@ -30,7 +30,6 @@ import zipfile
 
 import numpy as np
 from six.moves import urllib
-from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 
 from tensorflow.contrib.tensorboard.plugins import projector
@@ -254,7 +253,7 @@ def word2vec_basic(log_dir):
     print('Initialized')
 
     average_loss = 0
-    for step in xrange(num_steps):
+    for step in range(num_steps):
       batch_inputs, batch_labels = generate_batch(batch_size, num_skips,
                                                   skip_window)
       feed_dict = {train_inputs: batch_inputs, train_labels: batch_labels}
@@ -289,12 +288,12 @@ def word2vec_basic(log_dir):
       # Note that this is expensive (~20% slowdown if computed every 500 steps)
       if step % 10000 == 0:
         sim = similarity.eval()
-        for i in xrange(valid_size):
+        for i in range(valid_size):
           valid_word = reverse_dictionary[valid_examples[i]]
           top_k = 8  # number of nearest neighbors
           nearest = (-sim[i, :]).argsort()[1:top_k + 1]
           log_str = 'Nearest to %s:' % valid_word
-          for k in xrange(top_k):
+          for k in range(top_k):
             close_word = reverse_dictionary[nearest[k]]
             log_str = '%s %s,' % (log_str, close_word)
           print(log_str)
@@ -302,7 +301,7 @@ def word2vec_basic(log_dir):
 
     # Write corresponding labels for the embeddings.
     with open(log_dir + '/metadata.tsv', 'w') as f:
-      for i in xrange(vocabulary_size):
+      for i in range(vocabulary_size):
         f.write(reverse_dictionary[i] + '\n')
 
     # Save the model for checkpoints.
@@ -347,7 +346,7 @@ def word2vec_basic(log_dir):
         perplexity=30, n_components=2, init='pca', n_iter=5000, method='exact')
     plot_only = 500
     low_dim_embs = tsne.fit_transform(final_embeddings[:plot_only, :])
-    labels = [reverse_dictionary[i] for i in xrange(plot_only)]
+    labels = [reverse_dictionary[i] for i in range(plot_only)]
     plot_with_labels(low_dim_embs, labels, os.path.join(gettempdir(),
                                                         'tsne.png'))
 
diff --git a/tensorflow/lite/kernels/internal/spectrogram.cc b/tensorflow/lite/kernels/internal/spectrogram.cc
index 784e4bc99ef..0d016e27b3f 100644
--- a/tensorflow/lite/kernels/internal/spectrogram.cc
+++ b/tensorflow/lite/kernels/internal/spectrogram.cc
@@ -17,6 +17,7 @@ limitations under the License.
 
 #include <assert.h>
 #include <math.h>
+#include <cstdint>
 
 #include "third_party/fft2d/fft.h"
 
diff --git a/tensorflow/lite/testing/generate_examples_lib.py b/tensorflow/lite/testing/generate_examples_lib.py
index 292d7169e61..0264803da52 100644
--- a/tensorflow/lite/testing/generate_examples_lib.py
+++ b/tensorflow/lite/testing/generate_examples_lib.py
@@ -41,7 +41,6 @@ import traceback
 import zipfile
 import numpy as np
 from six import StringIO
-from six.moves import xrange
 
 # TODO(aselle): Disable GPU for now
 os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
@@ -3404,7 +3403,7 @@ def make_lstm_tests(options):
     input_vec_size = parameters["input_vec_size"]
     num_cells = parameters["num_cells"]
     inputs_after_split = []
-    for i in xrange(time_step_size):
+    for i in range(time_step_size):
       one_timestamp_input = tf.placeholder(
           dtype=parameters["dtype"],
           name="split_{}".format(i),
@@ -3437,7 +3436,7 @@ def make_lstm_tests(options):
     time_step_size = parameters["time_step_size"]
     input_vec_size = parameters["input_vec_size"]
     input_values = []
-    for _ in xrange(time_step_size):
+    for _ in range(time_step_size):
       tensor_data = create_tensor_data(parameters["dtype"],
                                        [num_batchs, input_vec_size], 0, 1)
       input_values.append(tensor_data)
diff --git a/tensorflow/python/autograph/impl/conversion.py b/tensorflow/python/autograph/impl/conversion.py
index a0275725ad1..780f3b9cc5f 100644
--- a/tensorflow/python/autograph/impl/conversion.py
+++ b/tensorflow/python/autograph/impl/conversion.py
@@ -20,7 +20,7 @@ from __future__ import print_function
 
 import collections
 import functools
-import imp
+import importlib
 import sys
 import threading
 import types
@@ -606,7 +606,7 @@ def _add_self_references(namespace, autograph_module):
   if ag_internal is None:
     # Craft a module that exposes parts of the external API as well as certain
     # internal modules.
-    ag_internal = imp.new_module('autograph')
+    ag_internal = importlib.import_module('autograph')
     ag_internal.__dict__.update(autograph_module.__dict__)
     ag_internal.ConversionOptions = converter.ConversionOptions
     ag_internal.STD = converter.STANDARD_OPTIONS
diff --git a/tensorflow/python/autograph/operators/py_builtins.py b/tensorflow/python/autograph/operators/py_builtins.py
index cd5f69bbce2..40d88fe18f7 100644
--- a/tensorflow/python/autograph/operators/py_builtins.py
+++ b/tensorflow/python/autograph/operators/py_builtins.py
@@ -359,7 +359,7 @@ def _py_zip(*iterables):
 SUPPORTED_BUILTINS = (abs, float, int, len, print, range, enumerate, zip)
 
 if six.PY2:
-  SUPPORTED_BUILTINS += (xrange,)
+  SUPPORTED_BUILTINS += (range,)
 
 BUILTIN_FUINCTIONS_MAP = {
     'abs': abs_,
@@ -369,7 +369,7 @@ BUILTIN_FUINCTIONS_MAP = {
     'print': print_,
     'range': range_,
     # TODO(mdan): This might make more sense as tf.data.range.
-    'xrange': range_,
+    'range': range_,
     'enumerate': enumerate_,
     'zip': zip_,
 }
diff --git a/tensorflow/python/autograph/pyct/compiler.py b/tensorflow/python/autograph/pyct/compiler.py
index 297f28cfeaf..4b0ac777b2b 100644
--- a/tensorflow/python/autograph/pyct/compiler.py
+++ b/tensorflow/python/autograph/pyct/compiler.py
@@ -25,7 +25,7 @@ from __future__ import print_function
 
 # TODO(mdan): Use six for compatibility here.
 import atexit
-import imp
+import importlib.util
 import os
 import tempfile
 
@@ -101,7 +101,10 @@ def source_to_entity(source, delete_on_exit):
 
   if delete_on_exit and ag_logging.get_verbosity() < 3:
     atexit.register(lambda: os.remove(f.name))
-  return imp.load_source(module_name, f.name), f.name
+  spec = importlib.util.spec_from_file_location(module_name, f.name)
+  module = importlib.util.module_from_spec(spec)
+  spec.loader.exec_module(module)
+  return module, f.name
 
 
 # TODO(mdan): Rename: ast_to_entity
diff --git a/tensorflow/python/autograph/pyct/inspect_utils.py b/tensorflow/python/autograph/pyct/inspect_utils.py
index 47c52d2e8bb..aacccdfeda9 100644
--- a/tensorflow/python/autograph/pyct/inspect_utils.py
+++ b/tensorflow/python/autograph/pyct/inspect_utils.py
@@ -54,7 +54,7 @@ SPECIAL_BUILTINS = {
 }
 
 if six.PY2:
-  SPECIAL_BUILTINS['xrange'] = xrange
+  SPECIAL_BUILTINS['range'] = range
 
 
 def islambda(f):
diff --git a/tensorflow/python/autograph/utils/testing.py b/tensorflow/python/autograph/utils/testing.py
index a59642c9577..7435a1f9018 100644
--- a/tensorflow/python/autograph/utils/testing.py
+++ b/tensorflow/python/autograph/utils/testing.py
@@ -18,7 +18,7 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-import imp
+import importlib
 
 from tensorflow.python.framework import ops
 from tensorflow.python.ops import gen_math_ops
@@ -27,7 +27,7 @@ from tensorflow.python.ops import math_ops
 
 def fake_tf():
   """Creates a fake module that looks like TensorFlow, for testing."""
-  mod = imp.new_module('tensorflow')
+  mod = importlib.import_module('tensorflow')
   mod_contents = {}
   mod_contents.update(gen_math_ops.__dict__)
   mod_contents.update(math_ops.__dict__)
diff --git a/tensorflow/python/client/session_benchmark.py b/tensorflow/python/client/session_benchmark.py
index 130b05ac4ce..8bca11b1fac 100644
--- a/tensorflow/python/client/session_benchmark.py
+++ b/tensorflow/python/client/session_benchmark.py
@@ -22,7 +22,6 @@ import time
 
 import numpy as np
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 from tensorflow.python.client import session
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
@@ -57,7 +56,7 @@ class SessionBenchmark(test.Benchmark):
       no_op = array_ops.identity(p).op
       with session.Session(target) as sess:
         sess.run(no_op, feed_dict={p: feed_val})  # Warm-up run.
-        for _ in xrange(iters):
+        for _ in range(iters):
           start_time = time.time()
           sess.run(no_op, feed_dict={p: feed_val})
           end_time = time.time()
@@ -85,7 +84,7 @@ class SessionBenchmark(test.Benchmark):
       with session.Session(target) as sess:
         sess.run(v.initializer)
         sess.run(v)  # Warm-up run.
-        for _ in xrange(iters):
+        for _ in range(iters):
           start_time = time.time()
           sess.run(v)
           end_time = time.time()
@@ -114,7 +113,7 @@ class SessionBenchmark(test.Benchmark):
         sess.run(v.initializer)
         runner = sess.make_callable(v)
         runner()  # Warm-up run.
-        for _ in xrange(iters):
+        for _ in range(iters):
           start_time = time.time()
           runner()
           end_time = time.time()
@@ -140,7 +139,7 @@ class SessionBenchmark(test.Benchmark):
       with session.Session(target) as sess:
         sess.run(v.initializer)
         sess.run(v.op)  # Warm-up run.
-        for _ in xrange(iters):
+        for _ in range(iters):
           start_time = time.time()
           sess.run(v.op)
           end_time = time.time()
@@ -167,7 +166,7 @@ class SessionBenchmark(test.Benchmark):
         sess.run(v.initializer)
         runner = sess.make_callable(v.op)
         runner()  # Warm-up run.
-        for _ in xrange(iters):
+        for _ in range(iters):
           start_time = time.time()
           runner()
           end_time = time.time()
diff --git a/tensorflow/python/client/session_partial_run_test.py b/tensorflow/python/client/session_partial_run_test.py
index a97930635af..0d00670eec4 100644
--- a/tensorflow/python/client/session_partial_run_test.py
+++ b/tensorflow/python/client/session_partial_run_test.py
@@ -19,7 +19,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.client import session
 from tensorflow.python.framework import common_shapes
@@ -96,18 +95,18 @@ class PartialRunTest(test_util.TensorFlowTestCase):
     inputs = []
     outputs = []
     a = constant_op.constant(2.0, dtypes.float32)
-    for i in xrange(steps):
+    for i in range(steps):
       inputs.append(array_ops.placeholder(dtypes.float32, shape=[]))
       a = math_ops.multiply(a, inputs[i])
       outputs.append(a)
 
     h = sess.partial_run_setup(outputs, inputs)
-    for i in xrange(steps):
+    for i in range(steps):
       res = sess.partial_run(h, outputs[i], feed_dict={inputs[i]: 1.0})
     self.assertEqual(2.0, res)
 
     feed_dict = {}
-    for i in xrange(steps):
+    for i in range(steps):
       feed_dict[inputs[i]] = 1.0
     res = sess.run(outputs, feed_dict)
     self.assertEqual(steps, len(res))
diff --git a/tensorflow/python/client/session_test.py b/tensorflow/python/client/session_test.py
index e407a541fcd..85fe81d467f 100644
--- a/tensorflow/python/client/session_test.py
+++ b/tensorflow/python/client/session_test.py
@@ -27,7 +27,6 @@ import warnings
 
 import numpy as np
 import six
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.framework import attr_value_pb2
 from tensorflow.core.lib.core import error_codes_pb2
@@ -1551,7 +1550,7 @@ class SessionTest(test_util.TensorFlowTestCase):
         for s in shape:
           size *= s
         c_list = np.array(
-            [compat.as_bytes(str(i)) for i in xrange(size)],
+            [compat.as_bytes(str(i)) for i in range(size)],
             dtype=np.object).reshape(shape) if size > 0 else []
         c = constant_op.constant(c_list)
         self.assertAllEqual(c.eval(), c_list)
@@ -1563,7 +1562,7 @@ class SessionTest(test_util.TensorFlowTestCase):
         for s in shape:
           size *= s
         c_list = np.array(
-            [compat.as_bytes(str(i)) for i in xrange(size)],
+            [compat.as_bytes(str(i)) for i in range(size)],
             dtype=np.object).reshape(shape)
         feed_t = array_ops.placeholder(dtype=dtypes.string, shape=shape)
         c = array_ops.identity(feed_t)
@@ -1760,7 +1759,7 @@ class SessionTest(test_util.TensorFlowTestCase):
         b = math_ops.add(a, a)
         c = array_ops.identity(b)
         d = math_ops.multiply(c, c)
-      for step in xrange(120):
+      for step in range(120):
         run_metadata = config_pb2.RunMetadata()
         sess.run(
             d,
diff --git a/tensorflow/python/compiler/xla/xla.py b/tensorflow/python/compiler/xla/xla.py
index 55bfaeb3931..bc54f40a5f4 100644
--- a/tensorflow/python/compiler/xla/xla.py
+++ b/tensorflow/python/compiler/xla/xla.py
@@ -20,7 +20,6 @@ from __future__ import print_function
 
 import collections
 import contextlib
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.compiler.jit.ops import xla_ops
 from tensorflow.compiler.jit.ops import xla_ops_grad  # pylint: disable=unused-import
@@ -220,7 +219,7 @@ class XLACompileContext(control_flow_ops.XLAControlFlowContext):
         op._add_control_input(self._pivot)
         # pylint: enable=protected-access
     else:
-      for index in xrange(len(op.inputs)):
+      for index in range(len(op.inputs)):
         x = op.inputs[index]
         real_x = self.AddValue(x)
         if real_x is not x:
diff --git a/tensorflow/python/data/experimental/benchmarks/rejection_resample_benchmark.py b/tensorflow/python/data/experimental/benchmarks/rejection_resample_benchmark.py
index 28253b19ece..52552d9d935 100644
--- a/tensorflow/python/data/experimental/benchmarks/rejection_resample_benchmark.py
+++ b/tensorflow/python/data/experimental/benchmarks/rejection_resample_benchmark.py
@@ -20,7 +20,6 @@ from __future__ import print_function
 import time
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.client import session
 from tensorflow.python.data.experimental.ops import resampling
@@ -46,7 +45,7 @@ def _time_resampling(data_np, target_dist, init_dist, num_to_sample):  # pylint:
 
   with session.Session() as sess:
     start_time = time.time()
-    for _ in xrange(num_to_sample):
+    for _ in range(num_to_sample):
       sess.run(get_next)
     end_time = time.time()
 
diff --git a/tensorflow/python/data/experimental/kernel_tests/parallel_interleave_test.py b/tensorflow/python/data/experimental/kernel_tests/parallel_interleave_test.py
index 1fe5655ec02..21d771bf05c 100644
--- a/tensorflow/python/data/experimental/kernel_tests/parallel_interleave_test.py
+++ b/tensorflow/python/data/experimental/kernel_tests/parallel_interleave_test.py
@@ -23,7 +23,7 @@ import threading
 import time
 
 import numpy as np
-from six.moves import zip_longest
+from itertools import zip_longest
 
 from tensorflow.python.data.experimental.ops import interleave_ops
 from tensorflow.python.data.kernel_tests import test_base
diff --git a/tensorflow/python/data/ops/dataset_ops.py b/tensorflow/python/data/ops/dataset_ops.py
index cad80d37a45..6b444c1e38e 100644
--- a/tensorflow/python/data/ops/dataset_ops.py
+++ b/tensorflow/python/data/ops/dataset_ops.py
@@ -26,7 +26,7 @@ import weakref
 
 import numpy as np
 import six
-from six.moves import queue as Queue  # pylint: disable=redefined-builtin
+import queue as Queue
 
 
 from tensorflow.core.framework import graph_pb2
@@ -700,7 +700,7 @@ class DatasetV2(tracking_base.Trackable, composite_tensor.CompositeTensor):
     ```
 
     Args:
-      *args: follows the same semantics as python's xrange.
+      *args: follows the same semantics as python's range.
         len(args) == 1 -> start = 0, stop = args[0], step = 1
         len(args) == 2 -> start = args[0], stop = args[1], step = 1
         len(args) == 3 -> start = args[0], stop = args[1, stop = args[2]
diff --git a/tensorflow/python/data/util/traverse.py b/tensorflow/python/data/util/traverse.py
index 12e576fb414..103021765d3 100644
--- a/tensorflow/python/data/util/traverse.py
+++ b/tensorflow/python/data/util/traverse.py
@@ -17,7 +17,7 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import queue as Queue  # pylint: disable=redefined-builtin
+import queue as Queue
 
 from tensorflow.python.framework import dtypes
 
diff --git a/tensorflow/python/debug/cli/analyzer_cli.py b/tensorflow/python/debug/cli/analyzer_cli.py
index 5c21b1518ed..d022525f2ad 100644
--- a/tensorflow/python/debug/cli/analyzer_cli.py
+++ b/tensorflow/python/debug/cli/analyzer_cli.py
@@ -25,7 +25,6 @@ import argparse
 import copy
 import re
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.debug.cli import cli_config
 from tensorflow.python.debug.cli import cli_shared
@@ -1422,7 +1421,7 @@ class DebugAnalyzer(object):
 
     # Create depth-dependent hanging indent for the line.
     hang = ""
-    for k in xrange(depth):
+    for k in range(depth):
       if k < depth - 1:
         if k + 1 in unfinished:
           hang += HANG_UNFINISHED
diff --git a/tensorflow/python/debug/cli/analyzer_cli_test.py b/tensorflow/python/debug/cli/analyzer_cli_test.py
index 795849843f4..9ffc98791ae 100644
--- a/tensorflow/python/debug/cli/analyzer_cli_test.py
+++ b/tensorflow/python/debug/cli/analyzer_cli_test.py
@@ -22,7 +22,6 @@ import shutil
 import tempfile
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.core.protobuf import rewriter_config_pb2
@@ -382,7 +381,7 @@ def assert_node_attribute_lines(tst,
     tst.assertEqual("", next(line_iter))
 
     dump_timestamps_ms = []
-    for _ in xrange(num_dumped_tensors):
+    for _ in range(num_dumped_tensors):
       line = next(line_iter)
 
       tst.assertStartsWith(line.strip(), "Slot 0 @ DebugIdentity @")
@@ -1516,10 +1515,10 @@ class AnalyzerCLISimpleMulAddTest(test_util.TensorFlowTestCase):
     out = self._registry.dispatch_command("list_source", [])
 
     non_tf_lib_files_start = [
-        i for i in xrange(len(out.lines))
+        i for i in range(len(out.lines))
         if out.lines[i].startswith("Source file path")][0] + 1
     non_tf_lib_files_end = [
-        i for i in xrange(len(out.lines))
+        i for i in range(len(out.lines))
         if out.lines[i].startswith("TensorFlow Python library file(s):")][0] - 1
     non_tf_lib_files = [
         line.split(" ")[0] for line
@@ -1527,7 +1526,7 @@ class AnalyzerCLISimpleMulAddTest(test_util.TensorFlowTestCase):
     self.assertIn(self._curr_file_path, non_tf_lib_files)
 
     # Check that the TF library files are marked with special color attribute.
-    for i in xrange(non_tf_lib_files_end + 1, len(out.lines)):
+    for i in range(non_tf_lib_files_end + 1, len(out.lines)):
       if not out.lines[i]:
         continue
       for attr_seg in  out.font_attr_segs[i]:
@@ -1541,10 +1540,10 @@ class AnalyzerCLISimpleMulAddTest(test_util.TensorFlowTestCase):
     self.assertStartsWith(out.lines[1], "Node name regex filter: \".*/read\"")
 
     non_tf_lib_files_start = [
-        i for i in xrange(len(out.lines))
+        i for i in range(len(out.lines))
         if out.lines[i].startswith("Source file path")][0] + 1
     non_tf_lib_files_end = [
-        i for i in xrange(len(out.lines))
+        i for i in range(len(out.lines))
         if out.lines[i].startswith("TensorFlow Python library file(s):")][0] - 1
     non_tf_lib_files = [
         line.split(" ")[0] for line
@@ -1552,7 +1551,7 @@ class AnalyzerCLISimpleMulAddTest(test_util.TensorFlowTestCase):
     self.assertIn(self._curr_file_path, non_tf_lib_files)
 
     # Check that the TF library files are marked with special color attribute.
-    for i in xrange(non_tf_lib_files_end + 1, len(out.lines)):
+    for i in range(non_tf_lib_files_end + 1, len(out.lines)):
       if not out.lines[i]:
         continue
       for attr_seg in  out.font_attr_segs[i]:
@@ -2071,7 +2070,7 @@ class AnalyzerCLIWhileLoopTest(test_util.TensorFlowTestCase):
     self.assertEqual("Tensor \"while/Identity:0\" generated 10 dumps:",
                      output.lines[0])
 
-    for i in xrange(10):
+    for i in range(10):
       self.assertTrue(output.lines[i + 1].startswith("#%d" % i))
       self.assertTrue(output.lines[i + 1].endswith(
           " ms] while/Identity:0:DebugIdentity"))
@@ -2083,7 +2082,7 @@ class AnalyzerCLIWhileLoopTest(test_util.TensorFlowTestCase):
     self.assertEqual("  print_tensor while/Identity:0 -n 0", output.lines[-1])
 
   def testMultipleDumpsPrintTensorWithNumber(self):
-    for i in xrange(5):
+    for i in range(5):
       output = self._registry.dispatch_command(
           "pt", ["while/Identity:0", "-n", "%d" % i])
 
diff --git a/tensorflow/python/debug/cli/curses_ui.py b/tensorflow/python/debug/cli/curses_ui.py
index 7b87972d694..ccbce81e661 100644
--- a/tensorflow/python/debug/cli/curses_ui.py
+++ b/tensorflow/python/debug/cli/curses_ui.py
@@ -25,7 +25,6 @@ import signal
 import sys
 import threading
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.debug.cli import base_ui
 from tensorflow.python.debug.cli import cli_shared
@@ -179,7 +178,7 @@ class ScrollBar(object):
 
       layout = debugger_cli_common.RichTextLines(
           [up_text], font_attr_segs={0: [(0, width, self.BASE_ATTR)]})
-      for i in xrange(1, self._scroll_bar_height - 1):
+      for i in range(1, self._scroll_bar_height - 1):
         font_attr_segs = foreground_font_attr_segs if i == block_y else None
         layout.append(empty_line, font_attr_segs=font_attr_segs)
       layout.append(down_text, font_attr_segs=foreground_font_attr_segs)
@@ -965,7 +964,7 @@ class CursesUI(base_ui.BaseUI):
     """Erase existing text in command textpad."""
 
     existing_len = len(self._command_textbox.gather())
-    for _ in xrange(existing_len):
+    for _ in range(existing_len):
       self._command_textbox.do_command(self.BACKSPACE_KEY)
 
   def _screen_draw_text_line(self, row, line, attr=curses.A_NORMAL, color=None):
@@ -1159,7 +1158,7 @@ class CursesUI(base_ui.BaseUI):
     # Create new output pad.
     pad = self._screen_new_output_pad(rows, cols)
 
-    for i in xrange(len(output.lines)):
+    for i in range(len(output.lines)):
       if i in output.font_attr_segs:
         self._screen_add_line_to_output_pad(
             pad, i, output.lines[i], color_segments=output.font_attr_segs[i])
diff --git a/tensorflow/python/debug/cli/tensor_format.py b/tensorflow/python/debug/cli/tensor_format.py
index 374fcd58bc3..cd2733f3212 100644
--- a/tensorflow/python/debug/cli/tensor_format.py
+++ b/tensorflow/python/debug/cli/tensor_format.py
@@ -21,7 +21,6 @@ import copy
 import re
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.debug.cli import debugger_cli_common
 from tensorflow.python.debug.lib import debug_data
@@ -273,7 +272,7 @@ def _annotate_ndarray_lines(
       else:
         if curr_dim > 0:
           curr_indices[curr_dim - 1] += 1
-          for k in xrange(curr_dim, ndims):
+          for k in range(curr_dim, ndims):
             curr_indices[k] = 0
 
   return annotations
@@ -348,7 +347,7 @@ def locate_tensor_element(formatted, indices):
 
   batch_pos = 0  # Current position in the batch.
 
-  for r in xrange(len(lines)):
+  for r in range(len(lines)):
     if r not in annot:
       continue
 
diff --git a/tensorflow/python/debug/examples/debug_fibonacci.py b/tensorflow/python/debug/examples/debug_fibonacci.py
index 870a6d6813a..6313d06c421 100644
--- a/tensorflow/python/debug/examples/debug_fibonacci.py
+++ b/tensorflow/python/debug/examples/debug_fibonacci.py
@@ -21,7 +21,6 @@ import argparse
 import sys
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow
 
 from tensorflow.python import debug as tf_debug
@@ -40,7 +39,7 @@ def main(_):
   n1 = tf.Variable(
       np.ones([FLAGS.tensor_size] * 2), dtype=tf.int32, name="node_01")
 
-  for i in xrange(2, FLAGS.length):
+  for i in range(2, FLAGS.length):
     n0, n1 = n1, tf.add(n0, n1, name="node_%.2d" % i)
 
   sess.run(tf.global_variables_initializer())
diff --git a/tensorflow/python/debug/lib/debug_utils.py b/tensorflow/python/debug/lib/debug_utils.py
index eb21694ba2f..f4b599b7038 100644
--- a/tensorflow/python/debug/lib/debug_utils.py
+++ b/tensorflow/python/debug/lib/debug_utils.py
@@ -20,7 +20,6 @@ from __future__ import print_function
 
 import re
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 
 def add_debug_tensor_watch(run_options,
@@ -163,7 +162,7 @@ def watch_graph(run_options,
     if op_type_pattern and not op_type_pattern.match(op_type):
       continue
 
-    for slot in xrange(len(op.outputs)):
+    for slot in range(len(op.outputs)):
       if (tensor_dtype_pattern and
           not tensor_dtype_pattern.match(op.outputs[slot].dtype.name)):
         continue
@@ -273,7 +272,7 @@ def watch_graph_with_blacklists(run_options,
     if op_type_pattern and op_type_pattern.match(op_type):
       continue
 
-    for slot in xrange(len(op.outputs)):
+    for slot in range(len(op.outputs)):
       if (tensor_dtype_pattern and
           tensor_dtype_pattern.match(op.outputs[slot].dtype.name)):
         continue
diff --git a/tensorflow/python/debug/lib/dist_session_debug_grpc_test.py b/tensorflow/python/debug/lib/dist_session_debug_grpc_test.py
index 0dc01748b08..129101329e4 100644
--- a/tensorflow/python/debug/lib/dist_session_debug_grpc_test.py
+++ b/tensorflow/python/debug/lib/dist_session_debug_grpc_test.py
@@ -26,7 +26,6 @@ import sys
 import time
 
 import portpicker
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.python.client import session
@@ -159,7 +158,7 @@ class DistributedSessionDebugTest(test_util.TensorFlowTestCase):
           debug_ops=["DebugIdentity(gated_grpc=True)"],
           debug_urls=[self.debug_server_url])
 
-      for i in xrange(4):
+      for i in range(4):
         if i % 2 == 0:
           self.debug_server.request_watch("p", 0, "DebugIdentity")
         else:
@@ -214,7 +213,7 @@ class DistributedSessionDebugTest(test_util.TensorFlowTestCase):
       sess = grpc_wrapper.GrpcDebugWrapperSession(
           sess, "localhost:%d" % self.debug_server_port, watch_fn=watch_fn)
 
-      for i in xrange(4):
+      for i in range(4):
         expected_p = (10.0 + 2.0 * (i + 1)) * (100.0 - 5.0 * (i + 1))
         self.assertAllClose(-expected_p, sess.run(self.q))
 
diff --git a/tensorflow/python/debug/lib/grpc_large_data_test.py b/tensorflow/python/debug/lib/grpc_large_data_test.py
index 64d5a241531..d362b40d80d 100644
--- a/tensorflow/python/debug/lib/grpc_large_data_test.py
+++ b/tensorflow/python/debug/lib/grpc_large_data_test.py
@@ -21,7 +21,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.debug.lib import grpc_debug_test_server
 from tensorflow.python.debug.lib import session_debug_testlib
@@ -63,7 +62,7 @@ class LargeGraphAndLargeTensorsDebugTest(test_util.TensorFlowTestCase):
         use_gpu=True,
         config=session_debug_testlib.no_rewrite_session_config()) as sess:
       u = variables.VariableV1(42.0, name="original_u")
-      for _ in xrange(50 * 1000):
+      for _ in range(50 * 1000):
         u = array_ops.identity(u)
       sess.run(variables.global_variables_initializer())
 
@@ -91,7 +90,7 @@ class LargeGraphAndLargeTensorsDebugTest(test_util.TensorFlowTestCase):
     with self.session(
         use_gpu=True,
         config=session_debug_testlib.no_rewrite_session_config()) as sess:
-      u_init_val_array = list(xrange(1200 * 1024))
+      u_init_val_array = list(range(1200 * 1024))
       # Size: 4 * 1200 * 1024 = 4800k > 4M
 
       u_init = constant_op.constant(
diff --git a/tensorflow/python/debug/lib/session_debug_grpc_test.py b/tensorflow/python/debug/lib/session_debug_grpc_test.py
index f3b187cb3e9..5cba4f8f3ef 100644
--- a/tensorflow/python/debug/lib/session_debug_grpc_test.py
+++ b/tensorflow/python/debug/lib/session_debug_grpc_test.py
@@ -24,7 +24,6 @@ from __future__ import print_function
 import os
 import shutil
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.python.client import session
@@ -401,7 +400,7 @@ class SessionDebugGrpcGatingTest(test_util.TensorFlowTestCase):
                      "DebugNumericSummary(gated_grpc=true)"],
           debug_urls=[self._debug_server_url_1])
 
-      for i in xrange(4):
+      for i in range(4):
         self._server_1.clear_data()
 
         if i % 2 == 0:
@@ -473,7 +472,7 @@ class SessionDebugGrpcGatingTest(test_util.TensorFlowTestCase):
           debug_ops=["DebugIdentity(gated_grpc=true)"],
           debug_urls=[debug_server_url])
 
-      for i in xrange(4):
+      for i in range(4):
         server.clear_data()
 
         sess.run([inc_v_1, inc_v_2],
@@ -507,7 +506,7 @@ class SessionDebugGrpcGatingTest(test_util.TensorFlowTestCase):
           debug_ops=["DebugIdentity(gated_grpc=true)"],
           debug_urls=[self._debug_server_url_1, self._debug_server_url_2])
 
-      for i in xrange(4):
+      for i in range(4):
         self._server_1.clear_data()
         self._server_2.clear_data()
 
@@ -553,7 +552,7 @@ class SessionDebugGrpcGatingTest(test_util.TensorFlowTestCase):
           debug_ops=["DebugIdentity(gated_grpc=true)"],
           debug_urls=[self._debug_server_url_1])
 
-      for i in xrange(4):
+      for i in range(4):
         self._server_1.clear_data()
 
         if i in (0, 2):
@@ -608,7 +607,7 @@ class SessionDebugGrpcGatingTest(test_util.TensorFlowTestCase):
       sess = grpc_wrapper.TensorBoardDebugWrapperSession(
           sess, self._debug_server_url_1)
 
-      for i in xrange(4):
+      for i in range(4):
         self._server_1.clear_data()
 
         if i in (0, 2):
@@ -680,7 +679,7 @@ class SessionDebugGrpcGatingTest(test_util.TensorFlowTestCase):
       sess = grpc_wrapper.TensorBoardDebugWrapperSession(
           sess, self._debug_server_url_1, send_traceback_and_source_code=False)
 
-      for i in xrange(4):
+      for i in range(4):
         self._server_1.clear_data()
 
         if i == 0:
diff --git a/tensorflow/python/debug/lib/session_debug_testlib.py b/tensorflow/python/debug/lib/session_debug_testlib.py
index e2740d80705..25f0384b681 100644
--- a/tensorflow/python/debug/lib/session_debug_testlib.py
+++ b/tensorflow/python/debug/lib/session_debug_testlib.py
@@ -26,7 +26,6 @@ import tempfile
 import threading
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.core.protobuf import rewriter_config_pb2
@@ -520,7 +519,7 @@ class SessionDebugTestBase(test_util.TensorFlowTestCase):
 
       while_id_tensors = dump.get_tensors("while/Identity", 0, "DebugIdentity")
       self.assertEqual(10, len(while_id_tensors))
-      for k in xrange(len(while_id_tensors)):
+      for k in range(len(while_id_tensors)):
         self.assertAllClose(np.array(k), while_id_tensors[k])
 
       # Verify ascending timestamps from the while loops.
@@ -1488,7 +1487,7 @@ class DebugConcurrentRunCallsTest(test_util.TensorFlowTestCase):
     with session.Session() as sess:
       v = variables.VariableV1(30.0, name="v")
       constants = []
-      for i in xrange(self._num_concurrent_runs):
+      for i in range(self._num_concurrent_runs):
         constants.append(constant_op.constant(1.0, name="c%d" % i))
       incs = [
           state_ops.assign_add(
@@ -1503,11 +1502,11 @@ class DebugConcurrentRunCallsTest(test_util.TensorFlowTestCase):
         run_options = config_pb2.RunOptions(output_partition_graphs=True)
         debug_utils.watch_graph(
             run_options, sess.graph, debug_urls=concurrent_debug_urls[index])
-        for _ in xrange(100):
+        for _ in range(100):
           sess.run(incs[index], options=run_options)
 
       inc_threads = []
-      for index in xrange(self._num_concurrent_runs):
+      for index in range(self._num_concurrent_runs):
         inc_thread = threading.Thread(target=functools.partial(inc_job, index))
         inc_thread.start()
         inc_threads.append(inc_thread)
@@ -1518,7 +1517,7 @@ class DebugConcurrentRunCallsTest(test_util.TensorFlowTestCase):
                           sess.run(v))
 
       all_session_run_indices = []
-      for index in xrange(self._num_concurrent_runs):
+      for index in range(self._num_concurrent_runs):
         dump = debug_data.DebugDumpDir(self._dump_roots[index])
         self.assertTrue(dump.loaded_partition_graphs())
 
@@ -1548,14 +1547,14 @@ class DebugConcurrentRunCallsTest(test_util.TensorFlowTestCase):
         executor_step_indices = zip(timestamps, executor_step_indices)
         executor_step_indices = sorted(
             executor_step_indices, key=lambda x: x[0])
-        for i in xrange(len(executor_step_indices) - 1):
+        for i in range(len(executor_step_indices) - 1):
           self.assertEquals(executor_step_indices[i][1] + 1,
                             executor_step_indices[i + 1][1])
 
         # Assert that session_run_index increase monotonically.
         session_run_indices = zip(timestamps, session_run_indices)
         session_run_indices = sorted(session_run_indices, key=lambda x: x[0])
-        for i in xrange(len(session_run_indices) - 1):
+        for i in range(len(session_run_indices) - 1):
           self.assertGreater(session_run_indices[i + 1][1],
                              session_run_indices[i][1])
 
diff --git a/tensorflow/python/distribute/cluster_resolver/tpu_cluster_resolver.py b/tensorflow/python/distribute/cluster_resolver/tpu_cluster_resolver.py
index 9035cd4681c..934bdddcbf6 100644
--- a/tensorflow/python/distribute/cluster_resolver/tpu_cluster_resolver.py
+++ b/tensorflow/python/distribute/cluster_resolver/tpu_cluster_resolver.py
@@ -22,10 +22,9 @@ import collections
 import os
 import re
 
-from six.moves import urllib
-from six.moves.urllib.error import URLError
-from six.moves.urllib.request import Request
-from six.moves.urllib.request import urlopen
+from urllib.error import URLError
+from urllib.request import Request
+from urllib.request import urlopen
 
 from tensorflow.python.distribute.cluster_resolver.cluster_resolver import ClusterResolver
 from tensorflow.python.distribute.cluster_resolver.cluster_resolver import format_master_url
diff --git a/tensorflow/python/eager/benchmarks_test.py b/tensorflow/python/eager/benchmarks_test.py
index 01d2134da92..3567da3bba0 100644
--- a/tensorflow/python/eager/benchmarks_test.py
+++ b/tensorflow/python/eager/benchmarks_test.py
@@ -31,7 +31,6 @@ import time
 
 import numpy as np
 import six
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python import keras
 from tensorflow.python import pywrap_tensorflow
@@ -146,7 +145,7 @@ def run_benchmark(func, num_iters, execution_mode=None):
     if execution_mode == context.ASYNC:
       ctx.executor.wait()
     start = time.time()
-    for _ in xrange(num_iters):
+    for _ in range(num_iters):
       func()
     if execution_mode == context.ASYNC:
       ctx.executor.wait()
diff --git a/tensorflow/python/eager/def_function_test.py b/tensorflow/python/eager/def_function_test.py
index 9ab42b63098..2951623efca 100644
--- a/tensorflow/python/eager/def_function_test.py
+++ b/tensorflow/python/eager/def_function_test.py
@@ -21,7 +21,6 @@ import functools
 import re
 import weakref
 
-from six.moves import range
 
 from tensorflow.python.eager import backprop
 from tensorflow.python.eager import def_function
diff --git a/tensorflow/python/eager/monitoring.py b/tensorflow/python/eager/monitoring.py
index 09838f143e0..9bc737199e8 100644
--- a/tensorflow/python/eager/monitoring.py
+++ b/tensorflow/python/eager/monitoring.py
@@ -19,6 +19,7 @@ from __future__ import division
 from __future__ import print_function
 
 import collections
+import weakref
 
 from tensorflow.core.framework import summary_pb2
 from tensorflow.python import pywrap_tensorflow
@@ -119,8 +120,10 @@ class Metric(object):
           self._metric_name, len(self._metric_methods)))
 
     self._metric = self._metric_methods[self._label_length].create(*args)
+    self._finalizer = weakref.finalize(self, self._cleanup, self)
 
-  def __del__(self):
+  @staticmethod
+  def _cleanup(self):
     try:
       deleter = self._metric_methods[self._label_length].delete
       metric = self._metric
diff --git a/tensorflow/python/eager/pywrap_tfe_src.cc b/tensorflow/python/eager/pywrap_tfe_src.cc
index ba9d8a36b0d..3cbeac1fa20 100644
--- a/tensorflow/python/eager/pywrap_tfe_src.cc
+++ b/tensorflow/python/eager/pywrap_tfe_src.cc
@@ -1319,20 +1319,20 @@ static PyTypeObject TFE_Py_Tape_Type = {
     sizeof(TFE_Py_Tape),                          /* tp_basicsize */
     0,                                            /* tp_itemsize */
     &TFE_Py_Tape_Delete,                          /* tp_dealloc */
-    nullptr,                                      /* tp_print */
-    nullptr,                                      /* tp_getattr */
-    nullptr,                                      /* tp_setattr */
-    nullptr,                                      /* tp_reserved */
-    nullptr,                                      /* tp_repr */
-    nullptr,                                      /* tp_as_number */
-    nullptr,                                      /* tp_as_sequence */
-    nullptr,                                      /* tp_as_mapping */
-    nullptr,                                      /* tp_hash  */
-    nullptr,                                      /* tp_call */
-    nullptr,                                      /* tp_str */
-    nullptr,                                      /* tp_getattro */
-    nullptr,                                      /* tp_setattro */
-    nullptr,                                      /* tp_as_buffer */
+    0,                                      /* tp_print */
+    0,                                      /* tp_getattr */
+    0,                                      /* tp_setattr */
+    0,                                      /* tp_reserved */
+    0,                                      /* tp_repr */
+    0,                                      /* tp_as_number */
+    0,                                      /* tp_as_sequence */
+    0,                                      /* tp_as_mapping */
+    0,                                      /* tp_hash  */
+    0,                                      /* tp_call */
+    0,                                      /* tp_str */
+    0,                                      /* tp_getattro */
+    0,                                      /* tp_setattro */
+    0,                                      /* tp_as_buffer */
     Py_TPFLAGS_DEFAULT,                           /* tp_flags */
     "TFE_Py_Tape objects",                        /* tp_doc */
 };
@@ -1353,20 +1353,20 @@ static PyTypeObject TFE_Py_ForwardAccumulator_Type = {
     sizeof(TFE_Py_ForwardAccumulator),                      /* tp_basicsize */
     0,                                                      /* tp_itemsize */
     &TFE_Py_ForwardAccumulatorDelete,                       /* tp_dealloc */
-    nullptr,                                                /* tp_print */
-    nullptr,                                                /* tp_getattr */
-    nullptr,                                                /* tp_setattr */
-    nullptr,                                                /* tp_reserved */
-    nullptr,                                                /* tp_repr */
-    nullptr,                                                /* tp_as_number */
-    nullptr,                                                /* tp_as_sequence */
-    nullptr,                                                /* tp_as_mapping */
-    nullptr,                                                /* tp_hash  */
-    nullptr,                                                /* tp_call */
-    nullptr,                                                /* tp_str */
-    nullptr,                                                /* tp_getattro */
-    nullptr,                                                /* tp_setattro */
-    nullptr,                                                /* tp_as_buffer */
+    0,                                                /* tp_print */
+    0,                                                /* tp_getattr */
+    0,                                                /* tp_setattr */
+    0,                                                /* tp_reserved */
+    0,                                                /* tp_repr */
+    0,                                                /* tp_as_number */
+    0,                                                /* tp_as_sequence */
+    0,                                                /* tp_as_mapping */
+    0,                                                /* tp_hash  */
+    0,                                                /* tp_call */
+    0,                                                /* tp_str */
+    0,                                                /* tp_getattro */
+    0,                                                /* tp_setattro */
+    0,                                                /* tp_as_buffer */
     Py_TPFLAGS_DEFAULT,                                     /* tp_flags */
     "TFE_Py_ForwardAccumulator objects",                    /* tp_doc */
 };
diff --git a/tensorflow/python/framework/common_shapes.py b/tensorflow/python/framework/common_shapes.py
index 11612295d92..1db0849f8ae 100644
--- a/tensorflow/python/framework/common_shapes.py
+++ b/tensorflow/python/framework/common_shapes.py
@@ -18,7 +18,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-import six.moves
 
 from tensorflow.python import pywrap_tensorflow
 from tensorflow.python.framework import cpp_shape_inference_pb2
diff --git a/tensorflow/python/framework/dtypes.py b/tensorflow/python/framework/dtypes.py
index e817c3172ee..eb7e8608569 100644
--- a/tensorflow/python/framework/dtypes.py
+++ b/tensorflow/python/framework/dtypes.py
@@ -18,7 +18,7 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import builtins
+import builtins
 
 from tensorflow.core.framework import types_pb2
 from tensorflow.python import pywrap_tensorflow
@@ -594,7 +594,7 @@ _TF_TO_NP = {
     # NOTE(touts): For strings we use np.object as it supports variable length
     # strings.
     types_pb2.DT_STRING:
-        np.object,
+        object,
     types_pb2.DT_COMPLEX64:
         np.complex64,
     types_pb2.DT_COMPLEX128:
@@ -602,7 +602,7 @@ _TF_TO_NP = {
     types_pb2.DT_INT64:
         np.int64,
     types_pb2.DT_BOOL:
-        np.bool,
+        bool,
     types_pb2.DT_QINT8:
         _np_qint8,
     types_pb2.DT_QUINT8:
@@ -636,7 +636,7 @@ _TF_TO_NP = {
     types_pb2.DT_INT8_REF:
         np.int8,
     types_pb2.DT_STRING_REF:
-        np.object,
+        object,
     types_pb2.DT_COMPLEX64_REF:
         np.complex64,
     types_pb2.DT_COMPLEX128_REF:
@@ -646,7 +646,7 @@ _TF_TO_NP = {
     types_pb2.DT_UINT64_REF:
         np.uint64,
     types_pb2.DT_BOOL_REF:
-        np.bool,
+        bool,
     types_pb2.DT_QINT8_REF:
         _np_qint8,
     types_pb2.DT_QUINT8_REF:
diff --git a/tensorflow/python/framework/errors_impl.py b/tensorflow/python/framework/errors_impl.py
index caaeab40254..9bc5e079f3c 100644
--- a/tensorflow/python/framework/errors_impl.py
+++ b/tensorflow/python/framework/errors_impl.py
@@ -506,7 +506,7 @@ _CODE_TO_EXCEPTION_CLASS = {
     DATA_LOSS: DataLossError,
 }
 
-c_api.PyExceptionRegistry_Init(_CODE_TO_EXCEPTION_CLASS)
+c_api.PyExceptionRegistry.Init(_CODE_TO_EXCEPTION_CLASS)
 
 _EXCEPTION_CLASS_TO_CODE = {
     class_: code for code, class_ in _CODE_TO_EXCEPTION_CLASS.items()}
diff --git a/tensorflow/python/framework/load_library.py b/tensorflow/python/framework/load_library.py
index 304745fe3c7..f4b334dbc7e 100644
--- a/tensorflow/python/framework/load_library.py
+++ b/tensorflow/python/framework/load_library.py
@@ -20,7 +20,7 @@ from __future__ import print_function
 
 import errno
 import hashlib
-import imp
+from importlib.util import spec_from_loader, module_from_spec
 import os
 import platform
 import sys
@@ -73,7 +73,8 @@ def load_op_library(library_filename):
   module_name = hashlib.md5(wrappers).hexdigest()
   if module_name in sys.modules:
     return sys.modules[module_name]
-  module = imp.new_module(module_name)
+  spec = spec_from_loader(module_name, loader=None)
+  module = module_from_spec(spec)
   # pylint: disable=exec-used
   exec(wrappers, module.__dict__)
   # Stash away the library handle for making calls into the dynamic library.
diff --git a/tensorflow/python/framework/ops.py b/tensorflow/python/framework/ops.py
index ec43d755c7b..cd5721a1a39 100644
--- a/tensorflow/python/framework/ops.py
+++ b/tensorflow/python/framework/ops.py
@@ -26,7 +26,6 @@ import threading
 
 import numpy as np
 import six
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.framework import attr_value_pb2
 from tensorflow.core.framework import function_pb2
@@ -172,7 +171,7 @@ def _override_helper(clazz_object, operator, func):
   if existing is not None:
     # Check to see if this is a default method-wrapper or slot wrapper which
     # will be true for the comparison operators.
-    if not isinstance(existing, type(object.__lt__)):
+    if not isinstance(existing, (type(object.__lt__), type(object().__lt__))):
       raise ValueError("operator %s cannot be overwritten again on class %s." %
                        (operator, clazz_object))
   if operator not in Tensor.OVERLOADABLE_OPERATORS:
@@ -554,7 +553,7 @@ class Tensor(_TensorLike):
     if shape[0] is None:
       raise TypeError(
           "Cannot iterate over a tensor with unknown first dimension.")
-    for i in xrange(shape[0]):
+    for i in range(shape[0]):
       yield self[i]
 
   def _shape_as_list(self):
@@ -1974,7 +1973,7 @@ class Operation(object):
     num_outputs = c_api.TF_OperationNumOutputs(self._c_op)
     output_types = [
         c_api.TF_OperationOutputType(self._tf_output(i))
-        for i in xrange(num_outputs)
+        for i in range(num_outputs)
     ]
     # In all the tests we have output_types that are passed into
     # Operation.__init__ are a list of ints (which is illegal according
@@ -2182,7 +2181,7 @@ class Operation(object):
     num_inputs = c_api.TF_OperationNumInputs(self._c_op)
     input_types = [
         dtypes.as_dtype(c_api.TF_OperationInputType(self._tf_input(i)))
-        for i in xrange(num_inputs)
+        for i in range(num_inputs)
     ]
     return input_types
 
diff --git a/tensorflow/python/framework/tensor_util.py b/tensorflow/python/framework/tensor_util.py
index 78807080616..9b520b94d84 100644
--- a/tensorflow/python/framework/tensor_util.py
+++ b/tensorflow/python/framework/tensor_util.py
@@ -103,9 +103,9 @@ if _FAST_TENSOR_UTIL_AVAILABLE:
           fast_tensor_util.AppendComplex64ArrayToTensorProto,
       np.complex128:
           fast_tensor_util.AppendComplex128ArrayToTensorProto,
-      np.object:
+      object:
           fast_tensor_util.AppendObjectArrayToTensorProto,
-      np.bool:
+      bool:
           fast_tensor_util.AppendBoolArrayToTensorProto,
       dtypes.qint8.as_numpy_dtype:
           fast_tensor_util.AppendInt8ArrayToTensorProto,
@@ -171,8 +171,8 @@ else:
       np.int16: SlowAppendIntArrayToTensorProto,
       np.complex64: SlowAppendComplex64ArrayToTensorProto,
       np.complex128: SlowAppendComplex128ArrayToTensorProto,
-      np.object: SlowAppendObjectArrayToTensorProto,
-      np.bool: SlowAppendBoolArrayToTensorProto,
+      object: SlowAppendObjectArrayToTensorProto,
+      bool: SlowAppendBoolArrayToTensorProto,
       dtypes.qint8.as_numpy_dtype: SlowAppendQIntArrayToTensorProto,
       dtypes.quint8.as_numpy_dtype: SlowAppendQIntArrayToTensorProto,
       dtypes.qint16.as_numpy_dtype: SlowAppendQIntArrayToTensorProto,
diff --git a/tensorflow/python/keras/datasets/cifar.py b/tensorflow/python/keras/datasets/cifar.py
index 02344897f77..eefa8a6065b 100644
--- a/tensorflow/python/keras/datasets/cifar.py
+++ b/tensorflow/python/keras/datasets/cifar.py
@@ -20,7 +20,7 @@ from __future__ import print_function
 
 import sys
 
-from six.moves import cPickle
+import pickle
 
 
 def load_batch(fpath, label_key='labels'):
@@ -36,9 +36,9 @@ def load_batch(fpath, label_key='labels'):
   """
   with open(fpath, 'rb') as f:
     if sys.version_info < (3,):
-      d = cPickle.load(f)
+      d = pickle.load(f)
     else:
-      d = cPickle.load(f, encoding='bytes')
+      d = pickle.load(f, encoding='bytes')
       # decode utf8
       d_decoded = {}
       for k, v in d.items():
diff --git a/tensorflow/python/keras/engine/base_layer.py b/tensorflow/python/keras/engine/base_layer.py
index 83e01dfa23e..15837d45cfb 100644
--- a/tensorflow/python/keras/engine/base_layer.py
+++ b/tensorflow/python/keras/engine/base_layer.py
@@ -25,7 +25,6 @@ import json
 import threading
 
 import numpy as np
-from six.moves import zip  # pylint: disable=redefined-builtin
 
 from google.protobuf import json_format
 from tensorflow.core.framework import node_def_pb2
diff --git a/tensorflow/python/keras/engine/input_spec.py b/tensorflow/python/keras/engine/input_spec.py
index d790af4798f..318f0a01749 100644
--- a/tensorflow/python/keras/engine/input_spec.py
+++ b/tensorflow/python/keras/engine/input_spec.py
@@ -18,7 +18,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import zip  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import tensor_shape
diff --git a/tensorflow/python/keras/engine/network.py b/tensorflow/python/keras/engine/network.py
index 49a97ea9137..e11aa6290c0 100644
--- a/tensorflow/python/keras/engine/network.py
+++ b/tensorflow/python/keras/engine/network.py
@@ -27,7 +27,6 @@ import os
 import threading
 
 import numpy as np
-from six.moves import zip  # pylint: disable=redefined-builtin
 
 from tensorflow.python import pywrap_tensorflow
 from tensorflow.python.eager import context
diff --git a/tensorflow/python/keras/engine/training_utils.py b/tensorflow/python/keras/engine/training_utils.py
index b45bcbc5b3d..baad59bdaf4 100644
--- a/tensorflow/python/keras/engine/training_utils.py
+++ b/tensorflow/python/keras/engine/training_utils.py
@@ -27,7 +27,6 @@ import time
 
 import numpy as np
 import six
-from six.moves import zip  # pylint: disable=redefined-builtin
 
 from tensorflow.python import tf2
 from tensorflow.python.data.experimental.ops import cardinality
diff --git a/tensorflow/python/keras/optimizers.py b/tensorflow/python/keras/optimizers.py
index 10ce8336abc..5c2b4ccf955 100644
--- a/tensorflow/python/keras/optimizers.py
+++ b/tensorflow/python/keras/optimizers.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import six
-from six.moves import zip  # pylint: disable=redefined-builtin
 
 from tensorflow.python.distribute import distribution_strategy_context
 from tensorflow.python.framework import ops
diff --git a/tensorflow/python/keras/saving/hdf5_format.py b/tensorflow/python/keras/saving/hdf5_format.py
index d23db5b7763..9e4b74f1df4 100644
--- a/tensorflow/python/keras/saving/hdf5_format.py
+++ b/tensorflow/python/keras/saving/hdf5_format.py
@@ -23,7 +23,6 @@ import json
 import os
 
 import numpy as np
-from six.moves import zip  # pylint: disable=redefined-builtin
 
 from tensorflow.python.keras import backend as K
 from tensorflow.python.keras import optimizers
diff --git a/tensorflow/python/keras/utils/conv_utils.py b/tensorflow/python/keras/utils/conv_utils.py
index 1d6256ea3d8..35116b3511f 100644
--- a/tensorflow/python/keras/utils/conv_utils.py
+++ b/tensorflow/python/keras/utils/conv_utils.py
@@ -19,7 +19,6 @@ from __future__ import print_function
 
 import itertools
 import numpy as np
-from six.moves import range  # pylint: disable=redefined-builtin
 
 from tensorflow.python.keras import backend
 
@@ -416,7 +415,7 @@ def conv_connected_inputs(input_shape, kernel_shape, output_position, strides,
         >>> padding = "valid"
         >>> conv_connected_inputs(input_shape, kernel_shape, output_position,
         >>>                       strides, padding)
-        [xrange(1, 3), xrange(1, 2)]
+        [range(1, 3), range(1, 2)]
     ```
   Args:
     input_shape: tuple of size N: `(d_in1, ..., d_inN)`, spatial shape of the
diff --git a/tensorflow/python/keras/utils/data_utils.py b/tensorflow/python/keras/utils/data_utils.py
index 98b4929acde..0f4f11c51e0 100644
--- a/tensorflow/python/keras/utils/data_utils.py
+++ b/tensorflow/python/keras/utils/data_utils.py
@@ -37,9 +37,9 @@ import zipfile
 
 import numpy as np
 import six
-from six.moves.urllib.error import HTTPError
-from six.moves.urllib.error import URLError
-from six.moves.urllib.request import urlopen
+from urllib.error import HTTPError
+from urllib.error import URLError
+from urllib.request import urlopen
 
 from tensorflow.python.keras.utils.generic_utils import Progbar
 from tensorflow.python.util import tf_inspect
@@ -93,7 +93,7 @@ if sys.version_info[0] == 2:
       for chunk in chunk_read(response, reporthook=reporthook):
         fd.write(chunk)
 else:
-  from six.moves.urllib.request import urlretrieve
+  from urllib.request import urlretrieve
 
 
 def is_generator_or_sequence(x):
diff --git a/tensorflow/python/kernel_tests/basic_gpu_test.py b/tensorflow/python/kernel_tests/basic_gpu_test.py
index 789f6e90c9f..1ec1d08fff3 100644
--- a/tensorflow/python/kernel_tests/basic_gpu_test.py
+++ b/tensorflow/python/kernel_tests/basic_gpu_test.py
@@ -22,7 +22,6 @@ import itertools
 import threading
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
@@ -251,7 +250,7 @@ class GpuMultiSessionMemoryTest(test_util.TensorFlowTestCase):
         x4 = math_ops.matmul(array_ops.transpose(x3), x3, name='x4')
         s.run(variables.global_variables_initializer())
 
-        for _ in xrange(n_iterations):
+        for _ in range(n_iterations):
           value = s.run(x4)
           results.add(value.flat[0])
           if len(results) != 1:
@@ -262,7 +261,7 @@ class GpuMultiSessionMemoryTest(test_util.TensorFlowTestCase):
     n_threads = 4
     threads = []
     results = []
-    for _ in xrange(n_threads):
+    for _ in range(n_threads):
       session = self.session(graph=ops.Graph(), use_gpu=True)
       results.append(set())
       args = (session, results[-1])
diff --git a/tensorflow/python/kernel_tests/cholesky_op_test.py b/tensorflow/python/kernel_tests/cholesky_op_test.py
index e17a029c5ff..a65873a9e89 100644
--- a/tensorflow/python/kernel_tests/cholesky_op_test.py
+++ b/tensorflow/python/kernel_tests/cholesky_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.client import session
 from tensorflow.python.framework import constant_op
@@ -130,13 +129,13 @@ class CholeskyOpTest(test.TestCase):
 
     # Generate random positive-definite matrices.
     matrices = np.random.rand(10, 5, 5)
-    for i in xrange(10):
+    for i in range(10):
       matrices[i] = np.dot(matrices[i].T, matrices[i])
     self._verifyCholesky(matrices)
 
     # Generate random complex valued positive-definite matrices.
     matrices = np.random.rand(10, 5, 5) + 1j * np.random.rand(10, 5, 5)
-    for i in xrange(10):
+    for i in range(10):
       matrices[i] = np.dot(matrices[i].T.conj(), matrices[i])
     self._verifyCholesky(matrices)
 
diff --git a/tensorflow/python/kernel_tests/confusion_matrix_test.py b/tensorflow/python/kernel_tests/confusion_matrix_test.py
index cd6bd29e0de..3f29642f5de 100644
--- a/tensorflow/python/kernel_tests/confusion_matrix_test.py
+++ b/tensorflow/python/kernel_tests/confusion_matrix_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import errors_impl
@@ -106,7 +105,7 @@ class ConfusionMatrixTest(test.TestCase):
       d, l, cm_out = sess.run([data, lab, cm], {m_neg: 0.0, m_pos: 1.0, s: 1.0})
 
       truth = np.zeros([2, 2], dtype=np_dtype)
-      for i in xrange(len(d)):
+      for i in range(len(d)):
         truth[l[i], d[i]] += 1
 
       self.assertEqual(cm_out.dtype, np_dtype)
diff --git a/tensorflow/python/kernel_tests/control_flow_ops_py_test.py b/tensorflow/python/kernel_tests/control_flow_ops_py_test.py
index 4b420e60b98..2abe3c227f9 100644
--- a/tensorflow/python/kernel_tests/control_flow_ops_py_test.py
+++ b/tensorflow/python/kernel_tests/control_flow_ops_py_test.py
@@ -28,7 +28,6 @@ import time
 
 from absl.testing import parameterized
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.python.client import device_lib
@@ -2728,7 +2727,7 @@ class ControlFlowTest(test.TestCase, parameterized.TestCase):
 
       r = control_flow_ops.while_loop(c, b, [i], parallel_iterations=1)
       self.assertEqual([10], self.evaluate(r))
-      for i in xrange(10):
+      for i in range(10):
         self.assertEqual([i], self.evaluate(q.dequeue()))
 
   @test_util.run_v1_only("b/120545219")
@@ -4886,7 +4885,7 @@ class WhileOpBenchmark(test.Benchmark):
       variables.global_variables_initializer().run()
 
       if static_unroll:
-        for _ in xrange(steps):
+        for _ in range(steps):
           i, x = loop_body(i, x)
       else:
         i, x = control_flow_ops.while_loop(
@@ -4900,12 +4899,12 @@ class WhileOpBenchmark(test.Benchmark):
       # Use group to avoid fetching back results.
       r = control_flow_ops.group(dx, dk)
 
-      for _ in xrange(3):
+      for _ in range(3):
         # exclude warm up time
         self.evaluate(r)
 
       start_time = time.time()
-      for _ in xrange(num_iters):
+      for _ in range(num_iters):
         self.evaluate(r)
       return (time.time() - start_time) / num_iters
 
diff --git a/tensorflow/python/kernel_tests/conv1d_test.py b/tensorflow/python/kernel_tests/conv1d_test.py
index 5ac8c11130b..55bccc83757 100644
--- a/tensorflow/python/kernel_tests/conv1d_test.py
+++ b/tensorflow/python/kernel_tests/conv1d_test.py
@@ -18,7 +18,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -79,9 +78,9 @@ class Conv1DTest(test.TestCase):
       # The amount of padding added
       pad = 1
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[1]):
-          for w in xrange(pad, y_shape[1] - pad):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[1]):
+          for w in range(pad, y_shape[1] - pad):
             target = 3.0
             # We add a case for locations divisible by the stride.
             w_in = w % stride == 0 and w > pad and w < y_shape[1] - 1 - pad
diff --git a/tensorflow/python/kernel_tests/conv1d_transpose_test.py b/tensorflow/python/kernel_tests/conv1d_transpose_test.py
index 02ac5af7aae..2100665cbb5 100644
--- a/tensorflow/python/kernel_tests/conv1d_transpose_test.py
+++ b/tensorflow/python/kernel_tests/conv1d_transpose_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -51,9 +50,9 @@ class Conv1DTransposeTest(test.TestCase):
           x, f, y_shape, strides=strides, padding="SAME")
       value = self.evaluate(output)
 
-      for n in xrange(y_shape[0]):
-        for w in xrange(y_shape[1]):
-          for c in xrange(y_shape[2]):
+      for n in range(y_shape[0]):
+        for w in range(y_shape[1]):
+          for c in range(y_shape[2]):
             target = 2 * 3.0
             w_in = w > 0 and w < y_shape[1] - 1
             if w_in:
@@ -79,9 +78,9 @@ class Conv1DTransposeTest(test.TestCase):
           x, f, y_shape, strides=strides, padding="SAME")
       value = self.evaluate(output)
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[1]):
-          for w in xrange(y_shape[1]):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[1]):
+          for w in range(y_shape[1]):
             target = 3.0
             # We add a case for locations divisible by the stride.
             w_in = w % strides[1] == 0 and w > 0 and w < y_shape[1] - 1
@@ -113,9 +112,9 @@ class Conv1DTransposeTest(test.TestCase):
       # The amount of padding added
       pad = 1
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[1]):
-          for w in xrange(pad, y_shape[1] - pad):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[1]):
+          for w in range(pad, y_shape[1] - pad):
             target = 3.0
             # We add a case for locations divisible by the stride.
             w_in = w % strides[1] == 0 and w > pad and w < y_shape[1] - 1 - pad
@@ -172,9 +171,9 @@ class Conv1DTransposeTest(test.TestCase):
             x, f, y_shape, strides=strides, padding="SAME", data_format="NCW")
 
         value = self.evaluate(output)
-        for n in xrange(x_shape[0]):
-          for k in xrange(f_shape[1]):
-            for w in xrange(y_shape[2]):
+        for n in range(x_shape[0]):
+          for k in range(f_shape[1]):
+            for w in range(y_shape[2]):
               target = 2 * 3.0
               w_in = w > 0 and w < y_shape[2] - 1
               if w_in:
@@ -203,9 +202,9 @@ class Conv1DTransposeTest(test.TestCase):
             x, f, y_shape, strides=strides, padding="SAME", data_format="NCW")
 
         value = self.evaluate(output)
-        for n in xrange(x_shape[0]):
-          for k in xrange(f_shape[1]):
-            for w in xrange(y_shape[2]):
+        for n in range(x_shape[0]):
+          for k in range(f_shape[1]):
+            for w in range(y_shape[2]):
               target = 3.0
               # We add a case for locations divisible by the stride.
               w_in = w % strides[2] == 0 and w > 0 and w < y_shape[2] - 1
@@ -237,9 +236,9 @@ class Conv1DTransposeTest(test.TestCase):
         cache_values = np.zeros(y_shape, dtype=np.float32)
         # The amount of padding added
         pad = 1
-        for n in xrange(x_shape[0]):
-          for k in xrange(f_shape[1]):
-            for w in xrange(pad, y_shape[2] - pad):
+        for n in range(x_shape[0]):
+          for k in range(f_shape[1]):
+            for w in range(pad, y_shape[2] - pad):
               target = 3.0
               # We add a case for locations divisible by the stride.
               w_in = w % strides[2] == 0 and w > pad and \
diff --git a/tensorflow/python/kernel_tests/conv2d_transpose_test.py b/tensorflow/python/kernel_tests/conv2d_transpose_test.py
index c603c086306..695dc30e854 100644
--- a/tensorflow/python/kernel_tests/conv2d_transpose_test.py
+++ b/tensorflow/python/kernel_tests/conv2d_transpose_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -61,10 +60,10 @@ class Conv2DTransposeTest(test.TestCase):
       # At the borders, #cells=ceil(kernel_height/2)*kernel_width or
       #                        kernel_height * ceil(kernel_width/2)
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[2]):
-          for w in xrange(y_shape[2]):
-            for h in xrange(y_shape[1]):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[2]):
+          for w in range(y_shape[2]):
+            for h in range(y_shape[1]):
               target = 4 * 3.0
               h_in = h > 0 and h < y_shape[1] - 1
               w_in = w > 0 and w < y_shape[2] - 1
@@ -93,10 +92,10 @@ class Conv2DTransposeTest(test.TestCase):
           x, f, y_shape, strides=strides, padding="SAME")
       value = self.evaluate(output)
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[2]):
-          for w in xrange(y_shape[2]):
-            for h in xrange(y_shape[1]):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[2]):
+          for w in range(y_shape[2]):
+            for h in range(y_shape[1]):
               target = 3.0
               # We add a case for locations divisible by the stride.
               h_in = h % strides[1] == 0 and h > 0 and h < y_shape[1] - 1
@@ -131,10 +130,10 @@ class Conv2DTransposeTest(test.TestCase):
       # The amount of padding added
       pad = 1
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[2]):
-          for w in xrange(pad, y_shape[2] - pad):
-            for h in xrange(pad, y_shape[1] - pad):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[2]):
+          for w in range(pad, y_shape[2] - pad):
+            for h in range(pad, y_shape[1] - pad):
               target = 3.0
               # We add a case for locations divisible by the stride.
               h_in = h % strides[1] == 0 and h > pad and h < y_shape[
@@ -197,10 +196,10 @@ class Conv2DTransposeTest(test.TestCase):
             x, f, y_shape, strides=strides, padding="SAME", data_format="NCHW")
 
         value = self.evaluate(output)
-        for n in xrange(x_shape[0]):
-          for k in xrange(f_shape[2]):
-            for w in xrange(y_shape[3]):
-              for h in xrange(y_shape[2]):
+        for n in range(x_shape[0]):
+          for k in range(f_shape[2]):
+            for w in range(y_shape[3]):
+              for h in range(y_shape[2]):
                 target = 4 * 3.0
                 h_in = h > 0 and h < y_shape[2] - 1
                 w_in = w > 0 and w < y_shape[3] - 1
@@ -232,10 +231,10 @@ class Conv2DTransposeTest(test.TestCase):
             x, f, y_shape, strides=strides, padding="SAME", data_format="NCHW")
 
         value = self.evaluate(output)
-        for n in xrange(x_shape[0]):
-          for k in xrange(f_shape[2]):
-            for w in xrange(y_shape[3]):
-              for h in xrange(y_shape[2]):
+        for n in range(x_shape[0]):
+          for k in range(f_shape[2]):
+            for w in range(y_shape[3]):
+              for h in range(y_shape[2]):
                 target = 3.0
                 # We add a case for locations divisible by the stride.
                 h_in = h % strides[2] == 0 and h > 0 and h < y_shape[2] - 1
@@ -270,10 +269,10 @@ class Conv2DTransposeTest(test.TestCase):
         cache_values = np.zeros(y_shape, dtype=np.float32)
         # The amount of padding added
         pad = 1
-        for n in xrange(x_shape[0]):
-          for k in xrange(f_shape[2]):
-            for w in xrange(pad, y_shape[3] - pad):
-              for h in xrange(pad, y_shape[2] - pad):
+        for n in range(x_shape[0]):
+          for k in range(f_shape[2]):
+            for w in range(pad, y_shape[3] - pad):
+              for h in range(pad, y_shape[2] - pad):
                 target = 3.0
                 # We add a case for locations divisible by the stride.
                 h_in = h % strides[2] == 0 and h > pad and h < y_shape[
diff --git a/tensorflow/python/kernel_tests/conv3d_transpose_test.py b/tensorflow/python/kernel_tests/conv3d_transpose_test.py
index 22ba5b90375..7cca69facbd 100644
--- a/tensorflow/python/kernel_tests/conv3d_transpose_test.py
+++ b/tensorflow/python/kernel_tests/conv3d_transpose_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -64,11 +63,11 @@ class Conv3DTransposeTest(test.TestCase):
       #   kernel_depth * ceil(kernel_height/2) * kernel_width or
       #   kernel_depth * kernel_height * ceil(kernel_width/2)
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[3]):
-          for w in xrange(y_shape[3]):
-            for h in xrange(y_shape[2]):
-              for d in xrange(y_shape[1]):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[3]):
+          for w in range(y_shape[3]):
+            for h in range(y_shape[2]):
+              for d in range(y_shape[1]):
                 d_in = d > 0 and d < y_shape[1] - 1
                 h_in = h > 0 and h < y_shape[2] - 1
                 w_in = w > 0 and w < y_shape[3] - 1
@@ -101,11 +100,11 @@ class Conv3DTransposeTest(test.TestCase):
           x, f, y_shape, strides=strides, padding="SAME")
       value = self.evaluate(output)
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[3]):
-          for w in xrange(y_shape[3]):
-            for h in xrange(y_shape[2]):
-              for d in xrange(y_shape[1]):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[3]):
+          for w in range(y_shape[3]):
+            for h in range(y_shape[2]):
+              for d in range(y_shape[1]):
                 # We add a case for locations divisible by the stride.
                 d_in = d % strides[1] == 0 and 0 < d < y_shape[1] - 1
                 h_in = h % strides[2] == 0 and 0 < h < y_shape[2] - 1
@@ -174,11 +173,11 @@ class Conv3DTransposeTest(test.TestCase):
       # The amount of padding added
       pad = 1
 
-      for n in xrange(x_shape[0]):
-        for k in xrange(f_shape[3]):
-          for w in xrange(y_shape[3]):
-            for h in xrange(y_shape[2]):
-              for d in xrange(y_shape[1]):
+      for n in range(x_shape[0]):
+        for k in range(f_shape[3]):
+          for w in range(y_shape[3]):
+            for h in range(y_shape[2]):
+              for d in range(y_shape[1]):
                 # We add a case for locations divisible by the stride.
                 d_in = d % strides[1] == 0 and pad < d < y_shape[1] - 1 - pad
                 h_in = h % strides[2] == 0 and pad < h < y_shape[2] - 1 - pad
diff --git a/tensorflow/python/kernel_tests/conv_ops_test.py b/tensorflow/python/kernel_tests/conv_ops_test.py
index b51d8ae61aa..446caa89428 100644
--- a/tensorflow/python/kernel_tests/conv_ops_test.py
+++ b/tensorflow/python/kernel_tests/conv_ops_test.py
@@ -23,7 +23,6 @@ import time
 
 import numpy as np
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.core.protobuf import rewriter_config_pb2
 from tensorflow.python.client import session as session_lib
@@ -2803,7 +2802,7 @@ class Conv2DBenchmark(test.Benchmark):
 
       variables.global_variables_initializer().run()
       num_iterations = 4
-      for iter_index in xrange(num_iterations):
+      for iter_index in range(num_iterations):
         start = time.time()
         session.run(outputs)
         wall_time = time.time() - start
diff --git a/tensorflow/python/kernel_tests/ctc_decoder_ops_test.py b/tensorflow/python/kernel_tests/ctc_decoder_ops_test.py
index 0d86d13c715..0450d6c026f 100644
--- a/tensorflow/python/kernel_tests/ctc_decoder_ops_test.py
+++ b/tensorflow/python/kernel_tests/ctc_decoder_ops_test.py
@@ -21,7 +21,7 @@ from __future__ import print_function
 import itertools
 
 import numpy as np
-from six.moves import zip_longest
+from itertools import zip_longest
 
 from tensorflow.python.framework import errors
 from tensorflow.python.framework import ops
diff --git a/tensorflow/python/kernel_tests/decode_jpeg_op_test.py b/tensorflow/python/kernel_tests/decode_jpeg_op_test.py
index f8fc28062f4..d5468547cc7 100644
--- a/tensorflow/python/kernel_tests/decode_jpeg_op_test.py
+++ b/tensorflow/python/kernel_tests/decode_jpeg_op_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import os
 import time
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 from tensorflow.python.client import session
 from tensorflow.python.framework import ops
 from tensorflow.python.ops import array_ops
@@ -82,7 +81,7 @@ class DecodeJpegBenchmark(test.Benchmark):
     with session.Session() as sess:
       self.evaluate(variables.global_variables_initializer())
       images = []
-      for _ in xrange(parallelism):
+      for _ in range(parallelism):
         if crop_window is None:
           # No crop.
           image = image_ops.decode_jpeg(image_content, channels=3)
@@ -103,12 +102,12 @@ class DecodeJpegBenchmark(test.Benchmark):
         images.append(image)
       r = control_flow_ops.group(*images)
 
-      for _ in xrange(3):
+      for _ in range(3):
         # Skip warm up time.
         self.evaluate(r)
 
       start_time = time.time()
-      for _ in xrange(num_iters):
+      for _ in range(num_iters):
         self.evaluate(r)
       end_time = time.time()
     return end_time - start_time
diff --git a/tensorflow/python/kernel_tests/dynamic_partition_op_test.py b/tensorflow/python/kernel_tests/dynamic_partition_op_test.py
index 8c448194076..8eae27152c5 100644
--- a/tensorflow/python/kernel_tests/dynamic_partition_op_test.py
+++ b/tensorflow/python/kernel_tests/dynamic_partition_op_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import unittest
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -319,8 +318,8 @@ class DynamicPartitionTest(test.TestCase):
       data = np.zeros(shape + (5,))
       partitions = data_flow_ops.dynamic_partition(
           data, indices, num_partitions=7)
-      for i in xrange(2):
-        for j in xrange(3):
+      for i in range(2):
+        for j in range(3):
           bad = np.zeros(shape, dtype=np.int32)
           bad[i, j] = 17
           with self.assertRaisesOpError(
diff --git a/tensorflow/python/kernel_tests/embedding_ops_test.py b/tensorflow/python/kernel_tests/embedding_ops_test.py
index fec6c310341..26efa60ba40 100644
--- a/tensorflow/python/kernel_tests/embedding_ops_test.py
+++ b/tensorflow/python/kernel_tests/embedding_ops_test.py
@@ -22,7 +22,6 @@ import itertools
 import math
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -561,7 +560,7 @@ class EmbeddingLookupTest(test.TestCase):
           for procs in 1, 2, 3:
             stride = procs * math_ops.range(params.shape[0] // procs)
             split_params = [
-                array_ops.gather(params, stride + p) for p in xrange(procs)
+                array_ops.gather(params, stride + p) for p in range(procs)
             ]
             sharded = embedding_ops.embedding_lookup(split_params, ids).eval()
             self.assertAllEqual(simple, sharded)
@@ -595,7 +594,7 @@ class EmbeddingLookupTest(test.TestCase):
           for procs in 1, 2, 3:
             stride = procs * math_ops.range(params.shape[0] // procs)
             split_params = [
-                array_ops.gather(params, stride + p) for p in xrange(procs)
+                array_ops.gather(params, stride + p) for p in range(procs)
             ]
             sharded = embedding_ops.embedding_lookup(
                 split_params, ids, max_norm=1.0).eval()
@@ -631,7 +630,7 @@ class EmbeddingLookupTest(test.TestCase):
         for procs in 1, 2, 3:
           stride = procs * math_ops.range(params.shape[0] // procs)
           split_params = [
-              array_ops.gather(params, stride + p) for p in xrange(procs)
+              array_ops.gather(params, stride + p) for p in range(procs)
           ]
           sharded = embedding_ops._embedding_lookup_and_transform(
               split_params, ids, max_norm=l2_norm,
diff --git a/tensorflow/python/kernel_tests/fifo_queue_test.py b/tensorflow/python/kernel_tests/fifo_queue_test.py
index 6ffff8bf350..765b9f79553 100644
--- a/tensorflow/python/kernel_tests/fifo_queue_test.py
+++ b/tensorflow/python/kernel_tests/fifo_queue_test.py
@@ -22,7 +22,6 @@ import random
 import time
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.python.client import session as session_lib
@@ -176,7 +175,7 @@ class FIFOQueueTest(test.TestCase):
 
       # Dequeue every element using a single thread.
       results = []
-      for _ in xrange(len(elems)):
+      for _ in range(len(elems)):
         results.append(dequeued_t.eval())
       self.assertItemsEqual(elems, results)
 
@@ -217,7 +216,7 @@ class FIFOQueueTest(test.TestCase):
       for enqueue_op in enqueue_ops:
         enqueue_op.run()
 
-      for i in xrange(len(elems)):
+      for i in range(len(elems)):
         vals = self.evaluate(dequeued_t)
         self.assertEqual([elems[i]], vals)
 
@@ -231,7 +230,7 @@ class FIFOQueueTest(test.TestCase):
       for enqueue_op in enqueue_ops:
         enqueue_op.run()
 
-      for i in xrange(len(elems)):
+      for i in range(len(elems)):
         vals = self.evaluate(dequeued_t)
         self.assertEqual([elems[i]], vals)
 
@@ -255,7 +254,7 @@ class FIFOQueueTest(test.TestCase):
       results = []
 
       def dequeue():
-        for _ in xrange(len(elems)):
+        for _ in range(len(elems)):
           results.append(self.evaluate(dequeued_t))
 
       enqueue_thread = self.checkedThread(target=enqueue)
@@ -278,7 +277,7 @@ class FIFOQueueTest(test.TestCase):
       for enqueue_op in enqueue_ops:
         enqueue_op.run()
 
-      for i in xrange(len(elems)):
+      for i in range(len(elems)):
         x_val, y_val = self.evaluate(dequeued_t)
         x, y = elems[i]
         self.assertEqual([x], x_val)
@@ -617,11 +616,11 @@ class FIFOQueueTest(test.TestCase):
       dequeued_t = q.dequeue()
 
       def enqueue():
-        for _ in xrange(100):
+        for _ in range(100):
           self.evaluate(enqueue_op)
 
       def dequeue():
-        for _ in xrange(100):
+        for _ in range(100):
           self.assertTrue(self.evaluate(dequeued_t) in (10.0, 20.0))
 
       enqueue_threads = [self.checkedThread(target=enqueue) for _ in range(10)]
@@ -653,7 +652,7 @@ class FIFOQueueTest(test.TestCase):
       close_op = q.close()
 
       def dequeue():
-        for i in xrange(250):
+        for i in range(250):
           self.assertEqual(i, self.evaluate(dequeued_t))
 
       dequeue_thread = self.checkedThread(target=dequeue)
@@ -1305,10 +1304,10 @@ class FIFOQueueTest(test.TestCase):
     with self.cached_session():
       num_queues = 10
       qlist = []
-      for _ in xrange(num_queues):
+      for _ in range(num_queues):
         qlist.append(data_flow_ops.FIFOQueue(10, dtypes_lib.float32))
       # Enqueue/Dequeue into a dynamically selected queue
-      for _ in xrange(20):
+      for _ in range(20):
         index = np.random.randint(num_queues)
         q = data_flow_ops.FIFOQueue.from_list(index, qlist)
         q.enqueue((10.,)).run()
diff --git a/tensorflow/python/kernel_tests/init_ops_test.py b/tensorflow/python/kernel_tests/init_ops_test.py
index 3822b4b89fc..b26d75420cb 100644
--- a/tensorflow/python/kernel_tests/init_ops_test.py
+++ b/tensorflow/python/kernel_tests/init_ops_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -202,7 +201,7 @@ class ConstantInitializersTest(test.TestCase):
 
       actual = array_ops.reshape(x, [-1]).eval()
       self.assertGreater(len(actual), len(expected))
-      for i in xrange(len(actual)):
+      for i in range(len(actual)):
         a = actual[i]
         e = expected[i] if i < len(expected) else expected[-1]
         self.assertEqual(a, e)
diff --git a/tensorflow/python/kernel_tests/inplace_ops_test.py b/tensorflow/python/kernel_tests/inplace_ops_test.py
index 9eaaac7a248..dd3f1b39c50 100644
--- a/tensorflow/python/kernel_tests/inplace_ops_test.py
+++ b/tensorflow/python/kernel_tests/inplace_ops_test.py
@@ -18,7 +18,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import errors
@@ -113,7 +112,7 @@ class InplaceOpsTest(test_util.TensorFlowTestCase):
       d0, d1, d2 = 100, 3, 5
       x = array_ops.zeros([d0, d1, d2])
       y = np.zeros([d0, d1, d2])
-      for _ in xrange(20):
+      for _ in range(20):
         idx = np.random.choice(d0, d0 // 10, replace=False)
         val = np.random.randint(10, size=(d0 // 10, d1, d2))
         op = np.random.randint(3)
@@ -134,7 +133,7 @@ class InplaceOpsTest(test_util.TensorFlowTestCase):
       d0 = 100
       x = array_ops.zeros([d0])
       y = np.zeros([d0])
-      for _ in xrange(20):
+      for _ in range(20):
         idx = np.random.choice(d0, d0 // 10, replace=False)
         val = np.random.randint(10, size=(d0 // 10))
         op = np.random.randint(3)
diff --git a/tensorflow/python/kernel_tests/listdiff_op_test.py b/tensorflow/python/kernel_tests/listdiff_op_test.py
index 28657107980..dff8dfed4b6 100644
--- a/tensorflow/python/kernel_tests/listdiff_op_test.py
+++ b/tensorflow/python/kernel_tests/listdiff_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
@@ -86,7 +85,7 @@ class ListDiffTest(test.TestCase):
     int_low = -7
     int_high = 8
     max_size = 50
-    for _ in xrange(num_random_tests):
+    for _ in range(num_random_tests):
       x_size = np.random.randint(max_size + 1)
       x = np.random.randint(int_low, int_high, size=x_size)
       y_size = np.random.randint(max_size + 1)
diff --git a/tensorflow/python/kernel_tests/metrics_test.py b/tensorflow/python/kernel_tests/metrics_test.py
index 2cf6e629339..72f3320e270 100644
--- a/tensorflow/python/kernel_tests/metrics_test.py
+++ b/tensorflow/python/kernel_tests/metrics_test.py
@@ -22,7 +22,6 @@ import functools
 import math
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes as dtypes_lib
@@ -643,7 +642,7 @@ class AccuracyTest(test.TestCase):
       accuracy, update_op = metrics.accuracy(labels, predictions)
 
       self.evaluate(variables.local_variables_initializer())
-      for _ in xrange(3):
+      for _ in range(3):
         self.evaluate(update_op)
       self.assertEqual(0.5, self.evaluate(update_op))
       self.assertEqual(0.5, accuracy.eval())
@@ -743,7 +742,7 @@ class AccuracyTest(test.TestCase):
       accuracy, update_op = metrics.accuracy(labels, predictions, weights)
 
       self.evaluate(variables.local_variables_initializer())
-      for _ in xrange(3):
+      for _ in range(3):
         self.evaluate(update_op)
       self.assertEqual(1.0, self.evaluate(update_op))
       self.assertEqual(1.0, accuracy.eval())
@@ -2176,7 +2175,7 @@ class MultiLabelPrecisionAtKTest(test.TestCase):
     precision_ex1 = (0.0 / 1, 1.0 / 2, 1.0 / 3, 2.0 / 4)
     avg_precision_ex1 = (0.0 / 1, precision_ex1[1] / 2, precision_ex1[1] / 3,
                          (precision_ex1[1] + precision_ex1[3]) / 4)
-    for i in xrange(4):
+    for i in range(4):
       k = i + 1
       self._test_precision_at_k(
           predictions, labels, k, expected=precision_ex1[i])
@@ -2194,7 +2193,7 @@ class MultiLabelPrecisionAtKTest(test.TestCase):
     precision_ex2 = (0.0 / 1, 0.0 / 2, 1.0 / 3, 2.0 / 4)
     avg_precision_ex2 = (0.0 / 1, 0.0 / 2, precision_ex2[2] / 3,
                          (precision_ex2[2] + precision_ex2[3]) / 4)
-    for i in xrange(4):
+    for i in range(4):
       k = i + 1
       self._test_precision_at_k(
           predictions, labels, k, expected=precision_ex2[i])
@@ -2213,7 +2212,7 @@ class MultiLabelPrecisionAtKTest(test.TestCase):
         (ex1 + ex2) / 2
         for ex1, ex2 in zip(avg_precision_ex1, avg_precision_ex2)
     ]
-    for i in xrange(4):
+    for i in range(4):
       k = i + 1
       predictions_idx = (predictions_idx_ex1[:k], predictions_idx_ex2[:k])
       self._test_precision_at_k(
@@ -2230,7 +2229,7 @@ class MultiLabelPrecisionAtKTest(test.TestCase):
         (weights[0] * ex1 + weights[1] * ex2) / (weights[0] + weights[1])
         for ex1, ex2 in zip(avg_precision_ex1, avg_precision_ex2)
     ]
-    for i in xrange(4):
+    for i in range(4):
       k = i + 1
       self._test_average_precision_at_k(
           predictions,
@@ -2250,7 +2249,7 @@ class MultiLabelPrecisionAtKTest(test.TestCase):
     precision_ex1 = (0.0 / 1, 1.0 / 2, 1.0 / 3, 2.0 / 4)
     avg_precision_ex1 = (0.0 / 1, precision_ex1[1] / 2, precision_ex1[1] / 3,
                          (precision_ex1[1] + precision_ex1[3]) / 4)
-    for i in xrange(4):
+    for i in range(4):
       k = i + 1
       self._test_precision_at_k(
           predictions, labels, k, expected=precision_ex1[i])
@@ -2275,7 +2274,7 @@ class MultiLabelPrecisionAtKTest(test.TestCase):
         (1.0 / 1, 1.0 / 1, 1.0 / 1, 1.0 / 1))
     mean_avg_precision_ex1 = np.mean(avg_precision_ex1, axis=0)
     for labels in (sparse_labels, dense_labels):
-      for i in xrange(4):
+      for i in range(4):
         k = i + 1
         self._test_precision_at_k(
             predictions, labels, k, expected=mean_precision_ex1[i])
@@ -2881,7 +2880,7 @@ class MultiLabel3dRecallAtKTest(test.TestCase):
 
   @test_util.run_deprecated_v1
   def test_3d_ignore_all(self):
-    for class_id in xrange(10):
+    for class_id in range(10):
       self._test_recall_at_k(
           self._predictions, self._labels, k=5, expected=NAN, class_id=class_id,
           weights=[[0], [0]])
diff --git a/tensorflow/python/kernel_tests/padding_fifo_queue_test.py b/tensorflow/python/kernel_tests/padding_fifo_queue_test.py
index 1825bebea8a..63716c70d4b 100644
--- a/tensorflow/python/kernel_tests/padding_fifo_queue_test.py
+++ b/tensorflow/python/kernel_tests/padding_fifo_queue_test.py
@@ -22,7 +22,6 @@ import random
 import time
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes as dtypes_lib
@@ -144,7 +143,7 @@ class PaddingFIFOQueueTest(test.TestCase):
 
       # Dequeue every element using a single thread.
       results = []
-      for _ in xrange(len(elems)):
+      for _ in range(len(elems)):
         results.append(dequeued_t.eval())
       self.assertItemsEqual(elems, results)
 
@@ -185,7 +184,7 @@ class PaddingFIFOQueueTest(test.TestCase):
       for enqueue_op in enqueue_ops:
         enqueue_op.run()
 
-      for i in xrange(len(elems)):
+      for i in range(len(elems)):
         vals = self.evaluate(dequeued_t)
         self.assertEqual([elems[i]], vals)
 
@@ -209,7 +208,7 @@ class PaddingFIFOQueueTest(test.TestCase):
       results = []
 
       def dequeue():
-        for _ in xrange(len(elems)):
+        for _ in range(len(elems)):
           results.append(self.evaluate(dequeued_t))
 
       enqueue_thread = self.checkedThread(target=enqueue)
@@ -234,7 +233,7 @@ class PaddingFIFOQueueTest(test.TestCase):
       for enqueue_op in enqueue_ops:
         enqueue_op.run()
 
-      for i in xrange(len(elems)):
+      for i in range(len(elems)):
         x_val, y_val = self.evaluate(dequeued_t)
         x, y = elems[i]
         self.assertEqual([x], x_val)
@@ -722,11 +721,11 @@ class PaddingFIFOQueueTest(test.TestCase):
       dequeued_t = q.dequeue()
 
       def enqueue():
-        for _ in xrange(100):
+        for _ in range(100):
           self.evaluate(enqueue_op)
 
       def dequeue():
-        for _ in xrange(100):
+        for _ in range(100):
           self.assertTrue(self.evaluate(dequeued_t) in (10.0, 20.0))
 
       enqueue_threads = [self.checkedThread(target=enqueue) for _ in range(10)]
@@ -758,7 +757,7 @@ class PaddingFIFOQueueTest(test.TestCase):
       close_op = q.close()
 
       def dequeue():
-        for i in xrange(250):
+        for i in range(250):
           self.assertEqual(i, self.evaluate(dequeued_t))
 
       dequeue_thread = self.checkedThread(target=dequeue)
@@ -1421,11 +1420,11 @@ class PaddingFIFOQueueTest(test.TestCase):
     with self.cached_session():
       num_queues = 10
       qlist = []
-      for _ in xrange(num_queues):
+      for _ in range(num_queues):
         qlist.append(
             data_flow_ops.PaddingFIFOQueue(10, dtypes_lib.float32, ((),)))
       # Enqueue/Dequeue into a dynamically selected queue
-      for _ in xrange(20):
+      for _ in range(20):
         index = np.random.randint(num_queues)
         q = data_flow_ops.PaddingFIFOQueue.from_list(index, qlist)
         q.enqueue((10.,)).run()
diff --git a/tensorflow/python/kernel_tests/parameterized_truncated_normal_op_test.py b/tensorflow/python/kernel_tests/parameterized_truncated_normal_op_test.py
index ac8ad7a2bd4..be4bdca2203 100644
--- a/tensorflow/python/kernel_tests/parameterized_truncated_normal_op_test.py
+++ b/tensorflow/python/kernel_tests/parameterized_truncated_normal_op_test.py
@@ -23,7 +23,6 @@ import math
 import timeit
 
 import numpy as np
-from six.moves import range  # pylint: disable=redefined-builtin
 
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.python.client import session
diff --git a/tensorflow/python/kernel_tests/partitioned_variables_test.py b/tensorflow/python/kernel_tests/partitioned_variables_test.py
index edcbc2967e2..52d88fe9cb5 100644
--- a/tensorflow/python/kernel_tests/partitioned_variables_test.py
+++ b/tensorflow/python/kernel_tests/partitioned_variables_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import os
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -320,7 +319,7 @@ class PartitionedVariablesTestCase(test.TestCase):
 
   def _TestSaveSpec(self, slices, expected_specs):
     self.assertEqual(len(expected_specs), len(slices))
-    for i in xrange(len(expected_specs)):
+    for i in range(len(expected_specs)):
       self.assertEquals(expected_specs[i], slices[i]._save_slice_info.spec)
 
   def testVecConstantInit(self):
@@ -433,7 +432,7 @@ class PartitionedVariablesTestCase(test.TestCase):
       var_lists = [
           partitioned_variables.create_partitioned_variables(
               rnd.get_shape(), [1, i], rnd.initialized_value())
-          for i in xrange(1, 10)
+          for i in range(1, 10)
       ]
       self.evaluate(variables.global_variables_initializer())
       rnd_val = self.evaluate(rnd)
diff --git a/tensorflow/python/kernel_tests/py_func_test.py b/tensorflow/python/kernel_tests/py_func_test.py
index a3cdb258b29..b045c10c32c 100644
--- a/tensorflow/python/kernel_tests/py_func_test.py
+++ b/tensorflow/python/kernel_tests/py_func_test.py
@@ -24,7 +24,6 @@ import re
 
 import numpy as np
 from six.moves import queue
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.client import session as session_lib
 from tensorflow.python.eager import backprop
@@ -250,7 +249,7 @@ class PyFuncTest(test.TestCase):
       x = array_ops.zeros([1000000], dtype=np.float32)
       y = script_ops.py_func(lambda x: x + 1, [x], [dtypes.float32])
       z = script_ops.py_func(lambda x: x * 2, [x], [dtypes.float32])
-      for _ in xrange(100):
+      for _ in range(100):
         sess.run([y[0].op, z[0].op])
 
   def testNoInput(self):
@@ -469,7 +468,7 @@ class PyFuncTest(test.TestCase):
     initial_size = script_ops._py_funcs.size()
     # Encapsulate the graph generation, so locals can be deleted.
     def make_graphs():
-      for _ in xrange(1000):
+      for _ in range(1000):
         g = ops.Graph()
         with g.as_default():
           c = constant_op.constant([1.], dtypes.float32)
diff --git a/tensorflow/python/kernel_tests/random/random_gamma_test.py b/tensorflow/python/kernel_tests/random/random_gamma_test.py
index 77e194d5fcf..bb7df678936 100644
--- a/tensorflow/python/kernel_tests/random/random_gamma_test.py
+++ b/tensorflow/python/kernel_tests/random/random_gamma_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
@@ -47,7 +46,7 @@ class RandomGammaTest(test.TestCase):
         rng = random_ops.random_gamma(
             [num], alpha, beta=beta, dtype=dtype, seed=seed)
         ret = np.empty([10, num])
-        for i in xrange(10):
+        for i in range(10):
           ret[i, :] = self.evaluate(rng)
       return ret
 
diff --git a/tensorflow/python/kernel_tests/random/random_ops_test.py b/tensorflow/python/kernel_tests/random/random_ops_test.py
index 1023b8f7901..47b544048b6 100644
--- a/tensorflow/python/kernel_tests/random/random_ops_test.py
+++ b/tensorflow/python/kernel_tests/random/random_ops_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.eager import context
 from tensorflow.python.framework import dtypes
@@ -69,7 +68,7 @@ class RandomNormalTest(RandomOpTestCommon):
         rng = random_ops.random_normal(
             [num], mean=mu, stddev=sigma, dtype=dtype, seed=seed)
         ret = np.empty([10, num])
-        for i in xrange(10):
+        for i in range(10):
           ret[i, :] = self.evaluate(rng)
       return ret
 
@@ -166,7 +165,7 @@ class TruncatedNormalTest(test.TestCase):
         rng = random_ops.truncated_normal(
             [num], mean=mu, stddev=sigma, dtype=dtype, seed=seed)
         ret = np.empty([10, num])
-        for i in xrange(10):
+        for i in range(10):
           ret[i, :] = self.evaluate(rng)
       return ret
 
@@ -268,7 +267,7 @@ class RandomUniformTest(RandomOpTestCommon):
         rng = random_ops.random_uniform(
             [num], minval=minv, maxval=maxv, dtype=dtype, seed=seed)
         ret = np.empty([10, num])
-        for i in xrange(10):
+        for i in range(10):
           ret[i, :] = self.evaluate(rng)
       return ret
 
diff --git a/tensorflow/python/kernel_tests/random/random_poisson_test.py b/tensorflow/python/kernel_tests/random/random_poisson_test.py
index 51dd4cb47ca..9c47b586406 100644
--- a/tensorflow/python/kernel_tests/random/random_poisson_test.py
+++ b/tensorflow/python/kernel_tests/random/random_poisson_test.py
@@ -18,7 +18,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -44,7 +43,7 @@ class RandomPoissonTest(test.TestCase):
       with self.session(use_gpu=use_gpu, graph=ops.Graph()) as sess:
         rng = random_ops.random_poisson(lam, [num], dtype=dtype, seed=seed)
         ret = np.empty([10, num])
-        for i in xrange(10):
+        for i in range(10):
           ret[i, :] = self.evaluate(rng)
       return ret
 
diff --git a/tensorflow/python/kernel_tests/random/random_shuffle_queue_test.py b/tensorflow/python/kernel_tests/random/random_shuffle_queue_test.py
index ba7ee16e7c2..f62049064a3 100644
--- a/tensorflow/python/kernel_tests/random/random_shuffle_queue_test.py
+++ b/tensorflow/python/kernel_tests/random/random_shuffle_queue_test.py
@@ -22,7 +22,6 @@ import random
 import time
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes as dtypes_lib
@@ -120,7 +119,7 @@ class RandomShuffleQueueTest(test.TestCase):
 
       # Dequeue every element using a single thread.
       results = []
-      for _ in xrange(len(elems)):
+      for _ in range(len(elems)):
         results.append(dequeued_t.eval())
       self.assertItemsEqual(elems, results)
 
@@ -158,7 +157,7 @@ class RandomShuffleQueueTest(test.TestCase):
       for enqueue_op in enqueue_ops:
         enqueue_op.run()
 
-      vals = [dequeued_t.eval() for _ in xrange(len(elems))]
+      vals = [dequeued_t.eval() for _ in range(len(elems))]
       self.assertItemsEqual(elems, vals)
 
   def testEnqueueAndBlockingDequeue(self):
@@ -178,7 +177,7 @@ class RandomShuffleQueueTest(test.TestCase):
       results = []
 
       def dequeue():
-        for _ in xrange(len(elems)):
+        for _ in range(len(elems)):
           results.append(self.evaluate(dequeued_t))
 
       enqueue_thread = self.checkedThread(target=enqueue)
@@ -202,7 +201,7 @@ class RandomShuffleQueueTest(test.TestCase):
         enqueue_op.run()
 
       results = []
-      for _ in xrange(len(elems)):
+      for _ in range(len(elems)):
         x, y = self.evaluate(dequeued_t)
         results.append((x, y))
       self.assertItemsEqual(elems, results)
@@ -506,12 +505,12 @@ class RandomShuffleQueueTest(test.TestCase):
 
   def testParallelDequeueUpToRandomPartition(self):
     with self.cached_session() as sess:
-      dequeue_sizes = [random.randint(50, 150) for _ in xrange(10)]
+      dequeue_sizes = [random.randint(50, 150) for _ in range(10)]
       total_elements = sum(dequeue_sizes)
       q = data_flow_ops.RandomShuffleQueue(
           total_elements, 0, dtypes_lib.float32, shapes=())
 
-      elems = [10.0 * x for x in xrange(total_elements)]
+      elems = [10.0 * x for x in range(total_elements)]
       enqueue_op = q.enqueue_many((elems,))
       dequeue_ops = [q.dequeue_up_to(size) for size in dequeue_sizes]
 
@@ -1202,11 +1201,11 @@ class RandomShuffleQueueTest(test.TestCase):
     with self.cached_session():
       num_queues = 10
       qlist = []
-      for _ in xrange(num_queues):
+      for _ in range(num_queues):
         qlist.append(
             data_flow_ops.RandomShuffleQueue(10, 0, dtypes_lib.float32))
       # Enqueue/Dequeue into a dynamically selected queue
-      for _ in xrange(20):
+      for _ in range(20):
         index = np.random.randint(num_queues)
         q = data_flow_ops.RandomShuffleQueue.from_list(index, qlist)
         q.enqueue((10.,)).run()
diff --git a/tensorflow/python/kernel_tests/reduce_benchmark_test.py b/tensorflow/python/kernel_tests/reduce_benchmark_test.py
index ef9c4c350fd..9f184802878 100644
--- a/tensorflow/python/kernel_tests/reduce_benchmark_test.py
+++ b/tensorflow/python/kernel_tests/reduce_benchmark_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import time
 
 import numpy as np
-from six.moves import range  # pylint: disable=redefined-builtin
 
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.python.client import session
diff --git a/tensorflow/python/kernel_tests/reduce_join_op_test.py b/tensorflow/python/kernel_tests/reduce_join_op_test.py
index 751e3e3648b..aa2937e4016 100644
--- a/tensorflow/python/kernel_tests/reduce_join_op_test.py
+++ b/tensorflow/python/kernel_tests/reduce_join_op_test.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import itertools
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -41,7 +40,7 @@ def _input_array(num_dims):
     An ndarray of shape [2] * num_dims.
   """
   formatter = "{:0%db}" % num_dims
-  strings = [formatter.format(i) for i in xrange(2**num_dims)]
+  strings = [formatter.format(i) for i in range(2**num_dims)]
   return np.array(strings, dtype="S%d" % num_dims).reshape([2] * num_dims)
 
 
@@ -58,10 +57,10 @@ def _joined_array(num_dims, reduce_dim):
   formatter = "{:0%db}" % (num_dims - 1)
   result = np.zeros(shape=[2] * (num_dims - 1), dtype="S%d" % (2 * num_dims))
   flat = result.ravel()
-  for i in xrange(2**(num_dims - 1)):
+  for i in range(2**(num_dims - 1)):
     dims = formatter.format(i)
     flat[i] = "".join([(dims[:reduce_dim] + "%d" + dims[reduce_dim:]) % j
-                       for j in xrange(2)])
+                       for j in range(2)])
   return result
 
 
@@ -78,7 +77,7 @@ class ReduceJoinTestHelperTest(UnicodeTestCase):
 
   def testInputArray(self):
     num_dims = 3
-    truth = ["{:03b}".format(i) for i in xrange(2**num_dims)]
+    truth = ["{:03b}".format(i) for i in range(2**num_dims)]
     output_array = _input_array(num_dims).reshape([-1])
     self.assertAllEqualUnicode(truth, output_array)
 
@@ -188,26 +187,26 @@ class ReduceJoinTest(UnicodeTestCase):
 
   def testRankFive(self):
     input_array = _input_array(num_dims=5)
-    truths = [_joined_array(num_dims=5, reduce_dim=i) for i in xrange(5)]
+    truths = [_joined_array(num_dims=5, reduce_dim=i) for i in range(5)]
     truth_shape = [2] * 4
-    for i in xrange(5):
+    for i in range(5):
       self._testReduceJoin(input_array, truths[i], truth_shape, axis=i)
 
   def testNegative(self):
     input_array = _input_array(num_dims=5)
-    truths = [_joined_array(num_dims=5, reduce_dim=i) for i in xrange(5)]
+    truths = [_joined_array(num_dims=5, reduce_dim=i) for i in range(5)]
     truth_shape = [2] * 4
-    for i in xrange(5):
+    for i in range(5):
       self._testReduceJoin(input_array, truths[i], truth_shape, axis=i - 5)
 
   def testSingletonDimension(self):
     input_arrays = [
         _input_array(num_dims=5).reshape([2] * i + [1] + [2] * (5 - i))
-        for i in xrange(6)
+        for i in range(6)
     ]
     truth = _input_array(num_dims=5)
     truth_shape = [2] * 5
-    for i in xrange(6):
+    for i in range(6):
       self._testReduceJoin(input_arrays[i], truth, truth_shape, axis=i)
 
   def testSeparator(self):
@@ -296,8 +295,8 @@ class ReduceJoinTest(UnicodeTestCase):
     num_dims = 3
     input_array = _input_array(num_dims=num_dims)
     # Also tests [].
-    for i in xrange(num_dims + 1):
-      for permutation in itertools.permutations(xrange(num_dims), i):
+    for i in range(num_dims + 1):
+      for permutation in itertools.permutations(range(num_dims), i):
         self._testMultipleReduceJoin(input_array, axis=permutation)
 
   @test_util.run_deprecated_v1
diff --git a/tensorflow/python/kernel_tests/relu_op_test.py b/tensorflow/python/kernel_tests/relu_op_test.py
index eed8bd7d258..b9f6161fb4a 100644
--- a/tensorflow/python/kernel_tests/relu_op_test.py
+++ b/tensorflow/python/kernel_tests/relu_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python import tf2
 from tensorflow.python.compat import compat
@@ -159,7 +158,7 @@ class ReluTest(test.TestCase):
 
     # Repeat the experiment for 100 times. All tensor shapes and its tensor
     # values are randomly generated for each run.
-    for _ in xrange(100):
+    for _ in range(100):
       d32, d16 = f()
       self.assertAllClose(d32, d16, atol=3e-4)
 
diff --git a/tensorflow/python/kernel_tests/reverse_sequence_op_test.py b/tensorflow/python/kernel_tests/reverse_sequence_op_test.py
index 05307c9834a..04c25e0a3ef 100644
--- a/tensorflow/python/kernel_tests/reverse_sequence_op_test.py
+++ b/tensorflow/python/kernel_tests/reverse_sequence_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -183,7 +182,7 @@ class ReverseSequenceTest(test.TestCase):
       with self.assertRaisesOpError("batch_dim == seq_dim"):
         output.eval(feed_dict={
             inputs: np.random.rand(32, 2, 3),
-            seq_lengths: xrange(32)
+            seq_lengths: range(32)
         })
 
 
diff --git a/tensorflow/python/kernel_tests/rnn_test.py b/tensorflow/python/kernel_tests/rnn_test.py
index bdb01700582..4d04a8e823d 100644
--- a/tensorflow/python/kernel_tests/rnn_test.py
+++ b/tensorflow/python/kernel_tests/rnn_test.py
@@ -24,7 +24,6 @@ import timeit
 
 import numpy as np
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.python import keras
 from tensorflow.python.client import session
@@ -1199,7 +1198,7 @@ class BenchmarkRNN(test.Benchmark):
       rnn_long_sequence_benchmark(batch_size, seqlen, num_units,
                                   dynamic, swap_memory, 2)
     # Measure the performance.
-    for slen in xrange(100, 1100, 100):
+    for slen in range(100, 1100, 100):
       rnn_long_sequence_benchmark(batch_size, slen, num_units, dynamic,
                                   swap_memory, 3)
 
diff --git a/tensorflow/python/kernel_tests/signal/fft_ops_test.py b/tensorflow/python/kernel_tests/signal/fft_ops_test.py
index b5858e3ac08..f11a8d46bc9 100644
--- a/tensorflow/python/kernel_tests/signal/fft_ops_test.py
+++ b/tensorflow/python/kernel_tests/signal/fft_ops_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.protobuf import config_pb2
 from tensorflow.python.framework import dtypes
@@ -163,7 +162,7 @@ class FFTOpsTest(BaseFFTOpsTest):
     with spectral_ops_test_util.fft_kernel_label_map():
       for np_type in (np.complex64, np.complex128):
         for rank in VALID_FFT_RANKS:
-          for dims in xrange(rank, rank + 3):
+          for dims in range(rank, rank + 3):
             x = np.zeros((0,) * dims).astype(np_type)
             self.assertEqual(x.shape, self._tfFFT(x, rank).shape)
             self.assertEqual(x.shape, self._tfIFFT(x, rank).shape)
@@ -173,7 +172,7 @@ class FFTOpsTest(BaseFFTOpsTest):
     with spectral_ops_test_util.fft_kernel_label_map():
       for np_type, tol in ((np.complex64, 1e-4), (np.complex128, 1e-8)):
         for rank in VALID_FFT_RANKS:
-          for dims in xrange(rank, rank + 3):
+          for dims in range(rank, rank + 3):
             self._compare(
                 np.mod(np.arange(np.power(4, dims)), 10).reshape(
                     (4,) * dims).astype(np_type), rank, rtol=tol, atol=tol)
@@ -181,7 +180,7 @@ class FFTOpsTest(BaseFFTOpsTest):
   def testLargeBatch(self):
     if test.is_gpu_available(cuda_only=True):
       rank = 1
-      for dims in xrange(rank, rank + 3):
+      for dims in range(rank, rank + 3):
         for np_type, tol in ((np.complex64, 1e-4), (np.complex128, 1e-5)):
           self._compare(
               np.mod(np.arange(np.power(128, dims)), 10).reshape(
@@ -192,7 +191,7 @@ class FFTOpsTest(BaseFFTOpsTest):
   # def testLargeBatchMemoryFail(self):
   #   if test.is_gpu_available(cuda_only=True):
   #     rank = 1
-  #     for dims in xrange(rank, rank + 3):
+  #     for dims in range(rank, rank + 3):
   #       self._checkMemoryFail(
   #           np.mod(np.arange(np.power(128, dims)), 64).reshape(
   #               (128,) * dims).astype(np.complex64), rank)
@@ -202,7 +201,7 @@ class FFTOpsTest(BaseFFTOpsTest):
     with spectral_ops_test_util.fft_kernel_label_map():
       for np_type, tol in ((np.complex64, 1e-4), (np.complex128, 1e-8)):
         for rank in VALID_FFT_RANKS:
-          for dims in xrange(rank, rank + 3):
+          for dims in range(rank, rank + 3):
             self._compare(
                 np.mod(np.arange(np.power(4, dims)), 10).reshape(
                     (4,) * dims).astype(np_type),
@@ -219,7 +218,7 @@ class FFTOpsTest(BaseFFTOpsTest):
           return (re + im * 1j).reshape(shape)
 
         for rank in VALID_FFT_RANKS:
-          for dims in xrange(rank, rank + 3):
+          for dims in range(rank, rank + 3):
             self._compare(gen((4,) * dims).astype(np_type), rank,
                           rtol=tol, atol=tol)
 
@@ -249,7 +248,7 @@ class FFTOpsTest(BaseFFTOpsTest):
   @test_util.run_deprecated_v1
   def testError(self):
     for rank in VALID_FFT_RANKS:
-      for dims in xrange(0, rank):
+      for dims in range(0, rank):
         x = np.zeros((1,) * dims).astype(np.complex64)
         with self.assertRaisesWithPredicateMatch(
             ValueError, "Shape must be .*rank {}.*".format(rank)):
@@ -263,7 +262,7 @@ class FFTOpsTest(BaseFFTOpsTest):
     with spectral_ops_test_util.fft_kernel_label_map():
       for np_type, tol in ((np.float32, 1e-4), (np.float64, 1e-10)):
         for rank in VALID_FFT_RANKS:
-          for dims in xrange(rank, rank + 2):
+          for dims in range(rank, rank + 2):
             re = np.ones(shape=(4,) * dims, dtype=np_type) / 10.0
             im = np.zeros(shape=(4,) * dims, dtype=np_type)
             self._checkGradComplex(self._tfFFTForRank(rank), re, im,
@@ -276,7 +275,7 @@ class FFTOpsTest(BaseFFTOpsTest):
     with spectral_ops_test_util.fft_kernel_label_map():
       for np_type, tol in ((np.float32, 1e-2), (np.float64, 1e-10)):
         for rank in VALID_FFT_RANKS:
-          for dims in xrange(rank, rank + 2):
+          for dims in range(rank, rank + 2):
             re = np.random.rand(*((3,) * dims)).astype(np_type) * 2 - 1
             im = np.random.rand(*((3,) * dims)).astype(np_type) * 2 - 1
             self._checkGradComplex(self._tfFFTForRank(rank), re, im,
@@ -345,7 +344,7 @@ class RFFTOpsTest(BaseFFTOpsTest):
   def testEmpty(self):
     with spectral_ops_test_util.fft_kernel_label_map():
       for rank in VALID_FFT_RANKS:
-        for dims in xrange(rank, rank + 3):
+        for dims in range(rank, rank + 3):
           x = np.zeros((0,) * dims).astype(np.float32)
           self.assertEqual(x.shape, self._tfFFT(x, rank).shape)
           x = np.zeros((0,) * dims).astype(np.complex64)
@@ -355,7 +354,7 @@ class RFFTOpsTest(BaseFFTOpsTest):
   def testBasic(self):
     with spectral_ops_test_util.fft_kernel_label_map():
       for rank in VALID_FFT_RANKS:
-        for dims in xrange(rank, rank + 3):
+        for dims in range(rank, rank + 3):
           for size in (5, 6):
             inner_dim = size // 2 + 1
             r2c = np.mod(np.arange(np.power(size, dims)), 10).reshape(
@@ -369,7 +368,7 @@ class RFFTOpsTest(BaseFFTOpsTest):
   def testLargeBatch(self):
     if test.is_gpu_available(cuda_only=True):
       rank = 1
-      for dims in xrange(rank, rank + 3):
+      for dims in range(rank, rank + 3):
         for size in (64, 128):
           inner_dim = size // 2 + 1
           r2c = np.mod(np.arange(np.power(size, dims)), 10).reshape(
@@ -383,7 +382,7 @@ class RFFTOpsTest(BaseFFTOpsTest):
   def testBasicPlaceholder(self):
     with spectral_ops_test_util.fft_kernel_label_map():
       for rank in VALID_FFT_RANKS:
-        for dims in xrange(rank, rank + 3):
+        for dims in range(rank, rank + 3):
           for size in (5, 6):
             inner_dim = size // 2 + 1
             r2c = np.mod(np.arange(np.power(size, dims)), 10).reshape(
@@ -404,7 +403,7 @@ class RFFTOpsTest(BaseFFTOpsTest):
     if test.is_gpu_available(cuda_only=True):
       with spectral_ops_test_util.fft_kernel_label_map():
         for rank in VALID_FFT_RANKS:
-          for dims in xrange(rank, rank + 3):
+          for dims in range(rank, rank + 3):
             for size in (5, 6):
               inner_dim = size // 2 + 1
               r2c = np.mod(np.arange(np.power(size, dims)), 10).reshape(
@@ -459,7 +458,7 @@ class RFFTOpsTest(BaseFFTOpsTest):
         return ret
 
       for rank in VALID_FFT_RANKS:
-        for dims in xrange(rank, rank + 3):
+        for dims in range(rank, rank + 3):
           for size in (5, 6):
             inner_dim = size // 2 + 1
             self._compareForward(gen_real((size,) * dims), rank, (size,) * rank)
@@ -471,7 +470,7 @@ class RFFTOpsTest(BaseFFTOpsTest):
   def testError(self):
     with spectral_ops_test_util.fft_kernel_label_map():
       for rank in VALID_FFT_RANKS:
-        for dims in xrange(0, rank):
+        for dims in range(0, rank):
           x = np.zeros((1,) * dims).astype(np.complex64)
           with self.assertRaisesWithPredicateMatch(
               ValueError, "Shape .* must have rank at least {}".format(rank)):
@@ -479,7 +478,7 @@ class RFFTOpsTest(BaseFFTOpsTest):
           with self.assertRaisesWithPredicateMatch(
               ValueError, "Shape .* must have rank at least {}".format(rank)):
             self._tfIFFT(x, rank)
-        for dims in xrange(rank, rank + 2):
+        for dims in range(rank, rank + 2):
           x = np.zeros((1,) * rank)
 
           # Test non-rank-1 fft_length produces an error.
@@ -531,7 +530,7 @@ class RFFTOpsTest(BaseFFTOpsTest):
         # rfft3d/irfft3d do not have gradients yet.
         if rank == 3:
           continue
-        for dims in xrange(rank, rank + 2):
+        for dims in range(rank, rank + 2):
           for size in (5, 6):
             re = np.ones(shape=(size,) * dims, dtype=np.float32)
             im = -np.ones(shape=(size,) * dims, dtype=np.float32)
@@ -546,7 +545,7 @@ class RFFTOpsTest(BaseFFTOpsTest):
         # rfft3d/irfft3d do not have gradients yet.
         if rank == 3:
           continue
-        for dims in xrange(rank, rank + 2):
+        for dims in range(rank, rank + 2):
           for size in (5, 6):
             re = np.random.rand(*((size,) * dims)).astype(np.float32) * 2 - 1
             im = np.random.rand(*((size,) * dims)).astype(np.float32) * 2 - 1
diff --git a/tensorflow/python/kernel_tests/slice_op_test.py b/tensorflow/python/kernel_tests/slice_op_test.py
index 39d88896416..bb2995cf3cb 100644
--- a/tensorflow/python/kernel_tests/slice_op_test.py
+++ b/tensorflow/python/kernel_tests/slice_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -36,7 +35,7 @@ class SliceTest(test.TestCase):
 
   def testEmpty(self):
     inp = np.random.rand(4, 4).astype("f")
-    for k in xrange(4):
+    for k in range(4):
       with self.cached_session(use_gpu=True):
         a = constant_op.constant(inp, shape=[4, 4], dtype=dtypes.float32)
         slice_t = a[2, k:k]
@@ -45,7 +44,7 @@ class SliceTest(test.TestCase):
 
   def testInt32(self):
     inp = np.random.rand(4, 4).astype("i")
-    for k in xrange(4):
+    for k in range(4):
       with self.cached_session(use_gpu=True):
         a = constant_op.constant(inp, shape=[4, 4], dtype=dtypes.int32)
         slice_t = a[2, k:k]
@@ -325,7 +324,7 @@ class SliceTest(test.TestCase):
     # the grads into the right location to compare against TensorFlow.
     np_ans = np.zeros(input_shape)
     slices = []
-    for i in xrange(len(input_shape)):
+    for i in range(len(input_shape)):
       slices.append(slice(slice_begin[i], slice_begin[i] + slice_size[i]))
     np_ans[slices] = grads
 
diff --git a/tensorflow/python/kernel_tests/summary_v1_audio_op_test.py b/tensorflow/python/kernel_tests/summary_v1_audio_op_test.py
index 1547c55f8b0..c9096a489df 100644
--- a/tensorflow/python/kernel_tests/summary_v1_audio_op_test.py
+++ b/tensorflow/python/kernel_tests/summary_v1_audio_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.framework import summary_pb2
 from tensorflow.python.framework import ops
@@ -44,7 +43,7 @@ class SummaryV1AudioOpTest(test.TestCase):
           tag: "snd/audio/%d"
           audio { content_type: "audio/wav" sample_rate: %d
                   num_channels: %d length_frames: %d }
-        }""" % (i, sample_rate, num_channels, length_frames) for i in xrange(3))
+        }""" % (i, sample_rate, num_channels, length_frames) for i in range(3))
     self.assertProtoEquals(expected, audio_summ)
 
   def testAudioSummary(self):
diff --git a/tensorflow/python/kernel_tests/summary_v1_image_op_test.py b/tensorflow/python/kernel_tests/summary_v1_image_op_test.py
index 56de2e933db..b1d88ad95a6 100644
--- a/tensorflow/python/kernel_tests/summary_v1_image_op_test.py
+++ b/tensorflow/python/kernel_tests/summary_v1_image_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.framework import summary_pb2
 from tensorflow.python.framework import dtypes
@@ -47,7 +46,7 @@ class SummaryV1ImageOpTest(test.TestCase):
         value {
           tag: "img/image/%d"
           image { height: %d width: %d colorspace: %d }
-        }""" % ((i,) + shape[1:]) for i in xrange(3))
+        }""" % ((i,) + shape[1:]) for i in range(3))
     self.assertProtoEquals(expected, image_summ)
 
   @test_util.run_deprecated_v1
diff --git a/tensorflow/python/kernel_tests/unstack_op_test.py b/tensorflow/python/kernel_tests/unstack_op_test.py
index 89885cf752b..12446696701 100644
--- a/tensorflow/python/kernel_tests/unstack_op_test.py
+++ b/tensorflow/python/kernel_tests/unstack_op_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import test_util
@@ -82,7 +81,7 @@ class UnstackOpTest(test.TestCase):
     for shape in (2,), (3,), (2, 3), (3, 2), (4, 3, 2):
       data = np.random.randn(*shape)
       shapes = [shape[1:]] * shape[0]
-      for i in xrange(shape[0]):
+      for i in range(shape[0]):
         with self.cached_session():
           x = constant_op.constant(data)
           cs = array_ops.unstack(x, num=shape[0])
@@ -96,7 +95,7 @@ class UnstackOpTest(test.TestCase):
       data = np.random.randn(*shape)
       out_shape = list(shape)
       del out_shape[1]
-      for i in xrange(shape[1]):
+      for i in range(shape[1]):
         with self.cached_session():
           x = constant_op.constant(data)
           cs = array_ops.unstack(x, num=shape[1], axis=1)
diff --git a/tensorflow/python/lib/core/bfloat16.cc b/tensorflow/python/lib/core/bfloat16.cc
index fde3a837702..83097bfc605 100644
--- a/tensorflow/python/lib/core/bfloat16.cc
+++ b/tensorflow/python/lib/core/bfloat16.cc
@@ -316,48 +316,48 @@ PyTypeObject PyBfloat16_Type = {
     "bfloat16",                                // tp_name
     sizeof(PyBfloat16),                        // tp_basicsize
     0,                                         // tp_itemsize
-    nullptr,                                   // tp_dealloc
-    nullptr,                                   // tp_print
-    nullptr,                                   // tp_getattr
-    nullptr,                                   // tp_setattr
-    nullptr,                                   // tp_compare / tp_reserved
+    0,                                   // tp_dealloc
+    0,                                   // tp_print
+    0,                                   // tp_getattr
+    0,                                   // tp_setattr
+    0,                                   // tp_compare / tp_reserved
     PyBfloat16_Repr,                           // tp_repr
     &PyBfloat16_AsNumber,                      // tp_as_number
-    nullptr,                                   // tp_as_sequence
-    nullptr,                                   // tp_as_mapping
+    0,                                   // tp_as_sequence
+    0,                                   // tp_as_mapping
     PyBfloat16_Hash,                           // tp_hash
-    nullptr,                                   // tp_call
+    0,                                   // tp_call
     PyBfloat16_Str,                            // tp_str
-    nullptr,                                   // tp_getattro
-    nullptr,                                   // tp_setattro
-    nullptr,                                   // tp_as_buffer
+    0,                                   // tp_getattro
+    0,                                   // tp_setattro
+    0,                                   // tp_as_buffer
     Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE,  // tp_flags
     "bfloat16 floating-point values",          // tp_doc
-    nullptr,                                   // tp_traverse
-    nullptr,                                   // tp_clear
+    0,                                   // tp_traverse
+    0,                                   // tp_clear
     PyBfloat16_RichCompare,                    // tp_richcompare
     0,                                         // tp_weaklistoffset
-    nullptr,                                   // tp_iter
-    nullptr,                                   // tp_iternext
-    nullptr,                                   // tp_methods
-    nullptr,                                   // tp_members
-    nullptr,                                   // tp_getset
-    nullptr,                                   // tp_base
-    nullptr,                                   // tp_dict
-    nullptr,                                   // tp_descr_get
-    nullptr,                                   // tp_descr_set
+    0,                                   // tp_iter
+    0,                                   // tp_iternext
+    0,                                   // tp_methods
+    0,                                   // tp_members
+    0,                                   // tp_getset
+    0,                                   // tp_base
+    0,                                   // tp_dict
+    0,                                   // tp_descr_get
+    0,                                   // tp_descr_set
     0,                                         // tp_dictoffset
-    nullptr,                                   // tp_init
-    nullptr,                                   // tp_alloc
+    0,                                   // tp_init
+    0,                                   // tp_alloc
     PyBfloat16_New,                            // tp_new
-    nullptr,                                   // tp_free
-    nullptr,                                   // tp_is_gc
-    nullptr,                                   // tp_bases
-    nullptr,                                   // tp_mro
-    nullptr,                                   // tp_cache
-    nullptr,                                   // tp_subclasses
-    nullptr,                                   // tp_weaklist
-    nullptr,                                   // tp_del
+    0,                                   // tp_free
+    0,                                   // tp_is_gc
+    0,                                   // tp_bases
+    0,                                   // tp_mro
+    0,                                   // tp_cache
+    0,                                   // tp_subclasses
+    0,                                   // tp_weaklist
+    0,                                   // tp_del
     0,                                         // tp_version_tag
 };
 
@@ -490,7 +490,7 @@ bool RegisterBfloat16Cast(int numpy_type, bool cast_is_safe) {
 }
 
 template <typename InType, typename OutType, typename Functor>
-void BinaryUFunc(char** args, npy_intp* dimensions, npy_intp* steps,
+void BinaryUFunc(char** args, npy_intp const* dimensions, npy_intp const* steps,
                  void* data) {
   const char* i0 = args[0];
   const char* i1 = args[1];
@@ -506,7 +506,7 @@ void BinaryUFunc(char** args, npy_intp* dimensions, npy_intp* steps,
 }
 
 template <typename Functor>
-void CompareUFunc(char** args, npy_intp* dimensions, npy_intp* steps,
+void CompareUFunc(char** args, npy_intp const* dimensions, npy_intp const* steps,
                   void* data) {
   BinaryUFunc<bfloat16, npy_bool, Functor>(args, dimensions, steps, data);
 }
@@ -560,7 +560,11 @@ bool Initialize() {
   NPyBfloat16_ArrFuncs.nonzero = NPyBfloat16_NonZero;
   NPyBfloat16_ArrFuncs.fill = NPyBfloat16_Fill;
 
+#if PY_MINOR_VERSION >= 9
+  Py_SET_TYPE(&NPyBfloat16_Descr, &PyArrayDescr_Type);
+#else
   Py_TYPE(&NPyBfloat16_Descr) = &PyArrayDescr_Type;
+#endif
   npy_bfloat16_ = PyArray_RegisterDataType(&NPyBfloat16_Descr);
   if (npy_bfloat16_ < 0) return false;
 
@@ -612,7 +616,7 @@ bool Initialize() {
       return false;
     }
     PyUFuncObject* ufunc = reinterpret_cast<PyUFuncObject*>(ufunc_obj.get());
-    if (types.size() != ufunc->nargs) {
+    if (types.size() != (size_t)ufunc->nargs) {
       PyErr_Format(PyExc_AssertionError,
                    "ufunc %s takes %d arguments, loop takes %lu", name,
                    ufunc->nargs, types.size());
diff --git a/tensorflow/python/lib/core/ndarray_tensor_bridge.cc b/tensorflow/python/lib/core/ndarray_tensor_bridge.cc
index 16f69341861..76f26fb0dc5 100644
--- a/tensorflow/python/lib/core/ndarray_tensor_bridge.cc
+++ b/tensorflow/python/lib/core/ndarray_tensor_bridge.cc
@@ -86,25 +86,25 @@ PyTypeObject TensorReleaserType = {
     0,                                /* tp_itemsize */
     /* methods */
     TensorReleaser_dealloc,      /* tp_dealloc */
-    nullptr,                     /* tp_print */
-    nullptr,                     /* tp_getattr */
-    nullptr,                     /* tp_setattr */
-    nullptr,                     /* tp_compare */
-    nullptr,                     /* tp_repr */
-    nullptr,                     /* tp_as_number */
-    nullptr,                     /* tp_as_sequence */
-    nullptr,                     /* tp_as_mapping */
-    nullptr,                     /* tp_hash */
-    nullptr,                     /* tp_call */
-    nullptr,                     /* tp_str */
-    nullptr,                     /* tp_getattro */
-    nullptr,                     /* tp_setattro */
-    nullptr,                     /* tp_as_buffer */
+    0,                     /* tp_print */
+    0,                     /* tp_getattr */
+    0,                     /* tp_setattr */
+    0,                     /* tp_compare */
+    0,                     /* tp_repr */
+    0,                     /* tp_as_number */
+    0,                     /* tp_as_sequence */
+    0,                     /* tp_as_mapping */
+    0,                     /* tp_hash */
+    0,                     /* tp_call */
+    0,                     /* tp_str */
+    0,                     /* tp_getattro */
+    0,                     /* tp_setattro */
+    0,                     /* tp_as_buffer */
     Py_TPFLAGS_DEFAULT,          /* tp_flags */
     "Wrapped TensorFlow Tensor", /* tp_doc */
-    nullptr,                     /* tp_traverse */
-    nullptr,                     /* tp_clear */
-    nullptr,                     /* tp_richcompare */
+    0,                     /* tp_traverse */
+    0,                     /* tp_clear */
+    0,                     /* tp_richcompare */
 };
 
 Status TF_DataType_to_PyArray_TYPE(TF_DataType tf_datatype,
diff --git a/tensorflow/python/lib/core/numpy.h b/tensorflow/python/lib/core/numpy.h
index 0098d938a08..1ecc9524703 100644
--- a/tensorflow/python/lib/core/numpy.h
+++ b/tensorflow/python/lib/core/numpy.h
@@ -32,6 +32,7 @@ limitations under the License.
 // Place `<locale>` before <Python.h> to avoid build failure in macOS.
 #include <locale>
 
+#define PY_SSIZE_T_CLEAN
 #include <Python.h>
 
 #include "numpy/arrayobject.h"
diff --git a/tensorflow/python/ops/accumulate_n_benchmark.py b/tensorflow/python/ops/accumulate_n_benchmark.py
index 08349003dc3..c67d0a1744f 100644
--- a/tensorflow/python/ops/accumulate_n_benchmark.py
+++ b/tensorflow/python/ops/accumulate_n_benchmark.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 import random
 import time
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.client import session
 from tensorflow.python.framework import ops
@@ -72,7 +71,7 @@ class AccumulateNBenchmark(test.Benchmark):
         validate_shape=True)
 
   def _GenerateUnorderedInputs(self, size, n):
-    inputs = [random_ops.random_uniform(shape=[size]) for _ in xrange(n)]
+    inputs = [random_ops.random_uniform(shape=[size]) for _ in range(n)]
     random.shuffle(inputs)
     return inputs
 
@@ -83,7 +82,7 @@ class AccumulateNBenchmark(test.Benchmark):
     inputs = self._GenerateUnorderedInputs(size, 1)
     queue = data_flow_ops.FIFOQueue(
         capacity=1, dtypes=[inputs[0].dtype], shapes=[inputs[0].get_shape()])
-    for _ in xrange(n - 1):
+    for _ in range(n - 1):
       op = queue.enqueue(inputs[-1])
       with ops.control_dependencies([op]):
         inputs.append(math_ops.tanh(1.0 + queue.dequeue()))
@@ -108,10 +107,10 @@ class AccumulateNBenchmark(test.Benchmark):
 
     with session.Session(graph=graph):
       for tag, op in test_ops:
-        for _ in xrange(100):
+        for _ in range(100):
           op.run()  # Run for warm up.
         start = time.time()
-        for _ in xrange(repeats):
+        for _ in range(repeats):
           op.run()
         duration = time.time() - start
         args = format_args + (tag, duration)
diff --git a/tensorflow/python/ops/control_flow_grad.py b/tensorflow/python/ops/control_flow_grad.py
index 6a551deb5ba..a5f4848bf7d 100644
--- a/tensorflow/python/ops/control_flow_grad.py
+++ b/tensorflow/python/ops/control_flow_grad.py
@@ -18,7 +18,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
@@ -131,10 +130,10 @@ def _MergeGrad(op, grad, _):
     # pylint: enable=protected-access
   else:
     num_inputs = len(op.inputs)
-    cond = [math_ops.equal(op.outputs[1], i) for i in xrange(num_inputs)]
+    cond = [math_ops.equal(op.outputs[1], i) for i in range(num_inputs)]
     # pylint: disable=protected-access
     return [control_flow_ops._SwitchRefOrTensor(grad, cond[i])[1]
-            for i in xrange(num_inputs)]
+            for i in range(num_inputs)]
     # pylint: enable=protected-access
 
 
diff --git a/tensorflow/python/ops/data_flow_grad.py b/tensorflow/python/ops/data_flow_grad.py
index 7e382af64f6..50edcd6fd4f 100644
--- a/tensorflow/python/ops/data_flow_grad.py
+++ b/tensorflow/python/ops/data_flow_grad.py
@@ -18,8 +18,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
-
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
 from tensorflow.python.ops import array_ops
@@ -55,7 +53,7 @@ def _DynamicStitchGrads(op, grad):
   def AsInt32(x):
     return (x if op.inputs[0].dtype == dtypes.int32 else
             math_ops.cast(x, dtypes.int32))
-  inputs = [AsInt32(op.inputs[i]) for i in xrange(num_values)]
+  inputs = [AsInt32(op.inputs[i]) for i in range(num_values)]
   if isinstance(grad, ops.IndexedSlices):
     output_shape = array_ops.shape(op.outputs[0])
     output_rows = output_shape[0]
diff --git a/tensorflow/python/ops/embedding_ops.py b/tensorflow/python/ops/embedding_ops.py
index 9c437d32b9e..01f9a4352ce 100644
--- a/tensorflow/python/ops/embedding_ops.py
+++ b/tensorflow/python/ops/embedding_ops.py
@@ -17,8 +17,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
-
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
@@ -158,14 +156,14 @@ def _embedding_lookup_and_transform(params,
         # if we already know the full shape statically.
         dim_0_size = tensor_shape.Dimension(
             tensor_shape.dimension_value(params[0].get_shape()[0]))
-        for p in xrange(1, np):
+        for p in range(1, np):
           dim_0_size += tensor_shape.Dimension(
               tensor_shape.dimension_value(params[p].get_shape()[0]))
         if dim_0_size.value:
           num_total_ids = constant_op.constant(dim_0_size.value, flat_ids.dtype)
         else:
           dim_0_sizes = []
-          for p in xrange(np):
+          for p in range(np):
             param_p_dim = tensor_shape.dimension_value(params[p].get_shape()[0])
             if param_p_dim is not None:
               dim_0_sizes.append(param_p_dim)
@@ -199,7 +197,7 @@ def _embedding_lookup_and_transform(params,
                                                  p_assignments, np)
       # Do np separate lookups, finding embeddings for plist[p] in params[p]
       partitioned_result = []
-      for p in xrange(np):
+      for p in range(np):
         pids = gather_ids[p]
         with ops.colocate_with(params[p]):
           result = array_ops.gather(params[p], pids)
diff --git a/tensorflow/python/ops/gradients_util.py b/tensorflow/python/ops/gradients_util.py
index 53520ec1075..0996fe47585 100644
--- a/tensorflow/python/ops/gradients_util.py
+++ b/tensorflow/python/ops/gradients_util.py
@@ -21,8 +21,6 @@ from __future__ import print_function
 import collections
 import contextlib
 
-from six.moves import xrange, zip  # pylint: disable=redefined-builtin
-
 from tensorflow.core.framework import attr_value_pb2
 from tensorflow.python.eager import backprop
 from tensorflow.python.eager import context
@@ -781,7 +779,7 @@ def _SetGrad(grads, t, grad):
   op = t.op
   op_grads = grads.get(op)
   if not op_grads:
-    op_grads = [[] for _ in xrange(len(op.outputs))]
+    op_grads = [[] for _ in range(len(op.outputs))]
     grads[op] = op_grads
   t_grads = op_grads[t.value_index]
   if isinstance(t_grads, list):
@@ -820,7 +818,7 @@ def _GetGrads(grads, op):
   if op in grads:
     return grads[op]
   else:
-    return [[] for _ in xrange(len(op.outputs))]
+    return [[] for _ in range(len(op.outputs))]
 
 
 def _AccumulatorShape(inputs):
diff --git a/tensorflow/python/ops/image_ops_test.py b/tensorflow/python/ops/image_ops_test.py
index 8dd6e35e7bf..66ea4996a63 100644
--- a/tensorflow/python/ops/image_ops_test.py
+++ b/tensorflow/python/ops/image_ops_test.py
@@ -26,7 +26,6 @@ import os
 import time
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.compat import compat
 from tensorflow.core.protobuf import config_pb2
@@ -159,9 +158,9 @@ class GrayscaleToRGBTest(test_util.TensorFlowTestCase):
       images = np.expand_dims(images, axis=0)
     out_shape = images.shape[0:3] + (1,)
     out = np.zeros(shape=out_shape, dtype=np.uint8)
-    for batch in xrange(images.shape[0]):
-      for y in xrange(images.shape[1]):
-        for x in xrange(images.shape[2]):
+    for batch in range(images.shape[0]):
+      for y in range(images.shape[1]):
+        for x in range(images.shape[2]):
           red = images[batch, y, x, 0]
           green = images[batch, y, x, 1]
           blue = images[batch, y, x, 2]
@@ -460,7 +459,7 @@ class AdjustHueTest(test_util.TensorFlowTestCase):
     x_v = x_np.reshape([-1, 3])
     y_v = np.ndarray(x_v.shape, dtype=x_v.dtype)
     channel_count = x_v.shape[0]
-    for i in xrange(channel_count):
+    for i in range(channel_count):
       r = x_v[i][0]
       g = x_v[i][1]
       b = x_v[i][2]
@@ -551,7 +550,7 @@ class FlipImageBenchmark(test.Benchmark):
             dtype=dtypes.float32)
         run_op = image_ops.flip_left_right(inputs)
         self.evaluate(variables.global_variables_initializer())
-        for i in xrange(warmup_rounds + benchmark_rounds):
+        for i in range(warmup_rounds + benchmark_rounds):
           if i == warmup_rounds:
             start = time.time()
           self.evaluate(run_op)
@@ -581,7 +580,7 @@ class FlipImageBenchmark(test.Benchmark):
             dtype=dtypes.float32)
         run_op = image_ops.random_flip_left_right(inputs)
         self.evaluate(variables.global_variables_initializer())
-        for i in xrange(warmup_rounds + benchmark_rounds):
+        for i in range(warmup_rounds + benchmark_rounds):
           if i == warmup_rounds:
             start = time.time()
           self.evaluate(run_op)
@@ -611,7 +610,7 @@ class FlipImageBenchmark(test.Benchmark):
             dtype=dtypes.float32)
         run_op = image_ops.random_flip_left_right(inputs)
         self.evaluate(variables.global_variables_initializer())
-        for i in xrange(warmup_rounds + benchmark_rounds):
+        for i in range(warmup_rounds + benchmark_rounds):
           if i == warmup_rounds:
             start = time.time()
           self.evaluate(run_op)
@@ -673,7 +672,7 @@ class AdjustHueBenchmark(test.Benchmark):
       outputs = image_ops.adjust_hue(inputs, delta)
       run_op = control_flow_ops.group(outputs)
       self.evaluate(variables.global_variables_initializer())
-      for i in xrange(warmup_rounds + benchmark_rounds):
+      for i in range(warmup_rounds + benchmark_rounds):
         if i == warmup_rounds:
           start = time.time()
         self.evaluate(run_op)
@@ -716,10 +715,10 @@ class AdjustSaturationBenchmark(test.Benchmark):
       outputs = image_ops.adjust_saturation(inputs, delta)
       run_op = control_flow_ops.group(outputs)
       self.evaluate(variables.global_variables_initializer())
-      for _ in xrange(warmup_rounds):
+      for _ in range(warmup_rounds):
         self.evaluate(run_op)
       start = time.time()
-      for _ in xrange(benchmark_rounds):
+      for _ in range(benchmark_rounds):
         self.evaluate(run_op)
     end = time.time()
     step_time = (end - start) / benchmark_rounds
@@ -752,7 +751,7 @@ class ResizeBilinearBenchmark(test.Benchmark):
         name="img")
 
     deps = []
-    for _ in xrange(num_ops):
+    for _ in range(num_ops):
       with ops.control_dependencies(deps):
         resize_op = image_ops.resize_bilinear(
             img, [299, 299], align_corners=False)
@@ -800,7 +799,7 @@ class ResizeBicubicBenchmark(test.Benchmark):
         name="img")
 
     deps = []
-    for _ in xrange(num_ops):
+    for _ in range(num_ops):
       with ops.control_dependencies(deps):
         resize_op = image_ops.resize_bicubic(
             img, [299, 299], align_corners=False)
@@ -858,7 +857,7 @@ class ResizeAreaBenchmark(test.Benchmark):
         name="img")
 
     deps = []
-    for _ in xrange(num_ops):
+    for _ in range(num_ops):
       with ops.control_dependencies(deps):
         resize_op = image_ops.resize_area(img, [299, 299], align_corners=False)
         deps = [resize_op]
@@ -946,7 +945,7 @@ class AdjustSaturationTest(test_util.TensorFlowTestCase):
     x_v = x_np.reshape([-1, 3])
     y_v = np.ndarray(x_v.shape, dtype=x_v.dtype)
     channel_count = x_v.shape[0]
-    for i in xrange(channel_count):
+    for i in range(channel_count):
       r = x_v[i][0]
       g = x_v[i][1]
       b = x_v[i][2]
@@ -1332,7 +1331,7 @@ class FlipTransposeRotateTest(test_util.TensorFlowTestCase):
     image = np.arange(24, dtype=np.uint8).reshape([2, 4, 3])
     with self.cached_session(use_gpu=True):
       rotated = image
-      for _ in xrange(4):
+      for _ in range(4):
         rotated = image_ops.rot90(rotated)
       self.assertAllEqual(image, self.evaluate(rotated))
 
@@ -1340,7 +1339,7 @@ class FlipTransposeRotateTest(test_util.TensorFlowTestCase):
     image = np.arange(48, dtype=np.uint8).reshape([2, 2, 4, 3])
     with self.cached_session(use_gpu=True):
       rotated = image
-      for _ in xrange(4):
+      for _ in range(4):
         rotated = image_ops.rot90(rotated)
       self.assertAllEqual(image, self.evaluate(rotated))
 
@@ -1350,7 +1349,7 @@ class FlipTransposeRotateTest(test_util.TensorFlowTestCase):
     with self.cached_session(use_gpu=True):
       k_placeholder = array_ops.placeholder(dtypes.int32, shape=[])
       y_tf = image_ops.rot90(image, k_placeholder)
-      for k in xrange(4):
+      for k in range(4):
         y_np = np.rot90(image, k=k)
         self.assertAllEqual(y_np, y_tf.eval({k_placeholder: k}))
 
@@ -1360,7 +1359,7 @@ class FlipTransposeRotateTest(test_util.TensorFlowTestCase):
     with self.cached_session(use_gpu=True):
       k_placeholder = array_ops.placeholder(dtypes.int32, shape=[])
       y_tf = image_ops.rot90(image, k_placeholder)
-      for k in xrange(4):
+      for k in range(4):
         y_np = np.rot90(image, k=k, axes=(1, 2))
         self.assertAllEqual(y_np, y_tf.eval({k_placeholder: k}))
 
@@ -2114,7 +2113,7 @@ class SelectDistortedCropBoxTest(test_util.TensorFlowTestCase):
           area_range=area_range)
       y = array_ops.strided_slice(image_tf, begin, begin + size)
 
-      for _ in xrange(num_iter):
+      for _ in range(num_iter):
         y_tf = self.evaluate(y)
         crop_height = y_tf.shape[0]
         crop_width = y_tf.shape[1]
@@ -2135,7 +2134,7 @@ class SelectDistortedCropBoxTest(test_util.TensorFlowTestCase):
           area_range=area_range)
       y = array_ops.strided_slice(image_tf, begin, begin + size)
 
-      for _ in xrange(num_iter):
+      for _ in range(num_iter):
         y_tf = y.eval(feed_dict={
             min_object_covered_placeholder: min_object_covered
         })
diff --git a/tensorflow/python/ops/math_ops.py b/tensorflow/python/ops/math_ops.py
index e849a687de8..ba07e6b5aa7 100644
--- a/tensorflow/python/ops/math_ops.py
+++ b/tensorflow/python/ops/math_ops.py
@@ -71,7 +71,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.compat import compat as fwd_compat
 from tensorflow.python.eager import context
@@ -3988,7 +3987,7 @@ def tensordot(a, b, axes, name=None):
     if a.get_shape().is_fully_defined() and isinstance(axes, (list, tuple)):
       shape_a = a.get_shape().as_list()
       axes = [i if i >= 0 else i + len(shape_a) for i in axes]
-      free = [i for i in xrange(len(shape_a)) if i not in axes]
+      free = [i for i in range(len(shape_a)) if i not in axes]
       free_dims = [shape_a[i] for i in free]
       prod_free = int(np.prod([shape_a[i] for i in free]))
       prod_axes = int(np.prod([shape_a[i] for i in axes]))
@@ -4000,7 +3999,7 @@ def tensordot(a, b, axes, name=None):
       if a.get_shape().ndims is not None and isinstance(axes, (list, tuple)):
         shape_a = a.get_shape().as_list()
         axes = [i if i >= 0 else i + len(shape_a) for i in axes]
-        free = [i for i in xrange(len(shape_a)) if i not in axes]
+        free = [i for i in range(len(shape_a)) if i not in axes]
         axes_dims = [shape_a[i] for i in axes]
         free_dims = [shape_a[i] for i in free]
         free_dims_static = free_dims
@@ -4037,8 +4036,8 @@ def tensordot(a, b, axes, name=None):
         if axes > a_shape.ndims:
           raise ValueError("'axes' must not be larger than the number of "
                            "dimensions of tensor %s." % a)
-        return (list(xrange(a_shape.ndims - axes,
-                            a_shape.ndims)), list(xrange(axes)))
+        return (list(range(a_shape.ndims - axes,
+                            a_shape.ndims)), list(range(axes)))
       else:
         rank = array_ops.rank(a)
         return (range(rank - axes, rank,
diff --git a/tensorflow/python/ops/nn_batchnorm_test.py b/tensorflow/python/ops/nn_batchnorm_test.py
index 5f0616b384f..216b9c56ba4 100644
--- a/tensorflow/python/ops/nn_batchnorm_test.py
+++ b/tensorflow/python/ops/nn_batchnorm_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -360,7 +359,7 @@ class SufficientStatisticsTest(test.TestCase):
       m_ss = np.sum(x, axis=axis, keepdims=keep_dims)
       v_ss = np.sum(x * x, axis=axis, keepdims=keep_dims)
     count = 1.0
-    for d in xrange(x.ndim):
+    for d in range(x.ndim):
       if d in set(axes):
         count *= x.shape[d]
     if not keep_dims:
diff --git a/tensorflow/python/ops/nn_test.py b/tensorflow/python/ops/nn_test.py
index 8bd90f01ba1..1a8d0d4c555 100644
--- a/tensorflow/python/ops/nn_test.py
+++ b/tensorflow/python/ops/nn_test.py
@@ -22,7 +22,6 @@ import math
 
 from absl.testing import parameterized
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.eager import def_function
 from tensorflow.python.framework import constant_op
@@ -317,7 +316,7 @@ class DropoutTest(test_lib.TestCase):
       dropout = nn_ops.dropout(t, rate=(1 - keep_prob))
       final_count = 0
       self.assertEqual([x_dim, y_dim], dropout.get_shape())
-      for _ in xrange(0, num_iter):
+      for _ in range(0, num_iter):
         value = self.evaluate(dropout)
         final_count += np.count_nonzero(value)
         # Verifies that there are only two values: 0 and 1/keep_prob.
@@ -344,7 +343,7 @@ class DropoutTest(test_lib.TestCase):
       dropout = nn_ops.dropout(t, rate=(1 - keep_prob), noise_shape=[x_dim, 1])
       self.assertEqual([x_dim, y_dim], dropout.get_shape())
       final_count = 0
-      for _ in xrange(0, num_iter):
+      for _ in range(0, num_iter):
         value = self.evaluate(dropout)
         final_count += np.count_nonzero(value)
         # Verifies that there are only two values: 0 and 1/keep_prob.
@@ -367,10 +366,10 @@ class DropoutTest(test_lib.TestCase):
       t = constant_op.constant(1.0, shape=[x_dim, y_dim], dtype=dtypes.float32)
       dropout = nn_ops.dropout(t, rate=(1 - keep_prob), noise_shape=[x_dim, 1])
       self.assertEqual([x_dim, y_dim], dropout.get_shape())
-      for _ in xrange(0, num_iter):
+      for _ in range(0, num_iter):
         value = self.evaluate(dropout)
         # Verifies that each y column as only one type of activation.
-        for i in xrange(x_dim):
+        for i in range(x_dim):
           sorted_value = np.unique(np.sort(value[i, :]))
           self.assertEqual(sorted_value.size, 1)
 
@@ -390,7 +389,7 @@ class DropoutTest(test_lib.TestCase):
         dropout = nn_ops.dropout(t, keep_prob_placeholder)
         final_count = 0
         self.assertEqual([x_dim, y_dim], dropout.get_shape())
-        for _ in xrange(0, num_iter):
+        for _ in range(0, num_iter):
           value = dropout.eval(feed_dict={keep_prob_placeholder: keep_prob})
           final_count += np.count_nonzero(value)
           # Verifies that there are only two values: 0 and 1/keep_prob.
@@ -425,7 +424,7 @@ class DropoutTest(test_lib.TestCase):
       dropout = nn_ops.dropout(t, rate=(1 - keep_prob), noise_shape=[None, 1])
       self.assertEqual([x_dim, y_dim], dropout.get_shape())
       final_count = 0
-      for _ in xrange(0, num_iter):
+      for _ in range(0, num_iter):
         value = self.evaluate(dropout)
         final_count += np.count_nonzero(value)
         # Verifies that there are only two values: 0 and 1/keep_prob.
@@ -706,9 +705,9 @@ class ComputeSampledLogitsTest(test_lib.TestCase):
       # First we need to find the hits in this random test run:
       labels_reshape = labels.reshape((batch_size, num_true))
       got_logits = self.evaluate(logits_tensor)
-      for row in xrange(batch_size):
+      for row in range(batch_size):
         row_labels = labels_reshape[row, :]
-        for col in xrange(len(sampled)):
+        for col in range(len(sampled)):
           if sampled[col] in row_labels:
             # We need to add the num_true_test offset into logits_*
             self.assertNear(
diff --git a/tensorflow/python/ops/special_math_ops.py b/tensorflow/python/ops/special_math_ops.py
index 5f3b179d5d2..8c2c3e8a6b1 100644
--- a/tensorflow/python/ops/special_math_ops.py
+++ b/tensorflow/python/ops/special_math_ops.py
@@ -24,8 +24,6 @@ from __future__ import print_function
 import re
 import string
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
-
 from tensorflow.compiler.tf2xla.ops import gen_xla_ops
 from tensorflow.python.framework import ops
 from tensorflow.python.ops import array_ops
@@ -283,7 +281,7 @@ def einsum(equation, *inputs, **kwargs):
           input_axis_labels[1] + '->' + output_axis_labels)
     temp = inputs[0]
     temp_axis_labels = input_axis_labels[0]
-    for i in xrange(len(inputs) - 1):
+    for i in range(len(inputs) - 1):
       axes_to_sum = (
           set(temp_axis_labels) &
           set(input_axis_labels[i + 1]) - set(output_axis_labels))
diff --git a/tensorflow/python/ops/variable_scope.py b/tensorflow/python/ops/variable_scope.py
index 553052550d2..d26ce3374a1 100644
--- a/tensorflow/python/ops/variable_scope.py
+++ b/tensorflow/python/ops/variable_scope.py
@@ -28,7 +28,6 @@ import traceback
 
 import six
 from six import iteritems
-from six.moves import xrange, zip  # pylint: disable=redefined-builtin
 
 from tensorflow.python import tf2
 from tensorflow.python.eager import context
@@ -162,7 +161,7 @@ class _PartitionInfo(object):
                                                        self.full_shape,
                                                        len(self.full_shape)))
 
-    for i in xrange(len(shape)):
+    for i in range(len(shape)):
       if self.var_offset[i] + shape[i] > self.full_shape[i]:
         raise ValueError(
             "With self.var_offset={}, a partition of shape={} would exceed "
@@ -170,7 +169,7 @@ class _PartitionInfo(object):
                 self.var_offset, shape, self.full_shape, i))
 
     slice_dim = None
-    for i in xrange(len(shape)):
+    for i in range(len(shape)):
       if shape[i] == self.full_shape[i]:
         continue
       if slice_dim is not None:
@@ -2453,7 +2452,7 @@ def _iter_slices(full_shape, num_slices, slice_dim):
   num_slices_with_excess = full_shape[slice_dim] % num_slices
   offset = [0] * len(full_shape)
   min_slice_len = full_shape[slice_dim] // num_slices
-  for i in xrange(num_slices):
+  for i in range(num_slices):
     shape = full_shape[:]
     shape[slice_dim] = min_slice_len + bool(i < num_slices_with_excess)
     yield offset[:], shape
diff --git a/tensorflow/python/platform/tf_logging.py b/tensorflow/python/platform/tf_logging.py
index 86a4957c9da..8924334a5df 100644
--- a/tensorflow/python/platform/tf_logging.py
+++ b/tensorflow/python/platform/tf_logging.py
@@ -60,8 +60,8 @@ def _get_caller(offset=3):
 
 # The definition of `findCaller` changed in Python 3.2
 if _sys.version_info.major >= 3 and _sys.version_info.minor >= 2:
-  def _logger_find_caller(stack_info=False):  # pylint: disable=g-wrong-blank-lines
-    code, frame = _get_caller(4)
+  def _logger_find_caller(stack_info=False, stacklevel=1):  # pylint: disable=g-wrong-blank-lines
+    code, frame = _get_caller(3+stacklevel)
     sinfo = None
     if stack_info:
       sinfo = '\n'.join(_traceback.format_stack())
diff --git a/tensorflow/python/profiler/model_analyzer.py b/tensorflow/python/profiler/model_analyzer.py
index f3b521a6bfb..5b1e6e2f674 100644
--- a/tensorflow/python/profiler/model_analyzer.py
+++ b/tensorflow/python/profiler/model_analyzer.py
@@ -133,7 +133,7 @@ class Profiler(object):
     # Currently we are only allowed to create 1 profiler per process.
     profiler = Profiler(sess.graph)
 
-    for i in xrange(total_steps):
+    for i in range(total_steps):
       if i % 10000 == 0:
         run_meta = tf.compat.v1.RunMetadata()
         _ = sess.run(...,
diff --git a/tensorflow/python/summary/summary_test.py b/tensorflow/python/summary/summary_test.py
index 64f0f315c58..1df412b8ed9 100644
--- a/tensorflow/python/summary/summary_test.py
+++ b/tensorflow/python/summary/summary_test.py
@@ -23,7 +23,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.framework import summary_pb2
 from tensorflow.python.framework import constant_op
@@ -105,7 +104,7 @@ class SummaryTest(test.TestCase):
     values = summary.value
     self.assertEqual(len(values), 3)
     tags = sorted(v.tag for v in values)
-    expected = sorted('outer/inner/image/{}'.format(i) for i in xrange(3))
+    expected = sorted('outer/inner/image/{}'.format(i) for i in range(3))
     self.assertEqual(tags, expected)
 
   @test_util.run_deprecated_v1
@@ -122,7 +121,7 @@ class SummaryTest(test.TestCase):
     self.assertEqual(len(values), 3)
     tags = sorted(v.tag for v in values)
     expected = sorted('family/outer/family/inner/image/{}'.format(i)
-                      for i in xrange(3))
+                      for i in range(3))
     self.assertEqual(tags, expected)
 
   @test_util.run_deprecated_v1
@@ -168,7 +167,7 @@ class SummaryTest(test.TestCase):
     values = summary.value
     self.assertEqual(len(values), 3)
     tags = sorted(v.tag for v in values)
-    expected = sorted('outer/inner/audio/{}'.format(i) for i in xrange(3))
+    expected = sorted('outer/inner/audio/{}'.format(i) for i in range(3))
     self.assertEqual(tags, expected)
 
   @test_util.run_deprecated_v1
@@ -185,7 +184,7 @@ class SummaryTest(test.TestCase):
     self.assertEqual(len(values), 3)
     tags = sorted(v.tag for v in values)
     expected = sorted('family/outer/family/inner/audio/{}'.format(i)
-                      for i in xrange(3))
+                      for i in range(3))
     self.assertEqual(tags, expected)
 
   @test_util.run_deprecated_v1
diff --git a/tensorflow/python/tpu/device_assignment.py b/tensorflow/python/tpu/device_assignment.py
index 863d5e61e69..580fbf48205 100644
--- a/tensorflow/python/tpu/device_assignment.py
+++ b/tensorflow/python/tpu/device_assignment.py
@@ -20,7 +20,6 @@ from __future__ import print_function
 
 import math
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.platform import tf_logging as logging
 from tensorflow.python.tpu.topology import Topology
@@ -33,8 +32,8 @@ SINGLE_CORE_ASSIGNMENT = [[[0, 0, 0]]]
 def _compute_task_and_cores_to_replicas(core_assignment, topology):
   """Computes a nested dict which maps task and logical core to replicas."""
   task_and_cores_to_replicas = {}
-  for replica in xrange(core_assignment.shape[0]):
-    for logical_core in xrange(core_assignment.shape[1]):
+  for replica in range(core_assignment.shape[0]):
+    for logical_core in range(core_assignment.shape[1]):
       coordinates = core_assignment[replica, logical_core, :]
       task_id = topology.task_ordinal_at_coordinates(coordinates)
       if task_id not in task_and_cores_to_replicas:
@@ -347,10 +346,10 @@ def device_assignment(topology,
     inner_ring = _ring_2d(computation_shape[0], computation_shape[1])
     outer_ring = _ring_2d(replica_shape[0], replica_shape[1])
 
-    for replica in xrange(num_replicas):
+    for replica in range(num_replicas):
       outer_x, outer_y = outer_ring[replica]
       per_replica_assignment = []
-      for index in xrange(np.prod(computation_shape)):
+      for index in range(np.prod(computation_shape)):
         inner_x, inner_y = inner_ring[index // 2]
         px = outer_x * computation_shape[0] + inner_x
         py = outer_y * computation_shape[1] + inner_y
@@ -358,7 +357,7 @@ def device_assignment(topology,
         per_replica_assignment.append([px, py, pz])
       assignment.append(per_replica_assignment)
   else:
-    for replica in xrange(num_replicas):
+    for replica in range(num_replicas):
       # Chooses a replica number in each axis.
       t = replica
       pos = []
@@ -375,7 +374,7 @@ def device_assignment(topology,
     # Computes a logical core -> physical core mapping for each replica.
     indices = [
         np.arange(0, computation_shape[i] * computation_stride[i],
-                  computation_stride[i]) for i in xrange(topology_rank)
+                  computation_stride[i]) for i in range(topology_rank)
     ]
     indices = np.concatenate(
         [i[..., np.newaxis] for i in np.meshgrid(*indices, indexing="ij")],
diff --git a/tensorflow/python/tpu/topology.py b/tensorflow/python/tpu/topology.py
index 00ee21e694d..ea5bb22a485 100644
--- a/tensorflow/python/tpu/topology.py
+++ b/tensorflow/python/tpu/topology.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.protobuf.tpu import topology_pb2
 
@@ -129,8 +128,8 @@ class Topology(object):
     """Inverts a [task,device,axis] topology to [x,y,z] -> task/device maps."""
     tasks = np.full(list(self.mesh_shape), -1, dtype=np.int32)
     devices = np.full(list(self.mesh_shape), -1, dtype=np.int32)
-    for task in xrange(self.device_coordinates.shape[0]):
-      for device in xrange(self.device_coordinates.shape[1]):
+    for task in range(self.device_coordinates.shape[0]):
+      for device in range(self.device_coordinates.shape[1]):
         x, y, z = self.device_coordinates[task, device, :]
         tasks[x, y, z] = task
         devices[x, y, z] = device
diff --git a/tensorflow/python/tpu/tpu.py b/tensorflow/python/tpu/tpu.py
index f65068b3f7f..e884c5c42a7 100644
--- a/tensorflow/python/tpu/tpu.py
+++ b/tensorflow/python/tpu/tpu.py
@@ -21,7 +21,6 @@ from __future__ import print_function
 
 from absl import logging
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.framework import attr_value_pb2
 from tensorflow.core.protobuf.tpu import dynamic_padding_pb2 as dynamic_padding
@@ -458,7 +457,7 @@ class TPUReplicateContext(control_flow_ops.XLAControlFlowContext):
         op._add_control_input(self.GetControlPivot())
         # pylint: enable=protected-access
     else:
-      for index in xrange(len(op.inputs)):
+      for index in range(len(op.inputs)):
         x = op.inputs[index]
         real_x = self.AddValue(x)
         if real_x is not x:
@@ -832,7 +831,7 @@ def split_compile_and_replicate(computation,
     return []
 
   # Checks all replicas have the same structure.
-  for i in xrange(1, num_replicas):
+  for i in range(1, num_replicas):
     nest.assert_same_structure(inputs[0], inputs[i])
 
   # Flatten inputs.
@@ -904,7 +903,7 @@ def split_compile_and_replicate(computation,
   # Fan-in: Builds a TPUReplicatedInput node for each input.
   flat_replicated_inputs = []
   for i in range(0, len(flat_inputs[0])):
-    replicas = [flat_inputs[replica][i] for replica in xrange(num_replicas)]
+    replicas = [flat_inputs[replica][i] for replica in range(num_replicas)]
     flat_replicated_inputs.append(
         tpu_ops.tpu_replicated_input(replicas, name="input{}".format(i)))
 
@@ -1043,7 +1042,7 @@ def split_compile_and_replicate(computation,
     ]
 
   # Fan-out: Builds a TPUReplicatedOutput node for each output.
-  replicated_outputs = [[] for i in xrange(num_replicas)]
+  replicated_outputs = [[] for i in range(num_replicas)]
   for i, t in enumerate(output_tensors):
     # Fan-out: Builds a TPUReplicatedOutput node for each output.
     ys = tpu_ops.tpu_replicated_output(
@@ -1052,7 +1051,7 @@ def split_compile_and_replicate(computation,
     # Wraps the outputs in identity operators so the names of any possible
     # `fetch` nodes are preserved by the replication rewrite.
     with ops.control_dependencies(control_deps):
-      for replica in xrange(num_replicas):
+      for replica in range(num_replicas):
         replicated_outputs[replica].append(
             array_ops.identity(
                 ys[replica], name="output_%d_shard_%d" % (i, replica)))
diff --git a/tensorflow/python/tpu/tpu_feed.py b/tensorflow/python/tpu/tpu_feed.py
index 2b01eeb3934..202bd2a02cd 100644
--- a/tensorflow/python/tpu/tpu_feed.py
+++ b/tensorflow/python/tpu/tpu_feed.py
@@ -23,7 +23,6 @@ from __future__ import print_function
 import itertools
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.compiler.xla.experimental.xla_sharding import xla_sharding
 from tensorflow.python.framework import dtypes
@@ -183,7 +182,7 @@ class InfeedQueue(object):
     # Make an empty sharding policy for each tuple element.
     self._sharding_policies = [
         tpu_sharding.ShardingPolicy()
-        for _ in xrange(number_of_tuple_elements)
+        for _ in range(number_of_tuple_elements)
     ]
     if tuple_types is not None:
       self.set_tuple_types(tuple_types)
@@ -416,14 +415,14 @@ class InfeedQueue(object):
     # Transpose the inputs to make a list of shard shapes for each tuple
     # element.
     sharded_shapes = [[t[i].shape for t in input_tensors]
-                      for i in xrange(self.number_of_tuple_elements)]
+                      for i in range(self.number_of_tuple_elements)]
     # For each tuple, get the unsharded shape using that tuple's policy.
     unsharded_shapes = [
         policy.get_unsharded_shape(s)
         for (policy, s) in zip(self._sharding_policies, sharded_shapes)
     ]
     self.set_tuple_shapes(unsharded_shapes)
-    for i in xrange(1, self.number_of_shards):
+    for i in range(1, self.number_of_shards):
       for (t1, t2) in zip(input_tensors[0], input_tensors[i]):
         if t1.dtype != t2.dtype:
           raise TypeError(
@@ -530,7 +529,7 @@ class InfeedQueue(object):
     shapes = [t.shape for t in inputs]
     if device is None:
       devices = [t.device for t in inputs]
-      for i in xrange(1, self.number_of_tuple_elements):
+      for i in range(1, self.number_of_tuple_elements):
         if devices[0] != devices[i]:
           raise ValueError(
               "input devices for shard %d are %s, but should all be the same" %
@@ -608,7 +607,7 @@ class InfeedQueue(object):
             index,
             tpu_ordinal=tpu_ordinal_function(index),
             device=placement_function(index) if placement_function else None)
-        for (shard, index) in zip(sharded_inputs, xrange(self.number_of_shards))
+        for (shard, index) in zip(sharded_inputs, range(self.number_of_shards))
     ]
 
   # TODO(misard) Generalize this to the case of systems that don't
@@ -714,10 +713,10 @@ class InfeedQueue(object):
               axis=policy.shard_dimension,
               name="%s/%d" % (split_name_prefix, index))
           for (inp, policy, index) in zip(inputs, self._sharding_policies,
-                                          xrange(self.number_of_tuple_elements))
+                                          range(self.number_of_tuple_elements))
       ]
     sharded_inputs = [[shard[i] for shard in transposed_sharded_inputs]
-                      for i in xrange(self.number_of_shards)]
+                      for i in range(self.number_of_shards)]
     name_prefix = "%s/enqueue" % self._name
     return [
         self._generate_enqueue_op(
@@ -726,7 +725,7 @@ class InfeedQueue(object):
             index,
             device=placement_function(index),
             tpu_ordinal=tpu_ordinal_function(index))
-        for (shard, index) in zip(sharded_inputs, xrange(self.number_of_shards))
+        for (shard, index) in zip(sharded_inputs, range(self.number_of_shards))
     ]
 
 
@@ -855,7 +854,7 @@ class _PartitionedInfeedQueue(InfeedQueue):
       # input_fn).
       replica_id = self._device_assignment.lookup_replicas(
           task_id=self._host_id, logical_core=0)[replica_index]
-      for logical_core in xrange(self._device_assignment.num_cores_per_replica):
+      for logical_core in range(self._device_assignment.num_cores_per_replica):
         # Places different partitions to different logic cores.
         # Since there can be multiple hosts per replica, we need to find
         # the actual host (device) of this logical core.
diff --git a/tensorflow/python/tpu/tpu_sharding.py b/tensorflow/python/tpu/tpu_sharding.py
index 2541cf241b8..7f5f370c046 100644
--- a/tensorflow/python/tpu/tpu_sharding.py
+++ b/tensorflow/python/tpu/tpu_sharding.py
@@ -19,7 +19,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import tensor_shape
 
@@ -242,7 +241,7 @@ class ShardingPolicy(object):
           "shapes is %s but must be a list of length number_of_shards=%d" % (
               str(shapes), self.number_of_shards))
     unsharded_shapes = [self._unshard_shape(s) for s in shapes]
-    for i in xrange(self.number_of_shards - 1):
+    for i in range(self.number_of_shards - 1):
       if not unsharded_shapes[i].is_compatible_with(
           unsharded_shapes[self.number_of_shards - 1]):
         raise ValueError(
diff --git a/tensorflow/python/training/input.py b/tensorflow/python/training/input.py
index 41c8c715d77..5e52323f02a 100644
--- a/tensorflow/python/training/input.py
+++ b/tensorflow/python/training/input.py
@@ -24,7 +24,6 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.eager import context
 from tensorflow.python.framework import constant_op
@@ -699,13 +698,13 @@ def _shapes(tensor_list_list, shapes, enqueue_many):
     len0 = len(tensor_list_list[0])
 
     for tl in tensor_list_list:
-      for i in xrange(len0):
+      for i in range(len0):
         if tl[i].shape.ndims is None:
           raise ValueError("Cannot infer Tensor's rank: %s" % tl[i])
 
     shapes = [_merge_shapes(
         [tl[i].shape.as_list() for tl in tensor_list_list], enqueue_many)
-              for i in xrange(len0)]
+              for i in range(len0)]
   return shapes
 
 
diff --git a/tensorflow/python/training/input_test.py b/tensorflow/python/training/input_test.py
index b82a2611d54..31b6496f08b 100644
--- a/tensorflow/python/training/input_test.py
+++ b/tensorflow/python/training/input_test.py
@@ -22,7 +22,6 @@ import itertools
 import os
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -290,7 +289,7 @@ class RangeInputProducerTest(test_lib.TestCase):
 
       # No randomness, so just see repeated copies of the input.
       output = self.evaluate(dequeue_many)
-      self.assertAllEqual(list(xrange(range_size)) * num_epochs, output)
+      self.assertAllEqual(list(range(range_size)) * num_epochs, output)
 
       # Reached the limit.
       with self.assertRaises(errors_impl.OutOfRangeError):
diff --git a/tensorflow/python/training/momentum_test.py b/tensorflow/python/training/momentum_test.py
index ba155fa6c64..4658410264a 100644
--- a/tensorflow/python/training/momentum_test.py
+++ b/tensorflow/python/training/momentum_test.py
@@ -19,7 +19,6 @@ from __future__ import division
 from __future__ import print_function
 
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.eager import context
 from tensorflow.python.framework import constant_op
@@ -453,7 +452,7 @@ class MomentumOptimizerTest(test.TestCase):
       mom_opt = momentum_lib.MomentumOptimizer(learning_rate=0.1, momentum=0.1)
       mom_update = mom_opt.apply_gradients(zip([grads0], [var0]))
       variables.global_variables_initializer().run()
-      for i in xrange(num_samples):
+      for i in range(num_samples):
         mom_update.run(feed_dict={grads0: db_grad[i]})
         self.assertAllClose(np.array(db_out[i]), self.evaluate(var0))
 
diff --git a/tensorflow/python/training/saver.py b/tensorflow/python/training/saver.py
index d65297fb30d..a84291aaab6 100644
--- a/tensorflow/python/training/saver.py
+++ b/tensorflow/python/training/saver.py
@@ -661,7 +661,7 @@ class Saver(object):
   saver = tf.compat.v1.train.Saver(...variables...)
   # Launch the graph and train, saving the model every 1,000 steps.
   sess = tf.compat.v1.Session()
-  for step in xrange(1000000):
+  for step in range(1000000):
       sess.run(..training_op..)
       if step % 1000 == 0:
           # Append the step number to the checkpoint name:
@@ -1368,7 +1368,7 @@ def import_meta_graph(meta_graph_or_file,
   # Remember the training_op we want to run by adding it to a collection.
   tf.compat.v1.add_to_collection('train_op', train_op)
   sess = tf.compat.v1.Session()
-  for step in xrange(1000000):
+  for step in range(1000000):
       sess.run(train_op)
       if step % 1000 == 0:
           # Saves checkpoint, which by default also exports a meta_graph
@@ -1387,7 +1387,7 @@ def import_meta_graph(meta_graph_or_file,
     # tf.get_collection() returns a list. In this example we only want
     # the first one.
     train_op = tf.get_collection('train_op')[0]
-    for step in xrange(1000000):
+    for step in range(1000000):
       sess.run(train_op)
   ```
 
diff --git a/tensorflow/python/training/supervisor.py b/tensorflow/python/training/supervisor.py
index 91960cb3f94..e1e94cb9fa5 100644
--- a/tensorflow/python/training/supervisor.py
+++ b/tensorflow/python/training/supervisor.py
@@ -179,7 +179,7 @@ class Supervisor(object):
   # As summary_op was None, managed_session() does not start the
   # summary thread.
   with sv.managed_session(FLAGS.master) as sess:
-    for step in xrange(1000000):
+    for step in range(1000000):
       if sv.should_stop():
         break
       if is_chief and step % 100 == 0:
@@ -951,7 +951,7 @@ class Supervisor(object):
     def train():
       sv = tf.compat.v1.train.Supervisor(...)
       with sv.managed_session(<master>) as sess:
-        for step in xrange(..):
+        for step in range(..):
           if sv.should_stop():
             break
           sess.run(<my training op>)
diff --git a/tensorflow/python/training/supervisor_test.py b/tensorflow/python/training/supervisor_test.py
index 180ddb52876..ef31a0748fc 100644
--- a/tensorflow/python/training/supervisor_test.py
+++ b/tensorflow/python/training/supervisor_test.py
@@ -24,7 +24,6 @@ import shutil
 import time
 import uuid
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.core.framework import graph_pb2
 from tensorflow.core.protobuf import config_pb2
@@ -100,7 +99,7 @@ class SupervisorTest(test.TestCase):
       my_op = constant_op.constant(1.0)
       sv = supervisor.Supervisor(logdir=logdir)
       sess = sv.prepare_or_wait_for_session("")
-      for _ in xrange(10):
+      for _ in range(10):
         self.evaluate(my_op)
       sess.close()
       sv.stop()
@@ -111,7 +110,7 @@ class SupervisorTest(test.TestCase):
       my_op = constant_op.constant(1.0)
       sv = supervisor.Supervisor(logdir=logdir)
       with sv.managed_session("") as sess:
-        for _ in xrange(10):
+        for _ in range(10):
           self.evaluate(my_op)
       # Supervisor has been stopped.
       self.assertTrue(sv.should_stop())
@@ -124,7 +123,7 @@ class SupervisorTest(test.TestCase):
       last_step = None
       with self.assertRaisesRegexp(RuntimeError, "failing here"):
         with sv.managed_session("") as sess:
-          for step in xrange(10):
+          for step in range(10):
             last_step = step
             if step == 1:
               raise RuntimeError("failing here")
@@ -141,7 +140,7 @@ class SupervisorTest(test.TestCase):
       sv = supervisor.Supervisor(logdir=logdir)
       last_step = None
       with sv.managed_session("") as sess:
-        for step in xrange(10):
+        for step in range(10):
           last_step = step
           if step == 3:
             raise errors_impl.OutOfRangeError(my_op.op.node_def, my_op.op,
@@ -335,7 +334,7 @@ class SupervisorTest(test.TestCase):
       sv = supervisor.Supervisor(logdir=logdir)
       sess = sv.prepare_or_wait_for_session(
           "", config=config_pb2.ConfigProto(device_count={"CPU": 2}))
-      for _ in xrange(10):
+      for _ in range(10):
         self.evaluate(my_op)
       sess.close()
       sv.stop()
diff --git a/tensorflow/python/training/tracking/benchmarks_test.py b/tensorflow/python/training/tracking/benchmarks_test.py
index 7514d9f54cf..632c823774a 100644
--- a/tensorflow/python/training/tracking/benchmarks_test.py
+++ b/tensorflow/python/training/tracking/benchmarks_test.py
@@ -77,7 +77,7 @@ class SavingBenchmarks(test.Benchmark):
   def _run(self, func, num_iters, execution_mode=None):
     func()
     start = time.time()
-    for _ in xrange(num_iters):
+    for _ in range(num_iters):
       func()
     end = time.time()
     mean_us = (end - start) * 1e6 / num_iters
diff --git a/tensorflow/python/util/nest_test.py b/tensorflow/python/util/nest_test.py
index 6ec4a5d61f9..b1a7237c139 100644
--- a/tensorflow/python/util/nest_test.py
+++ b/tensorflow/python/util/nest_test.py
@@ -23,7 +23,6 @@ import time
 
 from absl.testing import parameterized
 import numpy as np
-from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
@@ -1215,11 +1214,11 @@ class NestBenchmark(test.Benchmark):
   def run_and_report(self, s1, s2, name):
     burn_iter, test_iter = 100, 30000
 
-    for _ in xrange(burn_iter):
+    for _ in range(burn_iter):
       nest.assert_same_structure(s1, s2)
 
     t0 = time.time()
-    for _ in xrange(test_iter):
+    for _ in range(test_iter):
       nest.assert_same_structure(s1, s2)
     t1 = time.time()
 
diff --git a/tensorflow/python/util/tf_inspect.py b/tensorflow/python/util/tf_inspect.py
index ee2df3cb21b..793a12154e0 100644
--- a/tensorflow/python/util/tf_inspect.py
+++ b/tensorflow/python/util/tf_inspect.py
@@ -25,12 +25,15 @@ import six
 
 from tensorflow.python.util import tf_decorator
 
-ArgSpec = _inspect.ArgSpec
 
 
 if hasattr(_inspect, 'FullArgSpec'):
   FullArgSpec = _inspect.FullArgSpec  # pylint: disable=invalid-name
+  ArgSpec = collections.namedtuple('ArgSpec', [
+      'args', 'varargs', 'keywords', 'defaults',
+  ])
 else:
+  ArgSpec = _inspect.ArgSpec
   FullArgSpec = collections.namedtuple('FullArgSpec', [
       'args', 'varargs', 'varkw', 'defaults', 'kwonlyargs', 'kwonlydefaults',
       'annotations'
diff --git a/tensorflow/python/util/tf_stack.cc b/tensorflow/python/util/tf_stack.cc
index d6551ccb98c..8bdbde1c796 100644
--- a/tensorflow/python/util/tf_stack.cc
+++ b/tensorflow/python/util/tf_stack.cc
@@ -53,13 +53,22 @@ std::vector<StackFrame> ExtractStack(ssize_t limit, const py::list& mappers,
 
   const auto* tstate = PyThreadState_GET();
   // Drop extract_stack() wrapper-function frame from the result.
+#if PY_MINOR_VERSION >= 9
+  const PyFrameObject* f = PyFrame_GetBack(PyThreadState_GetFrame(const_cast<PyThreadState*>(tstate)));
+#else
   const PyFrameObject* f = tstate->frame->f_back;  // TODO(slebedev): INCREF?
+#endif
 
   std::vector<StackFrame> ret;
   // 16 is somewhat arbitrary, but TensorFlow stack traces tend to be deep.
   ret.reserve(limit < 0 ? 16 : static_cast<size_t>(limit));
+#if PY_MINOR_VERSION >= 9
+  for (; f != nullptr && (limit < 0 || ret.size() < limit); f = PyFrame_GetBack(const_cast<PyFrameObject*>(f))) {
+    PyCodeObject* co = PyFrame_GetCode(const_cast<PyFrameObject*>(f));
+#else
   for (; f != nullptr && (limit < 0 || ret.size() < limit); f = f->f_back) {
     PyCodeObject* co = f->f_code;
+#endif
     int lineno = PyFrame_GetLineNumber(const_cast<PyFrameObject*>(f));
     auto filename = py::reinterpret_borrow<py::str>(co->co_filename);
     auto name = py::reinterpret_borrow<py::str>(co->co_name);
@@ -81,7 +90,11 @@ std::vector<StackFrame> ExtractStack(ssize_t limit, const py::list& mappers,
         PySet_Contains(filtered_filenames.ptr(), filename.ptr()))
       continue;
 
+#if PY_MINOR_VERSION >= 11
+    const auto& globals = py::reinterpret_borrow<py::object>(PyFrame_GetGlobals(const_cast<PyFrameObject*>(f)));
+#else
     const auto& globals = py::reinterpret_borrow<py::object>(f->f_globals);
+#endif
     const int func_start_lineno = co->co_firstlineno;
     ret.push_back({std::move(filename), lineno, std::move(name), globals,
                    func_start_lineno});
diff --git a/tensorflow/stream_executor/cuda/BUILD b/tensorflow/stream_executor/cuda/BUILD
index 27b1364c6cb..700d9734ec0 100644
--- a/tensorflow/stream_executor/cuda/BUILD
+++ b/tensorflow/stream_executor/cuda/BUILD
@@ -88,7 +88,7 @@ cc_library(
 cc_library(
     name = "cuda_stub",
     srcs = if_cuda_is_configured(["cuda_stub.cc"]),
-    textual_hdrs = ["cuda_10_0.inc"],
+    textual_hdrs = ["cuda_12_0.inc"],
     deps = if_cuda_is_configured([
         "@local_config_cuda//cuda:cuda_headers",
         "//tensorflow/stream_executor/lib",
diff --git a/tensorflow/stream_executor/cuda/cublas_11_0.inc b/tensorflow/stream_executor/cuda/cublas_11_0.inc
new file mode 100644
index 00000000000..072df757ddb
--- /dev/null
+++ b/tensorflow/stream_executor/cuda/cublas_11_0.inc
@@ -0,0 +1,5236 @@
+// Auto-generated, do not edit.
+
+static inline cublasStatus_t cublasMigrateComputeType(cublasHandle_t handle,
+                                                      cudaDataType_t dataType,
+                                                      cublasComputeType_t* computeType) {
+  cublasMath_t mathMode = CUBLAS_DEFAULT_MATH;
+  cublasStatus_t status = CUBLAS_STATUS_SUCCESS;
+
+  status = cublasGetMathMode(handle, &mathMode);
+  if (status != CUBLAS_STATUS_SUCCESS) {
+    return status;
+  }
+
+  bool isPedantic = ((mathMode & 0xf) == CUBLAS_PEDANTIC_MATH);
+
+  switch (dataType) {
+    case CUDA_R_32F:
+    case CUDA_C_32F:
+      *computeType = isPedantic ? CUBLAS_COMPUTE_32F_PEDANTIC : CUBLAS_COMPUTE_32F;
+      return CUBLAS_STATUS_SUCCESS;
+    case CUDA_R_64F:
+    case CUDA_C_64F:
+      *computeType = isPedantic ? CUBLAS_COMPUTE_64F_PEDANTIC : CUBLAS_COMPUTE_64F;
+      return CUBLAS_STATUS_SUCCESS;
+    case CUDA_R_16F:
+      *computeType = isPedantic ? CUBLAS_COMPUTE_16F_PEDANTIC : CUBLAS_COMPUTE_16F;
+      return CUBLAS_STATUS_SUCCESS;
+    case CUDA_R_32I:
+      *computeType = isPedantic ? CUBLAS_COMPUTE_32I_PEDANTIC : CUBLAS_COMPUTE_32I;
+      return CUBLAS_STATUS_SUCCESS;
+    default:
+      return CUBLAS_STATUS_NOT_SUPPORTED;
+  }
+}
+
+extern "C" {
+cublasStatus_t CUBLASWINAPI cublasCreate_v2(cublasHandle_t *handle) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCreate_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDestroy_v2(cublasHandle_t handle) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDestroy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetVersion_v2(cublasHandle_t handle,
+                                                int *version) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetVersion_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, version);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetProperty(libraryPropertyType type,
+                                              int *value) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(libraryPropertyType, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetProperty");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(type, value);
+}
+
+size_t CUBLASWINAPI cublasGetCudartVersion(void) {
+  using FuncPtr = size_t(CUBLASWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetCudartVersion");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasGetCudartVersion");
+  return func_ptr();
+}
+
+cublasStatus_t CUBLASWINAPI cublasSetWorkspace_v2(cublasHandle_t handle,
+                                                  void *workspace,
+                                                  size_t workspaceSizeInBytes) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetWorkspace_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, workspace, workspaceSizeInBytes);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSetStream_v2(cublasHandle_t handle,
+                                               cudaStream_t streamId) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetStream_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, streamId);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetStream_v2(cublasHandle_t handle,
+                                               cudaStream_t *streamId) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, cudaStream_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetStream_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, streamId);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetPointerMode_v2(cublasHandle_t handle,
+                                                    cublasPointerMode_t *mode) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, cublasPointerMode_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetPointerMode_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSetPointerMode_v2(cublasHandle_t handle,
+                                                    cublasPointerMode_t mode) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, cublasPointerMode_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetPointerMode_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetAtomicsMode(cublasHandle_t handle,
+                                                 cublasAtomicsMode_t *mode) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, cublasAtomicsMode_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetAtomicsMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSetAtomicsMode(cublasHandle_t handle,
+                                                 cublasAtomicsMode_t mode) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, cublasAtomicsMode_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetAtomicsMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetMathMode(cublasHandle_t handle,
+                                              cublasMath_t *mode) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, cublasMath_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetMathMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSetMathMode(cublasHandle_t handle,
+                                              cublasMath_t mode) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, cublasMath_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetMathMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetSmCountTarget(cublasHandle_t handle,
+                                                   int *smCountTarget) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetSmCountTarget");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, smCountTarget);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSetSmCountTarget(cublasHandle_t handle,
+                                                   int smCountTarget) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetSmCountTarget");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, smCountTarget);
+}
+
+const char *CUBLASWINAPI cublasGetStatusName(cublasStatus_t status) {
+  using FuncPtr = const char *(CUBLASWINAPI *)(cublasStatus_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetStatusName");
+  if (!func_ptr) return "cublasGetStatusName symbol not found.";
+  return func_ptr(status);
+}
+
+const char *CUBLASWINAPI cublasGetStatusString(cublasStatus_t status) {
+  using FuncPtr = const char *(CUBLASWINAPI *)(cublasStatus_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetStatusString");
+  if (!func_ptr) return "cublasGetStatusString symbol not found.";
+  return func_ptr(status);
+}
+
+cublasStatus_t CUBLASWINAPI cublasLoggerConfigure(int logIsOn, int logToStdOut,
+                                                  int logToStdErr,
+                                                  const char *logFileName) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(int, int, int, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasLoggerConfigure");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(logIsOn, logToStdOut, logToStdErr, logFileName);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSetLoggerCallback(cublasLogCallback userCallback) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasLogCallback);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetLoggerCallback");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(userCallback);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasGetLoggerCallback(cublasLogCallback *userCallback) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasLogCallback *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetLoggerCallback");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(userCallback);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSetVector(int n, int elemSize, const void *x,
+                                            int incx, void *devicePtr,
+                                            int incy) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(int, int, const void *, int, void *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetVector");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(n, elemSize, x, incx, devicePtr, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetVector(int n, int elemSize, const void *x,
+                                            int incx, void *y, int incy) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(int, int, const void *, int, void *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetVector");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(n, elemSize, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSetMatrix(int rows, int cols, int elemSize,
+                                            const void *A, int lda, void *B,
+                                            int ldb) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(int, int, int, const void *,
+                                                 int, void *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetMatrix");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rows, cols, elemSize, A, lda, B, ldb);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetMatrix(int rows, int cols, int elemSize,
+                                            const void *A, int lda, void *B,
+                                            int ldb) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(int, int, int, const void *,
+                                                 int, void *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetMatrix");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rows, cols, elemSize, A, lda, B, ldb);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSetVectorAsync(int n, int elemSize,
+                                                 const void *hostPtr, int incx,
+                                                 void *devicePtr, int incy,
+                                                 cudaStream_t stream) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(int, int, const void *, int,
+                                                 void *, int, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetVectorAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(n, elemSize, hostPtr, incx, devicePtr, incy, stream);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetVectorAsync(int n, int elemSize,
+                                                 const void *devicePtr,
+                                                 int incx, void *hostPtr,
+                                                 int incy,
+                                                 cudaStream_t stream) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(int, int, const void *, int,
+                                                 void *, int, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetVectorAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(n, elemSize, devicePtr, incx, hostPtr, incy, stream);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSetMatrixAsync(int rows, int cols,
+                                                 int elemSize, const void *A,
+                                                 int lda, void *B, int ldb,
+                                                 cudaStream_t stream) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      int, int, int, const void *, int, void *, int, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetMatrixAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rows, cols, elemSize, A, lda, B, ldb, stream);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGetMatrixAsync(int rows, int cols,
+                                                 int elemSize, const void *A,
+                                                 int lda, void *B, int ldb,
+                                                 cudaStream_t stream) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      int, int, int, const void *, int, void *, int, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetMatrixAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rows, cols, elemSize, A, lda, B, ldb, stream);
+}
+
+void CUBLASWINAPI cublasXerbla(const char *srName, int info) {
+  using FuncPtr = void(CUBLASWINAPI *)(const char *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasXerbla");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasXerbla");
+  return func_ptr(srName, info);
+}
+
+cublasStatus_t CUBLASWINAPI cublasNrm2Ex(cublasHandle_t handle, int n,
+                                         const void *x, cudaDataType xType,
+                                         int incx, void *result,
+                                         cudaDataType resultType,
+                                         cudaDataType executionType) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const void *, cudaDataType, int, void *,
+      cudaDataType, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasNrm2Ex");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, xType, incx, result, resultType, executionType);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSnrm2_v2(cublasHandle_t handle, int n,
+                                           const float *x, int incx,
+                                           float *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const float *, int, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSnrm2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDnrm2_v2(cublasHandle_t handle, int n,
+                                           const double *x, int incx,
+                                           double *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const double *, int, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDnrm2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasScnrm2_v2(cublasHandle_t handle, int n,
+                                            const cuComplex *x, int incx,
+                                            float *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuComplex *, int, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasScnrm2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDznrm2_v2(cublasHandle_t handle, int n,
+                                            const cuDoubleComplex *x, int incx,
+                                            double *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuDoubleComplex *, int, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDznrm2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDotEx(cublasHandle_t handle, int n,
+                                        const void *x, cudaDataType xType,
+                                        int incx, const void *y,
+                                        cudaDataType yType, int incy,
+                                        void *result, cudaDataType resultType,
+                                        cudaDataType executionType) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const void *, cudaDataType, int, const void *,
+      cudaDataType, int, void *, cudaDataType, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDotEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, xType, incx, y, yType, incy, result, resultType,
+                  executionType);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDotcEx(cublasHandle_t handle, int n,
+                                         const void *x, cudaDataType xType,
+                                         int incx, const void *y,
+                                         cudaDataType yType, int incy,
+                                         void *result, cudaDataType resultType,
+                                         cudaDataType executionType) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const void *, cudaDataType, int, const void *,
+      cudaDataType, int, void *, cudaDataType, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDotcEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, xType, incx, y, yType, incy, result, resultType,
+                  executionType);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSdot_v2(cublasHandle_t handle, int n,
+                                          const float *x, int incx,
+                                          const float *y, int incy,
+                                          float *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const float *, int, const float *, int, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSdot_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDdot_v2(cublasHandle_t handle, int n,
+                                          const double *x, int incx,
+                                          const double *y, int incy,
+                                          double *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const double *, int, const double *, int, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDdot_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCdotu_v2(cublasHandle_t handle, int n,
+                                           const cuComplex *x, int incx,
+                                           const cuComplex *y, int incy,
+                                           cuComplex *result) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, const cuComplex *,
+                                     int, const cuComplex *, int, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCdotu_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCdotc_v2(cublasHandle_t handle, int n,
+                                           const cuComplex *x, int incx,
+                                           const cuComplex *y, int incy,
+                                           cuComplex *result) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, const cuComplex *,
+                                     int, const cuComplex *, int, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCdotc_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZdotu_v2(cublasHandle_t handle, int n,
+                                           const cuDoubleComplex *x, int incx,
+                                           const cuDoubleComplex *y, int incy,
+                                           cuDoubleComplex *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZdotu_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZdotc_v2(cublasHandle_t handle, int n,
+                                           const cuDoubleComplex *x, int incx,
+                                           const cuDoubleComplex *y, int incy,
+                                           cuDoubleComplex *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZdotc_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, result);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasScalEx(cublasHandle_t handle, int n,
+             const void *alpha, /* host or device pointer */
+             cudaDataType alphaType, void *x, cudaDataType xType, int incx,
+             cudaDataType executionType) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const void *, cudaDataType, void *, cudaDataType,
+      int, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasScalEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, alphaType, x, xType, incx, executionType);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSscal_v2(cublasHandle_t handle, int n,
+               const float *alpha, /* host or device pointer */
+               float *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSscal_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDscal_v2(cublasHandle_t handle, int n,
+               const double *alpha, /* host or device pointer */
+               double *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDscal_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCscal_v2(cublasHandle_t handle, int n,
+               const cuComplex *alpha, /* host or device pointer */
+               cuComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuComplex *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCscal_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCsscal_v2(cublasHandle_t handle, int n,
+                const float *alpha, /* host or device pointer */
+                cuComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const float *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsscal_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZscal_v2(cublasHandle_t handle, int n,
+               const cuDoubleComplex *alpha, /* host or device pointer */
+               cuDoubleComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZscal_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZdscal_v2(cublasHandle_t handle, int n,
+                const double *alpha, /* host or device pointer */
+                cuDoubleComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const double *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZdscal_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasAxpyEx(
+    cublasHandle_t handle, int n,
+    const void *alpha, /* host or device pointer */
+    cudaDataType alphaType, const void *x, cudaDataType xType, int incx,
+    void *y, cudaDataType yType, int incy, cudaDataType executiontype) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const void *, cudaDataType, const void *,
+      cudaDataType, int, void *, cudaDataType, int, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasAxpyEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, alphaType, x, xType, incx, y, yType, incy,
+                  executiontype);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSaxpy_v2(cublasHandle_t handle, int n,
+               const float *alpha, /* host or device pointer */
+               const float *x, int incx, float *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const float *, const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSaxpy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDaxpy_v2(cublasHandle_t handle, int n,
+               const double *alpha, /* host or device pointer */
+               const double *x, int incx, double *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const double *, const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDaxpy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCaxpy_v2(cublasHandle_t handle, int n,
+               const cuComplex *alpha, /* host or device pointer */
+               const cuComplex *x, int incx, cuComplex *y, int incy) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, const cuComplex *,
+                                     const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCaxpy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZaxpy_v2(
+    cublasHandle_t handle, int n,
+    const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *x, int incx, cuDoubleComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuDoubleComplex *, const cuDoubleComplex *,
+      int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZaxpy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, alpha, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCopyEx(cublasHandle_t handle, int n,
+                                         const void *x, cudaDataType xType,
+                                         int incx, void *y, cudaDataType yType,
+                                         int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const void *, cudaDataType, int, void *,
+      cudaDataType, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCopyEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, xType, incx, y, yType, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasScopy_v2(cublasHandle_t handle, int n,
+                                           const float *x, int incx, float *y,
+                                           int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasScopy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDcopy_v2(cublasHandle_t handle, int n,
+                                           const double *x, int incx, double *y,
+                                           int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDcopy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCcopy_v2(cublasHandle_t handle, int n,
+                                           const cuComplex *x, int incx,
+                                           cuComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCcopy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZcopy_v2(cublasHandle_t handle, int n,
+                                           const cuDoubleComplex *x, int incx,
+                                           cuDoubleComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const cuDoubleComplex *, int,
+                                                 cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZcopy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSswap_v2(cublasHandle_t handle, int n,
+                                           float *x, int incx, float *y,
+                                           int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, float *,
+                                                 int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSswap_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDswap_v2(cublasHandle_t handle, int n,
+                                           double *x, int incx, double *y,
+                                           int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, double *,
+                                                 int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDswap_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCswap_v2(cublasHandle_t handle, int n,
+                                           cuComplex *x, int incx, cuComplex *y,
+                                           int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCswap_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZswap_v2(cublasHandle_t handle, int n,
+                                           cuDoubleComplex *x, int incx,
+                                           cuDoubleComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZswap_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSwapEx(cublasHandle_t handle, int n, void *x,
+                                         cudaDataType xType, int incx, void *y,
+                                         cudaDataType yType, int incy) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, void *, cudaDataType,
+                                     int, void *, cudaDataType, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSwapEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, xType, incx, y, yType, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasIsamax_v2(cublasHandle_t handle, int n,
+                                            const float *x, int incx,
+                                            int *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const float *, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIsamax_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasIdamax_v2(cublasHandle_t handle, int n,
+                                            const double *x, int incx,
+                                            int *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const double *, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIdamax_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasIcamax_v2(cublasHandle_t handle, int n,
+                                            const cuComplex *x, int incx,
+                                            int *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const cuComplex *, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIcamax_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasIzamax_v2(cublasHandle_t handle, int n,
+                                            const cuDoubleComplex *x, int incx,
+                                            int *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuDoubleComplex *, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIzamax_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasIamaxEx(
+    cublasHandle_t handle, int n, const void *x, cudaDataType xType, int incx,
+    int *result /* host or device pointer */
+) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const void *, cudaDataType, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIamaxEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, xType, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasIsamin_v2(cublasHandle_t handle, int n,
+                                            const float *x, int incx,
+                                            int *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const float *, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIsamin_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasIdamin_v2(cublasHandle_t handle, int n,
+                                            const double *x, int incx,
+                                            int *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const double *, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIdamin_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasIcamin_v2(cublasHandle_t handle, int n,
+                                            const cuComplex *x, int incx,
+                                            int *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const cuComplex *, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIcamin_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasIzamin_v2(cublasHandle_t handle, int n,
+                                            const cuDoubleComplex *x, int incx,
+                                            int *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuDoubleComplex *, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIzamin_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasIaminEx(
+    cublasHandle_t handle, int n, const void *x, cudaDataType xType, int incx,
+    int *result /* host or device pointer */
+) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const void *, cudaDataType, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIaminEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, xType, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasAsumEx(
+    cublasHandle_t handle, int n, const void *x, cudaDataType xType, int incx,
+    void *result, cudaDataType resultType, /* host or device pointer */
+    cudaDataType executiontype) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const void *, cudaDataType, int, void *,
+      cudaDataType, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasAsumEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, xType, incx, result, resultType, executiontype);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSasum_v2(cublasHandle_t handle, int n,
+                                           const float *x, int incx,
+                                           float *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const float *, int, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSasum_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDasum_v2(cublasHandle_t handle, int n,
+                                           const double *x, int incx,
+                                           double *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int,
+                                                 const double *, int, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDasum_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasScasum_v2(cublasHandle_t handle, int n,
+                                            const cuComplex *x, int incx,
+                                            float *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuComplex *, int, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasScasum_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDzasum_v2(cublasHandle_t handle, int n,
+                                            const cuDoubleComplex *x, int incx,
+                                            double *result) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuDoubleComplex *, int, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDzasum_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, result);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSrot_v2(cublasHandle_t handle, int n, float *x, int incx, float *y,
+              int incy, const float *c, /* host or device pointer */
+              const float *s) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, float *, int, float *,
+                                     int, const float *, const float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSrot_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, c, s);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDrot_v2(cublasHandle_t handle, int n, double *x, int incx, double *y,
+              int incy, const double *c, /* host or device pointer */
+              const double *s) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, double *, int, double *, int, const double *,
+      const double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDrot_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, c, s);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCrot_v2(
+    cublasHandle_t handle, int n, cuComplex *x, int incx, cuComplex *y,
+    int incy, const float *c, /* host or device pointer */
+    const cuComplex *s) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, cuComplex *, int, cuComplex *, int, const float *,
+      const cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCrot_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, c, s);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCsrot_v2(
+    cublasHandle_t handle, int n, cuComplex *x, int incx, cuComplex *y,
+    int incy, const float *c, /* host or device pointer */
+    const float *s) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, cuComplex *, int, cuComplex *, int, const float *,
+      const float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsrot_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, c, s);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZrot_v2(
+    cublasHandle_t handle, int n, cuDoubleComplex *x, int incx,
+    cuDoubleComplex *y, int incy, const double *c, /* host or device pointer */
+    const cuDoubleComplex *s) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, cuDoubleComplex *, int, cuDoubleComplex *, int,
+      const double *, const cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZrot_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, c, s);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZdrot_v2(
+    cublasHandle_t handle, int n, cuDoubleComplex *x, int incx,
+    cuDoubleComplex *y, int incy, const double *c, /* host or device pointer */
+    const double *s) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, cuDoubleComplex *, int, cuDoubleComplex *, int,
+      const double *, const double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZdrot_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, c, s);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasRotEx(cublasHandle_t handle, int n, void *x, cudaDataType xType, int incx,
+            void *y, cudaDataType yType, int incy,
+            const void *c, /* host or device pointer */
+            const void *s, cudaDataType csType, cudaDataType executiontype) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, void *, cudaDataType, int, void *, cudaDataType, int,
+      const void *, const void *, cudaDataType, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasRotEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, xType, incx, y, yType, incy, c, s, csType,
+                  executiontype);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSrotg_v2(cublasHandle_t handle, float *a, /* host or device pointer */
+               float *b,                        /* host or device pointer */
+               float *c,                        /* host or device pointer */
+               float *s) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, float *,
+                                                 float *, float *, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSrotg_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, a, b, c, s);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDrotg_v2(cublasHandle_t handle, double *a, /* host or device pointer */
+               double *b,                        /* host or device pointer */
+               double *c,                        /* host or device pointer */
+               double *s) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, double *,
+                                                 double *, double *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDrotg_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, a, b, c, s);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCrotg_v2(cublasHandle_t handle, cuComplex *a, /* host or device pointer */
+               cuComplex *b,                        /* host or device pointer */
+               float *c,                            /* host or device pointer */
+               cuComplex *s) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cuComplex *, cuComplex *, float *, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCrotg_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, a, b, c, s);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZrotg_v2(
+    cublasHandle_t handle, cuDoubleComplex *a, /* host or device pointer */
+    cuDoubleComplex *b,                        /* host or device pointer */
+    double *c,                                 /* host or device pointer */
+    cuDoubleComplex *s) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cuDoubleComplex *, cuDoubleComplex *, double *,
+      cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZrotg_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, a, b, c, s);
+}
+
+cublasStatus_t CUBLASWINAPI cublasRotgEx(cublasHandle_t handle,
+                                         void *a, /* host or device pointer */
+                                         void *b, /* host or device pointer */
+                                         cudaDataType abType,
+                                         void *c, /* host or device pointer */
+                                         void *s, /* host or device pointer */
+                                         cudaDataType csType,
+                                         cudaDataType executiontype) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, void *, void *,
+                                                 cudaDataType, void *, void *,
+                                                 cudaDataType, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasRotgEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, a, b, abType, c, s, csType, executiontype);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSrotm_v2(cublasHandle_t handle, int n,
+                                           float *x, int incx, float *y,
+                                           int incy, const float *param) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, float *, int, float *, int, const float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSrotm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, param);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDrotm_v2(cublasHandle_t handle, int n,
+                                           double *x, int incx, double *y,
+                                           int incy, const double *param) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, double *, int, double *, int, const double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDrotm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, incx, y, incy, param);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasRotmEx(cublasHandle_t handle, int n, void *x, cudaDataType xType,
+             int incx, void *y, cudaDataType yType, int incy,
+             const void *param, /* host or device pointer */
+             cudaDataType paramType, cudaDataType executiontype) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, void *, cudaDataType, int, void *, cudaDataType, int,
+      const void *, cudaDataType, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasRotmEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, x, xType, incx, y, yType, incy, param, paramType,
+                  executiontype);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSrotmg_v2(cublasHandle_t handle, float *d1, /* host or device pointer */
+                float *d2,                        /* host or device pointer */
+                float *x1,                        /* host or device pointer */
+                const float *y1,                  /* host or device pointer */
+                float *param) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, float *, float *, float *, const float *, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSrotmg_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, d1, d2, x1, y1, param);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDrotmg_v2(cublasHandle_t handle, double *d1, /* host or device pointer */
+                double *d2,                        /* host or device pointer */
+                double *x1,                        /* host or device pointer */
+                const double *y1,                  /* host or device pointer */
+                double *param) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, double *, double *, double *, const double *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDrotmg_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, d1, d2, x1, y1, param);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasRotmgEx(cublasHandle_t handle, void *d1,     /* host or device pointer */
+              cudaDataType d1Type, void *d2,       /* host or device pointer */
+              cudaDataType d2Type, void *x1,       /* host or device pointer */
+              cudaDataType x1Type, const void *y1, /* host or device pointer */
+              cudaDataType y1Type, void *param,    /* host or device pointer */
+              cudaDataType paramType, cudaDataType executiontype) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, void *, cudaDataType, void *, cudaDataType, void *,
+      cudaDataType, const void *, cudaDataType, void *, cudaDataType,
+      cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasRotmgEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, d1, d1Type, d2, d2Type, x1, x1Type, y1, y1Type, param,
+                  paramType, executiontype);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSgemv_v2(cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+               const float *alpha, /* host or device pointer */
+               const float *A, int lda, const float *x, int incx,
+               const float *beta, /* host or device pointer */
+               float *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const float *, const float *,
+      int, const float *, int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgemv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDgemv_v2(cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+               const double *alpha, /* host or device pointer */
+               const double *A, int lda, const double *x, int incx,
+               const double *beta, /* host or device pointer */
+               double *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const double *,
+      const double *, int, const double *, int, const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgemv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCgemv_v2(cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+               const cuComplex *alpha, /* host or device pointer */
+               const cuComplex *A, int lda, const cuComplex *x, int incx,
+               const cuComplex *beta, /* host or device pointer */
+               cuComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const cuComplex *,
+      const cuComplex *, int, const cuComplex *, int, const cuComplex *,
+      cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZgemv_v2(
+    cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+    const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *x, int incx,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgemv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSgbmv_v2(cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+               int kl, int ku, const float *alpha, /* host or device pointer */
+               const float *A, int lda, const float *x, int incx,
+               const float *beta, /* host or device pointer */
+               float *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, int, int, const float *,
+      const float *, int, const float *, int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, kl, ku, alpha, A, lda, x, incx, beta, y,
+                  incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDgbmv_v2(cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+               int kl, int ku, const double *alpha, /* host or device pointer */
+               const double *A, int lda, const double *x, int incx,
+               const double *beta, /* host or device pointer */
+               double *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, int, int, const double *,
+      const double *, int, const double *, int, const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, kl, ku, alpha, A, lda, x, incx, beta, y,
+                  incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgbmv_v2(
+    cublasHandle_t handle, cublasOperation_t trans, int m, int n, int kl,
+    int ku, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, const cuComplex *x, int incx,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, int, int, const cuComplex *,
+      const cuComplex *, int, const cuComplex *, int, const cuComplex *,
+      cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, kl, ku, alpha, A, lda, x, incx, beta, y,
+                  incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZgbmv_v2(
+    cublasHandle_t handle, cublasOperation_t trans, int m, int n, int kl,
+    int ku, const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *x, int incx,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, cuDoubleComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, kl, ku, alpha, A, lda, x, incx, beta, y,
+                  incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStrmv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    cublasDiagType_t diag, int n, const float *A, int lda, float *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStrmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtrmv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n,
+                                           const double *A, int lda, double *x,
+                                           int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtrmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtrmv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n,
+                                           const cuComplex *A, int lda,
+                                           cuComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtrmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtrmv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n,
+                                           const cuDoubleComplex *A, int lda,
+                                           cuDoubleComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtrmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStbmv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n, int k,
+                                           const float *A, int lda, float *x,
+                                           int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, int, const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtbmv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n, int k,
+                                           const double *A, int lda, double *x,
+                                           int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, int, const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtbmv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n, int k,
+                                           const cuComplex *A, int lda,
+                                           cuComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, int, const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtbmv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n, int k,
+                                           const cuDoubleComplex *A, int lda,
+                                           cuDoubleComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, int, const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStpmv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    cublasDiagType_t diag, int n, const float *AP, float *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStpmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, AP, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtpmv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    cublasDiagType_t diag, int n, const double *AP, double *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtpmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, AP, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtpmv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    cublasDiagType_t diag, int n, const cuComplex *AP, cuComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const cuComplex *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtpmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, AP, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtpmv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n,
+                                           const cuDoubleComplex *AP,
+                                           cuDoubleComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtpmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, AP, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStrsv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    cublasDiagType_t diag, int n, const float *A, int lda, float *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStrsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtrsv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n,
+                                           const double *A, int lda, double *x,
+                                           int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtrsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtrsv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n,
+                                           const cuComplex *A, int lda,
+                                           cuComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtrsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtrsv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n,
+                                           const cuDoubleComplex *A, int lda,
+                                           cuDoubleComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtrsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStpsv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    cublasDiagType_t diag, int n, const float *AP, float *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStpsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, AP, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtpsv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    cublasDiagType_t diag, int n, const double *AP, double *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtpsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, AP, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtpsv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    cublasDiagType_t diag, int n, const cuComplex *AP, cuComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const cuComplex *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtpsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, AP, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtpsv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n,
+                                           const cuDoubleComplex *AP,
+                                           cuDoubleComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtpsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, AP, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStbsv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n, int k,
+                                           const float *A, int lda, float *x,
+                                           int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, int, const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStbsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtbsv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n, int k,
+                                           const double *A, int lda, double *x,
+                                           int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, int, const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtbsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtbsv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n, int k,
+                                           const cuComplex *A, int lda,
+                                           cuComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, int, const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtbsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtbsv_v2(cublasHandle_t handle,
+                                           cublasFillMode_t uplo,
+                                           cublasOperation_t trans,
+                                           cublasDiagType_t diag, int n, int k,
+                                           const cuDoubleComplex *A, int lda,
+                                           cuDoubleComplex *x, int incx) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, cublasDiagType_t,
+      int, int, const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtbsv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSsymv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const float *alpha, /* host or device pointer */
+               const float *A, int lda, const float *x, int incx,
+               const float *beta, /* host or device pointer */
+               float *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const float *, const float *, int,
+      const float *, int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsymv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDsymv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const double *alpha, /* host or device pointer */
+               const double *A, int lda, const double *x, int incx,
+               const double *beta, /* host or device pointer */
+               double *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const double *, const double *,
+      int, const double *, int, const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsymv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCsymv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const cuComplex *alpha, /* host or device pointer */
+               const cuComplex *A, int lda, const cuComplex *x, int incx,
+               const cuComplex *beta, /* host or device pointer */
+               cuComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuComplex *,
+      const cuComplex *, int, const cuComplex *, int, const cuComplex *,
+      cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsymv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZsymv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, int n,
+    const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *x, int incx,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZsymv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasChemv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const cuComplex *alpha, /* host or device pointer */
+               const cuComplex *A, int lda, const cuComplex *x, int incx,
+               const cuComplex *beta, /* host or device pointer */
+               cuComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuComplex *,
+      const cuComplex *, int, const cuComplex *, int, const cuComplex *,
+      cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChemv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZhemv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, int n,
+    const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *x, int incx,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhemv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSsbmv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n, int k,
+               const float *alpha, /* host or device pointer */
+               const float *A, int lda, const float *x, int incx,
+               const float *beta, /* host or device pointer */
+               float *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, int, const float *, const float *,
+      int, const float *, int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, k, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDsbmv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n, int k,
+               const double *alpha, /* host or device pointer */
+               const double *A, int lda, const double *x, int incx,
+               const double *beta, /* host or device pointer */
+               double *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, int, const double *,
+      const double *, int, const double *, int, const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, k, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasChbmv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n, int k,
+               const cuComplex *alpha, /* host or device pointer */
+               const cuComplex *A, int lda, const cuComplex *x, int incx,
+               const cuComplex *beta, /* host or device pointer */
+               cuComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, int, const cuComplex *,
+      const cuComplex *, int, const cuComplex *, int, const cuComplex *,
+      cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, k, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZhbmv_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, int n, int k,
+    const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *x, int incx,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhbmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, k, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSspmv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const float *alpha, /* host or device pointer */
+               const float *AP, const float *x, int incx,
+               const float *beta, /* host or device pointer */
+               float *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const float *, const float *,
+      const float *, int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSspmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, AP, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDspmv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const double *alpha, /* host or device pointer */
+               const double *AP, const double *x, int incx,
+               const double *beta, /* host or device pointer */
+               double *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const double *, const double *,
+      const double *, int, const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDspmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, AP, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasChpmv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const cuComplex *alpha, /* host or device pointer */
+               const cuComplex *AP, const cuComplex *x, int incx,
+               const cuComplex *beta, /* host or device pointer */
+               cuComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuComplex *,
+      const cuComplex *, const cuComplex *, int, const cuComplex *, cuComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChpmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, AP, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZhpmv_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const cuDoubleComplex *alpha, /* host or device pointer */
+               const cuDoubleComplex *AP, const cuDoubleComplex *x, int incx,
+               const cuDoubleComplex *beta, /* host or device pointer */
+               cuDoubleComplex *y, int incy) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhpmv_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, AP, x, incx, beta, y, incy);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSger_v2(
+    cublasHandle_t handle, int m, int n,
+    const float *alpha, /* host or device pointer */
+    const float *x, int incx, const float *y, int incy, float *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, int, const float *, const float *, int,
+      const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSger_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDger_v2(
+    cublasHandle_t handle, int m, int n,
+    const double *alpha, /* host or device pointer */
+    const double *x, int incx, const double *y, int incy, double *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, int, const double *, const double *, int,
+      const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDger_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCgeru_v2(cublasHandle_t handle, int m, int n,
+               const cuComplex *alpha, /* host or device pointer */
+               const cuComplex *x, int incx, const cuComplex *y, int incy,
+               cuComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, int, const cuComplex *, const cuComplex *, int,
+      const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgeru_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCgerc_v2(cublasHandle_t handle, int m, int n,
+               const cuComplex *alpha, /* host or device pointer */
+               const cuComplex *x, int incx, const cuComplex *y, int incy,
+               cuComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, int, const cuComplex *, const cuComplex *, int,
+      const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgerc_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZgeru_v2(cublasHandle_t handle, int m, int n,
+               const cuDoubleComplex *alpha, /* host or device pointer */
+               const cuDoubleComplex *x, int incx, const cuDoubleComplex *y,
+               int incy, cuDoubleComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, int,
+      cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgeru_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZgerc_v2(cublasHandle_t handle, int m, int n,
+               const cuDoubleComplex *alpha, /* host or device pointer */
+               const cuDoubleComplex *x, int incx, const cuDoubleComplex *y,
+               int incy, cuDoubleComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, int,
+      cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgerc_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSsyr_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+              const float *alpha, /* host or device pointer */
+              const float *x, int incx, float *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const float *, const float *, int,
+      float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsyr_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDsyr_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+              const double *alpha, /* host or device pointer */
+              const double *x, int incx, double *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const double *, const double *,
+      int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsyr_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCsyr_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+              const cuComplex *alpha, /* host or device pointer */
+              const cuComplex *x, int incx, cuComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuComplex *,
+      const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsyr_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZsyr_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+              const cuDoubleComplex *alpha, /* host or device pointer */
+              const cuDoubleComplex *x, int incx, cuDoubleComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZsyr_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCher_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+              const float *alpha, /* host or device pointer */
+              const cuComplex *x, int incx, cuComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const float *, const cuComplex *,
+      int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCher_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZher_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+              const double *alpha, /* host or device pointer */
+              const cuDoubleComplex *x, int incx, cuDoubleComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const double *,
+      const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZher_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSspr_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+              const float *alpha, /* host or device pointer */
+              const float *x, int incx, float *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const float *, const float *, int,
+      float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSspr_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, AP);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDspr_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+              const double *alpha, /* host or device pointer */
+              const double *x, int incx, double *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const double *, const double *,
+      int, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDspr_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, AP);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasChpr_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+              const float *alpha, /* host or device pointer */
+              const cuComplex *x, int incx, cuComplex *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const float *, const cuComplex *,
+      int, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChpr_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, AP);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZhpr_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+              const double *alpha, /* host or device pointer */
+              const cuDoubleComplex *x, int incx, cuDoubleComplex *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const double *,
+      const cuDoubleComplex *, int, cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhpr_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, AP);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSsyr2_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, int n,
+    const float *alpha, /* host or device pointer */
+    const float *x, int incx, const float *y, int incy, float *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const float *, const float *, int,
+      const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsyr2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDsyr2_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, int n,
+    const double *alpha, /* host or device pointer */
+    const double *x, int incx, const double *y, int incy, double *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const double *, const double *,
+      int, const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsyr2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCsyr2_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const cuComplex *alpha, /* host or device pointer */
+               const cuComplex *x, int incx, const cuComplex *y, int incy,
+               cuComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuComplex *,
+      const cuComplex *, int, const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsyr2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZsyr2_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const cuDoubleComplex *alpha, /* host or device pointer */
+               const cuDoubleComplex *x, int incx, const cuDoubleComplex *y,
+               int incy, cuDoubleComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, int,
+      cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZsyr2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCher2_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const cuComplex *alpha, /* host or device pointer */
+               const cuComplex *x, int incx, const cuComplex *y, int incy,
+               cuComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuComplex *,
+      const cuComplex *, int, const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCher2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZher2_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const cuDoubleComplex *alpha, /* host or device pointer */
+               const cuDoubleComplex *x, int incx, const cuDoubleComplex *y,
+               int incy, cuDoubleComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, int,
+      cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZher2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, y, incy, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSspr2_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const float *alpha, /* host or device pointer */
+               const float *x, int incx, const float *y, int incy, float *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const float *, const float *, int,
+      const float *, int, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSspr2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, y, incy, AP);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDspr2_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, int n,
+    const double *alpha, /* host or device pointer */
+    const double *x, int incx, const double *y, int incy, double *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const double *, const double *,
+      int, const double *, int, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDspr2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, y, incy, AP);
+}
+
+cublasStatus_t CUBLASWINAPI cublasChpr2_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, int n,
+    const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *x, int incx, const cuComplex *y, int incy, cuComplex *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuComplex *,
+      const cuComplex *, int, const cuComplex *, int, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChpr2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, y, incy, AP);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZhpr2_v2(cublasHandle_t handle, cublasFillMode_t uplo, int n,
+               const cuDoubleComplex *alpha, /* host or device pointer */
+               const cuDoubleComplex *x, int incx, const cuDoubleComplex *y,
+               int incy, cuDoubleComplex *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, int,
+      cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhpr2_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, alpha, x, incx, y, incy, AP);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSgemvBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+    const float *alpha, /* host or device pointer */
+    const float *const Aarray[], int lda, const float *const xarray[], int incx,
+    const float *beta, /* host or device pointer */
+    float *const yarray[], int incy, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const float *,
+      const float *const[], int, const float *const[], int, const float *,
+      float *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgemvBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, Aarray, lda, xarray, incx, beta,
+                  yarray, incy, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDgemvBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+    const double *alpha, /* host or device pointer */
+    const double *const Aarray[], int lda, const double *const xarray[],
+    int incx, const double *beta, /* host or device pointer */
+    double *const yarray[], int incy, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const double *,
+      const double *const[], int, const double *const[], int, const double *,
+      double *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgemvBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, Aarray, lda, xarray, incx, beta,
+                  yarray, incy, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgemvBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+    const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *const Aarray[], int lda, const cuComplex *const xarray[],
+    int incx, const cuComplex *beta, /* host or device pointer */
+    cuComplex *const yarray[], int incy, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const cuComplex *,
+      const cuComplex *const[], int, const cuComplex *const[], int,
+      const cuComplex *, cuComplex *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemvBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, Aarray, lda, xarray, incx, beta,
+                  yarray, incy, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZgemvBatched(cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+                   const cuDoubleComplex *alpha, /* host or device pointer */
+                   const cuDoubleComplex *const Aarray[], int lda,
+                   const cuDoubleComplex *const xarray[], int incx,
+                   const cuDoubleComplex *beta, /* host or device pointer */
+                   cuDoubleComplex *const yarray[], int incy, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *const[], int, const cuDoubleComplex *const[], int,
+      const cuDoubleComplex *, cuDoubleComplex *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgemvBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, Aarray, lda, xarray, incx, beta,
+                  yarray, incy, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSgemvStridedBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+    const float *alpha,                             /* host or device pointer */
+    const float *A, int lda, long long int strideA, /* purposely signed */
+    const float *x, int incx, long long int stridex,
+    const float *beta, /* host or device pointer */
+    float *y, int incy, long long int stridey, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const float *, const float *,
+      int, long long, const float *, int, long long, const float *, float *,
+      int, long long, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgemvStridedBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, A, lda, strideA, x, incx, stridex,
+                  beta, y, incy, stridey, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDgemvStridedBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+    const double *alpha, /* host or device pointer */
+    const double *A, int lda, long long int strideA, /* purposely signed */
+    const double *x, int incx, long long int stridex,
+    const double *beta, /* host or device pointer */
+    double *y, int incy, long long int stridey, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const double *,
+      const double *, int, long long, const double *, int, long long,
+      const double *, double *, int, long long, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgemvStridedBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, A, lda, strideA, x, incx, stridex,
+                  beta, y, incy, stridey, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgemvStridedBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+    const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, long long int strideA, /* purposely signed */
+    const cuComplex *x, int incx, long long int stridex,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *y, int incy, long long int stridey, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const cuComplex *,
+      const cuComplex *, int, long long, const cuComplex *, int, long long,
+      const cuComplex *, cuComplex *, int, long long, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemvStridedBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, A, lda, strideA, x, incx, stridex,
+                  beta, y, incy, stridey, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZgemvStridedBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+    const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda,
+    long long int strideA, /* purposely signed */
+    const cuDoubleComplex *x, int incx, long long int stridex,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *y, int incy, long long int stridey, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, long long, const cuDoubleComplex *, int,
+      long long, const cuDoubleComplex *, cuDoubleComplex *, int, long long,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgemvStridedBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, alpha, A, lda, strideA, x, incx, stridex,
+                  beta, y, incy, stridey, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSgemm_v2(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const float *alpha, /* host or device pointer */
+    const float *A, int lda, const float *B, int ldb,
+    const float *beta, /* host or device pointer */
+    float *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const float *, const float *, int, const float *, int, const float *,
+      float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgemm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, B, ldb, beta,
+                  C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDgemm_v2(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const double *alpha, /* host or device pointer */
+    const double *A, int lda, const double *B, int ldb,
+    const double *beta, /* host or device pointer */
+    double *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const double *, const double *, int, const double *, int, const double *,
+      double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgemm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, B, ldb, beta,
+                  C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgemm_v2(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, const cuComplex *B, int ldb,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuComplex *, const cuComplex *, int, const cuComplex *, int,
+      const cuComplex *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, B, ldb, beta,
+                  C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgemm3m(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, const cuComplex *B, int ldb,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuComplex *, const cuComplex *, int, const cuComplex *, int,
+      const cuComplex *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemm3m");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, B, ldb, beta,
+                  C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgemm3mEx(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const cuComplex *alpha, const void *A,
+    cudaDataType Atype, int lda, const void *B, cudaDataType Btype, int ldb,
+    const cuComplex *beta, void *C, cudaDataType Ctype, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuComplex *, const void *, cudaDataType, int, const void *,
+      cudaDataType, int, const cuComplex *, void *, cudaDataType, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemm3mEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, Atype, lda, B,
+                  Btype, ldb, beta, C, Ctype, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZgemm_v2(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k,
+    const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, cuDoubleComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgemm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, B, ldb, beta,
+                  C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZgemm3m(cublasHandle_t handle, cublasOperation_t transa,
+              cublasOperation_t transb, int m, int n, int k,
+              const cuDoubleComplex *alpha, /* host or device pointer */
+              const cuDoubleComplex *A, int lda, const cuDoubleComplex *B,
+              int ldb, const cuDoubleComplex *beta, /* host or device pointer */
+              cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, cuDoubleComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgemm3m");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, B, ldb, beta,
+                  C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSgemmEx(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const float *alpha, /* host or device pointer */
+    const void *A, cudaDataType Atype, int lda, const void *B,
+    cudaDataType Btype, int ldb, const float *beta, /* host or device pointer */
+    void *C, cudaDataType Ctype, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const float *, const void *, cudaDataType, int, const void *,
+      cudaDataType, int, const float *, void *, cudaDataType, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgemmEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, Atype, lda, B,
+                  Btype, ldb, beta, C, Ctype, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGemmEx(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const void *alpha, /* host or device pointer */
+    const void *A, cudaDataType Atype, int lda, const void *B,
+    cudaDataType Btype, int ldb, const void *beta, /* host or device pointer */
+    void *C, cudaDataType Ctype, int ldc, cudaDataType computeType,
+    cublasGemmAlgo_t algo) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const void *, const void *, cudaDataType, int, const void *, cudaDataType,
+      int, const void *, void *, cudaDataType, int, cublasComputeType_t,
+      cublasGemmAlgo_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGemmEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  cublasComputeType_t migratedComputeType = CUBLAS_COMPUTE_32F;
+  cublasMigrateComputeType(handle, computeType, &migratedComputeType);
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, Atype, lda, B,
+                  Btype, ldb, beta, C, Ctype, ldc, migratedComputeType, algo);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgemmEx(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const cuComplex *alpha, const void *A,
+    cudaDataType Atype, int lda, const void *B, cudaDataType Btype, int ldb,
+    const cuComplex *beta, void *C, cudaDataType Ctype, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuComplex *, const void *, cudaDataType, int, const void *,
+      cudaDataType, int, const cuComplex *, void *, cudaDataType, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemmEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, Atype, lda, B,
+                  Btype, ldb, beta, C, Ctype, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasUint8gemmBias(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    cublasOperation_t transc, int m, int n, int k, const unsigned char *A,
+    int A_bias, int lda, const unsigned char *B, int B_bias, int ldb,
+    unsigned char *C, int C_bias, int ldc, int C_mult, int C_shift) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, cublasOperation_t,
+      int, int, int, const unsigned char *, int, int, const unsigned char *,
+      int, int, unsigned char *, int, int, int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasUint8gemmBias");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, transc, m, n, k, A, A_bias, lda, B,
+                  B_bias, ldb, C, C_bias, ldc, C_mult, C_shift);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSsyrk_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const float *alpha,           /* host or device pointer */
+    const float *A, int lda, const float *beta, /* host or device pointer */
+    float *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const float *, const float *, int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsyrk_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDsyrk_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const double *alpha,            /* host or device pointer */
+    const double *A, int lda, const double *beta, /* host or device pointer */
+    double *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const double *, const double *, int, const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsyrk_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCsyrk_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuComplex *, const cuComplex *, int, const cuComplex *, cuComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsyrk_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZsyrk_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZsyrk_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCsyrkEx(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const void *A, cudaDataType Atype, int lda,
+    const cuComplex *beta, /* host or device pointer */
+    void *C, cudaDataType Ctype, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuComplex *, const void *, cudaDataType, int, const cuComplex *,
+      void *, cudaDataType, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsyrkEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, Atype, lda, beta, C,
+                  Ctype, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCsyrk3mEx(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuComplex *alpha, const void *A, cudaDataType Atype,
+    int lda, const cuComplex *beta, void *C, cudaDataType Ctype, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuComplex *, const void *, cudaDataType, int, const cuComplex *,
+      void *, cudaDataType, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsyrk3mEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, Atype, lda, beta, C,
+                  Ctype, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCherk_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const float *alpha,               /* host or device pointer */
+    const cuComplex *A, int lda, const float *beta, /* host or device pointer */
+    cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const float *, const cuComplex *, int, const float *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCherk_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZherk_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const double *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda,
+    const double *beta, /* host or device pointer */
+    cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const double *, const cuDoubleComplex *, int, const double *,
+      cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZherk_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCherkEx(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const float *alpha, /* host or device pointer */
+    const void *A, cudaDataType Atype, int lda,
+    const float *beta, /* host or device pointer */
+    void *C, cudaDataType Ctype, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const float *, const void *, cudaDataType, int, const float *, void *,
+      cudaDataType, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCherkEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, Atype, lda, beta, C,
+                  Ctype, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCherk3mEx(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const float *alpha, const void *A, cudaDataType Atype,
+    int lda, const float *beta, void *C, cudaDataType Ctype, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const float *, const void *, cudaDataType, int, const float *, void *,
+      cudaDataType, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCherk3mEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, Atype, lda, beta, C,
+                  Ctype, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSsyr2k_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const float *alpha, /* host or device pointer */
+    const float *A, int lda, const float *B, int ldb,
+    const float *beta, /* host or device pointer */
+    float *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const float *, const float *, int, const float *, int, const float *,
+      float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsyr2k_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDsyr2k_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const double *alpha, /* host or device pointer */
+    const double *A, int lda, const double *B, int ldb,
+    const double *beta, /* host or device pointer */
+    double *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const double *, const double *, int, const double *, int, const double *,
+      double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsyr2k_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCsyr2k_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, const cuComplex *B, int ldb,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuComplex *, const cuComplex *, int, const cuComplex *, int,
+      const cuComplex *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsyr2k_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZsyr2k_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, cuDoubleComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZsyr2k_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCher2k_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, const cuComplex *B, int ldb,
+    const float *beta, /* host or device pointer */
+    cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuComplex *, const cuComplex *, int, const cuComplex *, int,
+      const float *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCher2k_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZher2k_v2(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb,
+    const double *beta, /* host or device pointer */
+    cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, const double *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZher2k_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSsyrkx(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const float *alpha, /* host or device pointer */
+    const float *A, int lda, const float *B, int ldb,
+    const float *beta, /* host or device pointer */
+    float *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const float *, const float *, int, const float *, int, const float *,
+      float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsyrkx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDsyrkx(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const double *alpha, /* host or device pointer */
+    const double *A, int lda, const double *B, int ldb,
+    const double *beta, /* host or device pointer */
+    double *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const double *, const double *, int, const double *, int, const double *,
+      double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsyrkx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCsyrkx(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, const cuComplex *B, int ldb,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuComplex *, const cuComplex *, int, const cuComplex *, int,
+      const cuComplex *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsyrkx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZsyrkx(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, cuDoubleComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZsyrkx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCherkx(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, const cuComplex *B, int ldb,
+    const float *beta, /* host or device pointer */
+    cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuComplex *, const cuComplex *, int, const cuComplex *, int,
+      const float *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCherkx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZherkx(
+    cublasHandle_t handle, cublasFillMode_t uplo, cublasOperation_t trans,
+    int n, int k, const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb,
+    const double *beta, /* host or device pointer */
+    cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, cublasOperation_t, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, const double *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZherkx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSsymm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo, int m,
+    int n, const float *alpha, /* host or device pointer */
+    const float *A, int lda, const float *B, int ldb,
+    const float *beta, /* host or device pointer */
+    float *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, int, int,
+      const float *, const float *, int, const float *, int, const float *,
+      float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsymm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, m, n, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDsymm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo, int m,
+    int n, const double *alpha, /* host or device pointer */
+    const double *A, int lda, const double *B, int ldb,
+    const double *beta, /* host or device pointer */
+    double *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, int, int,
+      const double *, const double *, int, const double *, int, const double *,
+      double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsymm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, m, n, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCsymm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo, int m,
+    int n, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, const cuComplex *B, int ldb,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, int, int,
+      const cuComplex *, const cuComplex *, int, const cuComplex *, int,
+      const cuComplex *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsymm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, m, n, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZsymm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo, int m,
+    int n, const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, cuDoubleComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZsymm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, m, n, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasChemm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo, int m,
+    int n, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, const cuComplex *B, int ldb,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, int, int,
+      const cuComplex *, const cuComplex *, int, const cuComplex *, int,
+      const cuComplex *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChemm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, m, n, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZhemm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo, int m,
+    int n, const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, cuDoubleComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhemm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, m, n, alpha, A, lda, B, ldb, beta, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStrsm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const float *alpha, /* host or device pointer */
+    const float *A, int lda, float *B, int ldb) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const float *, const float *, int, float *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStrsm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtrsm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const double *alpha, /* host or device pointer */
+    const double *A, int lda, double *B, int ldb) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const double *, const double *, int, double *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtrsm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtrsm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, cuComplex *B, int ldb) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const cuComplex *, const cuComplex *, int,
+      cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtrsm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtrsm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, cuDoubleComplex *B, int ldb) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtrsm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStrmm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const float *alpha, /* host or device pointer */
+    const float *A, int lda, const float *B, int ldb, float *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const float *, const float *, int,
+      const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStrmm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb,
+                  C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtrmm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const double *alpha, /* host or device pointer */
+    const double *A, int lda, const double *B, int ldb, double *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const double *, const double *, int,
+      const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtrmm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb,
+                  C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtrmm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, const cuComplex *B, int ldb, cuComplex *C,
+    int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const cuComplex *, const cuComplex *, int,
+      const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtrmm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb,
+                  C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtrmm_v2(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb,
+    cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, const cuDoubleComplex *, int,
+      cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtrmm_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb,
+                  C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSgemmBatched(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const float *alpha, /* host or device pointer */
+    const float *const Aarray[], int lda, const float *const Barray[], int ldb,
+    const float *beta, /* host or device pointer */
+    float *const Carray[], int ldc, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const float *, const float *const[], int, const float *const[], int,
+      const float *, float *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgemmBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, Aarray, lda, Barray,
+                  ldb, beta, Carray, ldc, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDgemmBatched(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const double *alpha, /* host or device pointer */
+    const double *const Aarray[], int lda, const double *const Barray[],
+    int ldb, const double *beta, /* host or device pointer */
+    double *const Carray[], int ldc, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const double *, const double *const[], int, const double *const[], int,
+      const double *, double *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgemmBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, Aarray, lda, Barray,
+                  ldb, beta, Carray, ldc, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgemmBatched(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *const Aarray[], int lda, const cuComplex *const Barray[],
+    int ldb, const cuComplex *beta, /* host or device pointer */
+    cuComplex *const Carray[], int ldc, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuComplex *, const cuComplex *const[], int,
+      const cuComplex *const[], int, const cuComplex *, cuComplex *const[], int,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemmBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, Aarray, lda, Barray,
+                  ldb, beta, Carray, ldc, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgemm3mBatched(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *const Aarray[], int lda, const cuComplex *const Barray[],
+    int ldb, const cuComplex *beta, /* host or device pointer */
+    cuComplex *const Carray[], int ldc, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuComplex *, const cuComplex *const[], int,
+      const cuComplex *const[], int, const cuComplex *, cuComplex *const[], int,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemm3mBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, Aarray, lda, Barray,
+                  ldb, beta, Carray, ldc, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZgemmBatched(cublasHandle_t handle, cublasOperation_t transa,
+                   cublasOperation_t transb, int m, int n, int k,
+                   const cuDoubleComplex *alpha, /* host or device pointer */
+                   const cuDoubleComplex *const Aarray[], int lda,
+                   const cuDoubleComplex *const Barray[], int ldb,
+                   const cuDoubleComplex *beta, /* host or device pointer */
+                   cuDoubleComplex *const Carray[], int ldc, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *const[], int,
+      const cuDoubleComplex *const[], int, const cuDoubleComplex *,
+      cuDoubleComplex *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgemmBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, Aarray, lda, Barray,
+                  ldb, beta, Carray, ldc, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGemmBatchedEx(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const void *alpha, /* host or device pointer */
+    const void *const Aarray[], cudaDataType Atype, int lda,
+    const void *const Barray[], cudaDataType Btype, int ldb,
+    const void *beta, /* host or device pointer */
+    void *const Carray[], cudaDataType Ctype, int ldc, int batchCount,
+    cudaDataType computeType, cublasGemmAlgo_t algo) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const void *, const void *const[], cudaDataType, int, const void *const[],
+      cudaDataType, int, const void *, void *const[], cudaDataType, int, int,
+      cublasComputeType_t, cublasGemmAlgo_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGemmBatchedEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  cublasComputeType_t migratedComputeType = CUBLAS_COMPUTE_32F;
+  cublasMigrateComputeType(handle, computeType, &migratedComputeType);
+  return func_ptr(handle, transa, transb, m, n, k, alpha, Aarray, Atype, lda,
+                  Barray, Btype, ldb, beta, Carray, Ctype, ldc, batchCount,
+                  migratedComputeType, algo);
+}
+
+cublasStatus_t CUBLASWINAPI cublasGemmStridedBatchedEx(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const void *alpha, /* host or device pointer */
+    const void *A, cudaDataType Atype, int lda,
+    long long int strideA, /* purposely signed */
+    const void *B, cudaDataType Btype, int ldb, long long int strideB,
+    const void *beta, /* host or device pointer */
+    void *C, cudaDataType Ctype, int ldc, long long int strideC, int batchCount,
+    cudaDataType computeType, cublasGemmAlgo_t algo) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const void *, const void *, cudaDataType, int, long long, const void *,
+      cudaDataType, int, long long, const void *, void *, cudaDataType, int,
+      long long, int, cublasComputeType_t, cublasGemmAlgo_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGemmStridedBatchedEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  cublasComputeType_t migratedComputeType = CUBLAS_COMPUTE_32F;
+  cublasMigrateComputeType(handle, computeType, &migratedComputeType);
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, Atype, lda,
+                  strideA, B, Btype, ldb, strideB, beta, C, Ctype, ldc, strideC,
+                  batchCount, migratedComputeType, algo);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSgemmStridedBatched(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const float *alpha,        /* host or device pointer */
+    const float *A, int lda, long long int strideA, /* purposely signed */
+    const float *B, int ldb, long long int strideB,
+    const float *beta, /* host or device pointer */
+    float *C, int ldc, long long int strideC, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const float *, const float *, int, long long, const float *, int,
+      long long, const float *, float *, int, long long, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgemmStridedBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, strideA, B,
+                  ldb, strideB, beta, C, ldc, strideC, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDgemmStridedBatched(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const double *alpha, /* host or device pointer */
+    const double *A, int lda, long long int strideA, /* purposely signed */
+    const double *B, int ldb, long long int strideB,
+    const double *beta, /* host or device pointer */
+    double *C, int ldc, long long int strideC, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const double *, const double *, int, long long, const double *, int,
+      long long, const double *, double *, int, long long, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgemmStridedBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, strideA, B,
+                  ldb, strideB, beta, C, ldc, strideC, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgemmStridedBatched(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, long long int strideA, /* purposely signed */
+    const cuComplex *B, int ldb, long long int strideB,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *C, int ldc, long long int strideC, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuComplex *, const cuComplex *, int, long long, const cuComplex *,
+      int, long long, const cuComplex *, cuComplex *, int, long long, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemmStridedBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, strideA, B,
+                  ldb, strideB, beta, C, ldc, strideC, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgemm3mStridedBatched(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda, long long int strideA, /* purposely signed */
+    const cuComplex *B, int ldb, long long int strideB,
+    const cuComplex *beta, /* host or device pointer */
+    cuComplex *C, int ldc, long long int strideC, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuComplex *, const cuComplex *, int, long long, const cuComplex *,
+      int, long long, const cuComplex *, cuComplex *, int, long long, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemm3mStridedBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, strideA, B,
+                  ldb, strideB, beta, C, ldc, strideC, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZgemmStridedBatched(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, int k,
+    const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda,
+    long long int strideA, /* purposely signed */
+    const cuDoubleComplex *B, int ldb, long long int strideB,
+    const cuDoubleComplex *beta, /* host or device poi */
+    cuDoubleComplex *C, int ldc, long long int strideC, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int, long long,
+      const cuDoubleComplex *, int, long long, const cuDoubleComplex *,
+      cuDoubleComplex *, int, long long, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgemmStridedBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, k, alpha, A, lda, strideA, B,
+                  ldb, strideB, beta, C, ldc, strideC, batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSgeam(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, const float *alpha,           /* host or device pointer */
+    const float *A, int lda, const float *beta, /* host or device pointer */
+    const float *B, int ldb, float *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int,
+      const float *, const float *, int, const float *, const float *, int,
+      float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgeam");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, alpha, A, lda, beta, B, ldb, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDgeam(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, const double *alpha,            /* host or device pointer */
+    const double *A, int lda, const double *beta, /* host or device pointer */
+    const double *B, int ldb, double *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int,
+      const double *, const double *, int, const double *, const double *, int,
+      double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgeam");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, alpha, A, lda, beta, B, ldb, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgeam(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, const cuComplex *alpha, /* host or device pointer */
+    const cuComplex *A, int lda,
+    const cuComplex *beta, /* host or device pointer */
+    const cuComplex *B, int ldb, cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int,
+      const cuComplex *, const cuComplex *, int, const cuComplex *,
+      const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgeam");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, alpha, A, lda, beta, B, ldb, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZgeam(
+    cublasHandle_t handle, cublasOperation_t transa, cublasOperation_t transb,
+    int m, int n, const cuDoubleComplex *alpha, /* host or device pointer */
+    const cuDoubleComplex *A, int lda,
+    const cuDoubleComplex *beta, /* host or device pointer */
+    const cuDoubleComplex *B, int ldb, cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, int, cuDoubleComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgeam");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transa, transb, m, n, alpha, A, lda, beta, B, ldb, C,
+                  ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSgetrfBatched(
+    cublasHandle_t handle, int n, float *const A[], /*Device pointer*/
+    int lda, int *P,                                /*Device Pointer*/
+    int *info,                                      /*Device Pointer*/
+    int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, float *const[], int, int *, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgetrfBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, P, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDgetrfBatched(
+    cublasHandle_t handle, int n, double *const A[], /*Device pointer*/
+    int lda, int *P,                                 /*Device Pointer*/
+    int *info,                                       /*Device Pointer*/
+    int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, double *const[], int, int *, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgetrfBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, P, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgetrfBatched(
+    cublasHandle_t handle, int n, cuComplex *const A[], /*Device pointer*/
+    int lda, int *P,                                    /*Device Pointer*/
+    int *info,                                          /*Device Pointer*/
+    int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, cuComplex *const[], int, int *, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgetrfBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, P, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZgetrfBatched(
+    cublasHandle_t handle, int n, cuDoubleComplex *const A[], /*Device pointer*/
+    int lda, int *P,                                          /*Device Pointer*/
+    int *info,                                                /*Device Pointer*/
+    int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, cuDoubleComplex *const[], int, int *, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgetrfBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, P, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSgetriBatched(
+    cublasHandle_t handle, int n, const float *const A[], /*Device pointer*/
+    int lda, const int *P,                                /*Device pointer*/
+    float *const C[],                                     /*Device pointer*/
+    int ldc, int *info, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const float *const[], int, const int *,
+      float *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgetriBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, P, C, ldc, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDgetriBatched(
+    cublasHandle_t handle, int n, const double *const A[], /*Device pointer*/
+    int lda, const int *P,                                 /*Device pointer*/
+    double *const C[],                                     /*Device pointer*/
+    int ldc, int *info, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const double *const[], int, const int *,
+      double *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgetriBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, P, C, ldc, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgetriBatched(
+    cublasHandle_t handle, int n, const cuComplex *const A[], /*Device pointer*/
+    int lda, const int *P,                                    /*Device pointer*/
+    cuComplex *const C[],                                     /*Device pointer*/
+    int ldc, int *info, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuComplex *const[], int, const int *,
+      cuComplex *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgetriBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, P, C, ldc, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZgetriBatched(cublasHandle_t handle, int n,
+                    const cuDoubleComplex *const A[], /*Device pointer*/
+                    int lda, const int *P,            /*Device pointer*/
+                    cuDoubleComplex *const C[],       /*Device pointer*/
+                    int ldc, int *info, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuDoubleComplex *const[], int, const int *,
+      cuDoubleComplex *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgetriBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, P, C, ldc, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSgetrsBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int n, int nrhs,
+    const float *const Aarray[], int lda, const int *devIpiv,
+    float *const Barray[], int ldb, int *info, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const float *const[], int,
+      const int *, float *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgetrsBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, n, nrhs, Aarray, lda, devIpiv, Barray, ldb,
+                  info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDgetrsBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int n, int nrhs,
+    const double *const Aarray[], int lda, const int *devIpiv,
+    double *const Barray[], int ldb, int *info, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const double *const[], int,
+      const int *, double *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgetrsBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, n, nrhs, Aarray, lda, devIpiv, Barray, ldb,
+                  info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCgetrsBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int n, int nrhs,
+    const cuComplex *const Aarray[], int lda, const int *devIpiv,
+    cuComplex *const Barray[], int ldb, int *info, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, const cuComplex *const[],
+      int, const int *, cuComplex *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgetrsBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, n, nrhs, Aarray, lda, devIpiv, Barray, ldb,
+                  info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZgetrsBatched(
+    cublasHandle_t handle, cublasOperation_t trans, int n, int nrhs,
+    const cuDoubleComplex *const Aarray[], int lda, const int *devIpiv,
+    cuDoubleComplex *const Barray[], int ldb, int *info, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int,
+      const cuDoubleComplex *const[], int, const int *,
+      cuDoubleComplex *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgetrsBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, n, nrhs, Aarray, lda, devIpiv, Barray, ldb,
+                  info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStrsmBatched(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const float *alpha, /*Host or Device Pointer*/
+    const float *const A[], int lda, float *const B[], int ldb,
+    int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const float *, const float *const[], int,
+      float *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStrsmBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb,
+                  batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtrsmBatched(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const double *alpha, /*Host or Device Pointer*/
+    const double *const A[], int lda, double *const B[], int ldb,
+    int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const double *, const double *const[], int,
+      double *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtrsmBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb,
+                  batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtrsmBatched(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const cuComplex *alpha, /*Host or Device Pointer*/
+    const cuComplex *const A[], int lda, cuComplex *const B[], int ldb,
+    int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const cuComplex *, const cuComplex *const[],
+      int, cuComplex *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtrsmBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb,
+                  batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtrsmBatched(
+    cublasHandle_t handle, cublasSideMode_t side, cublasFillMode_t uplo,
+    cublasOperation_t trans, cublasDiagType_t diag, int m, int n,
+    const cuDoubleComplex *alpha, /*Host or Device Pointer*/
+    const cuDoubleComplex *const A[], int lda, cuDoubleComplex *const B[],
+    int ldb, int batchCount) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, cublasFillMode_t, cublasOperation_t,
+      cublasDiagType_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *const[], int, cuDoubleComplex *const[], int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtrsmBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, side, uplo, trans, diag, m, n, alpha, A, lda, B, ldb,
+                  batchCount);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSmatinvBatched(
+    cublasHandle_t handle, int n, const float *const A[], /*Device pointer*/
+    int lda, float *const Ainv[],                         /*Device pointer*/
+    int lda_inv, int *info,                               /*Device Pointer*/
+    int batchSize) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, const float *const[],
+                                     int, float *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSmatinvBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, Ainv, lda_inv, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDmatinvBatched(
+    cublasHandle_t handle, int n, const double *const A[], /*Device pointer*/
+    int lda, double *const Ainv[],                         /*Device pointer*/
+    int lda_inv, int *info,                                /*Device Pointer*/
+    int batchSize) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, const double *const[],
+                                     int, double *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDmatinvBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, Ainv, lda_inv, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCmatinvBatched(
+    cublasHandle_t handle, int n, const cuComplex *const A[], /*Device pointer*/
+    int lda, cuComplex *const Ainv[],                         /*Device pointer*/
+    int lda_inv, int *info,                                   /*Device Pointer*/
+    int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuComplex *const[], int, cuComplex *const[],
+      int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCmatinvBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, Ainv, lda_inv, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZmatinvBatched(cublasHandle_t handle, int n,
+                     const cuDoubleComplex *const A[],       /*Device pointer*/
+                     int lda, cuDoubleComplex *const Ainv[], /*Device pointer*/
+                     int lda_inv, int *info,                 /*Device Pointer*/
+                     int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, const cuDoubleComplex *const[], int,
+      cuDoubleComplex *const[], int, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZmatinvBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, A, lda, Ainv, lda_inv, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSgeqrfBatched(cublasHandle_t handle, int m, int n,
+                    float *const Aarray[],            /*Device pointer*/
+                    int lda, float *const TauArray[], /*Device pointer*/
+                    int *info, int batchSize) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, int, float *const[],
+                                     int, float *const[], int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgeqrfBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, Aarray, lda, TauArray, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDgeqrfBatched(cublasHandle_t handle, int m, int n,
+                    double *const Aarray[],            /*Device pointer*/
+                    int lda, double *const TauArray[], /*Device pointer*/
+                    int *info, int batchSize) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, int, int, double *const[],
+                                     int, double *const[], int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgeqrfBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, Aarray, lda, TauArray, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCgeqrfBatched(cublasHandle_t handle, int m, int n,
+                    cuComplex *const Aarray[],            /*Device pointer*/
+                    int lda, cuComplex *const TauArray[], /*Device pointer*/
+                    int *info, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, int, cuComplex *const[], int, cuComplex *const[],
+      int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgeqrfBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, Aarray, lda, TauArray, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZgeqrfBatched(
+    cublasHandle_t handle, int m, int n,
+    cuDoubleComplex *const Aarray[],            /*Device pointer*/
+    int lda, cuDoubleComplex *const TauArray[], /*Device pointer*/
+    int *info, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, int, int, cuDoubleComplex *const[], int,
+      cuDoubleComplex *const[], int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgeqrfBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, Aarray, lda, TauArray, info, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasSgelsBatched(cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+                   int nrhs, float *const Aarray[],       /*Device pointer*/
+                   int lda, float *const Carray[],        /*Device pointer*/
+                   int ldc, int *info, int *devInfoArray, /*Device pointer*/
+                   int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, int, float *const[], int,
+      float *const[], int, int *, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgelsBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, nrhs, Aarray, lda, Carray, ldc, info,
+                  devInfoArray, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasDgelsBatched(cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+                   int nrhs, double *const Aarray[],      /*Device pointer*/
+                   int lda, double *const Carray[],       /*Device pointer*/
+                   int ldc, int *info, int *devInfoArray, /*Device pointer*/
+                   int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, int, double *const[], int,
+      double *const[], int, int *, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgelsBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, nrhs, Aarray, lda, Carray, ldc, info,
+                  devInfoArray, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasCgelsBatched(cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+                   int nrhs, cuComplex *const Aarray[], /*Device pointer*/
+                   int lda, cuComplex *const Carray[],  /*Device pointer*/
+                   int ldc, int *info, int *devInfoArray, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, int, cuComplex *const[], int,
+      cuComplex *const[], int, int *, int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgelsBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, nrhs, Aarray, lda, Carray, ldc, info,
+                  devInfoArray, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI
+cublasZgelsBatched(cublasHandle_t handle, cublasOperation_t trans, int m, int n,
+                   int nrhs, cuDoubleComplex *const Aarray[], /*Device pointer*/
+                   int lda, cuDoubleComplex *const Carray[],  /*Device pointer*/
+                   int ldc, int *info, int *devInfoArray, int batchSize) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasOperation_t, int, int, int,
+      cuDoubleComplex *const[], int, cuDoubleComplex *const[], int, int *,
+      int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgelsBatched");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, trans, m, n, nrhs, Aarray, lda, Carray, ldc, info,
+                  devInfoArray, batchSize);
+}
+
+cublasStatus_t CUBLASWINAPI cublasSdgmm(cublasHandle_t handle,
+                                        cublasSideMode_t mode, int m, int n,
+                                        const float *A, int lda, const float *x,
+                                        int incx, float *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, int, int, const float *, int,
+      const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSdgmm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode, m, n, A, lda, x, incx, C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDdgmm(cublasHandle_t handle,
+                                        cublasSideMode_t mode, int m, int n,
+                                        const double *A, int lda,
+                                        const double *x, int incx, double *C,
+                                        int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, int, int, const double *, int,
+      const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDdgmm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode, m, n, A, lda, x, incx, C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCdgmm(cublasHandle_t handle,
+                                        cublasSideMode_t mode, int m, int n,
+                                        const cuComplex *A, int lda,
+                                        const cuComplex *x, int incx,
+                                        cuComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, int, int, const cuComplex *, int,
+      const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCdgmm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode, m, n, A, lda, x, incx, C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZdgmm(cublasHandle_t handle,
+                                        cublasSideMode_t mode, int m, int n,
+                                        const cuDoubleComplex *A, int lda,
+                                        const cuDoubleComplex *x, int incx,
+                                        cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasSideMode_t, int, int, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZdgmm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode, m, n, A, lda, x, incx, C, ldc);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStpttr(cublasHandle_t handle,
+                                         cublasFillMode_t uplo, int n,
+                                         const float *AP, float *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStpttr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, AP, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtpttr(cublasHandle_t handle,
+                                         cublasFillMode_t uplo, int n,
+                                         const double *AP, double *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtpttr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, AP, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtpttr(cublasHandle_t handle,
+                                         cublasFillMode_t uplo, int n,
+                                         const cuComplex *AP, cuComplex *A,
+                                         int lda) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, cublasFillMode_t, int,
+                                     const cuComplex *, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtpttr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, AP, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtpttr(cublasHandle_t handle,
+                                         cublasFillMode_t uplo, int n,
+                                         const cuDoubleComplex *AP,
+                                         cuDoubleComplex *A, int lda) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuDoubleComplex *,
+      cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtpttr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, AP, A, lda);
+}
+
+cublasStatus_t CUBLASWINAPI cublasStrttp(cublasHandle_t handle,
+                                         cublasFillMode_t uplo, int n,
+                                         const float *A, int lda, float *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const float *, int, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStrttp");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, A, lda, AP);
+}
+
+cublasStatus_t CUBLASWINAPI cublasDtrttp(cublasHandle_t handle,
+                                         cublasFillMode_t uplo, int n,
+                                         const double *A, int lda, double *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const double *, int, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtrttp");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, A, lda, AP);
+}
+
+cublasStatus_t CUBLASWINAPI cublasCtrttp(cublasHandle_t handle,
+                                         cublasFillMode_t uplo, int n,
+                                         const cuComplex *A, int lda,
+                                         cuComplex *AP) {
+  using FuncPtr =
+      cublasStatus_t(CUBLASWINAPI *)(cublasHandle_t, cublasFillMode_t, int,
+                                     const cuComplex *, int, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtrttp");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, A, lda, AP);
+}
+
+cublasStatus_t CUBLASWINAPI cublasZtrttp(cublasHandle_t handle,
+                                         cublasFillMode_t uplo, int n,
+                                         const cuDoubleComplex *A, int lda,
+                                         cuDoubleComplex *AP) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(
+      cublasHandle_t, cublasFillMode_t, int, const cuDoubleComplex *, int,
+      cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtrttp");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, uplo, n, A, lda, AP);
+}
+
+cublasStatus CUBLASWINAPI cublasInit(void) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasInit");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+cublasStatus CUBLASWINAPI cublasShutdown(void) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasShutdown");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+cublasStatus CUBLASWINAPI cublasGetError(void) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetError");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+cublasStatus CUBLASWINAPI cublasGetVersion(int *version) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasGetVersion");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(version);
+}
+
+cublasStatus CUBLASWINAPI cublasAlloc(int n, int elemSize, void **devicePtr) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(int, int, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasAlloc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(n, elemSize, devicePtr);
+}
+
+cublasStatus CUBLASWINAPI cublasFree(void *devicePtr) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasFree");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devicePtr);
+}
+
+cublasStatus CUBLASWINAPI cublasSetKernelStream(cudaStream_t stream) {
+  using FuncPtr = cublasStatus_t(CUBLASWINAPI *)(cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSetKernelStream");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream);
+}
+
+float CUBLASWINAPI cublasSnrm2(int n, const float *x, int incx) {
+  using FuncPtr = float(CUBLASWINAPI *)(int, const float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSnrm2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSnrm2");
+  return func_ptr(n, x, incx);
+}
+
+double CUBLASWINAPI cublasDnrm2(int n, const double *x, int incx) {
+  using FuncPtr = double(CUBLASWINAPI *)(int, const double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDnrm2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDnrm2");
+  return func_ptr(n, x, incx);
+}
+
+float CUBLASWINAPI cublasScnrm2(int n, const cuComplex *x, int incx) {
+  using FuncPtr = float(CUBLASWINAPI *)(int, const cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasScnrm2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasScnrm2");
+  return func_ptr(n, x, incx);
+}
+
+double CUBLASWINAPI cublasDznrm2(int n, const cuDoubleComplex *x, int incx) {
+  using FuncPtr = double(CUBLASWINAPI *)(int, const cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDznrm2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDznrm2");
+  return func_ptr(n, x, incx);
+}
+
+float CUBLASWINAPI cublasSdot(int n, const float *x, int incx, const float *y,
+                              int incy) {
+  using FuncPtr =
+      float(CUBLASWINAPI *)(int, const float *, int, const float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSdot");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSdot");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+double CUBLASWINAPI cublasDdot(int n, const double *x, int incx,
+                               const double *y, int incy) {
+  using FuncPtr =
+      double(CUBLASWINAPI *)(int, const double *, int, const double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDdot");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDdot");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+cuComplex CUBLASWINAPI cublasCdotu(int n, const cuComplex *x, int incx,
+                                   const cuComplex *y, int incy) {
+  using FuncPtr = cuComplex(CUBLASWINAPI *)(int, const cuComplex *, int,
+                                            const cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCdotu");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCdotu");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+cuComplex CUBLASWINAPI cublasCdotc(int n, const cuComplex *x, int incx,
+                                   const cuComplex *y, int incy) {
+  using FuncPtr = cuComplex(CUBLASWINAPI *)(int, const cuComplex *, int,
+                                            const cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCdotc");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCdotc");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+cuDoubleComplex CUBLASWINAPI cublasZdotu(int n, const cuDoubleComplex *x,
+                                         int incx, const cuDoubleComplex *y,
+                                         int incy) {
+  using FuncPtr = cuDoubleComplex(CUBLASWINAPI *)(
+      int, const cuDoubleComplex *, int, const cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZdotu");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZdotu");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+cuDoubleComplex CUBLASWINAPI cublasZdotc(int n, const cuDoubleComplex *x,
+                                         int incx, const cuDoubleComplex *y,
+                                         int incy) {
+  using FuncPtr = cuDoubleComplex(CUBLASWINAPI *)(
+      int, const cuDoubleComplex *, int, const cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZdotc");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZdotc");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasSscal(int n, float alpha, float *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, float, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSscal");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSscal");
+  return func_ptr(n, alpha, x, incx);
+}
+
+void CUBLASWINAPI cublasDscal(int n, double alpha, double *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, double, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDscal");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDscal");
+  return func_ptr(n, alpha, x, incx);
+}
+
+void CUBLASWINAPI cublasCscal(int n, cuComplex alpha, cuComplex *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCscal");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCscal");
+  return func_ptr(n, alpha, x, incx);
+}
+
+void CUBLASWINAPI cublasZscal(int n, cuDoubleComplex alpha, cuDoubleComplex *x,
+                              int incx) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZscal");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZscal");
+  return func_ptr(n, alpha, x, incx);
+}
+
+void CUBLASWINAPI cublasCsscal(int n, float alpha, cuComplex *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, float, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsscal");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCsscal");
+  return func_ptr(n, alpha, x, incx);
+}
+
+void CUBLASWINAPI cublasZdscal(int n, double alpha, cuDoubleComplex *x,
+                               int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, double, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZdscal");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZdscal");
+  return func_ptr(n, alpha, x, incx);
+}
+
+void CUBLASWINAPI cublasSaxpy(int n, float alpha, const float *x, int incx,
+                              float *y, int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, float, const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSaxpy");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSaxpy");
+  return func_ptr(n, alpha, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasDaxpy(int n, double alpha, const double *x, int incx,
+                              double *y, int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, double, const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDaxpy");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDaxpy");
+  return func_ptr(n, alpha, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasCaxpy(int n, cuComplex alpha, const cuComplex *x,
+                              int incx, cuComplex *y, int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, cuComplex, const cuComplex *, int,
+                                       cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCaxpy");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCaxpy");
+  return func_ptr(n, alpha, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasZaxpy(int n, cuDoubleComplex alpha,
+                              const cuDoubleComplex *x, int incx,
+                              cuDoubleComplex *y, int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, cuDoubleComplex, const cuDoubleComplex *, int,
+                           cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZaxpy");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZaxpy");
+  return func_ptr(n, alpha, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasScopy(int n, const float *x, int incx, float *y,
+                              int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasScopy");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasScopy");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasDcopy(int n, const double *x, int incx, double *y,
+                              int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDcopy");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDcopy");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasCcopy(int n, const cuComplex *x, int incx, cuComplex *y,
+                              int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCcopy");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCcopy");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasZcopy(int n, const cuDoubleComplex *x, int incx,
+                              cuDoubleComplex *y, int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, const cuDoubleComplex *, int,
+                                       cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZcopy");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZcopy");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasSswap(int n, float *x, int incx, float *y, int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSswap");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSswap");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasDswap(int n, double *x, int incx, double *y, int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDswap");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDswap");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasCswap(int n, cuComplex *x, int incx, cuComplex *y,
+                              int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCswap");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCswap");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+void CUBLASWINAPI cublasZswap(int n, cuDoubleComplex *x, int incx,
+                              cuDoubleComplex *y, int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZswap");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZswap");
+  return func_ptr(n, x, incx, y, incy);
+}
+
+int CUBLASWINAPI cublasIsamax(int n, const float *x, int incx) {
+  using FuncPtr = int(CUBLASWINAPI *)(int, const float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIsamax");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasIsamax");
+  return func_ptr(n, x, incx);
+}
+
+int CUBLASWINAPI cublasIdamax(int n, const double *x, int incx) {
+  using FuncPtr = int(CUBLASWINAPI *)(int, const double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIdamax");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasIdamax");
+  return func_ptr(n, x, incx);
+}
+
+int CUBLASWINAPI cublasIcamax(int n, const cuComplex *x, int incx) {
+  using FuncPtr = int(CUBLASWINAPI *)(int, const cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIcamax");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasIcamax");
+  return func_ptr(n, x, incx);
+}
+
+int CUBLASWINAPI cublasIzamax(int n, const cuDoubleComplex *x, int incx) {
+  using FuncPtr = int(CUBLASWINAPI *)(int, const cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIzamax");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasIzamax");
+  return func_ptr(n, x, incx);
+}
+
+int CUBLASWINAPI cublasIsamin(int n, const float *x, int incx) {
+  using FuncPtr = int(CUBLASWINAPI *)(int, const float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIsamin");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasIsamin");
+  return func_ptr(n, x, incx);
+}
+
+int CUBLASWINAPI cublasIdamin(int n, const double *x, int incx) {
+  using FuncPtr = int(CUBLASWINAPI *)(int, const double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIdamin");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasIdamin");
+  return func_ptr(n, x, incx);
+}
+
+int CUBLASWINAPI cublasIcamin(int n, const cuComplex *x, int incx) {
+  using FuncPtr = int(CUBLASWINAPI *)(int, const cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIcamin");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasIcamin");
+  return func_ptr(n, x, incx);
+}
+
+int CUBLASWINAPI cublasIzamin(int n, const cuDoubleComplex *x, int incx) {
+  using FuncPtr = int(CUBLASWINAPI *)(int, const cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasIzamin");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasIzamin");
+  return func_ptr(n, x, incx);
+}
+
+float CUBLASWINAPI cublasSasum(int n, const float *x, int incx) {
+  using FuncPtr = float(CUBLASWINAPI *)(int, const float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSasum");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSasum");
+  return func_ptr(n, x, incx);
+}
+
+double CUBLASWINAPI cublasDasum(int n, const double *x, int incx) {
+  using FuncPtr = double(CUBLASWINAPI *)(int, const double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDasum");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDasum");
+  return func_ptr(n, x, incx);
+}
+
+float CUBLASWINAPI cublasScasum(int n, const cuComplex *x, int incx) {
+  using FuncPtr = float(CUBLASWINAPI *)(int, const cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasScasum");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasScasum");
+  return func_ptr(n, x, incx);
+}
+
+double CUBLASWINAPI cublasDzasum(int n, const cuDoubleComplex *x, int incx) {
+  using FuncPtr = double(CUBLASWINAPI *)(int, const cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDzasum");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDzasum");
+  return func_ptr(n, x, incx);
+}
+
+void CUBLASWINAPI cublasSrot(int n, float *x, int incx, float *y, int incy,
+                             float sc, float ss) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, float *, int, float *, int, float, float);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSrot");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSrot");
+  return func_ptr(n, x, incx, y, incy, sc, ss);
+}
+
+void CUBLASWINAPI cublasDrot(int n, double *x, int incx, double *y, int incy,
+                             double sc, double ss) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, double *, int, double *, int, double, double);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDrot");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDrot");
+  return func_ptr(n, x, incx, y, incy, sc, ss);
+}
+
+void CUBLASWINAPI cublasCrot(int n, cuComplex *x, int incx, cuComplex *y,
+                             int incy, float c, cuComplex s) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, cuComplex *, int, cuComplex *, int,
+                                       float, cuComplex);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCrot");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCrot");
+  return func_ptr(n, x, incx, y, incy, c, s);
+}
+
+void CUBLASWINAPI cublasZrot(int n, cuDoubleComplex *x, int incx,
+                             cuDoubleComplex *y, int incy, double sc,
+                             cuDoubleComplex cs) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, cuDoubleComplex *, int, cuDoubleComplex *, int,
+                           double, cuDoubleComplex);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZrot");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZrot");
+  return func_ptr(n, x, incx, y, incy, sc, cs);
+}
+
+void CUBLASWINAPI cublasCsrot(int n, cuComplex *x, int incx, cuComplex *y,
+                              int incy, float c, float s) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, cuComplex *, int, cuComplex *, int,
+                                       float, float);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsrot");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCsrot");
+  return func_ptr(n, x, incx, y, incy, c, s);
+}
+
+void CUBLASWINAPI cublasZdrot(int n, cuDoubleComplex *x, int incx,
+                              cuDoubleComplex *y, int incy, double c,
+                              double s) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, cuDoubleComplex *, int,
+                                       cuDoubleComplex *, int, double, double);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZdrot");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZdrot");
+  return func_ptr(n, x, incx, y, incy, c, s);
+}
+
+void CUBLASWINAPI cublasSrotg(float *sa, float *sb, float *sc, float *ss) {
+  using FuncPtr = void(CUBLASWINAPI *)(float *, float *, float *, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSrotg");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSrotg");
+  return func_ptr(sa, sb, sc, ss);
+}
+
+void CUBLASWINAPI cublasDrotg(double *sa, double *sb, double *sc, double *ss) {
+  using FuncPtr = void(CUBLASWINAPI *)(double *, double *, double *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDrotg");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDrotg");
+  return func_ptr(sa, sb, sc, ss);
+}
+
+void CUBLASWINAPI cublasCrotg(cuComplex *ca, cuComplex cb, float *sc,
+                              cuComplex *cs) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(cuComplex *, cuComplex, float *, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCrotg");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCrotg");
+  return func_ptr(ca, cb, sc, cs);
+}
+
+void CUBLASWINAPI cublasZrotg(cuDoubleComplex *ca, cuDoubleComplex cb,
+                              double *sc, cuDoubleComplex *cs) {
+  using FuncPtr = void(CUBLASWINAPI *)(cuDoubleComplex *, cuDoubleComplex,
+                                       double *, cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZrotg");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZrotg");
+  return func_ptr(ca, cb, sc, cs);
+}
+
+void CUBLASWINAPI cublasSrotm(int n, float *x, int incx, float *y, int incy,
+                              const float *sparam) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, float *, int, float *, int, const float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSrotm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSrotm");
+  return func_ptr(n, x, incx, y, incy, sparam);
+}
+
+void CUBLASWINAPI cublasDrotm(int n, double *x, int incx, double *y, int incy,
+                              const double *sparam) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, double *, int, double *, int, const double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDrotm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDrotm");
+  return func_ptr(n, x, incx, y, incy, sparam);
+}
+
+void CUBLASWINAPI cublasSrotmg(float *sd1, float *sd2, float *sx1,
+                               const float *sy1, float *sparam) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(float *, float *, float *, const float *, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSrotmg");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSrotmg");
+  return func_ptr(sd1, sd2, sx1, sy1, sparam);
+}
+
+void CUBLASWINAPI cublasDrotmg(double *sd1, double *sd2, double *sx1,
+                               const double *sy1, double *sparam) {
+  using FuncPtr = void(CUBLASWINAPI *)(double *, double *, double *,
+                                       const double *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDrotmg");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDrotmg");
+  return func_ptr(sd1, sd2, sx1, sy1, sparam);
+}
+
+void CUBLASWINAPI cublasSgemv(char trans, int m, int n, float alpha,
+                              const float *A, int lda, const float *x, int incx,
+                              float beta, float *y, int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, int, float, const float *, int,
+                           const float *, int, float, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgemv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSgemv");
+  return func_ptr(trans, m, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasDgemv(char trans, int m, int n, double alpha,
+                              const double *A, int lda, const double *x,
+                              int incx, double beta, double *y, int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, int, double, const double *, int,
+                           const double *, int, double, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgemv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDgemv");
+  return func_ptr(trans, m, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasCgemv(char trans, int m, int n, cuComplex alpha,
+                              const cuComplex *A, int lda, const cuComplex *x,
+                              int incx, cuComplex beta, cuComplex *y,
+                              int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, int, cuComplex, const cuComplex *, int,
+                           const cuComplex *, int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCgemv");
+  return func_ptr(trans, m, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasZgemv(char trans, int m, int n, cuDoubleComplex alpha,
+                              const cuDoubleComplex *A, int lda,
+                              const cuDoubleComplex *x, int incx,
+                              cuDoubleComplex beta, cuDoubleComplex *y,
+                              int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, int, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgemv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZgemv");
+  return func_ptr(trans, m, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasSgbmv(char trans, int m, int n, int kl, int ku,
+                              float alpha, const float *A, int lda,
+                              const float *x, int incx, float beta, float *y,
+                              int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, int, int, int, float, const float *, int,
+                           const float *, int, float, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSgbmv");
+  return func_ptr(trans, m, n, kl, ku, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasDgbmv(char trans, int m, int n, int kl, int ku,
+                              double alpha, const double *A, int lda,
+                              const double *x, int incx, double beta, double *y,
+                              int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, int, int, int, double, const double *,
+                           int, const double *, int, double, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDgbmv");
+  return func_ptr(trans, m, n, kl, ku, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasCgbmv(char trans, int m, int n, int kl, int ku,
+                              cuComplex alpha, const cuComplex *A, int lda,
+                              const cuComplex *x, int incx, cuComplex beta,
+                              cuComplex *y, int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, int, int, int, int, cuComplex, const cuComplex *, int,
+      const cuComplex *, int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCgbmv");
+  return func_ptr(trans, m, n, kl, ku, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasZgbmv(char trans, int m, int n, int kl, int ku,
+                              cuDoubleComplex alpha, const cuDoubleComplex *A,
+                              int lda, const cuDoubleComplex *x, int incx,
+                              cuDoubleComplex beta, cuDoubleComplex *y,
+                              int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, int, int, int, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZgbmv");
+  return func_ptr(trans, m, n, kl, ku, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasStrmv(char uplo, char trans, char diag, int n,
+                              const float *A, int lda, float *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, const float *,
+                                       int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStrmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasStrmv");
+  return func_ptr(uplo, trans, diag, n, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasDtrmv(char uplo, char trans, char diag, int n,
+                              const double *A, int lda, double *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, const double *,
+                                       int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtrmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDtrmv");
+  return func_ptr(uplo, trans, diag, n, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasCtrmv(char uplo, char trans, char diag, int n,
+                              const cuComplex *A, int lda, cuComplex *x,
+                              int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, const cuComplex *,
+                                       int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtrmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCtrmv");
+  return func_ptr(uplo, trans, diag, n, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasZtrmv(char uplo, char trans, char diag, int n,
+                              const cuDoubleComplex *A, int lda,
+                              cuDoubleComplex *x, int incx) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, char, int, const cuDoubleComplex *, int,
+                           cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtrmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZtrmv");
+  return func_ptr(uplo, trans, diag, n, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasStbmv(char uplo, char trans, char diag, int n, int k,
+                              const float *A, int lda, float *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, int,
+                                       const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasStbmv");
+  return func_ptr(uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasDtbmv(char uplo, char trans, char diag, int n, int k,
+                              const double *A, int lda, double *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, int,
+                                       const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDtbmv");
+  return func_ptr(uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasCtbmv(char uplo, char trans, char diag, int n, int k,
+                              const cuComplex *A, int lda, cuComplex *x,
+                              int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, char, int, int, const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCtbmv");
+  return func_ptr(uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasZtbmv(char uplo, char trans, char diag, int n, int k,
+                              const cuDoubleComplex *A, int lda,
+                              cuDoubleComplex *x, int incx) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, char, int, int, const cuDoubleComplex *,
+                           int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZtbmv");
+  return func_ptr(uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasStpmv(char uplo, char trans, char diag, int n,
+                              const float *AP, float *x, int incx) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, char, int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStpmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasStpmv");
+  return func_ptr(uplo, trans, diag, n, AP, x, incx);
+}
+
+void CUBLASWINAPI cublasDtpmv(char uplo, char trans, char diag, int n,
+                              const double *AP, double *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, const double *,
+                                       double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtpmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDtpmv");
+  return func_ptr(uplo, trans, diag, n, AP, x, incx);
+}
+
+void CUBLASWINAPI cublasCtpmv(char uplo, char trans, char diag, int n,
+                              const cuComplex *AP, cuComplex *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, const cuComplex *,
+                                       cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtpmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCtpmv");
+  return func_ptr(uplo, trans, diag, n, AP, x, incx);
+}
+
+void CUBLASWINAPI cublasZtpmv(char uplo, char trans, char diag, int n,
+                              const cuDoubleComplex *AP, cuDoubleComplex *x,
+                              int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, char, int, const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtpmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZtpmv");
+  return func_ptr(uplo, trans, diag, n, AP, x, incx);
+}
+
+void CUBLASWINAPI cublasStrsv(char uplo, char trans, char diag, int n,
+                              const float *A, int lda, float *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, const float *,
+                                       int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStrsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasStrsv");
+  return func_ptr(uplo, trans, diag, n, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasDtrsv(char uplo, char trans, char diag, int n,
+                              const double *A, int lda, double *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, const double *,
+                                       int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtrsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDtrsv");
+  return func_ptr(uplo, trans, diag, n, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasCtrsv(char uplo, char trans, char diag, int n,
+                              const cuComplex *A, int lda, cuComplex *x,
+                              int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, const cuComplex *,
+                                       int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtrsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCtrsv");
+  return func_ptr(uplo, trans, diag, n, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasZtrsv(char uplo, char trans, char diag, int n,
+                              const cuDoubleComplex *A, int lda,
+                              cuDoubleComplex *x, int incx) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, char, int, const cuDoubleComplex *, int,
+                           cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtrsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZtrsv");
+  return func_ptr(uplo, trans, diag, n, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasStpsv(char uplo, char trans, char diag, int n,
+                              const float *AP, float *x, int incx) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, char, int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStpsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasStpsv");
+  return func_ptr(uplo, trans, diag, n, AP, x, incx);
+}
+
+void CUBLASWINAPI cublasDtpsv(char uplo, char trans, char diag, int n,
+                              const double *AP, double *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, const double *,
+                                       double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtpsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDtpsv");
+  return func_ptr(uplo, trans, diag, n, AP, x, incx);
+}
+
+void CUBLASWINAPI cublasCtpsv(char uplo, char trans, char diag, int n,
+                              const cuComplex *AP, cuComplex *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, const cuComplex *,
+                                       cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtpsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCtpsv");
+  return func_ptr(uplo, trans, diag, n, AP, x, incx);
+}
+
+void CUBLASWINAPI cublasZtpsv(char uplo, char trans, char diag, int n,
+                              const cuDoubleComplex *AP, cuDoubleComplex *x,
+                              int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, char, int, const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtpsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZtpsv");
+  return func_ptr(uplo, trans, diag, n, AP, x, incx);
+}
+
+void CUBLASWINAPI cublasStbsv(char uplo, char trans, char diag, int n, int k,
+                              const float *A, int lda, float *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, int,
+                                       const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStbsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasStbsv");
+  return func_ptr(uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasDtbsv(char uplo, char trans, char diag, int n, int k,
+                              const double *A, int lda, double *x, int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, int, int,
+                                       const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtbsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDtbsv");
+  return func_ptr(uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasCtbsv(char uplo, char trans, char diag, int n, int k,
+                              const cuComplex *A, int lda, cuComplex *x,
+                              int incx) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, char, int, int, const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtbsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCtbsv");
+  return func_ptr(uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasZtbsv(char uplo, char trans, char diag, int n, int k,
+                              const cuDoubleComplex *A, int lda,
+                              cuDoubleComplex *x, int incx) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, char, int, int, const cuDoubleComplex *,
+                           int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtbsv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZtbsv");
+  return func_ptr(uplo, trans, diag, n, k, A, lda, x, incx);
+}
+
+void CUBLASWINAPI cublasSsymv(char uplo, int n, float alpha, const float *A,
+                              int lda, const float *x, int incx, float beta,
+                              float *y, int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, int, float, const float *, int,
+                                       const float *, int, float, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsymv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSsymv");
+  return func_ptr(uplo, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasDsymv(char uplo, int n, double alpha, const double *A,
+                              int lda, const double *x, int incx, double beta,
+                              double *y, int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, double, const double *, int,
+                           const double *, int, double, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsymv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDsymv");
+  return func_ptr(uplo, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasChemv(char uplo, int n, cuComplex alpha,
+                              const cuComplex *A, int lda, const cuComplex *x,
+                              int incx, cuComplex beta, cuComplex *y,
+                              int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, cuComplex, const cuComplex *, int,
+                           const cuComplex *, int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChemv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasChemv");
+  return func_ptr(uplo, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasZhemv(char uplo, int n, cuDoubleComplex alpha,
+                              const cuDoubleComplex *A, int lda,
+                              const cuDoubleComplex *x, int incx,
+                              cuDoubleComplex beta, cuDoubleComplex *y,
+                              int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhemv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZhemv");
+  return func_ptr(uplo, n, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasSsbmv(char uplo, int n, int k, float alpha,
+                              const float *A, int lda, const float *x, int incx,
+                              float beta, float *y, int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, int, float, const float *, int,
+                           const float *, int, float, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSsbmv");
+  return func_ptr(uplo, n, k, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasDsbmv(char uplo, int n, int k, double alpha,
+                              const double *A, int lda, const double *x,
+                              int incx, double beta, double *y, int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, int, double, const double *, int,
+                           const double *, int, double, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDsbmv");
+  return func_ptr(uplo, n, k, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasChbmv(char uplo, int n, int k, cuComplex alpha,
+                              const cuComplex *A, int lda, const cuComplex *x,
+                              int incx, cuComplex beta, cuComplex *y,
+                              int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, int, cuComplex, const cuComplex *, int,
+                           const cuComplex *, int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasChbmv");
+  return func_ptr(uplo, n, k, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasZhbmv(char uplo, int n, int k, cuDoubleComplex alpha,
+                              const cuDoubleComplex *A, int lda,
+                              const cuDoubleComplex *x, int incx,
+                              cuDoubleComplex beta, cuDoubleComplex *y,
+                              int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, int, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhbmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZhbmv");
+  return func_ptr(uplo, n, k, alpha, A, lda, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasSspmv(char uplo, int n, float alpha, const float *AP,
+                              const float *x, int incx, float beta, float *y,
+                              int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, int, float, const float *,
+                                       const float *, int, float, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSspmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSspmv");
+  return func_ptr(uplo, n, alpha, AP, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasDspmv(char uplo, int n, double alpha, const double *AP,
+                              const double *x, int incx, double beta, double *y,
+                              int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, double, const double *, const double *,
+                           int, double, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDspmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDspmv");
+  return func_ptr(uplo, n, alpha, AP, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasChpmv(char uplo, int n, cuComplex alpha,
+                              const cuComplex *AP, const cuComplex *x, int incx,
+                              cuComplex beta, cuComplex *y, int incy) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, cuComplex, const cuComplex *,
+                           const cuComplex *, int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChpmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasChpmv");
+  return func_ptr(uplo, n, alpha, AP, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasZhpmv(char uplo, int n, cuDoubleComplex alpha,
+                              const cuDoubleComplex *AP,
+                              const cuDoubleComplex *x, int incx,
+                              cuDoubleComplex beta, cuDoubleComplex *y,
+                              int incy) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, int, cuDoubleComplex, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhpmv");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZhpmv");
+  return func_ptr(uplo, n, alpha, AP, x, incx, beta, y, incy);
+}
+
+void CUBLASWINAPI cublasSger(int m, int n, float alpha, const float *x,
+                             int incx, const float *y, int incy, float *A,
+                             int lda) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, int, float, const float *, int,
+                                       const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSger");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSger");
+  return func_ptr(m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+void CUBLASWINAPI cublasDger(int m, int n, double alpha, const double *x,
+                             int incx, const double *y, int incy, double *A,
+                             int lda) {
+  using FuncPtr = void(CUBLASWINAPI *)(int, int, double, const double *, int,
+                                       const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDger");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDger");
+  return func_ptr(m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+void CUBLASWINAPI cublasCgeru(int m, int n, cuComplex alpha, const cuComplex *x,
+                              int incx, const cuComplex *y, int incy,
+                              cuComplex *A, int lda) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, int, cuComplex, const cuComplex *, int,
+                           const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgeru");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCgeru");
+  return func_ptr(m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+void CUBLASWINAPI cublasCgerc(int m, int n, cuComplex alpha, const cuComplex *x,
+                              int incx, const cuComplex *y, int incy,
+                              cuComplex *A, int lda) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(int, int, cuComplex, const cuComplex *, int,
+                           const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgerc");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCgerc");
+  return func_ptr(m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+void CUBLASWINAPI cublasZgeru(int m, int n, cuDoubleComplex alpha,
+                              const cuDoubleComplex *x, int incx,
+                              const cuDoubleComplex *y, int incy,
+                              cuDoubleComplex *A, int lda) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      int, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgeru");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZgeru");
+  return func_ptr(m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+void CUBLASWINAPI cublasZgerc(int m, int n, cuDoubleComplex alpha,
+                              const cuDoubleComplex *x, int incx,
+                              const cuDoubleComplex *y, int incy,
+                              cuDoubleComplex *A, int lda) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      int, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgerc");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZgerc");
+  return func_ptr(m, n, alpha, x, incx, y, incy, A, lda);
+}
+
+void CUBLASWINAPI cublasSsyr(char uplo, int n, float alpha, const float *x,
+                             int incx, float *A, int lda) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, float, const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsyr");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSsyr");
+  return func_ptr(uplo, n, alpha, x, incx, A, lda);
+}
+
+void CUBLASWINAPI cublasDsyr(char uplo, int n, double alpha, const double *x,
+                             int incx, double *A, int lda) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, int, double, const double *, int,
+                                       double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsyr");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDsyr");
+  return func_ptr(uplo, n, alpha, x, incx, A, lda);
+}
+
+void CUBLASWINAPI cublasCher(char uplo, int n, float alpha, const cuComplex *x,
+                             int incx, cuComplex *A, int lda) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, int, float, const cuComplex *, int,
+                                       cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCher");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCher");
+  return func_ptr(uplo, n, alpha, x, incx, A, lda);
+}
+
+void CUBLASWINAPI cublasZher(char uplo, int n, double alpha,
+                             const cuDoubleComplex *x, int incx,
+                             cuDoubleComplex *A, int lda) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, int, double, const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZher");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZher");
+  return func_ptr(uplo, n, alpha, x, incx, A, lda);
+}
+
+void CUBLASWINAPI cublasSspr(char uplo, int n, float alpha, const float *x,
+                             int incx, float *AP) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, float, const float *, int, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSspr");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSspr");
+  return func_ptr(uplo, n, alpha, x, incx, AP);
+}
+
+void CUBLASWINAPI cublasDspr(char uplo, int n, double alpha, const double *x,
+                             int incx, double *AP) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, double, const double *, int, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDspr");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDspr");
+  return func_ptr(uplo, n, alpha, x, incx, AP);
+}
+
+void CUBLASWINAPI cublasChpr(char uplo, int n, float alpha, const cuComplex *x,
+                             int incx, cuComplex *AP) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, int, float, const cuComplex *, int,
+                                       cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChpr");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasChpr");
+  return func_ptr(uplo, n, alpha, x, incx, AP);
+}
+
+void CUBLASWINAPI cublasZhpr(char uplo, int n, double alpha,
+                             const cuDoubleComplex *x, int incx,
+                             cuDoubleComplex *AP) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, int, double, const cuDoubleComplex *, int, cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhpr");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZhpr");
+  return func_ptr(uplo, n, alpha, x, incx, AP);
+}
+
+void CUBLASWINAPI cublasSsyr2(char uplo, int n, float alpha, const float *x,
+                              int incx, const float *y, int incy, float *A,
+                              int lda) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, int, float, const float *, int,
+                                       const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsyr2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSsyr2");
+  return func_ptr(uplo, n, alpha, x, incx, y, incy, A, lda);
+}
+
+void CUBLASWINAPI cublasDsyr2(char uplo, int n, double alpha, const double *x,
+                              int incx, const double *y, int incy, double *A,
+                              int lda) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, int, double, const double *, int,
+                                       const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsyr2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDsyr2");
+  return func_ptr(uplo, n, alpha, x, incx, y, incy, A, lda);
+}
+
+void CUBLASWINAPI cublasCher2(char uplo, int n, cuComplex alpha,
+                              const cuComplex *x, int incx, const cuComplex *y,
+                              int incy, cuComplex *A, int lda) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, cuComplex, const cuComplex *, int,
+                           const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCher2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCher2");
+  return func_ptr(uplo, n, alpha, x, incx, y, incy, A, lda);
+}
+
+void CUBLASWINAPI cublasZher2(char uplo, int n, cuDoubleComplex alpha,
+                              const cuDoubleComplex *x, int incx,
+                              const cuDoubleComplex *y, int incy,
+                              cuDoubleComplex *A, int lda) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZher2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZher2");
+  return func_ptr(uplo, n, alpha, x, incx, y, incy, A, lda);
+}
+
+void CUBLASWINAPI cublasSspr2(char uplo, int n, float alpha, const float *x,
+                              int incx, const float *y, int incy, float *AP) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, int, float, const float *, int,
+                                       const float *, int, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSspr2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSspr2");
+  return func_ptr(uplo, n, alpha, x, incx, y, incy, AP);
+}
+
+void CUBLASWINAPI cublasDspr2(char uplo, int n, double alpha, const double *x,
+                              int incx, const double *y, int incy, double *AP) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, int, double, const double *, int,
+                                       const double *, int, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDspr2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDspr2");
+  return func_ptr(uplo, n, alpha, x, incx, y, incy, AP);
+}
+
+void CUBLASWINAPI cublasChpr2(char uplo, int n, cuComplex alpha,
+                              const cuComplex *x, int incx, const cuComplex *y,
+                              int incy, cuComplex *AP) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, int, cuComplex, const cuComplex *, int,
+                           const cuComplex *, int, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChpr2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasChpr2");
+  return func_ptr(uplo, n, alpha, x, incx, y, incy, AP);
+}
+
+void CUBLASWINAPI cublasZhpr2(char uplo, int n, cuDoubleComplex alpha,
+                              const cuDoubleComplex *x, int incx,
+                              const cuDoubleComplex *y, int incy,
+                              cuDoubleComplex *AP) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhpr2");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZhpr2");
+  return func_ptr(uplo, n, alpha, x, incx, y, incy, AP);
+}
+
+void CUBLASWINAPI cublasSgemm(char transa, char transb, int m, int n, int k,
+                              float alpha, const float *A, int lda,
+                              const float *B, int ldb, float beta, float *C,
+                              int ldc) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, int, int, int, float, const float *, int,
+                           const float *, int, float, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSgemm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSgemm");
+  return func_ptr(transa, transb, m, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasDgemm(char transa, char transb, int m, int n, int k,
+                              double alpha, const double *A, int lda,
+                              const double *B, int ldb, double beta, double *C,
+                              int ldc) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, int, int, int, double, const double *,
+                           int, const double *, int, double, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDgemm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDgemm");
+  return func_ptr(transa, transb, m, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasCgemm(char transa, char transb, int m, int n, int k,
+                              cuComplex alpha, const cuComplex *A, int lda,
+                              const cuComplex *B, int ldb, cuComplex beta,
+                              cuComplex *C, int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, int, cuComplex, const cuComplex *, int,
+      const cuComplex *, int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCgemm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCgemm");
+  return func_ptr(transa, transb, m, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasZgemm(char transa, char transb, int m, int n, int k,
+                              cuDoubleComplex alpha, const cuDoubleComplex *A,
+                              int lda, const cuDoubleComplex *B, int ldb,
+                              cuDoubleComplex beta, cuDoubleComplex *C,
+                              int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZgemm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZgemm");
+  return func_ptr(transa, transb, m, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasSsyrk(char uplo, char trans, int n, int k, float alpha,
+                              const float *A, int lda, float beta, float *C,
+                              int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, int, int, float,
+                                       const float *, int, float, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsyrk");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSsyrk");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasDsyrk(char uplo, char trans, int n, int k, double alpha,
+                              const double *A, int lda, double beta, double *C,
+                              int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, double, const double *, int, double, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsyrk");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDsyrk");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasCsyrk(char uplo, char trans, int n, int k,
+                              cuComplex alpha, const cuComplex *A, int lda,
+                              cuComplex beta, cuComplex *C, int ldc) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, int, int, cuComplex, const cuComplex *,
+                           int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsyrk");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCsyrk");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasZsyrk(char uplo, char trans, int n, int k,
+                              cuDoubleComplex alpha, const cuDoubleComplex *A,
+                              int lda, cuDoubleComplex beta, cuDoubleComplex *C,
+                              int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, int, int, cuDoubleComplex,
+                                       const cuDoubleComplex *, int,
+                                       cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZsyrk");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZsyrk");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasCherk(char uplo, char trans, int n, int k, float alpha,
+                              const cuComplex *A, int lda, float beta,
+                              cuComplex *C, int ldc) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, int, int, float, const cuComplex *, int,
+                           float, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCherk");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCherk");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasZherk(char uplo, char trans, int n, int k, double alpha,
+                              const cuDoubleComplex *A, int lda, double beta,
+                              cuDoubleComplex *C, int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, int, int, double,
+                                       const cuDoubleComplex *, int, double,
+                                       cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZherk");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZherk");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasSsyr2k(char uplo, char trans, int n, int k, float alpha,
+                               const float *A, int lda, const float *B, int ldb,
+                               float beta, float *C, int ldc) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, int, int, float, const float *, int,
+                           const float *, int, float, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsyr2k");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSsyr2k");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasDsyr2k(char uplo, char trans, int n, int k,
+                               double alpha, const double *A, int lda,
+                               const double *B, int ldb, double beta, double *C,
+                               int ldc) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, int, int, double, const double *, int,
+                           const double *, int, double, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsyr2k");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDsyr2k");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasCsyr2k(char uplo, char trans, int n, int k,
+                               cuComplex alpha, const cuComplex *A, int lda,
+                               const cuComplex *B, int ldb, cuComplex beta,
+                               cuComplex *C, int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, cuComplex, const cuComplex *, int,
+      const cuComplex *, int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsyr2k");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCsyr2k");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasZsyr2k(char uplo, char trans, int n, int k,
+                               cuDoubleComplex alpha, const cuDoubleComplex *A,
+                               int lda, const cuDoubleComplex *B, int ldb,
+                               cuDoubleComplex beta, cuDoubleComplex *C,
+                               int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZsyr2k");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZsyr2k");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasCher2k(char uplo, char trans, int n, int k,
+                               cuComplex alpha, const cuComplex *A, int lda,
+                               const cuComplex *B, int ldb, float beta,
+                               cuComplex *C, int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, cuComplex, const cuComplex *, int,
+      const cuComplex *, int, float, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCher2k");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCher2k");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasZher2k(char uplo, char trans, int n, int k,
+                               cuDoubleComplex alpha, const cuDoubleComplex *A,
+                               int lda, const cuDoubleComplex *B, int ldb,
+                               double beta, cuDoubleComplex *C, int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, double, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZher2k");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZher2k");
+  return func_ptr(uplo, trans, n, k, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasSsymm(char side, char uplo, int m, int n, float alpha,
+                              const float *A, int lda, const float *B, int ldb,
+                              float beta, float *C, int ldc) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, int, int, float, const float *, int,
+                           const float *, int, float, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasSsymm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasSsymm");
+  return func_ptr(side, uplo, m, n, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasDsymm(char side, char uplo, int m, int n, double alpha,
+                              const double *A, int lda, const double *B,
+                              int ldb, double beta, double *C, int ldc) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, int, int, double, const double *, int,
+                           const double *, int, double, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDsymm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDsymm");
+  return func_ptr(side, uplo, m, n, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasCsymm(char side, char uplo, int m, int n,
+                              cuComplex alpha, const cuComplex *A, int lda,
+                              const cuComplex *B, int ldb, cuComplex beta,
+                              cuComplex *C, int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, cuComplex, const cuComplex *, int,
+      const cuComplex *, int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCsymm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCsymm");
+  return func_ptr(side, uplo, m, n, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasZsymm(char side, char uplo, int m, int n,
+                              cuDoubleComplex alpha, const cuDoubleComplex *A,
+                              int lda, const cuDoubleComplex *B, int ldb,
+                              cuDoubleComplex beta, cuDoubleComplex *C,
+                              int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZsymm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZsymm");
+  return func_ptr(side, uplo, m, n, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasChemm(char side, char uplo, int m, int n,
+                              cuComplex alpha, const cuComplex *A, int lda,
+                              const cuComplex *B, int ldb, cuComplex beta,
+                              cuComplex *C, int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, cuComplex, const cuComplex *, int,
+      const cuComplex *, int, cuComplex, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasChemm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasChemm");
+  return func_ptr(side, uplo, m, n, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasZhemm(char side, char uplo, int m, int n,
+                              cuDoubleComplex alpha, const cuDoubleComplex *A,
+                              int lda, const cuDoubleComplex *B, int ldb,
+                              cuDoubleComplex beta, cuDoubleComplex *C,
+                              int ldc) {
+  using FuncPtr = void(CUBLASWINAPI *)(
+      char, char, int, int, cuDoubleComplex, const cuDoubleComplex *, int,
+      const cuDoubleComplex *, int, cuDoubleComplex, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZhemm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZhemm");
+  return func_ptr(side, uplo, m, n, alpha, A, lda, B, ldb, beta, C, ldc);
+}
+
+void CUBLASWINAPI cublasStrsm(char side, char uplo, char transa, char diag,
+                              int m, int n, float alpha, const float *A,
+                              int lda, float *B, int ldb) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, char, int, int, float,
+                                       const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStrsm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasStrsm");
+  return func_ptr(side, uplo, transa, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+void CUBLASWINAPI cublasDtrsm(char side, char uplo, char transa, char diag,
+                              int m, int n, double alpha, const double *A,
+                              int lda, double *B, int ldb) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, char, int, int, double,
+                                       const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtrsm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDtrsm");
+  return func_ptr(side, uplo, transa, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+void CUBLASWINAPI cublasCtrsm(char side, char uplo, char transa, char diag,
+                              int m, int n, cuComplex alpha, const cuComplex *A,
+                              int lda, cuComplex *B, int ldb) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, char, char, int, int, cuComplex,
+                           const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtrsm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCtrsm");
+  return func_ptr(side, uplo, transa, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+void CUBLASWINAPI cublasZtrsm(char side, char uplo, char transa, char diag,
+                              int m, int n, cuDoubleComplex alpha,
+                              const cuDoubleComplex *A, int lda,
+                              cuDoubleComplex *B, int ldb) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, char, int, int,
+                                       cuDoubleComplex, const cuDoubleComplex *,
+                                       int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtrsm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZtrsm");
+  return func_ptr(side, uplo, transa, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+void CUBLASWINAPI cublasStrmm(char side, char uplo, char transa, char diag,
+                              int m, int n, float alpha, const float *A,
+                              int lda, float *B, int ldb) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, char, int, int, float,
+                                       const float *, int, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasStrmm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasStrmm");
+  return func_ptr(side, uplo, transa, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+void CUBLASWINAPI cublasDtrmm(char side, char uplo, char transa, char diag,
+                              int m, int n, double alpha, const double *A,
+                              int lda, double *B, int ldb) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, char, int, int, double,
+                                       const double *, int, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasDtrmm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasDtrmm");
+  return func_ptr(side, uplo, transa, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+void CUBLASWINAPI cublasCtrmm(char side, char uplo, char transa, char diag,
+                              int m, int n, cuComplex alpha, const cuComplex *A,
+                              int lda, cuComplex *B, int ldb) {
+  using FuncPtr =
+      void(CUBLASWINAPI *)(char, char, char, char, int, int, cuComplex,
+                           const cuComplex *, int, cuComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasCtrmm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasCtrmm");
+  return func_ptr(side, uplo, transa, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+void CUBLASWINAPI cublasZtrmm(char side, char uplo, char transa, char diag,
+                              int m, int n, cuDoubleComplex alpha,
+                              const cuDoubleComplex *A, int lda,
+                              cuDoubleComplex *B, int ldb) {
+  using FuncPtr = void(CUBLASWINAPI *)(char, char, char, char, int, int,
+                                       cuDoubleComplex, const cuDoubleComplex *,
+                                       int, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cublasZtrmm");
+  if (!func_ptr) LogFatalSymbolNotFound("cublasZtrmm");
+  return func_ptr(side, uplo, transa, diag, m, n, alpha, A, lda, B, ldb);
+}
+
+}
diff --git a/tensorflow/stream_executor/cuda/cublas_stub.cc b/tensorflow/stream_executor/cuda/cublas_stub.cc
index b8e203fe235..f9cfcc9d4b3 100644
--- a/tensorflow/stream_executor/cuda/cublas_stub.cc
+++ b/tensorflow/stream_executor/cuda/cublas_stub.cc
@@ -61,5 +61,5 @@ typedef enum {} cublasMath_t;
 #if CUDA_VERSION < 9020
 #include "tensorflow/stream_executor/cuda/cublas_9_0.inc"
 #else
-#include "tensorflow/stream_executor/cuda/cublas_10_0.inc"
+#include "tensorflow/stream_executor/cuda/cublas_11_0.inc"
 #endif
diff --git a/tensorflow/stream_executor/cuda/cuda_12_0.inc b/tensorflow/stream_executor/cuda/cuda_12_0.inc
new file mode 100644
index 00000000000..9d97438c68f
--- /dev/null
+++ b/tensorflow/stream_executor/cuda/cuda_12_0.inc
@@ -0,0 +1,3324 @@
+// Auto-generated, do not edit.
+
+extern "C" {
+
+CUresult CUDAAPI cuGetErrorString(CUresult error, const char **pStr) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUresult, const char **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGetErrorString");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(error, pStr);
+}
+
+CUresult CUDAAPI cuGetErrorName(CUresult error, const char **pStr) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUresult, const char **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGetErrorName");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(error, pStr);
+}
+
+CUresult CUDAAPI cuInit(unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuInit");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(Flags);
+}
+
+CUresult CUDAAPI cuDriverGetVersion(int *driverVersion) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDriverGetVersion");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(driverVersion);
+}
+
+CUresult CUDAAPI cuDeviceGet(CUdevice *device, int ordinal) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device, ordinal);
+}
+
+CUresult CUDAAPI cuDeviceGetCount(int *count) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetCount");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(count);
+}
+
+CUresult CUDAAPI cuDeviceGetName(char *name, int len, CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(char *, int, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetName");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(name, len, dev);
+}
+
+CUresult CUDAAPI cuDeviceGetUuid(CUuuid *uuid, CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUuuid *, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetUuid");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(uuid, dev);
+}
+
+CUresult CUDAAPI cuDeviceGetUuid_v2(CUuuid *uuid, CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUuuid *, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetUuid_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(uuid, dev);
+}
+
+CUresult CUDAAPI cuDeviceGetLuid(char *luid, unsigned int *deviceNodeMask,
+                                 CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(char *, unsigned int *, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetLuid");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(luid, deviceNodeMask, dev);
+}
+
+CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(size_t *, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceTotalMem_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(bytes, dev);
+}
+
+CUresult CUDAAPI cuDeviceGetTexture1DLinearMaxWidth(size_t *maxWidthInElements,
+                                                    CUarray_format format,
+                                                    unsigned numChannels,
+                                                    CUdevice dev) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(size_t *, CUarray_format, unsigned int, CUdevice);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuDeviceGetTexture1DLinearMaxWidth");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(maxWidthInElements, format, numChannels, dev);
+}
+
+CUresult CUDAAPI cuDeviceGetAttribute(int *pi, CUdevice_attribute attrib,
+                                      CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *, CUdevice_attribute, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pi, attrib, dev);
+}
+
+CUresult CUDAAPI cuDeviceGetNvSciSyncAttributes(void *nvSciSyncAttrList,
+                                                CUdevice dev, int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(void *, CUdevice, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetNvSciSyncAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(nvSciSyncAttrList, dev, flags);
+}
+
+CUresult CUDAAPI cuDeviceSetMemPool(CUdevice dev, CUmemoryPool pool) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice, CUmemoryPool);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceSetMemPool");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dev, pool);
+}
+
+CUresult CUDAAPI cuDeviceGetMemPool(CUmemoryPool *pool, CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmemoryPool *, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetMemPool");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pool, dev);
+}
+
+CUresult CUDAAPI cuDeviceGetDefaultMemPool(CUmemoryPool *pool_out,
+                                           CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmemoryPool *, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetDefaultMemPool");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pool_out, dev);
+}
+
+CUresult CUDAAPI cuDeviceGetExecAffinitySupport(int *pi,
+                                                CUexecAffinityType type,
+                                                CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *, CUexecAffinityType, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetExecAffinitySupport");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pi, type, dev);
+}
+
+CUresult CUDAAPI
+cuFlushGPUDirectRDMAWrites(CUflushGPUDirectRDMAWritesTarget target,
+                           CUflushGPUDirectRDMAWritesScope scope) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUflushGPUDirectRDMAWritesTarget,
+                                      CUflushGPUDirectRDMAWritesScope);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuFlushGPUDirectRDMAWrites");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(target, scope);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuDeviceGetProperties(CUdevprop *prop,
+                                                         CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevprop *, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetProperties");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(prop, dev);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuDeviceComputeCapability(int *major,
+                                                             int *minor,
+                                                             CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *, int *, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceComputeCapability");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(major, minor, dev);
+}
+
+CUresult CUDAAPI cuDevicePrimaryCtxRetain(CUcontext *pctx, CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext *, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDevicePrimaryCtxRetain");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pctx, dev);
+}
+
+CUresult CUDAAPI cuDevicePrimaryCtxRelease(CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDevicePrimaryCtxRelease_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dev);
+}
+
+CUresult CUDAAPI cuDevicePrimaryCtxSetFlags(CUdevice dev, unsigned int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDevicePrimaryCtxSetFlags_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dev, flags);
+}
+
+CUresult CUDAAPI cuDevicePrimaryCtxGetState(CUdevice dev, unsigned int *flags,
+                                            int *active) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice, unsigned int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDevicePrimaryCtxGetState");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dev, flags, active);
+}
+
+CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDevicePrimaryCtxReset_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dev);
+}
+
+CUresult CUDAAPI cuCtxCreate(CUcontext *pctx, unsigned int flags,
+                             CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext *, unsigned int, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxCreate_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pctx, flags, dev);
+}
+
+CUresult CUDAAPI cuCtxCreate_v3(CUcontext *pctx,
+                                CUexecAffinityParam *paramsArray, int numParams,
+                                unsigned int flags, CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext *, CUexecAffinityParam *, int,
+                                      unsigned int, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxCreate_v3");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pctx, paramsArray, numParams, flags, dev);
+}
+
+CUresult CUDAAPI cuCtxDestroy(CUcontext ctx) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxDestroy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctx);
+}
+
+CUresult CUDAAPI cuCtxPushCurrent(CUcontext ctx) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxPushCurrent_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctx);
+}
+
+CUresult CUDAAPI cuCtxPopCurrent(CUcontext *pctx) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxPopCurrent_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pctx);
+}
+
+CUresult CUDAAPI cuCtxSetCurrent(CUcontext ctx) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxSetCurrent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctx);
+}
+
+CUresult CUDAAPI cuCtxGetCurrent(CUcontext *pctx) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxGetCurrent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pctx);
+}
+
+CUresult CUDAAPI cuCtxGetDevice(CUdevice *device) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxGetDevice");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device);
+}
+
+CUresult CUDAAPI cuCtxGetFlags(unsigned int *flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(unsigned int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxGetFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(flags);
+}
+
+CUresult CUDAAPI cuCtxGetId(CUcontext ctx, unsigned long long *ctxId) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext, unsigned long long *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxGetId");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctx, ctxId);
+}
+
+CUresult CUDAAPI cuCtxSynchronize(void) {
+  using FuncPtr = CUresult(CUDAAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxSynchronize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+CUresult CUDAAPI cuCtxSetLimit(CUlimit limit, size_t value) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUlimit, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxSetLimit");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(limit, value);
+}
+
+CUresult CUDAAPI cuCtxGetLimit(size_t *pvalue, CUlimit limit) {
+  using FuncPtr = CUresult(CUDAAPI *)(size_t *, CUlimit);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxGetLimit");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pvalue, limit);
+}
+
+CUresult CUDAAPI cuCtxGetCacheConfig(CUfunc_cache *pconfig) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunc_cache *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxGetCacheConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pconfig);
+}
+
+CUresult CUDAAPI cuCtxSetCacheConfig(CUfunc_cache config) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunc_cache);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxSetCacheConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(config);
+}
+
+CUresult CUDAAPI cuCtxGetSharedMemConfig(CUsharedconfig *pConfig) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUsharedconfig *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxGetSharedMemConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pConfig);
+}
+
+CUresult CUDAAPI cuCtxSetSharedMemConfig(CUsharedconfig config) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUsharedconfig);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxSetSharedMemConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(config);
+}
+
+CUresult CUDAAPI cuCtxGetApiVersion(CUcontext ctx, unsigned int *version) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext, unsigned int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxGetApiVersion");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctx, version);
+}
+
+CUresult CUDAAPI cuCtxGetStreamPriorityRange(int *leastPriority,
+                                             int *greatestPriority) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxGetStreamPriorityRange");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(leastPriority, greatestPriority);
+}
+
+CUresult CUDAAPI cuCtxResetPersistingL2Cache(void) {
+  using FuncPtr = CUresult(CUDAAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxResetPersistingL2Cache");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+CUresult CUDAAPI cuCtxGetExecAffinity(CUexecAffinityParam *pExecAffinity,
+                                      CUexecAffinityType type) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUexecAffinityParam *, CUexecAffinityType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxGetExecAffinity");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pExecAffinity, type);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuCtxAttach(CUcontext *pctx,
+                                               unsigned int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxAttach");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pctx, flags);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuCtxDetach(CUcontext ctx) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxDetach");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctx);
+}
+
+CUresult CUDAAPI cuModuleLoad(CUmodule *module, const char *fname) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmodule *, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuModuleLoad");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(module, fname);
+}
+
+CUresult CUDAAPI cuModuleLoadData(CUmodule *module, const void *image) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmodule *, const void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuModuleLoadData");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(module, image);
+}
+
+CUresult CUDAAPI cuModuleLoadDataEx(CUmodule *module, const void *image,
+                                    unsigned int numOptions,
+                                    CUjit_option *options,
+                                    void **optionValues) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmodule *, const void *, unsigned int,
+                                      CUjit_option *, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuModuleLoadDataEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(module, image, numOptions, options, optionValues);
+}
+
+CUresult CUDAAPI cuModuleLoadFatBinary(CUmodule *module, const void *fatCubin) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmodule *, const void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuModuleLoadFatBinary");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(module, fatCubin);
+}
+
+CUresult CUDAAPI cuModuleUnload(CUmodule hmod) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmodule);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuModuleUnload");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hmod);
+}
+
+CUresult CUDAAPI cuModuleGetLoadingMode(CUmoduleLoadingMode *mode) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmoduleLoadingMode *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuModuleGetLoadingMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mode);
+}
+
+CUresult CUDAAPI cuModuleGetFunction(CUfunction *hfunc, CUmodule hmod,
+                                     const char *name) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction *, CUmodule, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuModuleGetFunction");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, hmod, name);
+}
+
+CUresult CUDAAPI cuModuleGetGlobal(CUdeviceptr *dptr, size_t *bytes,
+                                   CUmodule hmod, const char *name) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr *, size_t *, CUmodule, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuModuleGetGlobal_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr, bytes, hmod, name);
+}
+
+CUresult CUDAAPI cuLinkCreate(unsigned int numOptions, CUjit_option *options,
+                              void **optionValues, CUlinkState *stateOut) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(unsigned int, CUjit_option *, void **, CUlinkState *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLinkCreate_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(numOptions, options, optionValues, stateOut);
+}
+
+CUresult CUDAAPI cuLinkAddData(CUlinkState state, CUjitInputType type,
+                               void *data, size_t size, const char *name,
+                               unsigned int numOptions, CUjit_option *options,
+                               void **optionValues) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUlinkState, CUjitInputType, void *, size_t,
+                          const char *, unsigned int, CUjit_option *, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLinkAddData_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(state, type, data, size, name, numOptions, options,
+                  optionValues);
+}
+
+CUresult CUDAAPI cuLinkAddFile(CUlinkState state, CUjitInputType type,
+                               const char *path, unsigned int numOptions,
+                               CUjit_option *options, void **optionValues) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUlinkState, CUjitInputType, const char *,
+                                      unsigned int, CUjit_option *, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLinkAddFile_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(state, type, path, numOptions, options, optionValues);
+}
+
+CUresult CUDAAPI cuLinkComplete(CUlinkState state, void **cubinOut,
+                                size_t *sizeOut) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUlinkState, void **, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLinkComplete");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(state, cubinOut, sizeOut);
+}
+
+CUresult CUDAAPI cuLinkDestroy(CUlinkState state) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUlinkState);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLinkDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(state);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuModuleGetTexRef(CUtexref *pTexRef,
+                                                     CUmodule hmod,
+                                                     const char *name) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref *, CUmodule, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuModuleGetTexRef");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pTexRef, hmod, name);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuModuleGetSurfRef(CUsurfref *pSurfRef,
+                                                      CUmodule hmod,
+                                                      const char *name) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUsurfref *, CUmodule, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuModuleGetSurfRef");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pSurfRef, hmod, name);
+}
+
+CUresult CUDAAPI cuLibraryLoadData(CUlibrary *library, const void *code,
+                                   CUjit_option *jitOptions,
+                                   void **jitOptionsValues,
+                                   unsigned int numJitOptions,
+                                   CUlibraryOption *libraryOptions,
+                                   void **libraryOptionValues,
+                                   unsigned int numLibraryOptions) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUlibrary *, const void *, CUjit_option *,
+                                      void **, unsigned int, CUlibraryOption *,
+                                      void **, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLibraryLoadData");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(library, code, jitOptions, jitOptionsValues, numJitOptions,
+                  libraryOptions, libraryOptionValues, numLibraryOptions);
+}
+
+CUresult CUDAAPI cuLibraryLoadFromFile(CUlibrary *library, const char *fileName,
+                                       CUjit_option *jitOptions,
+                                       void **jitOptionsValues,
+                                       unsigned int numJitOptions,
+                                       CUlibraryOption *libraryOptions,
+                                       void **libraryOptionValues,
+                                       unsigned int numLibraryOptions) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUlibrary *, const char *, CUjit_option *,
+                                      void **, unsigned int, CUlibraryOption *,
+                                      void **, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLibraryLoadFromFile");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(library, fileName, jitOptions, jitOptionsValues,
+                  numJitOptions, libraryOptions, libraryOptionValues,
+                  numLibraryOptions);
+}
+
+CUresult CUDAAPI cuLibraryUnload(CUlibrary library) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUlibrary);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLibraryUnload");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(library);
+}
+
+CUresult CUDAAPI cuLibraryGetKernel(CUkernel *pKernel, CUlibrary library,
+                                    const char *name) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUkernel *, CUlibrary, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLibraryGetKernel");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pKernel, library, name);
+}
+
+CUresult CUDAAPI cuLibraryGetModule(CUmodule *pMod, CUlibrary library) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmodule *, CUlibrary);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLibraryGetModule");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pMod, library);
+}
+
+CUresult CUDAAPI cuKernelGetFunction(CUfunction *pFunc, CUkernel kernel) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction *, CUkernel);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuKernelGetFunction");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pFunc, kernel);
+}
+
+CUresult CUDAAPI cuLibraryGetGlobal(CUdeviceptr *dptr, size_t *bytes,
+                                    CUlibrary library, const char *name) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr *, size_t *, CUlibrary, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLibraryGetGlobal");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr, bytes, library, name);
+}
+
+CUresult CUDAAPI cuLibraryGetManaged(CUdeviceptr *dptr, size_t *bytes,
+                                     CUlibrary library, const char *name) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr *, size_t *, CUlibrary, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLibraryGetManaged");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr, bytes, library, name);
+}
+
+CUresult CUDAAPI cuLibraryGetUnifiedFunction(void **fptr, CUlibrary library,
+                                             const char *symbol) {
+  using FuncPtr = CUresult(CUDAAPI *)(void **, CUlibrary, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLibraryGetUnifiedFunction");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(fptr, library, symbol);
+}
+
+CUresult CUDAAPI cuKernelGetAttribute(int *pi, CUfunction_attribute attrib,
+                                      CUkernel kernel, CUdevice dev) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(int *, CUfunction_attribute, CUkernel, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuKernelGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pi, attrib, kernel, dev);
+}
+
+CUresult CUDAAPI cuKernelSetAttribute(CUfunction_attribute attrib, int val,
+                                      CUkernel kernel, CUdevice dev) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUfunction_attribute, int, CUkernel, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuKernelSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(attrib, val, kernel, dev);
+}
+
+CUresult CUDAAPI cuKernelSetCacheConfig(CUkernel kernel, CUfunc_cache config,
+                                        CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUkernel, CUfunc_cache, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuKernelSetCacheConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(kernel, config, dev);
+}
+
+CUresult CUDAAPI cuMemGetInfo(size_t *free, size_t *total) {
+  using FuncPtr = CUresult(CUDAAPI *)(size_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemGetInfo_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(free, total);
+}
+
+CUresult CUDAAPI cuMemAlloc(CUdeviceptr *dptr, size_t bytesize) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemAlloc_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr, bytesize);
+}
+
+CUresult CUDAAPI cuMemAllocPitch(CUdeviceptr *dptr, size_t *pPitch,
+                                 size_t WidthInBytes, size_t Height,
+                                 unsigned int ElementSizeBytes) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr *, size_t *, size_t, size_t,
+                                      unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemAllocPitch_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr, pPitch, WidthInBytes, Height, ElementSizeBytes);
+}
+
+CUresult CUDAAPI cuMemFree(CUdeviceptr dptr) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemFree_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr);
+}
+
+CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, size_t *psize,
+                                      CUdeviceptr dptr) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr *, size_t *, CUdeviceptr);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemGetAddressRange_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pbase, psize, dptr);
+}
+
+CUresult CUDAAPI cuMemAllocHost(void **pp, size_t bytesize) {
+  using FuncPtr = CUresult(CUDAAPI *)(void **, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemAllocHost_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pp, bytesize);
+}
+
+CUresult CUDAAPI cuMemFreeHost(void *p) {
+  using FuncPtr = CUresult(CUDAAPI *)(void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemFreeHost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(p);
+}
+
+CUresult CUDAAPI cuMemHostAlloc(void **pp, size_t bytesize,
+                                unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(void **, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemHostAlloc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pp, bytesize, Flags);
+}
+
+CUresult CUDAAPI cuMemHostGetDevicePointer(CUdeviceptr *pdptr, void *p,
+                                           unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr *, void *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemHostGetDevicePointer_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pdptr, p, Flags);
+}
+
+CUresult CUDAAPI cuMemHostGetFlags(unsigned int *pFlags, void *p) {
+  using FuncPtr = CUresult(CUDAAPI *)(unsigned int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemHostGetFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pFlags, p);
+}
+
+CUresult CUDAAPI cuMemAllocManaged(CUdeviceptr *dptr, size_t bytesize,
+                                   unsigned int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr *, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemAllocManaged");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr, bytesize, flags);
+}
+
+CUresult CUDAAPI cuDeviceGetByPCIBusId(CUdevice *dev, const char *pciBusId) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice *, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetByPCIBusId");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dev, pciBusId);
+}
+
+CUresult CUDAAPI cuDeviceGetPCIBusId(char *pciBusId, int len, CUdevice dev) {
+  using FuncPtr = CUresult(CUDAAPI *)(char *, int, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetPCIBusId");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pciBusId, len, dev);
+}
+
+CUresult CUDAAPI cuIpcGetEventHandle(CUipcEventHandle *pHandle, CUevent event) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUipcEventHandle *, CUevent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuIpcGetEventHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pHandle, event);
+}
+
+CUresult CUDAAPI cuIpcOpenEventHandle(CUevent *phEvent,
+                                      CUipcEventHandle handle) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUevent *, CUipcEventHandle);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuIpcOpenEventHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phEvent, handle);
+}
+
+CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUipcMemHandle *, CUdeviceptr);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuIpcGetMemHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pHandle, dptr);
+}
+
+CUresult CUDAAPI cuIpcOpenMemHandle(CUdeviceptr *pdptr, CUipcMemHandle handle,
+                                    unsigned int Flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr *, CUipcMemHandle, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuIpcOpenMemHandle_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pdptr, handle, Flags);
+}
+
+CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuIpcCloseMemHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr);
+}
+
+CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize,
+                                   unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(void *, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemHostRegister_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(p, bytesize, Flags);
+}
+
+CUresult CUDAAPI cuMemHostUnregister(void *p) {
+  using FuncPtr = CUresult(CUDAAPI *)(void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemHostUnregister");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(p);
+}
+
+CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, CUdeviceptr, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, src, ByteCount);
+}
+
+CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext,
+                              CUdeviceptr srcDevice, CUcontext srcContext,
+                              size_t ByteCount) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, CUcontext, CUdeviceptr,
+                                      CUcontext, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyPeer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, dstContext, srcDevice, srcContext, ByteCount);
+}
+
+CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost,
+                              size_t ByteCount) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, const void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyHtoD_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, srcHost, ByteCount);
+}
+
+CUresult CUDAAPI cuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice,
+                              size_t ByteCount) {
+  using FuncPtr = CUresult(CUDAAPI *)(void *, CUdeviceptr, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyDtoH_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstHost, srcDevice, ByteCount);
+}
+
+CUresult CUDAAPI cuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice,
+                              size_t ByteCount) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, CUdeviceptr, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyDtoD_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, srcDevice, ByteCount);
+}
+
+CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, size_t dstOffset,
+                              CUdeviceptr srcDevice, size_t ByteCount) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarray, size_t, CUdeviceptr, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyDtoA_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstArray, dstOffset, srcDevice, ByteCount);
+}
+
+CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray,
+                              size_t srcOffset, size_t ByteCount) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, CUarray, size_t, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyAtoD_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, srcArray, srcOffset, ByteCount);
+}
+
+CUresult CUDAAPI cuMemcpyHtoA(CUarray dstArray, size_t dstOffset,
+                              const void *srcHost, size_t ByteCount) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarray, size_t, const void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyHtoA_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstArray, dstOffset, srcHost, ByteCount);
+}
+
+CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset,
+                              size_t ByteCount) {
+  using FuncPtr = CUresult(CUDAAPI *)(void *, CUarray, size_t, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyAtoH_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstHost, srcArray, srcOffset, ByteCount);
+}
+
+CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, size_t dstOffset,
+                              CUarray srcArray, size_t srcOffset,
+                              size_t ByteCount) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarray, size_t, CUarray, size_t, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyAtoA_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstArray, dstOffset, srcArray, srcOffset, ByteCount);
+}
+
+CUresult CUDAAPI cuMemcpy2D(const CUDA_MEMCPY2D *pCopy) {
+  using FuncPtr = CUresult(CUDAAPI *)(const CUDA_MEMCPY2D *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpy2D_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pCopy);
+}
+
+CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy) {
+  using FuncPtr = CUresult(CUDAAPI *)(const CUDA_MEMCPY2D *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpy2DUnaligned_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pCopy);
+}
+
+CUresult CUDAAPI cuMemcpy3D(const CUDA_MEMCPY3D *pCopy) {
+  using FuncPtr = CUresult(CUDAAPI *)(const CUDA_MEMCPY3D *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpy3D_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pCopy);
+}
+
+CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy) {
+  using FuncPtr = CUresult(CUDAAPI *)(const CUDA_MEMCPY3D_PEER *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpy3DPeer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pCopy);
+}
+
+CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src,
+                               size_t ByteCount, CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, CUdeviceptr, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, src, ByteCount, hStream);
+}
+
+CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext,
+                                   CUdeviceptr srcDevice, CUcontext srcContext,
+                                   size_t ByteCount, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, CUcontext, CUdeviceptr,
+                                      CUcontext, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyPeerAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, dstContext, srcDevice, srcContext, ByteCount,
+                  hStream);
+}
+
+CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost,
+                                   size_t ByteCount, CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, const void *, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyHtoDAsync_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, srcHost, ByteCount, hStream);
+}
+
+CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice,
+                                   size_t ByteCount, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(void *, CUdeviceptr, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyDtoHAsync_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstHost, srcDevice, ByteCount, hStream);
+}
+
+CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice,
+                                   size_t ByteCount, CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, CUdeviceptr, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyDtoDAsync_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, srcDevice, ByteCount, hStream);
+}
+
+CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset,
+                                   const void *srcHost, size_t ByteCount,
+                                   CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUarray, size_t, const void *, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyHtoAAsync_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstArray, dstOffset, srcHost, ByteCount, hStream);
+}
+
+CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray,
+                                   size_t srcOffset, size_t ByteCount,
+                                   CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(void *, CUarray, size_t, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpyAtoHAsync_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstHost, srcArray, srcOffset, ByteCount, hStream);
+}
+
+CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(const CUDA_MEMCPY2D *, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpy2DAsync_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pCopy, hStream);
+}
+
+CUresult CUDAAPI cuMemcpy3DAsync(const CUDA_MEMCPY3D *pCopy, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(const CUDA_MEMCPY3D *, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpy3DAsync_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pCopy, hStream);
+}
+
+CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy,
+                                     CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(const CUDA_MEMCPY3D_PEER *, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemcpy3DPeerAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pCopy, hStream);
+}
+
+CUresult CUDAAPI cuMemsetD8(CUdeviceptr dstDevice, unsigned char uc, size_t N) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, unsigned char, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD8_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, uc, N);
+}
+
+CUresult CUDAAPI cuMemsetD16(CUdeviceptr dstDevice, unsigned short us,
+                             size_t N) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, unsigned short, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD16_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, us, N);
+}
+
+CUresult CUDAAPI cuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, size_t N) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, unsigned int, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD32_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, ui, N);
+}
+
+CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, size_t dstPitch,
+                              unsigned char uc, size_t Width, size_t Height) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, size_t, unsigned char, size_t, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD2D8_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, dstPitch, uc, Width, Height);
+}
+
+CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, size_t dstPitch,
+                               unsigned short us, size_t Width, size_t Height) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, size_t, unsigned short, size_t, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD2D16_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, dstPitch, us, Width, Height);
+}
+
+CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, size_t dstPitch,
+                               unsigned int ui, size_t Width, size_t Height) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, size_t, unsigned int, size_t, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD2D32_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, dstPitch, ui, Width, Height);
+}
+
+CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc,
+                                 size_t N, CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, unsigned char, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD8Async");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, uc, N, hStream);
+}
+
+CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us,
+                                  size_t N, CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, unsigned short, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD16Async");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, us, N, hStream);
+}
+
+CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui,
+                                  size_t N, CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, unsigned int, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD32Async");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, ui, N, hStream);
+}
+
+CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch,
+                                   unsigned char uc, size_t Width,
+                                   size_t Height, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, size_t, unsigned char,
+                                      size_t, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD2D8Async");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, dstPitch, uc, Width, Height, hStream);
+}
+
+CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch,
+                                    unsigned short us, size_t Width,
+                                    size_t Height, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, size_t, unsigned short,
+                                      size_t, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD2D16Async");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, dstPitch, us, Width, Height, hStream);
+}
+
+CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch,
+                                    unsigned int ui, size_t Width,
+                                    size_t Height, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, size_t, unsigned int, size_t,
+                                      size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemsetD2D32Async");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dstDevice, dstPitch, ui, Width, Height, hStream);
+}
+
+CUresult CUDAAPI cuArrayCreate(CUarray *pHandle,
+                               const CUDA_ARRAY_DESCRIPTOR *pAllocateArray) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarray *, const CUDA_ARRAY_DESCRIPTOR *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuArrayCreate_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pHandle, pAllocateArray);
+}
+
+CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor,
+                                      CUarray hArray) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUDA_ARRAY_DESCRIPTOR *, CUarray);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuArrayGetDescriptor_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pArrayDescriptor, hArray);
+}
+
+CUresult CUDAAPI cuArrayGetSparseProperties(
+    CUDA_ARRAY_SPARSE_PROPERTIES *sparseProperties, CUarray array) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUDA_ARRAY_SPARSE_PROPERTIES *, CUarray);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuArrayGetSparseProperties");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(sparseProperties, array);
+}
+
+CUresult CUDAAPI cuMipmappedArrayGetSparseProperties(
+    CUDA_ARRAY_SPARSE_PROPERTIES *sparseProperties, CUmipmappedArray mipmap) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUDA_ARRAY_SPARSE_PROPERTIES *, CUmipmappedArray);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuMipmappedArrayGetSparseProperties");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(sparseProperties, mipmap);
+}
+
+CUresult CUDAAPI
+cuArrayGetMemoryRequirements(CUDA_ARRAY_MEMORY_REQUIREMENTS *memoryRequirements,
+                             CUarray array, CUdevice device) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUDA_ARRAY_MEMORY_REQUIREMENTS *, CUarray, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuArrayGetMemoryRequirements");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memoryRequirements, array, device);
+}
+
+CUresult CUDAAPI cuMipmappedArrayGetMemoryRequirements(
+    CUDA_ARRAY_MEMORY_REQUIREMENTS *memoryRequirements, CUmipmappedArray mipmap,
+    CUdevice device) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUDA_ARRAY_MEMORY_REQUIREMENTS *,
+                                      CUmipmappedArray, CUdevice);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuMipmappedArrayGetMemoryRequirements");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memoryRequirements, mipmap, device);
+}
+
+CUresult CUDAAPI cuArrayGetPlane(CUarray *pPlaneArray, CUarray hArray,
+                                 unsigned int planeIdx) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarray *, CUarray, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuArrayGetPlane");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pPlaneArray, hArray, planeIdx);
+}
+
+CUresult CUDAAPI cuArrayDestroy(CUarray hArray) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarray);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuArrayDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hArray);
+}
+
+CUresult CUDAAPI cuArray3DCreate(
+    CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pAllocateArray) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUarray *, const CUDA_ARRAY3D_DESCRIPTOR *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuArray3DCreate_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pHandle, pAllocateArray);
+}
+
+CUresult CUDAAPI cuArray3DGetDescriptor(
+    CUDA_ARRAY3D_DESCRIPTOR *pArrayDescriptor, CUarray hArray) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUDA_ARRAY3D_DESCRIPTOR *, CUarray);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuArray3DGetDescriptor_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pArrayDescriptor, hArray);
+}
+
+CUresult CUDAAPI
+cuMipmappedArrayCreate(CUmipmappedArray *pHandle,
+                       const CUDA_ARRAY3D_DESCRIPTOR *pMipmappedArrayDesc,
+                       unsigned int numMipmapLevels) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      CUmipmappedArray *, const CUDA_ARRAY3D_DESCRIPTOR *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMipmappedArrayCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pHandle, pMipmappedArrayDesc, numMipmapLevels);
+}
+
+CUresult CUDAAPI cuMipmappedArrayGetLevel(CUarray *pLevelArray,
+                                          CUmipmappedArray hMipmappedArray,
+                                          unsigned int level) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUarray *, CUmipmappedArray, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMipmappedArrayGetLevel");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pLevelArray, hMipmappedArray, level);
+}
+
+CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmipmappedArray);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMipmappedArrayDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hMipmappedArray);
+}
+
+CUresult CUDAAPI cuMemGetHandleForAddressRange(void *handle, CUdeviceptr dptr,
+                                               size_t size,
+                                               CUmemRangeHandleType handleType,
+                                               unsigned long long flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(void *, CUdeviceptr, size_t,
+                                      CUmemRangeHandleType, unsigned long long);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemGetHandleForAddressRange");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dptr, size, handleType, flags);
+}
+
+CUresult CUDAAPI cuMemAddressReserve(CUdeviceptr *ptr, size_t size,
+                                     size_t alignment, CUdeviceptr addr,
+                                     unsigned long long flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr *, size_t, size_t,
+                                      CUdeviceptr, unsigned long long);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemAddressReserve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr, size, alignment, addr, flags);
+}
+
+CUresult CUDAAPI cuMemAddressFree(CUdeviceptr ptr, size_t size) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemAddressFree");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr, size);
+}
+
+CUresult CUDAAPI cuMemCreate(CUmemGenericAllocationHandle *handle, size_t size,
+                             const CUmemAllocationProp *prop,
+                             unsigned long long flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUmemGenericAllocationHandle *, size_t,
+                          const CUmemAllocationProp *, unsigned long long);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, size, prop, flags);
+}
+
+CUresult CUDAAPI cuMemRelease(CUmemGenericAllocationHandle handle) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmemGenericAllocationHandle);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemRelease");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle);
+}
+
+CUresult CUDAAPI cuMemMap(CUdeviceptr ptr, size_t size, size_t offset,
+                          CUmemGenericAllocationHandle handle,
+                          unsigned long long flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, size_t, size_t,
+                          CUmemGenericAllocationHandle, unsigned long long);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemMap");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr, size, offset, handle, flags);
+}
+
+CUresult CUDAAPI cuMemMapArrayAsync(CUarrayMapInfo *mapInfoList,
+                                    unsigned int count, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarrayMapInfo *, unsigned int, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemMapArrayAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mapInfoList, count, hStream);
+}
+
+CUresult CUDAAPI cuMemUnmap(CUdeviceptr ptr, size_t size) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemUnmap");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr, size);
+}
+
+CUresult CUDAAPI cuMemSetAccess(CUdeviceptr ptr, size_t size,
+                                const CUmemAccessDesc *desc, size_t count) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, size_t, const CUmemAccessDesc *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemSetAccess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr, size, desc, count);
+}
+
+CUresult CUDAAPI cuMemGetAccess(unsigned long long *flags,
+                                const CUmemLocation *location,
+                                CUdeviceptr ptr) {
+  using FuncPtr = CUresult(CUDAAPI *)(unsigned long long *,
+                                      const CUmemLocation *, CUdeviceptr);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemGetAccess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(flags, location, ptr);
+}
+
+CUresult CUDAAPI cuMemExportToShareableHandle(
+    void *shareableHandle, CUmemGenericAllocationHandle handle,
+    CUmemAllocationHandleType handleType, unsigned long long flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(void *, CUmemGenericAllocationHandle,
+                          CUmemAllocationHandleType, unsigned long long);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemExportToShareableHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(shareableHandle, handle, handleType, flags);
+}
+
+CUresult CUDAAPI cuMemImportFromShareableHandle(
+    CUmemGenericAllocationHandle *handle, void *osHandle,
+    CUmemAllocationHandleType shHandleType) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmemGenericAllocationHandle *, void *,
+                                      CUmemAllocationHandleType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemImportFromShareableHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, osHandle, shHandleType);
+}
+
+CUresult CUDAAPI cuMemGetAllocationGranularity(
+    size_t *granularity, const CUmemAllocationProp *prop,
+    CUmemAllocationGranularity_flags option) {
+  using FuncPtr = CUresult(CUDAAPI *)(size_t *, const CUmemAllocationProp *,
+                                      CUmemAllocationGranularity_flags);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemGetAllocationGranularity");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(granularity, prop, option);
+}
+
+CUresult CUDAAPI cuMemGetAllocationPropertiesFromHandle(
+    CUmemAllocationProp *prop, CUmemGenericAllocationHandle handle) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUmemAllocationProp *, CUmemGenericAllocationHandle);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuMemGetAllocationPropertiesFromHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(prop, handle);
+}
+
+CUresult CUDAAPI
+cuMemRetainAllocationHandle(CUmemGenericAllocationHandle *handle, void *addr) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmemGenericAllocationHandle *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemRetainAllocationHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, addr);
+}
+
+CUresult CUDAAPI cuMemFreeAsync(CUdeviceptr dptr, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemFreeAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr, hStream);
+}
+
+CUresult CUDAAPI cuMemAllocAsync(CUdeviceptr *dptr, size_t bytesize,
+                                 CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr *, size_t, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemAllocAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr, bytesize, hStream);
+}
+
+CUresult CUDAAPI cuMemPoolTrimTo(CUmemoryPool pool, size_t minBytesToKeep) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmemoryPool, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemPoolTrimTo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pool, minBytesToKeep);
+}
+
+CUresult CUDAAPI cuMemPoolSetAttribute(CUmemoryPool pool,
+                                       CUmemPool_attribute attr, void *value) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUmemoryPool, CUmemPool_attribute, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemPoolSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pool, attr, value);
+}
+
+CUresult CUDAAPI cuMemPoolGetAttribute(CUmemoryPool pool,
+                                       CUmemPool_attribute attr, void *value) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUmemoryPool, CUmemPool_attribute, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemPoolGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pool, attr, value);
+}
+
+CUresult CUDAAPI cuMemPoolSetAccess(CUmemoryPool pool,
+                                    const CUmemAccessDesc *map, size_t count) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUmemoryPool, const CUmemAccessDesc *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemPoolSetAccess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pool, map, count);
+}
+
+CUresult CUDAAPI cuMemPoolGetAccess(CUmemAccess_flags *flags,
+                                    CUmemoryPool memPool,
+                                    CUmemLocation *location) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUmemAccess_flags *, CUmemoryPool, CUmemLocation *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemPoolGetAccess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(flags, memPool, location);
+}
+
+CUresult CUDAAPI cuMemPoolCreate(CUmemoryPool *pool,
+                                 const CUmemPoolProps *poolProps) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmemoryPool *, const CUmemPoolProps *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemPoolCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pool, poolProps);
+}
+
+CUresult CUDAAPI cuMemPoolDestroy(CUmemoryPool pool) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmemoryPool);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemPoolDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pool);
+}
+
+CUresult CUDAAPI cuMemAllocFromPoolAsync(CUdeviceptr *dptr, size_t bytesize,
+                                         CUmemoryPool pool, CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr *, size_t, CUmemoryPool, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemAllocFromPoolAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dptr, bytesize, pool, hStream);
+}
+
+CUresult CUDAAPI cuMemPoolExportToShareableHandle(
+    void *handle_out, CUmemoryPool pool, CUmemAllocationHandleType handleType,
+    unsigned long long flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      void *, CUmemoryPool, CUmemAllocationHandleType, unsigned long long);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuMemPoolExportToShareableHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle_out, pool, handleType, flags);
+}
+
+CUresult CUDAAPI cuMemPoolImportFromShareableHandle(
+    CUmemoryPool *pool_out, void *handle, CUmemAllocationHandleType handleType,
+    unsigned long long flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      CUmemoryPool *, void *, CUmemAllocationHandleType, unsigned long long);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuMemPoolImportFromShareableHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pool_out, handle, handleType, flags);
+}
+
+CUresult CUDAAPI cuMemPoolExportPointer(CUmemPoolPtrExportData *shareData_out,
+                                        CUdeviceptr ptr) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmemPoolPtrExportData *, CUdeviceptr);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemPoolExportPointer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(shareData_out, ptr);
+}
+
+CUresult CUDAAPI cuMemPoolImportPointer(CUdeviceptr *ptr_out, CUmemoryPool pool,
+                                        CUmemPoolPtrExportData *shareData) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr *, CUmemoryPool,
+                                      CUmemPoolPtrExportData *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemPoolImportPointer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr_out, pool, shareData);
+}
+
+CUresult CUDAAPI cuPointerGetAttribute(void *data,
+                                       CUpointer_attribute attribute,
+                                       CUdeviceptr ptr) {
+  using FuncPtr = CUresult(CUDAAPI *)(void *, CUpointer_attribute, CUdeviceptr);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuPointerGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(data, attribute, ptr);
+}
+
+CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count,
+                                    CUdevice dstDevice, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr, size_t, CUdevice, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemPrefetchAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, count, dstDevice, hStream);
+}
+
+CUresult CUDAAPI cuMemAdvise(CUdeviceptr devPtr, size_t count,
+                             CUmem_advise advice, CUdevice device) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr, size_t, CUmem_advise, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemAdvise");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, count, advice, device);
+}
+
+CUresult CUDAAPI cuMemRangeGetAttribute(void *data, size_t dataSize,
+                                        CUmem_range_attribute attribute,
+                                        CUdeviceptr devPtr, size_t count) {
+  using FuncPtr = CUresult(CUDAAPI *)(void *, size_t, CUmem_range_attribute,
+                                      CUdeviceptr, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemRangeGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(data, dataSize, attribute, devPtr, count);
+}
+
+CUresult CUDAAPI cuMemRangeGetAttributes(void **data, size_t *dataSizes,
+                                         CUmem_range_attribute *attributes,
+                                         size_t numAttributes,
+                                         CUdeviceptr devPtr, size_t count) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      void **, size_t *, CUmem_range_attribute *, size_t, CUdeviceptr, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuMemRangeGetAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(data, dataSizes, attributes, numAttributes, devPtr, count);
+}
+
+CUresult CUDAAPI cuPointerSetAttribute(const void *value,
+                                       CUpointer_attribute attribute,
+                                       CUdeviceptr ptr) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(const void *, CUpointer_attribute, CUdeviceptr);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuPointerSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(value, attribute, ptr);
+}
+
+CUresult CUDAAPI cuPointerGetAttributes(unsigned int numAttributes,
+                                        CUpointer_attribute *attributes,
+                                        void **data, CUdeviceptr ptr) {
+  using FuncPtr = CUresult(CUDAAPI *)(unsigned int, CUpointer_attribute *,
+                                      void **, CUdeviceptr);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuPointerGetAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(numAttributes, attributes, data, ptr);
+}
+
+CUresult CUDAAPI cuStreamCreate(CUstream *phStream, unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phStream, Flags);
+}
+
+CUresult CUDAAPI cuStreamCreateWithPriority(CUstream *phStream,
+                                            unsigned int flags, int priority) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream *, unsigned int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamCreateWithPriority");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phStream, flags, priority);
+}
+
+CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamGetPriority");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, priority);
+}
+
+CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, unsigned int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamGetFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, flags);
+}
+
+CUresult CUDAAPI cuStreamGetId(CUstream hStream, unsigned long long *streamId) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, unsigned long long *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamGetId");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, streamId);
+}
+
+CUresult CUDAAPI cuStreamGetCtx(CUstream hStream, CUcontext *pctx) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, CUcontext *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamGetCtx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, pctx);
+}
+
+CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent,
+                                   unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, CUevent, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamWaitEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, hEvent, Flags);
+}
+
+CUresult CUDAAPI cuStreamAddCallback(CUstream hStream,
+                                     CUstreamCallback callback, void *userData,
+                                     unsigned int flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUstream, CUstreamCallback, void *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamAddCallback");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, callback, userData, flags);
+}
+
+CUresult CUDAAPI cuStreamBeginCapture(CUstream hStream,
+                                      CUstreamCaptureMode mode) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, CUstreamCaptureMode);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamBeginCapture_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, mode);
+}
+
+CUresult CUDAAPI cuThreadExchangeStreamCaptureMode(CUstreamCaptureMode *mode) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstreamCaptureMode *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuThreadExchangeStreamCaptureMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mode);
+}
+
+CUresult CUDAAPI cuStreamEndCapture(CUstream hStream, CUgraph *phGraph) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, CUgraph *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamEndCapture");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, phGraph);
+}
+
+CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream,
+                                     CUstreamCaptureStatus *captureStatus) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, CUstreamCaptureStatus *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamIsCapturing");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, captureStatus);
+}
+
+CUresult CUDAAPI cuStreamGetCaptureInfo(
+    CUstream hStream, CUstreamCaptureStatus *captureStatus_out,
+    cuuint64_t *id_out, CUgraph *graph_out,
+    const CUgraphNode **dependencies_out, size_t *numDependencies_out) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUstream, CUstreamCaptureStatus *, cuuint64_t *,
+                          CUgraph *, const CUgraphNode **, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamGetCaptureInfo_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, captureStatus_out, id_out, graph_out,
+                  dependencies_out, numDependencies_out);
+}
+
+CUresult CUDAAPI cuStreamUpdateCaptureDependencies(CUstream hStream,
+                                                   CUgraphNode *dependencies,
+                                                   size_t numDependencies,
+                                                   unsigned int flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUstream, CUgraphNode *, size_t, unsigned int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuStreamUpdateCaptureDependencies");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, dependencies, numDependencies, flags);
+}
+
+CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr,
+                                        size_t length, unsigned int flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUstream, CUdeviceptr, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamAttachMemAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, dptr, length, flags);
+}
+
+CUresult CUDAAPI cuStreamQuery(CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamQuery");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream);
+}
+
+CUresult CUDAAPI cuStreamSynchronize(CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamSynchronize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream);
+}
+
+CUresult CUDAAPI cuStreamDestroy(CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamDestroy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream);
+}
+
+CUresult CUDAAPI cuStreamCopyAttributes(CUstream dst, CUstream src) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamCopyAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, src);
+}
+
+CUresult CUDAAPI cuStreamGetAttribute(CUstream hStream, CUstreamAttrID attr,
+                                      CUstreamAttrValue *value_out) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUstream, CUstreamAttrID, CUstreamAttrValue *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, attr, value_out);
+}
+
+CUresult CUDAAPI cuStreamSetAttribute(CUstream hStream, CUstreamAttrID attr,
+                                      const CUstreamAttrValue *value) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUstream, CUstreamAttrID, const CUstreamAttrValue *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, attr, value);
+}
+
+CUresult CUDAAPI cuEventCreate(CUevent *phEvent, unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUevent *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuEventCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phEvent, Flags);
+}
+
+CUresult CUDAAPI cuEventRecord(CUevent hEvent, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUevent, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuEventRecord");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hEvent, hStream);
+}
+
+CUresult CUDAAPI cuEventRecordWithFlags(CUevent hEvent, CUstream hStream,
+                                        unsigned int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUevent, CUstream, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuEventRecordWithFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hEvent, hStream, flags);
+}
+
+CUresult CUDAAPI cuEventQuery(CUevent hEvent) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUevent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuEventQuery");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hEvent);
+}
+
+CUresult CUDAAPI cuEventSynchronize(CUevent hEvent) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUevent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuEventSynchronize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hEvent);
+}
+
+CUresult CUDAAPI cuEventDestroy(CUevent hEvent) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUevent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuEventDestroy_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hEvent);
+}
+
+CUresult CUDAAPI cuEventElapsedTime(float *pMilliseconds, CUevent hStart,
+                                    CUevent hEnd) {
+  using FuncPtr = CUresult(CUDAAPI *)(float *, CUevent, CUevent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuEventElapsedTime");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pMilliseconds, hStart, hEnd);
+}
+
+CUresult CUDAAPI
+cuImportExternalMemory(CUexternalMemory *extMem_out,
+                       const CUDA_EXTERNAL_MEMORY_HANDLE_DESC *memHandleDesc) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUexternalMemory *,
+                                      const CUDA_EXTERNAL_MEMORY_HANDLE_DESC *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuImportExternalMemory");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extMem_out, memHandleDesc);
+}
+
+CUresult CUDAAPI cuExternalMemoryGetMappedBuffer(
+    CUdeviceptr *devPtr, CUexternalMemory extMem,
+    const CUDA_EXTERNAL_MEMORY_BUFFER_DESC *bufferDesc) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr *, CUexternalMemory,
+                                      const CUDA_EXTERNAL_MEMORY_BUFFER_DESC *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuExternalMemoryGetMappedBuffer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, extMem, bufferDesc);
+}
+
+CUresult CUDAAPI cuExternalMemoryGetMappedMipmappedArray(
+    CUmipmappedArray *mipmap, CUexternalMemory extMem,
+    const CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC *mipmapDesc) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUmipmappedArray *, CUexternalMemory,
+                          const CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuExternalMemoryGetMappedMipmappedArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mipmap, extMem, mipmapDesc);
+}
+
+CUresult CUDAAPI cuDestroyExternalMemory(CUexternalMemory extMem) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUexternalMemory);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDestroyExternalMemory");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extMem);
+}
+
+CUresult CUDAAPI cuImportExternalSemaphore(
+    CUexternalSemaphore *extSem_out,
+    const CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC *semHandleDesc) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      CUexternalSemaphore *, const CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuImportExternalSemaphore");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extSem_out, semHandleDesc);
+}
+
+CUresult CUDAAPI cuSignalExternalSemaphoresAsync(
+    const CUexternalSemaphore *extSemArray,
+    const CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS *paramsArray,
+    unsigned int numExtSems, CUstream stream) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      const CUexternalSemaphore *,
+      const CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS *, unsigned int, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuSignalExternalSemaphoresAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extSemArray, paramsArray, numExtSems, stream);
+}
+
+CUresult CUDAAPI cuWaitExternalSemaphoresAsync(
+    const CUexternalSemaphore *extSemArray,
+    const CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS *paramsArray,
+    unsigned int numExtSems, CUstream stream) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      const CUexternalSemaphore *, const CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS *,
+      unsigned int, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuWaitExternalSemaphoresAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extSemArray, paramsArray, numExtSems, stream);
+}
+
+CUresult CUDAAPI cuDestroyExternalSemaphore(CUexternalSemaphore extSem) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUexternalSemaphore);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDestroyExternalSemaphore");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extSem);
+}
+
+CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr,
+                                     cuuint32_t value, unsigned int flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUstream, CUdeviceptr, cuuint32_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamWaitValue32_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, addr, value, flags);
+}
+
+CUresult CUDAAPI cuStreamWaitValue64(CUstream stream, CUdeviceptr addr,
+                                     cuuint64_t value, unsigned int flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUstream, CUdeviceptr, cuuint64_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamWaitValue64_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, addr, value, flags);
+}
+
+CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr,
+                                      cuuint32_t value, unsigned int flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUstream, CUdeviceptr, cuuint32_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamWriteValue32_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, addr, value, flags);
+}
+
+CUresult CUDAAPI cuStreamWriteValue64(CUstream stream, CUdeviceptr addr,
+                                      cuuint64_t value, unsigned int flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUstream, CUdeviceptr, cuuint64_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamWriteValue64_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, addr, value, flags);
+}
+
+CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count,
+                                    CUstreamBatchMemOpParams *paramArray,
+                                    unsigned int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, unsigned int,
+                                      CUstreamBatchMemOpParams *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuStreamBatchMemOp_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, count, paramArray, flags);
+}
+
+CUresult CUDAAPI cuFuncGetAttribute(int *pi, CUfunction_attribute attrib,
+                                    CUfunction hfunc) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *, CUfunction_attribute, CUfunction);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuFuncGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pi, attrib, hfunc);
+}
+
+CUresult CUDAAPI cuFuncSetAttribute(CUfunction hfunc,
+                                    CUfunction_attribute attrib, int value) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, CUfunction_attribute, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuFuncSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, attrib, value);
+}
+
+CUresult CUDAAPI cuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache config) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, CUfunc_cache);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuFuncSetCacheConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, config);
+}
+
+CUresult CUDAAPI cuFuncSetSharedMemConfig(CUfunction hfunc,
+                                          CUsharedconfig config) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, CUsharedconfig);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuFuncSetSharedMemConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, config);
+}
+
+CUresult CUDAAPI cuFuncGetModule(CUmodule *hmod, CUfunction hfunc) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmodule *, CUfunction);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuFuncGetModule");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hmod, hfunc);
+}
+
+CUresult CUDAAPI cuLaunchKernel(CUfunction f, unsigned int gridDimX,
+                                unsigned int gridDimY, unsigned int gridDimZ,
+                                unsigned int blockDimX, unsigned int blockDimY,
+                                unsigned int blockDimZ,
+                                unsigned int sharedMemBytes, CUstream hStream,
+                                void **kernelParams, void **extra) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      CUfunction, unsigned int, unsigned int, unsigned int, unsigned int,
+      unsigned int, unsigned int, unsigned int, CUstream, void **, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLaunchKernel");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(f, gridDimX, gridDimY, gridDimZ, blockDimX, blockDimY,
+                  blockDimZ, sharedMemBytes, hStream, kernelParams, extra);
+}
+
+CUresult CUDAAPI cuLaunchKernelEx(const CUlaunchConfig *config, CUfunction f,
+                                  void **kernelParams, void **extra) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(const CUlaunchConfig *, CUfunction, void **, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLaunchKernelEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(config, f, kernelParams, extra);
+}
+
+CUresult CUDAAPI cuLaunchCooperativeKernel(
+    CUfunction f, unsigned int gridDimX, unsigned int gridDimY,
+    unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY,
+    unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream,
+    void **kernelParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      CUfunction, unsigned int, unsigned int, unsigned int, unsigned int,
+      unsigned int, unsigned int, unsigned int, CUstream, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLaunchCooperativeKernel");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(f, gridDimX, gridDimY, gridDimZ, blockDimX, blockDimY,
+                  blockDimZ, sharedMemBytes, hStream, kernelParams);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuLaunchCooperativeKernelMultiDevice(
+    CUDA_LAUNCH_PARAMS *launchParamsList, unsigned int numDevices,
+    unsigned int flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUDA_LAUNCH_PARAMS *, unsigned int, unsigned int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuLaunchCooperativeKernelMultiDevice");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(launchParamsList, numDevices, flags);
+}
+
+CUresult CUDAAPI cuLaunchHostFunc(CUstream hStream, CUhostFn fn,
+                                  void *userData) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUstream, CUhostFn, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLaunchHostFunc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, fn, userData);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuFuncSetBlockShape(CUfunction hfunc, int x,
+                                                       int y, int z) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, int, int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuFuncSetBlockShape");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, x, y, z);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuFuncSetSharedSize(CUfunction hfunc,
+                                                       unsigned int bytes) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuFuncSetSharedSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, bytes);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetSize(CUfunction hfunc,
+                                                  unsigned int numbytes) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuParamSetSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, numbytes);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuParamSeti(CUfunction hfunc, int offset,
+                                               unsigned int value) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, int, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuParamSeti");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, offset, value);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetf(CUfunction hfunc, int offset,
+                                               float value) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, int, float);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuParamSetf");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, offset, value);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetv(CUfunction hfunc, int offset,
+                                               void *ptr,
+                                               unsigned int numbytes) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, int, void *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuParamSetv");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, offset, ptr, numbytes);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuLaunch(CUfunction f) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLaunch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(f);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width,
+                                                int grid_height) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLaunchGrid");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(f, grid_width, grid_height);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGridAsync(CUfunction f,
+                                                     int grid_width,
+                                                     int grid_height,
+                                                     CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, int, int, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuLaunchGridAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(f, grid_width, grid_height, hStream);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetTexRef(CUfunction hfunc,
+                                                    int texunit,
+                                                    CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfunction, int, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuParamSetTexRef");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hfunc, texunit, hTexRef);
+}
+
+CUresult CUDAAPI cuGraphCreate(CUgraph *phGraph, unsigned int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraph *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraph, flags);
+}
+
+CUresult CUDAAPI cuGraphAddKernelNode(
+    CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies,
+    size_t numDependencies, const CUDA_KERNEL_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode *, CUgraph, const CUgraphNode *, size_t,
+                          const CUDA_KERNEL_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddKernelNode_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies,
+                  nodeParams);
+}
+
+CUresult CUDAAPI cuGraphKernelNodeGetParams(
+    CUgraphNode hNode, CUDA_KERNEL_NODE_PARAMS *nodeParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUDA_KERNEL_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphKernelNodeGetParams_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphKernelNodeSetParams(
+    CUgraphNode hNode, const CUDA_KERNEL_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode, const CUDA_KERNEL_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphKernelNodeSetParams_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphAddMemcpyNode(CUgraphNode *phGraphNode, CUgraph hGraph,
+                                      const CUgraphNode *dependencies,
+                                      size_t numDependencies,
+                                      const CUDA_MEMCPY3D *copyParams,
+                                      CUcontext ctx) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode *, CUgraph, const CUgraphNode *, size_t,
+                          const CUDA_MEMCPY3D *, CUcontext);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddMemcpyNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies,
+                  copyParams, ctx);
+}
+
+CUresult CUDAAPI cuGraphMemcpyNodeGetParams(CUgraphNode hNode,
+                                            CUDA_MEMCPY3D *nodeParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUDA_MEMCPY3D *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphMemcpyNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphMemcpyNodeSetParams(CUgraphNode hNode,
+                                            const CUDA_MEMCPY3D *nodeParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, const CUDA_MEMCPY3D *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphMemcpyNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphAddMemsetNode(
+    CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies,
+    size_t numDependencies, const CUDA_MEMSET_NODE_PARAMS *memsetParams,
+    CUcontext ctx) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode *, CUgraph, const CUgraphNode *, size_t,
+                          const CUDA_MEMSET_NODE_PARAMS *, CUcontext);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddMemsetNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies,
+                  memsetParams, ctx);
+}
+
+CUresult CUDAAPI cuGraphMemsetNodeGetParams(
+    CUgraphNode hNode, CUDA_MEMSET_NODE_PARAMS *nodeParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUDA_MEMSET_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphMemsetNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphMemsetNodeSetParams(
+    CUgraphNode hNode, const CUDA_MEMSET_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode, const CUDA_MEMSET_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphMemsetNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphAddHostNode(CUgraphNode *phGraphNode, CUgraph hGraph,
+                                    const CUgraphNode *dependencies,
+                                    size_t numDependencies,
+                                    const CUDA_HOST_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode *, CUgraph, const CUgraphNode *, size_t,
+                          const CUDA_HOST_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddHostNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies,
+                  nodeParams);
+}
+
+CUresult CUDAAPI cuGraphHostNodeGetParams(CUgraphNode hNode,
+                                          CUDA_HOST_NODE_PARAMS *nodeParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUDA_HOST_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphHostNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphHostNodeSetParams(
+    CUgraphNode hNode, const CUDA_HOST_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode, const CUDA_HOST_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphHostNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphAddChildGraphNode(CUgraphNode *phGraphNode,
+                                          CUgraph hGraph,
+                                          const CUgraphNode *dependencies,
+                                          size_t numDependencies,
+                                          CUgraph childGraph) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode *, CUgraph,
+                                      const CUgraphNode *, size_t, CUgraph);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddChildGraphNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies,
+                  childGraph);
+}
+
+CUresult CUDAAPI cuGraphChildGraphNodeGetGraph(CUgraphNode hNode,
+                                               CUgraph *phGraph) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUgraph *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphChildGraphNodeGetGraph");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, phGraph);
+}
+
+CUresult CUDAAPI cuGraphAddEmptyNode(CUgraphNode *phGraphNode, CUgraph hGraph,
+                                     const CUgraphNode *dependencies,
+                                     size_t numDependencies) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode *, CUgraph, const CUgraphNode *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddEmptyNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies);
+}
+
+CUresult CUDAAPI cuGraphAddEventRecordNode(CUgraphNode *phGraphNode,
+                                           CUgraph hGraph,
+                                           const CUgraphNode *dependencies,
+                                           size_t numDependencies,
+                                           CUevent event) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode *, CUgraph,
+                                      const CUgraphNode *, size_t, CUevent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddEventRecordNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies, event);
+}
+
+CUresult CUDAAPI cuGraphEventRecordNodeGetEvent(CUgraphNode hNode,
+                                                CUevent *event_out) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUevent *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphEventRecordNodeGetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, event_out);
+}
+
+CUresult CUDAAPI cuGraphEventRecordNodeSetEvent(CUgraphNode hNode,
+                                                CUevent event) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUevent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphEventRecordNodeSetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, event);
+}
+
+CUresult CUDAAPI cuGraphAddEventWaitNode(CUgraphNode *phGraphNode,
+                                         CUgraph hGraph,
+                                         const CUgraphNode *dependencies,
+                                         size_t numDependencies,
+                                         CUevent event) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode *, CUgraph,
+                                      const CUgraphNode *, size_t, CUevent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddEventWaitNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies, event);
+}
+
+CUresult CUDAAPI cuGraphEventWaitNodeGetEvent(CUgraphNode hNode,
+                                              CUevent *event_out) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUevent *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphEventWaitNodeGetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, event_out);
+}
+
+CUresult CUDAAPI cuGraphEventWaitNodeSetEvent(CUgraphNode hNode,
+                                              CUevent event) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUevent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphEventWaitNodeSetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, event);
+}
+
+CUresult CUDAAPI cuGraphAddExternalSemaphoresSignalNode(
+    CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies,
+    size_t numDependencies, const CUDA_EXT_SEM_SIGNAL_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode *, CUgraph, const CUgraphNode *, size_t,
+                          const CUDA_EXT_SEM_SIGNAL_NODE_PARAMS *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphAddExternalSemaphoresSignalNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies,
+                  nodeParams);
+}
+
+CUresult CUDAAPI cuGraphExternalSemaphoresSignalNodeGetParams(
+    CUgraphNode hNode, CUDA_EXT_SEM_SIGNAL_NODE_PARAMS *params_out) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode, CUDA_EXT_SEM_SIGNAL_NODE_PARAMS *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExternalSemaphoresSignalNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, params_out);
+}
+
+CUresult CUDAAPI cuGraphExternalSemaphoresSignalNodeSetParams(
+    CUgraphNode hNode, const CUDA_EXT_SEM_SIGNAL_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode, const CUDA_EXT_SEM_SIGNAL_NODE_PARAMS *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExternalSemaphoresSignalNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphAddExternalSemaphoresWaitNode(
+    CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies,
+    size_t numDependencies, const CUDA_EXT_SEM_WAIT_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode *, CUgraph, const CUgraphNode *, size_t,
+                          const CUDA_EXT_SEM_WAIT_NODE_PARAMS *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphAddExternalSemaphoresWaitNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies,
+                  nodeParams);
+}
+
+CUresult CUDAAPI cuGraphExternalSemaphoresWaitNodeGetParams(
+    CUgraphNode hNode, CUDA_EXT_SEM_WAIT_NODE_PARAMS *params_out) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode, CUDA_EXT_SEM_WAIT_NODE_PARAMS *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExternalSemaphoresWaitNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, params_out);
+}
+
+CUresult CUDAAPI cuGraphExternalSemaphoresWaitNodeSetParams(
+    CUgraphNode hNode, const CUDA_EXT_SEM_WAIT_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode, const CUDA_EXT_SEM_WAIT_NODE_PARAMS *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExternalSemaphoresWaitNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphAddBatchMemOpNode(
+    CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies,
+    size_t numDependencies, const CUDA_BATCH_MEM_OP_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode *, CUgraph, const CUgraphNode *, size_t,
+                          const CUDA_BATCH_MEM_OP_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddBatchMemOpNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies,
+                  nodeParams);
+}
+
+CUresult CUDAAPI cuGraphBatchMemOpNodeGetParams(
+    CUgraphNode hNode, CUDA_BATCH_MEM_OP_NODE_PARAMS *nodeParams_out) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode, CUDA_BATCH_MEM_OP_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphBatchMemOpNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams_out);
+}
+
+CUresult CUDAAPI cuGraphBatchMemOpNodeSetParams(
+    CUgraphNode hNode, const CUDA_BATCH_MEM_OP_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode, const CUDA_BATCH_MEM_OP_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphBatchMemOpNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphExecBatchMemOpNodeSetParams(
+    CUgraphExec hGraphExec, CUgraphNode hNode,
+    const CUDA_BATCH_MEM_OP_NODE_PARAMS *nodeParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode,
+                                      const CUDA_BATCH_MEM_OP_NODE_PARAMS *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExecBatchMemOpNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphAddMemAllocNode(
+    CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies,
+    size_t numDependencies, CUDA_MEM_ALLOC_NODE_PARAMS *nodeParams) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode *, CUgraph, const CUgraphNode *, size_t,
+                          CUDA_MEM_ALLOC_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddMemAllocNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies,
+                  nodeParams);
+}
+
+CUresult CUDAAPI cuGraphMemAllocNodeGetParams(
+    CUgraphNode hNode, CUDA_MEM_ALLOC_NODE_PARAMS *params_out) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphNode, CUDA_MEM_ALLOC_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphMemAllocNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, params_out);
+}
+
+CUresult CUDAAPI cuGraphAddMemFreeNode(CUgraphNode *phGraphNode, CUgraph hGraph,
+                                       const CUgraphNode *dependencies,
+                                       size_t numDependencies,
+                                       CUdeviceptr dptr) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode *, CUgraph,
+                                      const CUgraphNode *, size_t, CUdeviceptr);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddMemFreeNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphNode, hGraph, dependencies, numDependencies, dptr);
+}
+
+CUresult CUDAAPI cuGraphMemFreeNodeGetParams(CUgraphNode hNode,
+                                             CUdeviceptr *dptr_out) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUdeviceptr *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphMemFreeNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, dptr_out);
+}
+
+CUresult CUDAAPI cuDeviceGraphMemTrim(CUdevice device) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGraphMemTrim");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device);
+}
+
+CUresult CUDAAPI cuDeviceGetGraphMemAttribute(CUdevice device,
+                                              CUgraphMem_attribute attr,
+                                              void *value) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice, CUgraphMem_attribute, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetGraphMemAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device, attr, value);
+}
+
+CUresult CUDAAPI cuDeviceSetGraphMemAttribute(CUdevice device,
+                                              CUgraphMem_attribute attr,
+                                              void *value) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdevice, CUgraphMem_attribute, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceSetGraphMemAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device, attr, value);
+}
+
+CUresult CUDAAPI cuGraphClone(CUgraph *phGraphClone, CUgraph originalGraph) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraph *, CUgraph);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphClone");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphClone, originalGraph);
+}
+
+CUresult CUDAAPI cuGraphNodeFindInClone(CUgraphNode *phNode,
+                                        CUgraphNode hOriginalNode,
+                                        CUgraph hClonedGraph) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode *, CUgraphNode, CUgraph);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphNodeFindInClone");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phNode, hOriginalNode, hClonedGraph);
+}
+
+CUresult CUDAAPI cuGraphNodeGetType(CUgraphNode hNode, CUgraphNodeType *type) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUgraphNodeType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphNodeGetType");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, type);
+}
+
+CUresult CUDAAPI cuGraphGetNodes(CUgraph hGraph, CUgraphNode *nodes,
+                                 size_t *numNodes) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraph, CUgraphNode *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphGetNodes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraph, nodes, numNodes);
+}
+
+CUresult CUDAAPI cuGraphGetRootNodes(CUgraph hGraph, CUgraphNode *rootNodes,
+                                     size_t *numRootNodes) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraph, CUgraphNode *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphGetRootNodes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraph, rootNodes, numRootNodes);
+}
+
+CUresult CUDAAPI cuGraphGetEdges(CUgraph hGraph, CUgraphNode *from,
+                                 CUgraphNode *to, size_t *numEdges) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraph, CUgraphNode *, CUgraphNode *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphGetEdges");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraph, from, to, numEdges);
+}
+
+CUresult CUDAAPI cuGraphNodeGetDependencies(CUgraphNode hNode,
+                                            CUgraphNode *dependencies,
+                                            size_t *numDependencies) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUgraphNode *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphNodeGetDependencies");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, dependencies, numDependencies);
+}
+
+CUresult CUDAAPI cuGraphNodeGetDependentNodes(CUgraphNode hNode,
+                                              CUgraphNode *dependentNodes,
+                                              size_t *numDependentNodes) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUgraphNode *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphNodeGetDependentNodes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, dependentNodes, numDependentNodes);
+}
+
+CUresult CUDAAPI cuGraphAddDependencies(CUgraph hGraph, const CUgraphNode *from,
+                                        const CUgraphNode *to,
+                                        size_t numDependencies) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraph, const CUgraphNode *,
+                                      const CUgraphNode *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphAddDependencies");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraph, from, to, numDependencies);
+}
+
+CUresult CUDAAPI cuGraphRemoveDependencies(CUgraph hGraph,
+                                           const CUgraphNode *from,
+                                           const CUgraphNode *to,
+                                           size_t numDependencies) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraph, const CUgraphNode *,
+                                      const CUgraphNode *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphRemoveDependencies");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraph, from, to, numDependencies);
+}
+
+CUresult CUDAAPI cuGraphDestroyNode(CUgraphNode hNode) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphDestroyNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode);
+}
+
+CUresult CUDAAPI cuGraphInstantiate(CUgraphExec *phGraphExec, CUgraph hGraph,
+                                    unsigned long long flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphExec *, CUgraph, unsigned long long);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphInstantiateWithFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphExec, hGraph, flags);
+}
+
+CUresult CUDAAPI
+cuGraphInstantiateWithParams(CUgraphExec *phGraphExec, CUgraph hGraph,
+                             CUDA_GRAPH_INSTANTIATE_PARAMS *instantiateParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec *, CUgraph,
+                                      CUDA_GRAPH_INSTANTIATE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphInstantiateWithParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phGraphExec, hGraph, instantiateParams);
+}
+
+CUresult CUDAAPI cuGraphExecGetFlags(CUgraphExec hGraphExec,
+                                     cuuint64_t *flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, cuuint64_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphExecGetFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, flags);
+}
+
+CUresult CUDAAPI
+cuGraphExecKernelNodeSetParams(CUgraphExec hGraphExec, CUgraphNode hNode,
+                               const CUDA_KERNEL_NODE_PARAMS *nodeParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode,
+                                      const CUDA_KERNEL_NODE_PARAMS *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExecKernelNodeSetParams_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphExecMemcpyNodeSetParams(CUgraphExec hGraphExec,
+                                                CUgraphNode hNode,
+                                                const CUDA_MEMCPY3D *copyParams,
+                                                CUcontext ctx) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode,
+                                      const CUDA_MEMCPY3D *, CUcontext);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphExecMemcpyNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, copyParams, ctx);
+}
+
+CUresult CUDAAPI cuGraphExecMemsetNodeSetParams(
+    CUgraphExec hGraphExec, CUgraphNode hNode,
+    const CUDA_MEMSET_NODE_PARAMS *memsetParams, CUcontext ctx) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      CUgraphExec, CUgraphNode, const CUDA_MEMSET_NODE_PARAMS *, CUcontext);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphExecMemsetNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, memsetParams, ctx);
+}
+
+CUresult CUDAAPI
+cuGraphExecHostNodeSetParams(CUgraphExec hGraphExec, CUgraphNode hNode,
+                             const CUDA_HOST_NODE_PARAMS *nodeParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode,
+                                      const CUDA_HOST_NODE_PARAMS *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphExecHostNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphExecChildGraphNodeSetParams(CUgraphExec hGraphExec,
+                                                    CUgraphNode hNode,
+                                                    CUgraph childGraph) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode, CUgraph);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExecChildGraphNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, childGraph);
+}
+
+CUresult CUDAAPI cuGraphExecEventRecordNodeSetEvent(CUgraphExec hGraphExec,
+                                                    CUgraphNode hNode,
+                                                    CUevent event) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode, CUevent);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExecEventRecordNodeSetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, event);
+}
+
+CUresult CUDAAPI cuGraphExecEventWaitNodeSetEvent(CUgraphExec hGraphExec,
+                                                  CUgraphNode hNode,
+                                                  CUevent event) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode, CUevent);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExecEventWaitNodeSetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, event);
+}
+
+CUresult CUDAAPI cuGraphExecExternalSemaphoresSignalNodeSetParams(
+    CUgraphExec hGraphExec, CUgraphNode hNode,
+    const CUDA_EXT_SEM_SIGNAL_NODE_PARAMS *nodeParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode,
+                                      const CUDA_EXT_SEM_SIGNAL_NODE_PARAMS *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExecExternalSemaphoresSignalNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphExecExternalSemaphoresWaitNodeSetParams(
+    CUgraphExec hGraphExec, CUgraphNode hNode,
+    const CUDA_EXT_SEM_WAIT_NODE_PARAMS *nodeParams) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode,
+                                      const CUDA_EXT_SEM_WAIT_NODE_PARAMS *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphExecExternalSemaphoresWaitNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, nodeParams);
+}
+
+CUresult CUDAAPI cuGraphNodeSetEnabled(CUgraphExec hGraphExec,
+                                       CUgraphNode hNode,
+                                       unsigned int isEnabled) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphNodeSetEnabled");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, isEnabled);
+}
+
+CUresult CUDAAPI cuGraphNodeGetEnabled(CUgraphExec hGraphExec,
+                                       CUgraphNode hNode,
+                                       unsigned int *isEnabled) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUgraphNode, unsigned int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphNodeGetEnabled");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, isEnabled);
+}
+
+CUresult CUDAAPI cuGraphUpload(CUgraphExec hGraphExec, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphUpload");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hStream);
+}
+
+CUresult CUDAAPI cuGraphLaunch(CUgraphExec hGraphExec, CUstream hStream) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphLaunch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hStream);
+}
+
+CUresult CUDAAPI cuGraphExecDestroy(CUgraphExec hGraphExec) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphExec);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphExecDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec);
+}
+
+CUresult CUDAAPI cuGraphDestroy(CUgraph hGraph) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraph);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraph);
+}
+
+CUresult CUDAAPI cuGraphExecUpdate(CUgraphExec hGraphExec, CUgraph hGraph,
+                                   CUgraphExecUpdateResultInfo *resultInfo) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraphExec, CUgraph, CUgraphExecUpdateResultInfo *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphExecUpdate_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hGraph, resultInfo);
+}
+
+CUresult CUDAAPI cuGraphKernelNodeCopyAttributes(CUgraphNode dst,
+                                                 CUgraphNode src) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUgraphNode);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphKernelNodeCopyAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, src);
+}
+
+CUresult CUDAAPI
+cuGraphKernelNodeGetAttribute(CUgraphNode hNode, CUkernelNodeAttrID attr,
+                              CUkernelNodeAttrValue *value_out) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUkernelNodeAttrID,
+                                      CUkernelNodeAttrValue *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphKernelNodeGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, attr, value_out);
+}
+
+CUresult CUDAAPI
+cuGraphKernelNodeSetAttribute(CUgraphNode hNode, CUkernelNodeAttrID attr,
+                              const CUkernelNodeAttrValue *value) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphNode, CUkernelNodeAttrID,
+                                      const CUkernelNodeAttrValue *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphKernelNodeSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, attr, value);
+}
+
+CUresult CUDAAPI cuGraphDebugDotPrint(CUgraph hGraph, const char *path,
+                                      unsigned int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraph, const char *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphDebugDotPrint");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraph, path, flags);
+}
+
+CUresult CUDAAPI cuUserObjectCreate(CUuserObject *object_out, void *ptr,
+                                    CUhostFn destroy,
+                                    unsigned int initialRefcount,
+                                    unsigned int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUuserObject *, void *, CUhostFn,
+                                      unsigned int, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuUserObjectCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(object_out, ptr, destroy, initialRefcount, flags);
+}
+
+CUresult CUDAAPI cuUserObjectRetain(CUuserObject object, unsigned int count) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUuserObject, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuUserObjectRetain");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(object, count);
+}
+
+CUresult CUDAAPI cuUserObjectRelease(CUuserObject object, unsigned int count) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUuserObject, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuUserObjectRelease");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(object, count);
+}
+
+CUresult CUDAAPI cuGraphRetainUserObject(CUgraph graph, CUuserObject object,
+                                         unsigned int count,
+                                         unsigned int flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUgraph, CUuserObject, unsigned int, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphRetainUserObject");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph, object, count, flags);
+}
+
+CUresult CUDAAPI cuGraphReleaseUserObject(CUgraph graph, CUuserObject object,
+                                          unsigned int count) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraph, CUuserObject, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphReleaseUserObject");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph, object, count);
+}
+
+CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessor(
+    int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *, CUfunction, int, size_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuOccupancyMaxActiveBlocksPerMultiprocessor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(numBlocks, func, blockSize, dynamicSMemSize);
+}
+
+CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
+    int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize,
+    unsigned int flags) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(int *, CUfunction, int, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>(
+      "cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(numBlocks, func, blockSize, dynamicSMemSize, flags);
+}
+
+CUresult CUDAAPI cuOccupancyMaxPotentialBlockSize(
+    int *minGridSize, int *blockSize, CUfunction func,
+    CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize,
+    int blockSizeLimit) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *, int *, CUfunction,
+                                      CUoccupancyB2DSize, size_t, int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuOccupancyMaxPotentialBlockSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(minGridSize, blockSize, func, blockSizeToDynamicSMemSize,
+                  dynamicSMemSize, blockSizeLimit);
+}
+
+CUresult CUDAAPI cuOccupancyMaxPotentialBlockSizeWithFlags(
+    int *minGridSize, int *blockSize, CUfunction func,
+    CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize,
+    int blockSizeLimit, unsigned int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      int *, int *, CUfunction, CUoccupancyB2DSize, size_t, int, unsigned int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuOccupancyMaxPotentialBlockSizeWithFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(minGridSize, blockSize, func, blockSizeToDynamicSMemSize,
+                  dynamicSMemSize, blockSizeLimit, flags);
+}
+
+CUresult CUDAAPI cuOccupancyAvailableDynamicSMemPerBlock(
+    size_t *dynamicSmemSize, CUfunction func, int numBlocks, int blockSize) {
+  using FuncPtr = CUresult(CUDAAPI *)(size_t *, CUfunction, int, int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuOccupancyAvailableDynamicSMemPerBlock");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dynamicSmemSize, func, numBlocks, blockSize);
+}
+
+CUresult CUDAAPI cuOccupancyMaxPotentialClusterSize(
+    int *clusterSize, CUfunction func, const CUlaunchConfig *config) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(int *, CUfunction, const CUlaunchConfig *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuOccupancyMaxPotentialClusterSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(clusterSize, func, config);
+}
+
+CUresult CUDAAPI cuOccupancyMaxActiveClusters(int *numClusters, CUfunction func,
+                                              const CUlaunchConfig *config) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(int *, CUfunction, const CUlaunchConfig *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuOccupancyMaxActiveClusters");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(numClusters, func, config);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefSetArray(CUtexref hTexRef,
+                                                    CUarray hArray,
+                                                    unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, CUarray, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, hArray, Flags);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefSetMipmappedArray(
+    CUtexref hTexRef, CUmipmappedArray hMipmappedArray, unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, CUmipmappedArray, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetMipmappedArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, hMipmappedArray, Flags);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefSetAddress(size_t *ByteOffset,
+                                                      CUtexref hTexRef,
+                                                      CUdeviceptr dptr,
+                                                      size_t bytes) {
+  using FuncPtr = CUresult(CUDAAPI *)(size_t *, CUtexref, CUdeviceptr, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetAddress_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ByteOffset, hTexRef, dptr, bytes);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI
+cuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc,
+                     CUdeviceptr dptr, size_t Pitch) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, const CUDA_ARRAY_DESCRIPTOR *,
+                                      CUdeviceptr, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetAddress2D_v3");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, desc, dptr, Pitch);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefSetFormat(CUtexref hTexRef,
+                                                     CUarray_format fmt,
+                                                     int NumPackedComponents) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, CUarray_format, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetFormat");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, fmt, NumPackedComponents);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefSetAddressMode(CUtexref hTexRef,
+                                                          int dim,
+                                                          CUaddress_mode am) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, int, CUaddress_mode);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetAddressMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, dim, am);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefSetFilterMode(CUtexref hTexRef,
+                                                         CUfilter_mode fm) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, CUfilter_mode);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetFilterMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, fm);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI
+cuTexRefSetMipmapFilterMode(CUtexref hTexRef, CUfilter_mode fm) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, CUfilter_mode);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetMipmapFilterMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, fm);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefSetMipmapLevelBias(CUtexref hTexRef,
+                                                              float bias) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, float);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetMipmapLevelBias");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, bias);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefSetMipmapLevelClamp(
+    CUtexref hTexRef, float minMipmapLevelClamp, float maxMipmapLevelClamp) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, float, float);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetMipmapLevelClamp");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, minMipmapLevelClamp, maxMipmapLevelClamp);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI
+cuTexRefSetMaxAnisotropy(CUtexref hTexRef, unsigned int maxAniso) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetMaxAnisotropy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, maxAniso);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefSetBorderColor(CUtexref hTexRef,
+                                                          float *pBorderColor) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetBorderColor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, pBorderColor);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefSetFlags(CUtexref hTexRef,
+                                                    unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefSetFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef, Flags);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefGetAddress(CUdeviceptr *pdptr,
+                                                      CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUdeviceptr *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetAddress_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pdptr, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefGetArray(CUarray *phArray,
+                                                    CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarray *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phArray, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefGetMipmappedArray(
+    CUmipmappedArray *phMipmappedArray, CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmipmappedArray *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetMipmappedArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phMipmappedArray, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefGetAddressMode(CUaddress_mode *pam,
+                                                          CUtexref hTexRef,
+                                                          int dim) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUaddress_mode *, CUtexref, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetAddressMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pam, hTexRef, dim);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefGetFilterMode(CUfilter_mode *pfm,
+                                                         CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfilter_mode *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetFilterMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pfm, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefGetFormat(CUarray_format *pFormat,
+                                                     int *pNumChannels,
+                                                     CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarray_format *, int *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetFormat");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pFormat, pNumChannels, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI
+cuTexRefGetMipmapFilterMode(CUfilter_mode *pfm, CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUfilter_mode *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetMipmapFilterMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pfm, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI
+cuTexRefGetMipmapLevelBias(float *pbias, CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(float *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetMipmapLevelBias");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pbias, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI
+cuTexRefGetMipmapLevelClamp(float *pminMipmapLevelClamp,
+                            float *pmaxMipmapLevelClamp, CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(float *, float *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetMipmapLevelClamp");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pminMipmapLevelClamp, pmaxMipmapLevelClamp, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefGetMaxAnisotropy(int *pmaxAniso,
+                                                            CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetMaxAnisotropy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pmaxAniso, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefGetBorderColor(float *pBorderColor,
+                                                          CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(float *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetBorderColor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pBorderColor, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefGetFlags(unsigned int *pFlags,
+                                                    CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(unsigned int *, CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefGetFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pFlags, hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefCreate(CUtexref *pTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuTexRefDestroy(CUtexref hTexRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexRefDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hTexRef);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuSurfRefSetArray(CUsurfref hSurfRef,
+                                                     CUarray hArray,
+                                                     unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUsurfref, CUarray, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuSurfRefSetArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hSurfRef, hArray, Flags);
+}
+
+__CUDA_DEPRECATED CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray,
+                                                     CUsurfref hSurfRef) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarray *, CUsurfref);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuSurfRefGetArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(phArray, hSurfRef);
+}
+
+CUresult CUDAAPI
+cuTexObjectCreate(CUtexObject *pTexObject, const CUDA_RESOURCE_DESC *pResDesc,
+                  const CUDA_TEXTURE_DESC *pTexDesc,
+                  const CUDA_RESOURCE_VIEW_DESC *pResViewDesc) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexObject *, const CUDA_RESOURCE_DESC *,
+                                      const CUDA_TEXTURE_DESC *,
+                                      const CUDA_RESOURCE_VIEW_DESC *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexObjectCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pTexObject, pResDesc, pTexDesc, pResViewDesc);
+}
+
+CUresult CUDAAPI cuTexObjectDestroy(CUtexObject texObject) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtexObject);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexObjectDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(texObject);
+}
+
+CUresult CUDAAPI cuTexObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc,
+                                            CUtexObject texObject) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUDA_RESOURCE_DESC *, CUtexObject);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexObjectGetResourceDesc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pResDesc, texObject);
+}
+
+CUresult CUDAAPI cuTexObjectGetTextureDesc(CUDA_TEXTURE_DESC *pTexDesc,
+                                           CUtexObject texObject) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUDA_TEXTURE_DESC *, CUtexObject);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexObjectGetTextureDesc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pTexDesc, texObject);
+}
+
+CUresult CUDAAPI cuTexObjectGetResourceViewDesc(
+    CUDA_RESOURCE_VIEW_DESC *pResViewDesc, CUtexObject texObject) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUDA_RESOURCE_VIEW_DESC *, CUtexObject);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTexObjectGetResourceViewDesc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pResViewDesc, texObject);
+}
+
+CUresult CUDAAPI cuSurfObjectCreate(CUsurfObject *pSurfObject,
+                                    const CUDA_RESOURCE_DESC *pResDesc) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUsurfObject *, const CUDA_RESOURCE_DESC *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuSurfObjectCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pSurfObject, pResDesc);
+}
+
+CUresult CUDAAPI cuSurfObjectDestroy(CUsurfObject surfObject) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUsurfObject);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuSurfObjectDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(surfObject);
+}
+
+CUresult CUDAAPI cuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc,
+                                             CUsurfObject surfObject) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUDA_RESOURCE_DESC *, CUsurfObject);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuSurfObjectGetResourceDesc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pResDesc, surfObject);
+}
+
+CUresult CUDAAPI cuTensorMapEncodeTiled(
+    CUtensorMap *tensorMap, CUtensorMapDataType tensorDataType,
+    cuuint32_t tensorRank, void *globalAddress, const cuuint64_t *globalDim,
+    const cuuint64_t *globalStrides, const cuuint32_t *boxDim,
+    const cuuint32_t *elementStrides, CUtensorMapInterleave interleave,
+    CUtensorMapSwizzle swizzle, CUtensorMapL2promotion l2Promotion,
+    CUtensorMapFloatOOBfill oobFill) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      CUtensorMap *, CUtensorMapDataType, cuuint32_t, void *,
+      const cuuint64_t *, const cuuint64_t *, const cuuint32_t *,
+      const cuuint32_t *, CUtensorMapInterleave, CUtensorMapSwizzle,
+      CUtensorMapL2promotion, CUtensorMapFloatOOBfill);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTensorMapEncodeTiled");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorMap, tensorDataType, tensorRank, globalAddress,
+                  globalDim, globalStrides, boxDim, elementStrides, interleave,
+                  swizzle, l2Promotion, oobFill);
+}
+
+CUresult CUDAAPI cuTensorMapEncodeIm2col(
+    CUtensorMap *tensorMap, CUtensorMapDataType tensorDataType,
+    cuuint32_t tensorRank, void *globalAddress, const cuuint64_t *globalDim,
+    const cuuint64_t *globalStrides, const int *pixelBoxLowerCorner,
+    const int *pixelBoxUpperCorner, cuuint32_t channelsPerPixel,
+    cuuint32_t pixelsPerColumn, const cuuint32_t *elementStrides,
+    CUtensorMapInterleave interleave, CUtensorMapSwizzle swizzle,
+    CUtensorMapL2promotion l2Promotion, CUtensorMapFloatOOBfill oobFill) {
+  using FuncPtr = CUresult(CUDAAPI *)(
+      CUtensorMap *, CUtensorMapDataType, cuuint32_t, void *,
+      const cuuint64_t *, const cuuint64_t *, const int *, const int *,
+      cuuint32_t, cuuint32_t, const cuuint32_t *, CUtensorMapInterleave,
+      CUtensorMapSwizzle, CUtensorMapL2promotion, CUtensorMapFloatOOBfill);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTensorMapEncodeIm2col");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorMap, tensorDataType, tensorRank, globalAddress,
+                  globalDim, globalStrides, pixelBoxLowerCorner,
+                  pixelBoxUpperCorner, channelsPerPixel, pixelsPerColumn,
+                  elementStrides, interleave, swizzle, l2Promotion, oobFill);
+}
+
+CUresult CUDAAPI cuTensorMapReplaceAddress(CUtensorMap *tensorMap,
+                                           void *globalAddress) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUtensorMap *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuTensorMapReplaceAddress");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorMap, globalAddress);
+}
+
+CUresult CUDAAPI cuDeviceCanAccessPeer(int *canAccessPeer, CUdevice dev,
+                                       CUdevice peerDev) {
+  using FuncPtr = CUresult(CUDAAPI *)(int *, CUdevice, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceCanAccessPeer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(canAccessPeer, dev, peerDev);
+}
+
+CUresult CUDAAPI cuCtxEnablePeerAccess(CUcontext peerContext,
+                                       unsigned int Flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxEnablePeerAccess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(peerContext, Flags);
+}
+
+CUresult CUDAAPI cuCtxDisablePeerAccess(CUcontext peerContext) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUcontext);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuCtxDisablePeerAccess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(peerContext);
+}
+
+CUresult CUDAAPI cuDeviceGetP2PAttribute(int *value,
+                                         CUdevice_P2PAttribute attrib,
+                                         CUdevice srcDevice,
+                                         CUdevice dstDevice) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(int *, CUdevice_P2PAttribute, CUdevice, CUdevice);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuDeviceGetP2PAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(value, attrib, srcDevice, dstDevice);
+}
+
+CUresult CUDAAPI cuGraphicsUnregisterResource(CUgraphicsResource resource) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphicsResource);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphicsUnregisterResource");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(resource);
+}
+
+CUresult CUDAAPI cuGraphicsSubResourceGetMappedArray(
+    CUarray *pArray, CUgraphicsResource resource, unsigned int arrayIndex,
+    unsigned int mipLevel) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUarray *, CUgraphicsResource,
+                                      unsigned int, unsigned int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphicsSubResourceGetMappedArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pArray, resource, arrayIndex, mipLevel);
+}
+
+CUresult CUDAAPI cuGraphicsResourceGetMappedMipmappedArray(
+    CUmipmappedArray *pMipmappedArray, CUgraphicsResource resource) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUmipmappedArray *, CUgraphicsResource);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphicsResourceGetMappedMipmappedArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pMipmappedArray, resource);
+}
+
+CUresult CUDAAPI cuGraphicsResourceGetMappedPointer(
+    CUdeviceptr *pDevPtr, size_t *pSize, CUgraphicsResource resource) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(CUdeviceptr *, size_t *, CUgraphicsResource);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphicsResourceGetMappedPointer_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pDevPtr, pSize, resource);
+}
+
+CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource,
+                                               unsigned int flags) {
+  using FuncPtr = CUresult(CUDAAPI *)(CUgraphicsResource, unsigned int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cuGraphicsResourceSetMapFlags_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(resource, flags);
+}
+
+CUresult CUDAAPI cuGraphicsMapResources(unsigned int count,
+                                        CUgraphicsResource *resources,
+                                        CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(unsigned int, CUgraphicsResource *, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphicsMapResources");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(count, resources, hStream);
+}
+
+CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count,
+                                          CUgraphicsResource *resources,
+                                          CUstream hStream) {
+  using FuncPtr =
+      CUresult(CUDAAPI *)(unsigned int, CUgraphicsResource *, CUstream);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGraphicsUnmapResources");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(count, resources, hStream);
+}
+
+CUresult CUDAAPI cuGetProcAddress(
+    const char *symbol, void **pfn, int cudaVersion, cuuint64_t flags,
+    CUdriverProcAddressQueryResult *symbolStatus) {
+  using FuncPtr = CUresult(CUDAAPI *)(const char *, void **, int, cuuint64_t,
+                                      CUdriverProcAddressQueryResult *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGetProcAddress_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(symbol, pfn, cudaVersion, flags, symbolStatus);
+}
+
+CUresult CUDAAPI cuGetExportTable(const void **ppExportTable,
+                                  const CUuuid *pExportTableId) {
+  using FuncPtr = CUresult(CUDAAPI *)(const void **, const CUuuid *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cuGetExportTable");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ppExportTable, pExportTableId);
+}
+
+}
diff --git a/tensorflow/stream_executor/cuda/cuda_dnn.cc b/tensorflow/stream_executor/cuda/cuda_dnn.cc
index 076831671ea..760aa4c71f4 100755
--- a/tensorflow/stream_executor/cuda/cuda_dnn.cc
+++ b/tensorflow/stream_executor/cuda/cuda_dnn.cc
@@ -181,7 +181,7 @@ class CudnnAccess {
   // therefore a bad idea (performance wise) to call any cuDNN APIs that
   // enqueue work in the stream.
   CudnnHandle GetHandle(GpuExecutor* executor, Stream* stream) {
-    auto lock = absl::make_unique<absl::MutexLock>(&mutex_);
+    auto lock = std::make_unique<absl::MutexLock>(&mutex_);
     mutex_.AssertHeld();
     gpu::ScopedActivateExecutorContext context(executor);
     CUstream cu_stream = stream ? AsGpuStreamValue(stream) : cudaStreamLegacy;
@@ -1208,7 +1208,7 @@ port::Status CheckAndFetchProjectionWeights(
   cudnnRNNMode_t mode;
   cudnnRNNAlgo_t algo;
   cudnnDataType_t data_type;
-  RETURN_IF_CUDNN_ERROR(cudnnGetRNNDescriptor(
+  RETURN_IF_CUDNN_ERROR(cudnnGetRNNDescriptor_v6(
       /*handle=*/cudnn.handle(), /*rnnDesc=*/rnn_desc,
       /*hiddenSize=*/&hidden_size_v,
       /*numLayers=*/&num_layers_v,
@@ -2314,7 +2314,7 @@ namespace {
 
 // TODO(csigg): Merge a lot of duplicate code below for forward, backward data,
 // and backward filter.
-
+/*
 port::StatusOr<cudnnConvolutionFwdAlgo_t> GetCudnnConvolutionForwardAlgo(
     const CudnnHandle& cudnn, const CudnnTensorDescriptor& input_nd,
     const CudnnFilterDescriptor& filter, const CudnnConvolutionDescriptor& conv,
@@ -2367,7 +2367,7 @@ GetCudnnConvolutionBackwardFilterAlgo(const CudnnHandle& cudnn,
       filter.handle(), preference, memory_limit_bytes, &algo_to_use));
   return algo_to_use;
 }
-
+*/
 port::StatusOr<DeviceMemory<uint8>> AllocateCudnnConvolutionForwardWorkspace(
     Stream* stream, const CudnnHandle& cudnn,
     const CudnnTensorDescriptor& input_nd, const CudnnFilterDescriptor& filter,
@@ -2515,6 +2515,7 @@ port::StatusOr<dnn::AlgorithmDesc> GetCudnnConvolutionForwardAlgorithm(
   if (!algo_desc.has_value()) {
     // Pick fastest algorithm within memory limit according to cuDNN's
     // heuristics.
+    /*
     bool specify_workspace_limit = scratch_allocator != nullptr;
     auto memory_limit_bytes =
         specify_workspace_limit
@@ -2524,12 +2525,13 @@ port::StatusOr<dnn::AlgorithmDesc> GetCudnnConvolutionForwardAlgorithm(
                         GetCudnnConvolutionForwardAlgo(
                             cudnn, input_nd, filter, conv, output_nd,
                             specify_workspace_limit, memory_limit_bytes));
+    */
+    cudnnConvolutionFwdAlgo_t algo = CUDNN_CONVOLUTION_FWD_ALGO_IMPLICIT_GEMM;
     int cc_major, cc_minor;
     std::tie(cc_major, cc_minor) = GetCcMajorMinor(stream);
     algo_desc = dnn::AlgorithmDesc(
         algo, /*use_tensor_ops=*/TensorOpMathAvailable(cc_major));
   }
-
   const auto scratch_or = AllocateCudnnConvolutionForwardWorkspace(
       stream, cudnn, input_nd, filter, conv, output_nd, *algo_desc,
       scratch_allocator);
@@ -2568,6 +2570,7 @@ port::StatusOr<dnn::AlgorithmDesc> GetCudnnConvolutionBackwardDataAlgorithm(
   if (!algo_desc.has_value()) {
     // Pick fastest algorithm within memory limit according to cuDNN's
     // heuristics.
+    /*
     bool specify_workspace_limit = scratch_allocator != nullptr;
     auto memory_limit_bytes =
         specify_workspace_limit
@@ -2577,12 +2580,13 @@ port::StatusOr<dnn::AlgorithmDesc> GetCudnnConvolutionBackwardDataAlgorithm(
                         GetCudnnConvolutionBackwardDataAlgo(
                             cudnn, input_nd, filter, conv, output_nd,
                             specify_workspace_limit, memory_limit_bytes));
+    */
+    cudnnConvolutionBwdDataAlgo_t algo = CUDNN_CONVOLUTION_BWD_DATA_ALGO_0;
     int cc_major, cc_minor;
     std::tie(cc_major, cc_minor) = GetCcMajorMinor(stream);
     algo_desc = dnn::AlgorithmDesc(
         algo, /*use_tensor_ops=*/TensorOpMathAvailable(cc_major));
   }
-
   const auto scratch_or = AllocateCudnnConvolutionBackwardDataWorkspace(
       stream, cudnn, input_nd, filter, conv, output_nd, *algo_desc,
       scratch_allocator);
@@ -2620,6 +2624,7 @@ port::StatusOr<dnn::AlgorithmDesc> GetCudnnConvolutionBackwardFilterAlgorithm(
   if (!algo_desc.has_value()) {
     // Pick fastest algorithm within memory limit according to cuDNN's
     // heuristics.
+    /*
     bool specify_workspace_limit = scratch_allocator != nullptr;
     auto memory_limit_bytes =
         specify_workspace_limit
@@ -2629,12 +2634,13 @@ port::StatusOr<dnn::AlgorithmDesc> GetCudnnConvolutionBackwardFilterAlgorithm(
                         GetCudnnConvolutionBackwardFilterAlgo(
                             cudnn, input_nd, filter, conv, output_nd,
                             specify_workspace_limit, memory_limit_bytes));
+    */
+    cudnnConvolutionBwdFilterAlgo_t algo = CUDNN_CONVOLUTION_BWD_FILTER_ALGO_0;
     int cc_major, cc_minor;
     std::tie(cc_major, cc_minor) = GetCcMajorMinor(stream);
     algo_desc = dnn::AlgorithmDesc(
         algo, /*use_tensor_ops=*/TensorOpMathAvailable(cc_major));
   }
-
   auto scratch_or = AllocateCudnnConvolutionBackwardFilterWorkspace(
       stream, cudnn, input_nd, filter, conv, output_nd, *algo_desc,
       scratch_allocator);
diff --git a/tensorflow/stream_executor/cuda/cuda_driver.cc b/tensorflow/stream_executor/cuda/cuda_driver.cc
index f7a69fc086a..92a4e911da0 100644
--- a/tensorflow/stream_executor/cuda/cuda_driver.cc
+++ b/tensorflow/stream_executor/cuda/cuda_driver.cc
@@ -178,7 +178,7 @@ void CheckPointerIsValid(const PtrT ptr, absl::string_view name) {
   // If we failed, reset cuda error status to avoid poisoning cuda streams.
   if (err != cudaSuccess) cudaGetLastError();
   bool points_to_host_memory = (err == cudaErrorInvalidValue ||
-                                attributes.memoryType != cudaMemoryTypeDevice);
+                                attributes.type != cudaMemoryTypeDevice);
   CHECK_EQ(is_host_ptr, points_to_host_memory) << absl::StreamFormat(
       "%s pointer is not actually on %s: %p", name, is_host_ptr ? "CPU" : "GPU",
       reinterpret_cast<const void*>(ptr));
diff --git a/tensorflow/stream_executor/cuda/cuda_runtime_12_0.inc b/tensorflow/stream_executor/cuda/cuda_runtime_12_0.inc
new file mode 100644
index 00000000000..343db231322
--- /dev/null
+++ b/tensorflow/stream_executor/cuda/cuda_runtime_12_0.inc
@@ -0,0 +1,2676 @@
+// Auto-generated, do not edit.
+
+extern "C" {
+extern __host__ cudaError_t CUDARTAPI cudaDeviceReset(void) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceReset");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaDeviceSynchronize(void) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceSynchronize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaDeviceSetLimit(enum cudaLimit limit,
+                                                         size_t value) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(enum cudaLimit, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceSetLimit");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(limit, value);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaDeviceGetLimit(size_t *pValue, enum cudaLimit limit) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(size_t *, enum cudaLimit);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGetLimit");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pValue, limit);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaDeviceGetTexture1DLinearMaxWidth(
+    size_t *maxWidthInElements, const struct cudaChannelFormatDesc *fmtDesc,
+    int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      size_t *, const struct cudaChannelFormatDesc *, int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaDeviceGetTexture1DLinearMaxWidth");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(maxWidthInElements, fmtDesc, device);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaDeviceGetCacheConfig(enum cudaFuncCache *pCacheConfig) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(enum cudaFuncCache *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGetCacheConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pCacheConfig);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaDeviceGetStreamPriorityRange(int *leastPriority, int *greatestPriority) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int *, int *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaDeviceGetStreamPriorityRange");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(leastPriority, greatestPriority);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDeviceSetCacheConfig(enum cudaFuncCache cacheConfig) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(enum cudaFuncCache);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceSetCacheConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(cacheConfig);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaDeviceGetSharedMemConfig(enum cudaSharedMemConfig *pConfig) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(enum cudaSharedMemConfig *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGetSharedMemConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pConfig);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDeviceSetSharedMemConfig(enum cudaSharedMemConfig config) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(enum cudaSharedMemConfig);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceSetSharedMemConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(config);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDeviceGetByPCIBusId(int *device, const char *pciBusId) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int *, const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGetByPCIBusId");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device, pciBusId);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaDeviceGetPCIBusId(char *pciBusId,
+                                                            int len,
+                                                            int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(char *, int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGetPCIBusId");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pciBusId, len, device);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaIpcGetEventHandle(cudaIpcEventHandle_t *handle, cudaEvent_t event) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaIpcEventHandle_t *, cudaEvent_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaIpcGetEventHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, event);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaIpcOpenEventHandle(cudaEvent_t *event, cudaIpcEventHandle_t handle) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaEvent_t *, cudaIpcEventHandle_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaIpcOpenEventHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(event, handle);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaIpcGetMemHandle(cudaIpcMemHandle_t *handle, void *devPtr) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaIpcMemHandle_t *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaIpcGetMemHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, devPtr);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaIpcOpenMemHandle(
+    void **devPtr, cudaIpcMemHandle_t handle, unsigned int flags) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(void **, cudaIpcMemHandle_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaIpcOpenMemHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, handle, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaIpcCloseMemHandle(void *devPtr) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaIpcCloseMemHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaDeviceFlushGPUDirectRDMAWrites(
+    enum cudaFlushGPUDirectRDMAWritesTarget target,
+    enum cudaFlushGPUDirectRDMAWritesScope scope) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(enum cudaFlushGPUDirectRDMAWritesTarget,
+                               enum cudaFlushGPUDirectRDMAWritesScope);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaDeviceFlushGPUDirectRDMAWrites");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(target, scope);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI cudaThreadExit(void) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaThreadExit");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaThreadSynchronize(void) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaThreadSynchronize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaThreadSetLimit(enum cudaLimit limit, size_t value) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(enum cudaLimit, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaThreadSetLimit");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(limit, value);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaThreadGetLimit(size_t *pValue, enum cudaLimit limit) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(size_t *, enum cudaLimit);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaThreadGetLimit");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pValue, limit);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaThreadGetCacheConfig(enum cudaFuncCache *pCacheConfig) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(enum cudaFuncCache *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaThreadGetCacheConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pCacheConfig);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaThreadSetCacheConfig(enum cudaFuncCache cacheConfig) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(enum cudaFuncCache);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaThreadSetCacheConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(cacheConfig);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaGetLastError(void) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetLastError");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaPeekAtLastError(void) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaPeekAtLastError");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+extern __host__ __cudart_builtin__ const char *CUDARTAPI
+cudaGetErrorName(cudaError_t error) {
+  using FuncPtr = const char *(CUDARTAPI *)(cudaError_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetErrorName");
+  if (!func_ptr) return "cudaGetErrorName symbol not found.";
+  return func_ptr(error);
+}
+
+extern __host__ __cudart_builtin__ const char *CUDARTAPI
+cudaGetErrorString(cudaError_t error) {
+  using FuncPtr = const char *(CUDARTAPI *)(cudaError_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetErrorString");
+  if (!func_ptr) return "cudaGetErrorString symbol not found.";
+  return func_ptr(error);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaGetDeviceCount(int *count) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetDeviceCount");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(count);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaGetDeviceProperties(struct cudaDeviceProp *prop, int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(struct cudaDeviceProp *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetDeviceProperties_v2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(prop, device);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaDeviceGetAttribute(int *value, enum cudaDeviceAttr attr, int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int *, enum cudaDeviceAttr, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(value, attr, device);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDeviceGetDefaultMemPool(cudaMemPool_t *memPool, int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaMemPool_t *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGetDefaultMemPool");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memPool, device);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDeviceSetMemPool(int device, cudaMemPool_t memPool) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int, cudaMemPool_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceSetMemPool");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device, memPool);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDeviceGetMemPool(cudaMemPool_t *memPool, int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaMemPool_t *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGetMemPool");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memPool, device);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaDeviceGetNvSciSyncAttributes(
+    void *nvSciSyncAttrList, int device, int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, int, int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaDeviceGetNvSciSyncAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(nvSciSyncAttrList, device, flags);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaDeviceGetP2PAttribute(int *value, enum cudaDeviceP2PAttr attr,
+                          int srcDevice, int dstDevice) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(int *, enum cudaDeviceP2PAttr, int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGetP2PAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(value, attr, srcDevice, dstDevice);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaChooseDevice(int *device, const struct cudaDeviceProp *prop) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(int *, const struct cudaDeviceProp *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaChooseDevice");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device, prop);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaInitDevice(int device,
+                                                     unsigned int deviceFlags,
+                                                     unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int, unsigned int, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaInitDevice");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device, deviceFlags, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaSetDevice(int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaSetDevice");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaGetDevice(int *device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetDevice");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaSetValidDevices(int *device_arr,
+                                                          int len) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaSetValidDevices");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device_arr, len);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaSetDeviceFlags(unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaSetDeviceFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetDeviceFlags(unsigned int *flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(unsigned int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetDeviceFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaStreamCreate(cudaStream_t *pStream) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pStream);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaStreamCreateWithFlags(cudaStream_t *pStream, unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamCreateWithFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pStream, flags);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaStreamCreateWithPriority(cudaStream_t *pStream, unsigned int flags,
+                             int priority) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t *, unsigned int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamCreateWithPriority");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pStream, flags, priority);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaStreamGetPriority(cudaStream_t hStream, int *priority) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamGetPriority");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, priority);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaStreamGetFlags(cudaStream_t hStream, unsigned int *flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t, unsigned int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamGetFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, flags);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaStreamGetId(cudaStream_t hStream, unsigned long long *streamId) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t, unsigned long long *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamGetId");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, streamId);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaCtxResetPersistingL2Cache(void) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaCtxResetPersistingL2Cache");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaStreamCopyAttributes(cudaStream_t dst, cudaStream_t src) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamCopyAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, src);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaStreamGetAttribute(cudaStream_t hStream, cudaStreamAttrID attr,
+                       cudaStreamAttrValue *value_out) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t, cudaStreamAttrID,
+                                           cudaStreamAttrValue *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, attr, value_out);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaStreamSetAttribute(cudaStream_t hStream, cudaStreamAttrID attr,
+                       const cudaStreamAttrValue *value) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t, cudaStreamAttrID,
+                                           const cudaStreamAttrValue *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hStream, attr, value);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaStreamDestroy(cudaStream_t stream) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI cudaStreamWaitEvent(
+    cudaStream_t stream, cudaEvent_t event, unsigned int flags __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaStream_t, cudaEvent_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamWaitEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, event, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaStreamAddCallback(cudaStream_t stream, cudaStreamCallback_t callback,
+                      void *userData, unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t, cudaStreamCallback_t,
+                                           void *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamAddCallback");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, callback, userData, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaStreamSynchronize(cudaStream_t stream) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamSynchronize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaStreamQuery(cudaStream_t stream) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamQuery");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaStreamAttachMemAsync(cudaStream_t stream, void *devPtr,
+                         size_t length __dv(0), unsigned int flags) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaStream_t, void *, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamAttachMemAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, devPtr, length, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaStreamBeginCapture(cudaStream_t stream, enum cudaStreamCaptureMode mode) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaStream_t, enum cudaStreamCaptureMode);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamBeginCapture");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, mode);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaThreadExchangeStreamCaptureMode(enum cudaStreamCaptureMode *mode) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(enum cudaStreamCaptureMode *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaThreadExchangeStreamCaptureMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mode);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaStreamEndCapture(cudaStream_t stream, cudaGraph_t *pGraph) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t, cudaGraph_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamEndCapture");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, pGraph);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaStreamIsCapturing(
+    cudaStream_t stream, enum cudaStreamCaptureStatus *pCaptureStatus) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaStream_t, enum cudaStreamCaptureStatus *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaStreamIsCapturing");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, pCaptureStatus);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaStreamGetCaptureInfo(
+    cudaStream_t stream, enum cudaStreamCaptureStatus *captureStatus_out,
+    unsigned long long *id_out __dv(0), cudaGraph_t *graph_out __dv(0),
+    const cudaGraphNode_t **dependencies_out __dv(0),
+    size_t *numDependencies_out __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaStream_t, enum cudaStreamCaptureStatus *, unsigned long long *,
+      cudaGraph_t *, const cudaGraphNode_t **, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("__CUDART_API_PTSZ(cudaStreamGetCaptureInfo_v2)");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, captureStatus_out, id_out, graph_out,
+                  dependencies_out, numDependencies_out);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaStreamUpdateCaptureDependencies(
+    cudaStream_t stream, cudaGraphNode_t *dependencies, size_t numDependencies,
+    unsigned int flags __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t, cudaGraphNode_t *,
+                                           size_t, unsigned int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaStreamUpdateCaptureDependencies");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, dependencies, numDependencies, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaEventCreate(cudaEvent_t *event) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaEvent_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaEventCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(event);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaEventCreateWithFlags(cudaEvent_t *event, unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaEvent_t *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaEventCreateWithFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(event, flags);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaEventRecord(cudaEvent_t event, cudaStream_t stream __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaEvent_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaEventRecord");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(event, stream);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaEventRecordWithFlags(cudaEvent_t event, cudaStream_t stream __dv(0),
+                         unsigned int flags __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaEvent_t, cudaStream_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaEventRecordWithFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(event, stream, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaEventQuery(cudaEvent_t event) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaEvent_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaEventQuery");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(event);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaEventSynchronize(cudaEvent_t event) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaEvent_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaEventSynchronize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(event);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaEventDestroy(cudaEvent_t event) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaEvent_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaEventDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(event);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaEventElapsedTime(float *ms,
+                                                           cudaEvent_t start,
+                                                           cudaEvent_t end) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(float *, cudaEvent_t, cudaEvent_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaEventElapsedTime");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ms, start, end);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaImportExternalMemory(
+    cudaExternalMemory_t *extMem_out,
+    const struct cudaExternalMemoryHandleDesc *memHandleDesc) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaExternalMemory_t *, const struct cudaExternalMemoryHandleDesc *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaImportExternalMemory");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extMem_out, memHandleDesc);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaExternalMemoryGetMappedBuffer(
+    void **devPtr, cudaExternalMemory_t extMem,
+    const struct cudaExternalMemoryBufferDesc *bufferDesc) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(void **, cudaExternalMemory_t,
+                               const struct cudaExternalMemoryBufferDesc *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaExternalMemoryGetMappedBuffer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, extMem, bufferDesc);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaExternalMemoryGetMappedMipmappedArray(
+    cudaMipmappedArray_t *mipmap, cudaExternalMemory_t extMem,
+    const struct cudaExternalMemoryMipmappedArrayDesc *mipmapDesc) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaMipmappedArray_t *, cudaExternalMemory_t,
+      const struct cudaExternalMemoryMipmappedArrayDesc *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaExternalMemoryGetMappedMipmappedArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mipmap, extMem, mipmapDesc);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDestroyExternalMemory(cudaExternalMemory_t extMem) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaExternalMemory_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDestroyExternalMemory");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extMem);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaImportExternalSemaphore(
+    cudaExternalSemaphore_t *extSem_out,
+    const struct cudaExternalSemaphoreHandleDesc *semHandleDesc) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaExternalSemaphore_t *,
+                               const struct cudaExternalSemaphoreHandleDesc *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaImportExternalSemaphore");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extSem_out, semHandleDesc);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaSignalExternalSemaphoresAsync(
+    const cudaExternalSemaphore_t *extSemArray,
+    const struct cudaExternalSemaphoreSignalParams *paramsArray,
+    unsigned int numExtSems, cudaStream_t stream __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(const cudaExternalSemaphore_t *,
+                               const struct cudaExternalSemaphoreSignalParams *,
+                               unsigned int, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>(
+      "__CUDART_API_PTSZ(cudaSignalExternalSemaphoresAsync_v2)");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extSemArray, paramsArray, numExtSems, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaWaitExternalSemaphoresAsync(
+    const cudaExternalSemaphore_t *extSemArray,
+    const struct cudaExternalSemaphoreWaitParams *paramsArray,
+    unsigned int numExtSems, cudaStream_t stream __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(const cudaExternalSemaphore_t *,
+                               const struct cudaExternalSemaphoreWaitParams *,
+                               unsigned int, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>(
+      "__CUDART_API_PTSZ(cudaWaitExternalSemaphoresAsync_v2)");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extSemArray, paramsArray, numExtSems, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDestroyExternalSemaphore(cudaExternalSemaphore_t extSem) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaExternalSemaphore_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDestroyExternalSemaphore");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(extSem);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaLaunchKernel(const void *func, dim3 gridDim, dim3 blockDim, void **args,
+                 size_t sharedMem, cudaStream_t stream) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(const void *, dim3, dim3, void **,
+                                           size_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaLaunchKernel");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(func, gridDim, blockDim, args, sharedMem, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaLaunchKernelExC(
+    const cudaLaunchConfig_t *config, const void *func, void **args) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(const cudaLaunchConfig_t *,
+                                           const void *, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaLaunchKernelExC");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(config, func, args);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaLaunchCooperativeKernel(
+    const void *func, dim3 gridDim, dim3 blockDim, void **args,
+    size_t sharedMem, cudaStream_t stream) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(const void *, dim3, dim3, void **,
+                                           size_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaLaunchCooperativeKernel");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(func, gridDim, blockDim, args, sharedMem, stream);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaLaunchCooperativeKernelMultiDevice(
+    struct cudaLaunchParams *launchParamsList, unsigned int numDevices,
+    unsigned int flags __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(struct cudaLaunchParams *,
+                                           unsigned int, unsigned int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaLaunchCooperativeKernelMultiDevice");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(launchParamsList, numDevices, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaFuncSetCacheConfig(const void *func, enum cudaFuncCache cacheConfig) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(const void *, enum cudaFuncCache);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaFuncSetCacheConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(func, cacheConfig);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaFuncSetSharedMemConfig(const void *func, enum cudaSharedMemConfig config) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(const void *, enum cudaSharedMemConfig);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaFuncSetSharedMemConfig");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(func, config);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaFuncGetAttributes(struct cudaFuncAttributes *attr, const void *func) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(struct cudaFuncAttributes *, const void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaFuncGetAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(attr, func);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaFuncSetAttribute(const void *func, enum cudaFuncAttribute attr, int value) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(const void *, enum cudaFuncAttribute, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaFuncSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(func, attr, value);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaSetDoubleForDevice(double *d) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaSetDoubleForDevice");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(d);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaSetDoubleForHost(double *d) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaSetDoubleForHost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(d);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaLaunchHostFunc(cudaStream_t stream,
+                                                         cudaHostFn_t fn,
+                                                         void *userData) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaStream_t, cudaHostFn_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaLaunchHostFunc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stream, fn, userData);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaOccupancyMaxActiveBlocksPerMultiprocessor(int *numBlocks, const void *func,
+                                              int blockSize,
+                                              size_t dynamicSMemSize) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int *, const void *, int, size_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaOccupancyMaxActiveBlocksPerMultiprocessor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(numBlocks, func, blockSize, dynamicSMemSize);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaOccupancyAvailableDynamicSMemPerBlock(size_t *dynamicSmemSize,
+                                          const void *func, int numBlocks,
+                                          int blockSize) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(size_t *, const void *, int, int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaOccupancyAvailableDynamicSMemPerBlock");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dynamicSmemSize, func, numBlocks, blockSize);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int *numBlocks,
+                                                       const void *func,
+                                                       int blockSize,
+                                                       size_t dynamicSMemSize,
+                                                       unsigned int flags) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(int *, const void *, int, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>(
+      "cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(numBlocks, func, blockSize, dynamicSMemSize, flags);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaOccupancyMaxPotentialClusterSize(int *clusterSize, const void *func,
+                                     const cudaLaunchConfig_t *launchConfig) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(int *, const void *, const cudaLaunchConfig_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaOccupancyMaxPotentialClusterSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(clusterSize, func, launchConfig);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaOccupancyMaxActiveClusters(int *numClusters, const void *func,
+                               const cudaLaunchConfig_t *launchConfig) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(int *, const void *, const cudaLaunchConfig_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaOccupancyMaxActiveClusters");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(numClusters, func, launchConfig);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaMallocManaged(void **devPtr, size_t size, unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void **, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMallocManaged");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, size, flags);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaMalloc(void **devPtr, size_t size) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void **, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMalloc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, size);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void **, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMallocHost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr, size);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMallocPitch(void **devPtr,
+                                                      size_t *pitch,
+                                                      size_t width,
+                                                      size_t height) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void **, size_t *, size_t, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMallocPitch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, pitch, width, height);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMallocArray(
+    cudaArray_t *array, const struct cudaChannelFormatDesc *desc, size_t width,
+    size_t height __dv(0), unsigned int flags __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaArray_t *,
+                                           const struct cudaChannelFormatDesc *,
+                                           size_t, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMallocArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(array, desc, width, height, flags);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaFree(void *devPtr) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaFree");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaFreeHost(void *ptr) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaFreeHost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaFreeArray(cudaArray_t array) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaArray_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaFreeArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(array);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaFreeMipmappedArray(cudaMipmappedArray_t mipmappedArray) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaMipmappedArray_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaFreeMipmappedArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mipmappedArray);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaHostAlloc(void **pHost, size_t size,
+                                                    unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void **, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaHostAlloc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pHost, size, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaHostRegister(void *ptr, size_t size,
+                                                       unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, size_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaHostRegister");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr, size, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaHostUnregister(void *ptr) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaHostUnregister");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaHostGetDevicePointer(void **pDevice, void *pHost, unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void **, void *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaHostGetDevicePointer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pDevice, pHost, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaHostGetFlags(unsigned int *pFlags,
+                                                       void *pHost) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(unsigned int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaHostGetFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pFlags, pHost);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMalloc3D(struct cudaPitchedPtr *pitchedDevPtr, struct cudaExtent extent) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(struct cudaPitchedPtr *, struct cudaExtent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMalloc3D");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pitchedDevPtr, extent);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMalloc3DArray(cudaArray_t *array, const struct cudaChannelFormatDesc *desc,
+                  struct cudaExtent extent, unsigned int flags __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaArray_t *,
+                                           const struct cudaChannelFormatDesc *,
+                                           struct cudaExtent, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMalloc3DArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(array, desc, extent, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMallocMipmappedArray(
+    cudaMipmappedArray_t *mipmappedArray,
+    const struct cudaChannelFormatDesc *desc, struct cudaExtent extent,
+    unsigned int numLevels, unsigned int flags __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaMipmappedArray_t *, const struct cudaChannelFormatDesc *,
+      struct cudaExtent, unsigned int, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMallocMipmappedArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mipmappedArray, desc, extent, numLevels, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetMipmappedArrayLevel(
+    cudaArray_t *levelArray, cudaMipmappedArray_const_t mipmappedArray,
+    unsigned int level) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaArray_t *, cudaMipmappedArray_const_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetMipmappedArrayLevel");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(levelArray, mipmappedArray, level);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMemcpy3D(const struct cudaMemcpy3DParms *p) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(const struct cudaMemcpy3DParms *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy3D");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(p);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMemcpy3DPeer(const struct cudaMemcpy3DPeerParms *p) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(const struct cudaMemcpy3DPeerParms *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy3DPeer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(p);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI cudaMemcpy3DAsync(
+    const struct cudaMemcpy3DParms *p, cudaStream_t stream __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(const struct cudaMemcpy3DParms *, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy3DAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(p, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpy3DPeerAsync(
+    const struct cudaMemcpy3DPeerParms *p, cudaStream_t stream __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(const struct cudaMemcpy3DPeerParms *,
+                                           cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy3DPeerAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(p, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemGetInfo(size_t *free,
+                                                     size_t *total) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(size_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemGetInfo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(free, total);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaArrayGetInfo(struct cudaChannelFormatDesc *desc, struct cudaExtent *extent,
+                 unsigned int *flags, cudaArray_t array) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(struct cudaChannelFormatDesc *,
+                                           struct cudaExtent *, unsigned int *,
+                                           cudaArray_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaArrayGetInfo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(desc, extent, flags, array);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaArrayGetPlane(
+    cudaArray_t *pPlaneArray, cudaArray_t hArray, unsigned int planeIdx) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaArray_t *, cudaArray_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaArrayGetPlane");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pPlaneArray, hArray, planeIdx);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaArrayGetMemoryRequirements(
+    struct cudaArrayMemoryRequirements *memoryRequirements, cudaArray_t array,
+    int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(struct cudaArrayMemoryRequirements *,
+                                           cudaArray_t, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaArrayGetMemoryRequirements");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memoryRequirements, array, device);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMipmappedArrayGetMemoryRequirements(
+    struct cudaArrayMemoryRequirements *memoryRequirements,
+    cudaMipmappedArray_t mipmap, int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(struct cudaArrayMemoryRequirements *,
+                                           cudaMipmappedArray_t, int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaMipmappedArrayGetMemoryRequirements");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memoryRequirements, mipmap, device);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaArrayGetSparseProperties(
+    struct cudaArraySparseProperties *sparseProperties, cudaArray_t array) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(struct cudaArraySparseProperties *, cudaArray_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaArrayGetSparseProperties");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(sparseProperties, array);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMipmappedArrayGetSparseProperties(
+    struct cudaArraySparseProperties *sparseProperties,
+    cudaMipmappedArray_t mipmap) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(struct cudaArraySparseProperties *,
+                                           cudaMipmappedArray_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaMipmappedArrayGetSparseProperties");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(sparseProperties, mipmap);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src,
+                                                 size_t count,
+                                                 enum cudaMemcpyKind kind) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, const void *, size_t,
+                                           enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, src, count, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpyPeer(void *dst, int dstDevice,
+                                                     const void *src,
+                                                     int srcDevice,
+                                                     size_t count) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(void *, int, const void *, int, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyPeer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, dstDevice, src, srcDevice, count);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpy2D(void *dst, size_t dpitch,
+                                                   const void *src,
+                                                   size_t spitch, size_t width,
+                                                   size_t height,
+                                                   enum cudaMemcpyKind kind) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, size_t, const void *, size_t,
+                                           size_t, size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy2D");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, dpitch, src, spitch, width, height, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpy2DToArray(
+    cudaArray_t dst, size_t wOffset, size_t hOffset, const void *src,
+    size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaArray_t, size_t, size_t, const void *,
+                               size_t, size_t, size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy2DToArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, wOffset, hOffset, src, spitch, width, height, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArray(
+    void *dst, size_t dpitch, cudaArray_const_t src, size_t wOffset,
+    size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(void *, size_t, cudaArray_const_t, size_t,
+                               size_t, size_t, size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy2DFromArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, dpitch, src, wOffset, hOffset, width, height, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpy2DArrayToArray(
+    cudaArray_t dst, size_t wOffsetDst, size_t hOffsetDst,
+    cudaArray_const_t src, size_t wOffsetSrc, size_t hOffsetSrc, size_t width,
+    size_t height, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaArray_t, size_t, size_t,
+                                           cudaArray_const_t, size_t, size_t,
+                                           size_t, size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy2DArrayToArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, wOffsetDst, hOffsetDst, src, wOffsetSrc, hOffsetSrc,
+                  width, height, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol(
+    const void *symbol, const void *src, size_t count, size_t offset __dv(0),
+    enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(const void *, const void *, size_t,
+                                           size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyToSymbol");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(symbol, src, count, offset, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbol(
+    void *dst, const void *symbol, size_t count, size_t offset __dv(0),
+    enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, const void *, size_t, size_t,
+                                           enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyFromSymbol");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, symbol, count, offset, kind);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaMemcpyAsync(void *dst, const void *src, size_t count,
+                enum cudaMemcpyKind kind, cudaStream_t stream __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, const void *, size_t,
+                                           enum cudaMemcpyKind, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, src, count, kind, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMemcpyPeerAsync(void *dst, int dstDevice, const void *src, int srcDevice,
+                    size_t count, cudaStream_t stream __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, int, const void *, int,
+                                           size_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyPeerAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, dstDevice, src, srcDevice, count, stream);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI cudaMemcpy2DAsync(
+    void *dst, size_t dpitch, const void *src, size_t spitch, size_t width,
+    size_t height, enum cudaMemcpyKind kind, cudaStream_t stream __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(void *, size_t, const void *, size_t, size_t,
+                               size_t, enum cudaMemcpyKind, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy2DAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, dpitch, src, spitch, width, height, kind, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayAsync(
+    cudaArray_t dst, size_t wOffset, size_t hOffset, const void *src,
+    size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind,
+    cudaStream_t stream __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaArray_t, size_t, size_t,
+                                           const void *, size_t, size_t, size_t,
+                                           enum cudaMemcpyKind, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy2DToArrayAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, wOffset, hOffset, src, spitch, width, height, kind,
+                  stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArrayAsync(
+    void *dst, size_t dpitch, cudaArray_const_t src, size_t wOffset,
+    size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind,
+    cudaStream_t stream __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, size_t, cudaArray_const_t,
+                                           size_t, size_t, size_t, size_t,
+                                           enum cudaMemcpyKind, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpy2DFromArrayAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, dpitch, src, wOffset, hOffset, width, height, kind,
+                  stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpyToSymbolAsync(
+    const void *symbol, const void *src, size_t count, size_t offset,
+    enum cudaMemcpyKind kind, cudaStream_t stream __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(const void *, const void *, size_t, size_t,
+                               enum cudaMemcpyKind, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyToSymbolAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(symbol, src, count, offset, kind, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbolAsync(
+    void *dst, const void *symbol, size_t count, size_t offset,
+    enum cudaMemcpyKind kind, cudaStream_t stream __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, const void *, size_t, size_t,
+                                           enum cudaMemcpyKind, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyFromSymbolAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, symbol, count, offset, kind, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemset(void *devPtr, int value,
+                                                 size_t count) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, int, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemset");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, value, count);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemset2D(void *devPtr, size_t pitch,
+                                                   int value, size_t width,
+                                                   size_t height) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, size_t, int, size_t, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemset2D");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, pitch, value, width, height);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemset3D(
+    struct cudaPitchedPtr pitchedDevPtr, int value, struct cudaExtent extent) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(struct cudaPitchedPtr, int, struct cudaExtent);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemset3D");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pitchedDevPtr, value, extent);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI cudaMemsetAsync(
+    void *devPtr, int value, size_t count, cudaStream_t stream __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, int, size_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemsetAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, value, count, stream);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaMemset2DAsync(void *devPtr, size_t pitch, int value, size_t width,
+                  size_t height, cudaStream_t stream __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, size_t, int, size_t, size_t,
+                                           cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemset2DAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, pitch, value, width, height, stream);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaMemset3DAsync(struct cudaPitchedPtr pitchedDevPtr, int value,
+                  struct cudaExtent extent, cudaStream_t stream __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(struct cudaPitchedPtr, int,
+                                           struct cudaExtent, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemset3DAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pitchedDevPtr, value, extent, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetSymbolAddress(void **devPtr,
+                                                           const void *symbol) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void **, const void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetSymbolAddress");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, symbol);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetSymbolSize(size_t *size,
+                                                        const void *symbol) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(size_t *, const void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetSymbolSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(size, symbol);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMemPrefetchAsync(const void *devPtr, size_t count, int dstDevice,
+                     cudaStream_t stream __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(const void *, size_t, int, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemPrefetchAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, count, dstDevice, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMemAdvise(const void *devPtr, size_t count, enum cudaMemoryAdvise advice,
+              int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(const void *, size_t,
+                                           enum cudaMemoryAdvise, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemAdvise");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, count, advice, device);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemRangeGetAttribute(
+    void *data, size_t dataSize, enum cudaMemRangeAttribute attribute,
+    const void *devPtr, size_t count) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      void *, size_t, enum cudaMemRangeAttribute, const void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemRangeGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(data, dataSize, attribute, devPtr, count);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemRangeGetAttributes(
+    void **data, size_t *dataSizes, enum cudaMemRangeAttribute *attributes,
+    size_t numAttributes, const void *devPtr, size_t count) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(void **, size_t *, enum cudaMemRangeAttribute *,
+                               size_t, const void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemRangeGetAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(data, dataSizes, attributes, numAttributes, devPtr, count);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaMemcpyToArray(cudaArray_t dst, size_t wOffset, size_t hOffset,
+                  const void *src, size_t count, enum cudaMemcpyKind kind) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaArray_t, size_t, size_t, const void *, size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyToArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, wOffset, hOffset, src, count, kind);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaMemcpyFromArray(void *dst, cudaArray_const_t src, size_t wOffset,
+                    size_t hOffset, size_t count, enum cudaMemcpyKind kind) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, cudaArray_const_t, size_t,
+                                           size_t, size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyFromArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, src, wOffset, hOffset, count, kind);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI cudaMemcpyArrayToArray(
+    cudaArray_t dst, size_t wOffsetDst, size_t hOffsetDst,
+    cudaArray_const_t src, size_t wOffsetSrc, size_t hOffsetSrc, size_t count,
+    enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaArray_t, size_t, size_t, cudaArray_const_t,
+                               size_t, size_t, size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyArrayToArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, wOffsetDst, hOffsetDst, src, wOffsetSrc, hOffsetSrc,
+                  count, kind);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI cudaMemcpyToArrayAsync(
+    cudaArray_t dst, size_t wOffset, size_t hOffset, const void *src,
+    size_t count, enum cudaMemcpyKind kind, cudaStream_t stream __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaArray_t, size_t, size_t, const void *,
+                               size_t, enum cudaMemcpyKind, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyToArrayAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, wOffset, hOffset, src, count, kind, stream);
+}
+
+extern __CUDA_DEPRECATED __host__ cudaError_t CUDARTAPI
+cudaMemcpyFromArrayAsync(void *dst, cudaArray_const_t src, size_t wOffset,
+                         size_t hOffset, size_t count, enum cudaMemcpyKind kind,
+                         cudaStream_t stream __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(void *, cudaArray_const_t, size_t, size_t,
+                               size_t, enum cudaMemcpyKind, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemcpyFromArrayAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dst, src, wOffset, hOffset, count, kind, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMallocAsync(void **devPtr,
+                                                      size_t size,
+                                                      cudaStream_t hStream) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void **, size_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMallocAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, size, hStream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaFreeAsync(void *devPtr,
+                                                    cudaStream_t hStream) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void *, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaFreeAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, hStream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemPoolTrimTo(cudaMemPool_t memPool,
+                                                        size_t minBytesToKeep) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaMemPool_t, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemPoolTrimTo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memPool, minBytesToKeep);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemPoolSetAttribute(
+    cudaMemPool_t memPool, enum cudaMemPoolAttr attr, void *value) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaMemPool_t, enum cudaMemPoolAttr, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemPoolSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memPool, attr, value);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemPoolGetAttribute(
+    cudaMemPool_t memPool, enum cudaMemPoolAttr attr, void *value) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaMemPool_t, enum cudaMemPoolAttr, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemPoolGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memPool, attr, value);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMemPoolSetAccess(cudaMemPool_t memPool,
+                     const struct cudaMemAccessDesc *descList, size_t count) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaMemPool_t, const struct cudaMemAccessDesc *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemPoolSetAccess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memPool, descList, count);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMemPoolGetAccess(enum cudaMemAccessFlags *flags, cudaMemPool_t memPool,
+                     struct cudaMemLocation *location) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      enum cudaMemAccessFlags *, cudaMemPool_t, struct cudaMemLocation *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemPoolGetAccess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(flags, memPool, location);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemPoolCreate(
+    cudaMemPool_t *memPool, const struct cudaMemPoolProps *poolProps) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaMemPool_t *,
+                                           const struct cudaMemPoolProps *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemPoolCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memPool, poolProps);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMemPoolDestroy(cudaMemPool_t memPool) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaMemPool_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemPoolDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memPool);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMallocFromPoolAsync(
+    void **ptr, size_t size, cudaMemPool_t memPool, cudaStream_t stream) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(void **, size_t, cudaMemPool_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMallocFromPoolAsync");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr, size, memPool, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemPoolExportToShareableHandle(
+    void *shareableHandle, cudaMemPool_t memPool,
+    enum cudaMemAllocationHandleType handleType, unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      void *, cudaMemPool_t, enum cudaMemAllocationHandleType, unsigned int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaMemPoolExportToShareableHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(shareableHandle, memPool, handleType, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemPoolImportFromShareableHandle(
+    cudaMemPool_t *memPool, void *shareableHandle,
+    enum cudaMemAllocationHandleType handleType, unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaMemPool_t *, void *, enum cudaMemAllocationHandleType, unsigned int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaMemPoolImportFromShareableHandle");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(memPool, shareableHandle, handleType, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaMemPoolExportPointer(
+    struct cudaMemPoolPtrExportData *exportData, void *ptr) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(struct cudaMemPoolPtrExportData *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemPoolExportPointer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(exportData, ptr);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaMemPoolImportPointer(void **ptr, cudaMemPool_t memPool,
+                         struct cudaMemPoolPtrExportData *exportData) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(void **, cudaMemPool_t,
+                                           struct cudaMemPoolPtrExportData *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaMemPoolImportPointer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ptr, memPool, exportData);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaPointerGetAttributes(
+    struct cudaPointerAttributes *attributes, const void *ptr) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(struct cudaPointerAttributes *, const void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaPointerGetAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(attributes, ptr);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDeviceCanAccessPeer(int *canAccessPeer, int device, int peerDevice) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int *, int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceCanAccessPeer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(canAccessPeer, device, peerDevice);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDeviceEnablePeerAccess(int peerDevice, unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceEnablePeerAccess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(peerDevice, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDeviceDisablePeerAccess(int peerDevice) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceDisablePeerAccess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(peerDevice);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphicsUnregisterResource(cudaGraphicsResource_t resource) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphicsResource_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphicsUnregisterResource");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(resource);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphicsResourceSetMapFlags(
+    cudaGraphicsResource_t resource, unsigned int flags) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphicsResource_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphicsResourceSetMapFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(resource, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphicsMapResources(
+    int count, cudaGraphicsResource_t *resources, cudaStream_t stream __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(int, cudaGraphicsResource_t *, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphicsMapResources");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(count, resources, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphicsUnmapResources(
+    int count, cudaGraphicsResource_t *resources, cudaStream_t stream __dv(0)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(int, cudaGraphicsResource_t *, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphicsUnmapResources");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(count, resources, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphicsResourceGetMappedPointer(
+    void **devPtr, size_t *size, cudaGraphicsResource_t resource) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(void **, size_t *, cudaGraphicsResource_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphicsResourceGetMappedPointer");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(devPtr, size, resource);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphicsSubResourceGetMappedArray(
+    cudaArray_t *array, cudaGraphicsResource_t resource,
+    unsigned int arrayIndex, unsigned int mipLevel) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaArray_t *, cudaGraphicsResource_t, unsigned int, unsigned int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphicsSubResourceGetMappedArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(array, resource, arrayIndex, mipLevel);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphicsResourceGetMappedMipmappedArray(
+    cudaMipmappedArray_t *mipmappedArray, cudaGraphicsResource_t resource) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaMipmappedArray_t *, cudaGraphicsResource_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphicsResourceGetMappedMipmappedArray");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mipmappedArray, resource);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetChannelDesc(
+    struct cudaChannelFormatDesc *desc, cudaArray_const_t array) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(struct cudaChannelFormatDesc *,
+                                           cudaArray_const_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetChannelDesc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(desc, array);
+}
+
+extern __host__ struct cudaChannelFormatDesc CUDARTAPI cudaCreateChannelDesc(
+    int x, int y, int z, int w, enum cudaChannelFormatKind f) {
+  using FuncPtr = struct cudaChannelFormatDesc(CUDARTAPI *)(
+      int, int, int, int, enum cudaChannelFormatKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaCreateChannelDesc");
+  return func_ptr(x, y, z, w, f);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaCreateTextureObject(
+    cudaTextureObject_t *pTexObject, const struct cudaResourceDesc *pResDesc,
+    const struct cudaTextureDesc *pTexDesc,
+    const struct cudaResourceViewDesc *pResViewDesc) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaTextureObject_t *, const struct cudaResourceDesc *,
+      const struct cudaTextureDesc *, const struct cudaResourceViewDesc *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaCreateTextureObject");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pTexObject, pResDesc, pTexDesc, pResViewDesc);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDestroyTextureObject(cudaTextureObject_t texObject) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaTextureObject_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDestroyTextureObject");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(texObject);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetTextureObjectResourceDesc(
+    struct cudaResourceDesc *pResDesc, cudaTextureObject_t texObject) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(struct cudaResourceDesc *, cudaTextureObject_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGetTextureObjectResourceDesc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pResDesc, texObject);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetTextureObjectTextureDesc(
+    struct cudaTextureDesc *pTexDesc, cudaTextureObject_t texObject) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(struct cudaTextureDesc *, cudaTextureObject_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetTextureObjectTextureDesc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pTexDesc, texObject);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetTextureObjectResourceViewDesc(
+    struct cudaResourceViewDesc *pResViewDesc, cudaTextureObject_t texObject) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(struct cudaResourceViewDesc *,
+                                           cudaTextureObject_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGetTextureObjectResourceViewDesc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pResViewDesc, texObject);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaCreateSurfaceObject(
+    cudaSurfaceObject_t *pSurfObject, const struct cudaResourceDesc *pResDesc) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaSurfaceObject_t *,
+                                           const struct cudaResourceDesc *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaCreateSurfaceObject");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pSurfObject, pResDesc);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaDestroySurfaceObject(cudaSurfaceObject_t surfObject) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaSurfaceObject_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDestroySurfaceObject");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(surfObject);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetSurfaceObjectResourceDesc(
+    struct cudaResourceDesc *pResDesc, cudaSurfaceObject_t surfObject) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(struct cudaResourceDesc *, cudaSurfaceObject_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGetSurfaceObjectResourceDesc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pResDesc, surfObject);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaDriverGetVersion(int *driverVersion) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDriverGetVersion");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(driverVersion);
+}
+
+extern __host__ __cudart_builtin__ cudaError_t CUDARTAPI
+cudaRuntimeGetVersion(int *runtimeVersion) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaRuntimeGetVersion");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(runtimeVersion);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphCreate(cudaGraph_t *pGraph,
+                                                      unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraph_t *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraph, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddKernelNode(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies,
+    const struct cudaKernelNodeParams *pNodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t *, cudaGraph_t,
+                                           const cudaGraphNode_t *, size_t,
+                                           const struct cudaKernelNodeParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddKernelNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies,
+                  pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphKernelNodeGetParams(
+    cudaGraphNode_t node, struct cudaKernelNodeParams *pNodeParams) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t, struct cudaKernelNodeParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphKernelNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphKernelNodeSetParams(
+    cudaGraphNode_t node, const struct cudaKernelNodeParams *pNodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t,
+                                           const struct cudaKernelNodeParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphKernelNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphKernelNodeCopyAttributes(cudaGraphNode_t hSrc, cudaGraphNode_t hDst) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t, cudaGraphNode_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphKernelNodeCopyAttributes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hSrc, hDst);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphKernelNodeGetAttribute(
+    cudaGraphNode_t hNode, cudaKernelNodeAttrID attr,
+    cudaKernelNodeAttrValue *value_out) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t, cudaKernelNodeAttrID, cudaKernelNodeAttrValue *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphKernelNodeGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, attr, value_out);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphKernelNodeSetAttribute(
+    cudaGraphNode_t hNode, cudaKernelNodeAttrID attr,
+    const cudaKernelNodeAttrValue *value) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t, cudaKernelNodeAttrID, const cudaKernelNodeAttrValue *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphKernelNodeSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, attr, value);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddMemcpyNode(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies,
+    const struct cudaMemcpy3DParms *pCopyParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t *, cudaGraph_t,
+                                           const cudaGraphNode_t *, size_t,
+                                           const struct cudaMemcpy3DParms *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddMemcpyNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies,
+                  pCopyParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddMemcpyNodeToSymbol(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies,
+    const void *symbol, const void *src, size_t count, size_t offset,
+    enum cudaMemcpyKind kind) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t *, cudaGraph_t, const cudaGraphNode_t *, size_t,
+      const void *, const void *, size_t, size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddMemcpyNodeToSymbol");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies, symbol,
+                  src, count, offset, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddMemcpyNodeFromSymbol(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies, void *dst,
+    const void *symbol, size_t count, size_t offset, enum cudaMemcpyKind kind) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t *, cudaGraph_t, const cudaGraphNode_t *, size_t, void *,
+      const void *, size_t, size_t, enum cudaMemcpyKind);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphAddMemcpyNodeFromSymbol");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies, dst,
+                  symbol, count, offset, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddMemcpyNode1D(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies, void *dst,
+    const void *src, size_t count, enum cudaMemcpyKind kind) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t *, cudaGraph_t, const cudaGraphNode_t *, size_t, void *,
+      const void *, size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddMemcpyNode1D");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies, dst, src,
+                  count, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphMemcpyNodeGetParams(
+    cudaGraphNode_t node, struct cudaMemcpy3DParms *pNodeParams) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t, struct cudaMemcpy3DParms *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphMemcpyNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphMemcpyNodeSetParams(
+    cudaGraphNode_t node, const struct cudaMemcpy3DParms *pNodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t,
+                                           const struct cudaMemcpy3DParms *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphMemcpyNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphMemcpyNodeSetParamsToSymbol(
+    cudaGraphNode_t node, const void *symbol, const void *src, size_t count,
+    size_t offset, enum cudaMemcpyKind kind) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t, const void *, const void *,
+                               size_t, size_t, enum cudaMemcpyKind);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphMemcpyNodeSetParamsToSymbol");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, symbol, src, count, offset, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphMemcpyNodeSetParamsFromSymbol(
+    cudaGraphNode_t node, void *dst, const void *symbol, size_t count,
+    size_t offset, enum cudaMemcpyKind kind) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t, void *, const void *, size_t,
+                               size_t, enum cudaMemcpyKind);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphMemcpyNodeSetParamsFromSymbol");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, dst, symbol, count, offset, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphMemcpyNodeSetParams1D(cudaGraphNode_t node, void *dst, const void *src,
+                               size_t count, enum cudaMemcpyKind kind) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t, void *, const void *, size_t, enum cudaMemcpyKind);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphMemcpyNodeSetParams1D");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, dst, src, count, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddMemsetNode(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies,
+    const struct cudaMemsetParams *pMemsetParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t *, cudaGraph_t,
+                                           const cudaGraphNode_t *, size_t,
+                                           const struct cudaMemsetParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddMemsetNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies,
+                  pMemsetParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphMemsetNodeGetParams(
+    cudaGraphNode_t node, struct cudaMemsetParams *pNodeParams) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t, struct cudaMemsetParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphMemsetNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphMemsetNodeSetParams(
+    cudaGraphNode_t node, const struct cudaMemsetParams *pNodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t,
+                                           const struct cudaMemsetParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphMemsetNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddHostNode(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies,
+    const struct cudaHostNodeParams *pNodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t *, cudaGraph_t,
+                                           const cudaGraphNode_t *, size_t,
+                                           const struct cudaHostNodeParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddHostNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies,
+                  pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphHostNodeGetParams(
+    cudaGraphNode_t node, struct cudaHostNodeParams *pNodeParams) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t, struct cudaHostNodeParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphHostNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphHostNodeSetParams(
+    cudaGraphNode_t node, const struct cudaHostNodeParams *pNodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t,
+                                           const struct cudaHostNodeParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphHostNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphAddChildGraphNode(cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+                           const cudaGraphNode_t *pDependencies,
+                           size_t numDependencies, cudaGraph_t childGraph) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t *, cudaGraph_t,
+                               const cudaGraphNode_t *, size_t, cudaGraph_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddChildGraphNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies,
+                  childGraph);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphChildGraphNodeGetGraph(cudaGraphNode_t node, cudaGraph_t *pGraph) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t, cudaGraph_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphChildGraphNodeGetGraph");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pGraph);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddEmptyNode(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t *, cudaGraph_t,
+                                           const cudaGraphNode_t *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddEmptyNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphAddEventRecordNode(cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+                            const cudaGraphNode_t *pDependencies,
+                            size_t numDependencies, cudaEvent_t event) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t *, cudaGraph_t,
+                               const cudaGraphNode_t *, size_t, cudaEvent_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddEventRecordNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies, event);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphEventRecordNodeGetEvent(cudaGraphNode_t node, cudaEvent_t *event_out) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t, cudaEvent_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphEventRecordNodeGetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, event_out);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphEventRecordNodeSetEvent(cudaGraphNode_t node, cudaEvent_t event) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t, cudaEvent_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphEventRecordNodeSetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, event);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphAddEventWaitNode(cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+                          const cudaGraphNode_t *pDependencies,
+                          size_t numDependencies, cudaEvent_t event) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t *, cudaGraph_t,
+                               const cudaGraphNode_t *, size_t, cudaEvent_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddEventWaitNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies, event);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphEventWaitNodeGetEvent(cudaGraphNode_t node, cudaEvent_t *event_out) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t, cudaEvent_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphEventWaitNodeGetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, event_out);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphEventWaitNodeSetEvent(cudaGraphNode_t node, cudaEvent_t event) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t, cudaEvent_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphEventWaitNodeSetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, event);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddExternalSemaphoresSignalNode(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies,
+    const struct cudaExternalSemaphoreSignalNodeParams *nodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t *, cudaGraph_t, const cudaGraphNode_t *, size_t,
+      const struct cudaExternalSemaphoreSignalNodeParams *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphAddExternalSemaphoresSignalNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies,
+                  nodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExternalSemaphoresSignalNodeGetParams(
+    cudaGraphNode_t hNode,
+    struct cudaExternalSemaphoreSignalNodeParams *params_out) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t, struct cudaExternalSemaphoreSignalNodeParams *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExternalSemaphoresSignalNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, params_out);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExternalSemaphoresSignalNodeSetParams(
+    cudaGraphNode_t hNode,
+    const struct cudaExternalSemaphoreSignalNodeParams *nodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t, const struct cudaExternalSemaphoreSignalNodeParams *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExternalSemaphoresSignalNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddExternalSemaphoresWaitNode(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies,
+    const struct cudaExternalSemaphoreWaitNodeParams *nodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t *, cudaGraph_t, const cudaGraphNode_t *, size_t,
+      const struct cudaExternalSemaphoreWaitNodeParams *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphAddExternalSemaphoresWaitNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies,
+                  nodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExternalSemaphoresWaitNodeGetParams(
+    cudaGraphNode_t hNode,
+    struct cudaExternalSemaphoreWaitNodeParams *params_out) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t, struct cudaExternalSemaphoreWaitNodeParams *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExternalSemaphoresWaitNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, params_out);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExternalSemaphoresWaitNodeSetParams(
+    cudaGraphNode_t hNode,
+    const struct cudaExternalSemaphoreWaitNodeParams *nodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t, const struct cudaExternalSemaphoreWaitNodeParams *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExternalSemaphoresWaitNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hNode, nodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddMemAllocNode(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies,
+    struct cudaMemAllocNodeParams *nodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t *, cudaGraph_t,
+                                           const cudaGraphNode_t *, size_t,
+                                           struct cudaMemAllocNodeParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddMemAllocNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies,
+                  nodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphMemAllocNodeGetParams(
+    cudaGraphNode_t node, struct cudaMemAllocNodeParams *params_out) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t,
+                                           struct cudaMemAllocNodeParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphMemAllocNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, params_out);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphAddMemFreeNode(
+    cudaGraphNode_t *pGraphNode, cudaGraph_t graph,
+    const cudaGraphNode_t *pDependencies, size_t numDependencies, void *dptr) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphNode_t *, cudaGraph_t, const cudaGraphNode_t *, size_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddMemFreeNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphNode, graph, pDependencies, numDependencies, dptr);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphMemFreeNodeGetParams(cudaGraphNode_t node, void *dptr_out) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphMemFreeNodeGetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, dptr_out);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaDeviceGraphMemTrim(int device) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGraphMemTrim");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaDeviceGetGraphMemAttribute(
+    int device, enum cudaGraphMemAttributeType attr, void *value) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(int, enum cudaGraphMemAttributeType, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceGetGraphMemAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device, attr, value);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaDeviceSetGraphMemAttribute(
+    int device, enum cudaGraphMemAttributeType attr, void *value) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(int, enum cudaGraphMemAttributeType, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaDeviceSetGraphMemAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(device, attr, value);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphClone(cudaGraph_t *pGraphClone, cudaGraph_t originalGraph) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraph_t *, cudaGraph_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphClone");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphClone, originalGraph);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphNodeFindInClone(cudaGraphNode_t *pNode, cudaGraphNode_t originalNode,
+                         cudaGraph_t clonedGraph) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t *, cudaGraphNode_t, cudaGraph_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphNodeFindInClone");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pNode, originalNode, clonedGraph);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphNodeGetType(cudaGraphNode_t node, enum cudaGraphNodeType *pType) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t, enum cudaGraphNodeType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphNodeGetType");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pType);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphGetNodes(cudaGraph_t graph,
+                                                        cudaGraphNode_t *nodes,
+                                                        size_t *numNodes) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraph_t, cudaGraphNode_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphGetNodes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph, nodes, numNodes);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphGetRootNodes(
+    cudaGraph_t graph, cudaGraphNode_t *pRootNodes, size_t *pNumRootNodes) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraph_t, cudaGraphNode_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphGetRootNodes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph, pRootNodes, pNumRootNodes);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphGetEdges(cudaGraph_t graph,
+                                                        cudaGraphNode_t *from,
+                                                        cudaGraphNode_t *to,
+                                                        size_t *numEdges) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraph_t, cudaGraphNode_t *,
+                                           cudaGraphNode_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphGetEdges");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph, from, to, numEdges);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphNodeGetDependencies(
+    cudaGraphNode_t node, cudaGraphNode_t *pDependencies,
+    size_t *pNumDependencies) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t, cudaGraphNode_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphNodeGetDependencies");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pDependencies, pNumDependencies);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphNodeGetDependentNodes(
+    cudaGraphNode_t node, cudaGraphNode_t *pDependentNodes,
+    size_t *pNumDependentNodes) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphNode_t, cudaGraphNode_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphNodeGetDependentNodes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node, pDependentNodes, pNumDependentNodes);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphAddDependencies(cudaGraph_t graph, const cudaGraphNode_t *from,
+                         const cudaGraphNode_t *to, size_t numDependencies) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraph_t, const cudaGraphNode_t *,
+                                           const cudaGraphNode_t *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphAddDependencies");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph, from, to, numDependencies);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphRemoveDependencies(cudaGraph_t graph, const cudaGraphNode_t *from,
+                            const cudaGraphNode_t *to, size_t numDependencies) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraph_t, const cudaGraphNode_t *,
+                                           const cudaGraphNode_t *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphRemoveDependencies");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph, from, to, numDependencies);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphDestroyNode(cudaGraphNode_t node) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphNode_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphDestroyNode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(node);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphInstantiate(cudaGraphExec_t *pGraphExec, cudaGraph_t graph,
+                     unsigned long long flags __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t *, cudaGraph_t,
+                                           unsigned long long);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphInstantiate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphExec, graph, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphInstantiateWithFlags(cudaGraphExec_t *pGraphExec, cudaGraph_t graph,
+                              unsigned long long flags __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t *, cudaGraph_t,
+                                           unsigned long long);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphInstantiateWithFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphExec, graph, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphInstantiateWithParams(cudaGraphExec_t *pGraphExec, cudaGraph_t graph,
+                               cudaGraphInstantiateParams *instantiateParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t *, cudaGraph_t,
+                                           cudaGraphInstantiateParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphInstantiateWithParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(pGraphExec, graph, instantiateParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExecGetFlags(cudaGraphExec_t graphExec, unsigned long long *flags) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphExec_t, unsigned long long *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphExecGetFlags");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graphExec, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphExecKernelNodeSetParams(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t node,
+    const struct cudaKernelNodeParams *pNodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t,
+                                           const struct cudaKernelNodeParams *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecKernelNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphExecMemcpyNodeSetParams(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t node,
+    const struct cudaMemcpy3DParms *pNodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t,
+                                           const struct cudaMemcpy3DParms *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecMemcpyNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphExecMemcpyNodeSetParamsToSymbol(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t node, const void *symbol,
+    const void *src, size_t count, size_t offset, enum cudaMemcpyKind kind) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t,
+                                           const void *, const void *, size_t,
+                                           size_t, enum cudaMemcpyKind);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecMemcpyNodeSetParamsToSymbol");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, node, symbol, src, count, offset, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExecMemcpyNodeSetParamsFromSymbol(cudaGraphExec_t hGraphExec,
+                                           cudaGraphNode_t node, void *dst,
+                                           const void *symbol, size_t count,
+                                           size_t offset,
+                                           enum cudaMemcpyKind kind) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t,
+                                           void *, const void *, size_t, size_t,
+                                           enum cudaMemcpyKind);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecMemcpyNodeSetParamsFromSymbol");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, node, dst, symbol, count, offset, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphExecMemcpyNodeSetParams1D(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t node, void *dst,
+    const void *src, size_t count, enum cudaMemcpyKind kind) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t, void *,
+                               const void *, size_t, enum cudaMemcpyKind);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecMemcpyNodeSetParams1D");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, node, dst, src, count, kind);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphExecMemsetNodeSetParams(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t node,
+    const struct cudaMemsetParams *pNodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t,
+                                           const struct cudaMemsetParams *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecMemsetNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExecHostNodeSetParams(cudaGraphExec_t hGraphExec, cudaGraphNode_t node,
+                               const struct cudaHostNodeParams *pNodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t,
+                                           const struct cudaHostNodeParams *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphExecHostNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, node, pNodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphExecChildGraphNodeSetParams(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t node, cudaGraph_t childGraph) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t, cudaGraph_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecChildGraphNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, node, childGraph);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphExecEventRecordNodeSetEvent(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t hNode, cudaEvent_t event) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t, cudaEvent_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecEventRecordNodeSetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, event);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphExecEventWaitNodeSetEvent(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t hNode, cudaEvent_t event) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t, cudaEvent_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecEventWaitNodeSetEvent");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, event);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExecExternalSemaphoresSignalNodeSetParams(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t hNode,
+    const struct cudaExternalSemaphoreSignalNodeParams *nodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphExec_t, cudaGraphNode_t,
+      const struct cudaExternalSemaphoreSignalNodeParams *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecExternalSemaphoresSignalNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, nodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExecExternalSemaphoresWaitNodeSetParams(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t hNode,
+    const struct cudaExternalSemaphoreWaitNodeParams *nodeParams) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaGraphExec_t, cudaGraphNode_t,
+      const struct cudaExternalSemaphoreWaitNodeParams *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudaGraphExecExternalSemaphoresWaitNodeSetParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, nodeParams);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphNodeSetEnabled(
+    cudaGraphExec_t hGraphExec, cudaGraphNode_t hNode, unsigned int isEnabled) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphNodeSetEnabled");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, isEnabled);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphNodeGetEnabled(cudaGraphExec_t hGraphExec, cudaGraphNode_t hNode,
+                        unsigned int *isEnabled) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraphNode_t,
+                                           unsigned int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphNodeGetEnabled");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hNode, isEnabled);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExecUpdate(cudaGraphExec_t hGraphExec, cudaGraph_t hGraph,
+                    cudaGraphExecUpdateResultInfo *resultInfo) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaGraph_t,
+                                           cudaGraphExecUpdateResultInfo *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphExecUpdate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(hGraphExec, hGraph, resultInfo);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphUpload(cudaGraphExec_t graphExec,
+                                                      cudaStream_t stream) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphUpload");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graphExec, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphLaunch(cudaGraphExec_t graphExec,
+                                                      cudaStream_t stream) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphLaunch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graphExec, stream);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaGraphExecDestroy(cudaGraphExec_t graphExec) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraphExec_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphExecDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graphExec);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphDestroy(cudaGraph_t graph) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraph_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphDebugDotPrint(
+    cudaGraph_t graph, const char *path, unsigned int flags) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraph_t, const char *, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphDebugDotPrint");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph, path, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaUserObjectCreate(
+    cudaUserObject_t *object_out, void *ptr, cudaHostFn_t destroy,
+    unsigned int initialRefcount, unsigned int flags) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(
+      cudaUserObject_t *, void *, cudaHostFn_t, unsigned int, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaUserObjectCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(object_out, ptr, destroy, initialRefcount, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaUserObjectRetain(cudaUserObject_t object, unsigned int count __dv(1)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaUserObject_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaUserObjectRetain");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(object, count);
+}
+
+extern __host__ cudaError_t CUDARTAPI
+cudaUserObjectRelease(cudaUserObject_t object, unsigned int count __dv(1)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaUserObject_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaUserObjectRelease");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(object, count);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphRetainUserObject(
+    cudaGraph_t graph, cudaUserObject_t object, unsigned int count __dv(1),
+    unsigned int flags __dv(0)) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(cudaGraph_t, cudaUserObject_t,
+                                           unsigned int, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphRetainUserObject");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph, object, count, flags);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGraphReleaseUserObject(
+    cudaGraph_t graph, cudaUserObject_t object, unsigned int count __dv(1)) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(cudaGraph_t, cudaUserObject_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGraphReleaseUserObject");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(graph, object, count);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetDriverEntryPoint(
+    const char *symbol, void **funcPtr, unsigned long long flags,
+    enum cudaDriverEntryPointQueryResult *driverStatus) {
+  using FuncPtr =
+      cudaError_t(CUDARTAPI *)(const char *, void **, unsigned long long,
+                               enum cudaDriverEntryPointQueryResult *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetDriverEntryPoint");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(symbol, funcPtr, flags, driverStatus);
+}
+
+extern __host__ cudaError_t CUDARTAPI cudaGetExportTable(
+    const void **ppExportTable, const cudaUUID_t *pExportTableId) {
+  using FuncPtr = cudaError_t(CUDARTAPI *)(const void **, const cudaUUID_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudaGetExportTable");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ppExportTable, pExportTableId);
+}
+
+}  // extern "C"
diff --git a/tensorflow/stream_executor/cuda/cuda_stub.cc b/tensorflow/stream_executor/cuda/cuda_stub.cc
index 3248c9ddefd..db9948d5a07 100644
--- a/tensorflow/stream_executor/cuda/cuda_stub.cc
+++ b/tensorflow/stream_executor/cuda/cuda_stub.cc
@@ -61,6 +61,7 @@ typedef struct CUDA_LAUNCH_PARAMS_st CUDA_LAUNCH_PARAMS;
 #define __CUDA_DEPRECATED
 #endif
 
+#if CUDA_VERSION < 12000
 #if CUDA_VERSION < 10000
 // Define fake enums introduced in CUDA 10.0.
 typedef enum CUgraphNodeType_enum {} CUgraphNodeType;
@@ -93,7 +94,9 @@ typedef struct CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS_st
     CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS;
 typedef void(CUDA_CB* CUhostFn)(void* userData);
 
+#endif
+
 // For now only one stub implementation is needed. If a function that is not
 // available in the given CUDA release, the corresponding wrapper returns
 // CUDA_ERROR_SHARED_OBJECT_INIT_FAILED.
-#include "tensorflow/stream_executor/cuda/cuda_10_0.inc"
+#include "tensorflow/stream_executor/cuda/cuda_12_0.inc"
diff --git a/tensorflow/stream_executor/cuda/cudart_stub.cc b/tensorflow/stream_executor/cuda/cudart_stub.cc
index 0c6b274f88b..d19eeea7dc9 100644
--- a/tensorflow/stream_executor/cuda/cudart_stub.cc
+++ b/tensorflow/stream_executor/cuda/cudart_stub.cc
@@ -56,7 +56,7 @@ cudaError_t GetSymbolNotFoundError() {
 #elif CUDART_VERSION < 10010
 #include "tensorflow/stream_executor/cuda/cuda_runtime_10_0.inc"
 #else
-#include "tensorflow/stream_executor/cuda/cuda_runtime_10_1.inc"
+#include "tensorflow/stream_executor/cuda/cuda_runtime_12_0.inc"
 #endif
 #undef __dv
 #undef __CUDA_DEPRECATED
diff --git a/tensorflow/stream_executor/cuda/cudnn_8_0.inc b/tensorflow/stream_executor/cuda/cudnn_8_0.inc
new file mode 100644
index 00000000000..d9bf35184e4
--- /dev/null
+++ b/tensorflow/stream_executor/cuda/cudnn_8_0.inc
@@ -0,0 +1,3213 @@
+// Auto-generated, do not edit.
+
+extern "C" {
+size_t CUDNNWINAPI cudnnGetVersion(void) {
+  using FuncPtr = size_t(CUDNNWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetVersion");
+  if (!func_ptr) return 0;
+  return func_ptr();
+}
+
+size_t CUDNNWINAPI cudnnGetCudartVersion(void) {
+  using FuncPtr = size_t(CUDNNWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetCudartVersion");
+  if (!func_ptr) return 0;
+  return func_ptr();
+}
+
+const char *CUDNNWINAPI cudnnGetErrorString(cudnnStatus_t status) {
+  using FuncPtr = const char *(CUDNNWINAPI *)(cudnnStatus_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetErrorString");
+  if (!func_ptr) return "cudnnGetErrorString symbol not found.";
+  return func_ptr(status);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnQueryRuntimeError(cudnnHandle_t handle,
+                                                 cudnnStatus_t *rstatus,
+                                                 cudnnErrQueryMode_t mode,
+                                                 cudnnRuntimeTag_t *tag) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnStatus_t *, cudnnErrQueryMode_t, cudnnRuntimeTag_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnQueryRuntimeError");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rstatus, mode, tag);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetProperty(libraryPropertyType type,
+                                           int *value) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(libraryPropertyType, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetProperty");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(type, value);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCreate(cudnnHandle_t *handle) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDestroy(cudnnHandle_t handle) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle);
+}
+
+
+#if CUDNN_MAJOR>=8 && (CUDNN_MINOR > 0 || CUDNN_PATCHLEVEL >= 4)
+cudnnStatus_t CUDNNWINAPI cudnnCnnInferVersionCheck(void) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCnnInferVersionCheck");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCnnTrainVersionCheck(void) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCnnTrainVersionCheck");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+#endif
+
+cudnnStatus_t CUDNNWINAPI cudnnSetStream(cudnnHandle_t handle,
+                                         cudaStream_t streamId) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetStream");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, streamId);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetStream(cudnnHandle_t handle,
+                                         cudaStream_t *streamId) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t, cudaStream_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetStream");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, streamId);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateTensorDescriptor(cudnnTensorDescriptor_t *tensorDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnTensorDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetTensor4dDescriptor(
+    cudnnTensorDescriptor_t tensorDesc, cudnnTensorFormat_t format,
+    cudnnDataType_t dataType, /* image data type */
+    int n,                    /* number of inputs (batch size) */
+    int c,                    /* number of input feature maps */
+    int h,                    /* height of input section */
+    int w) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnTensorDescriptor_t, cudnnTensorFormat_t,
+                                   cudnnDataType_t, int, int, int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetTensor4dDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorDesc, format, dataType, n, c, h, w);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetTensor4dDescriptorEx(
+    cudnnTensorDescriptor_t tensorDesc,
+    cudnnDataType_t dataType, /* image data type */
+    int n,                    /* number of inputs (batch size) */
+    int c,                    /* number of input feature maps */
+    int h,                    /* height of input section */
+    int w,                    /* width of input section */
+    int nStride, int cStride, int hStride, int wStride) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnTensorDescriptor_t, cudnnDataType_t,
+                                   int, int, int, int, int, int, int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetTensor4dDescriptorEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorDesc, dataType, n, c, h, w, nStride, cStride, hStride,
+                  wStride);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetTensor4dDescriptor(
+    const cudnnTensorDescriptor_t tensorDesc,
+    cudnnDataType_t *dataType, /* image data type */
+    int *n,                    /* number of inputs (batch size) */
+    int *c,                    /* number of input feature maps  */
+    int *h,                    /* height of input section */
+    int *w,                    /* width of input section */
+    int *nStride, int *cStride, int *hStride, int *wStride) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnTensorDescriptor_t, cudnnDataType_t *, int *, int *, int *,
+      int *, int *, int *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetTensor4dDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorDesc, dataType, n, c, h, w, nStride, cStride, hStride,
+                  wStride);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetTensorNdDescriptor(
+    cudnnTensorDescriptor_t tensorDesc, cudnnDataType_t dataType, int nbDims,
+    const int dimA[], const int strideA[]) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnTensorDescriptor_t, cudnnDataType_t, int, const int[], const int[]);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetTensorNdDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorDesc, dataType, nbDims, dimA, strideA);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetTensorNdDescriptorEx(
+    cudnnTensorDescriptor_t tensorDesc, cudnnTensorFormat_t format,
+    cudnnDataType_t dataType, int nbDims, const int dimA[]) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnTensorDescriptor_t, cudnnTensorFormat_t,
+                                   cudnnDataType_t, int, const int[]);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetTensorNdDescriptorEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorDesc, format, dataType, nbDims, dimA);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetTensorNdDescriptor(
+    const cudnnTensorDescriptor_t tensorDesc, int nbDimsRequested,
+    cudnnDataType_t *dataType, int *nbDims, int dimA[], int strideA[]) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(const cudnnTensorDescriptor_t, int,
+                                   cudnnDataType_t *, int *, int[], int[]);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetTensorNdDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorDesc, nbDimsRequested, dataType, nbDims, dimA, strideA);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetTensorSizeInBytes(
+    const cudnnTensorDescriptor_t tensorDesc, size_t *size) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(const cudnnTensorDescriptor_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetTensorSizeInBytes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorDesc, size);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyTensorDescriptor(cudnnTensorDescriptor_t tensorDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnTensorDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(tensorDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnInitTransformDest(
+    const cudnnTensorTransformDescriptor_t transformDesc,
+    const cudnnTensorDescriptor_t srcDesc, cudnnTensorDescriptor_t destDesc,
+    size_t *destSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnTensorTransformDescriptor_t, const cudnnTensorDescriptor_t,
+      cudnnTensorDescriptor_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnInitTransformDest");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(transformDesc, srcDesc, destDesc, destSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCreateTensorTransformDescriptor(
+    cudnnTensorTransformDescriptor_t *transformDesc) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnTensorTransformDescriptor_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnCreateTensorTransformDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(transformDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetTensorTransformDescriptor(
+    cudnnTensorTransformDescriptor_t transformDesc, const uint32_t nbDims,
+    const cudnnTensorFormat_t destFormat, const int32_t padBeforeA[],
+    const int32_t padAfterA[], const uint32_t foldA[],
+    const cudnnFoldingDirection_t direction) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnTensorTransformDescriptor_t, const uint32_t,
+      const cudnnTensorFormat_t, const int32_t[], const int32_t[],
+      const uint32_t[], const cudnnFoldingDirection_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnSetTensorTransformDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(transformDesc, nbDims, destFormat, padBeforeA, padAfterA,
+                  foldA, direction);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetTensorTransformDescriptor(
+    cudnnTensorTransformDescriptor_t transformDesc, uint32_t nbDimsRequested,
+    cudnnTensorFormat_t *destFormat, int32_t padBeforeA[], int32_t padAfterA[],
+    uint32_t foldA[], cudnnFoldingDirection_t *direction) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnTensorTransformDescriptor_t, uint32_t, cudnnTensorFormat_t *,
+      int32_t[], int32_t[], uint32_t[], cudnnFoldingDirection_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetTensorTransformDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(transformDesc, nbDimsRequested, destFormat, padBeforeA,
+                  padAfterA, foldA, direction);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDestroyTensorTransformDescriptor(
+    cudnnTensorTransformDescriptor_t transformDesc) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnTensorTransformDescriptor_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnDestroyTensorTransformDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(transformDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnTransformTensor(
+    cudnnHandle_t handle, const void *alpha,
+    const cudnnTensorDescriptor_t xDesc, const void *x, const void *beta,
+    const cudnnTensorDescriptor_t yDesc, void *y) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const void *, const cudnnTensorDescriptor_t, const void *,
+      const void *, const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnTransformTensor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, alpha, xDesc, x, beta, yDesc, y);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnTransformTensorEx(
+    cudnnHandle_t handle, const cudnnTensorTransformDescriptor_t transDesc,
+    const void *alpha, const cudnnTensorDescriptor_t srcDesc,
+    const void *srcData, const void *beta,
+    const cudnnTensorDescriptor_t destDesc, void *destData) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorTransformDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnTransformTensorEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transDesc, alpha, srcDesc, srcData, beta, destDesc,
+                  destData);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnAddTensor(cudnnHandle_t handle,
+                                         const void *alpha,
+                                         const cudnnTensorDescriptor_t aDesc,
+                                         const void *A, const void *beta,
+                                         const cudnnTensorDescriptor_t cDesc,
+                                         void *C) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const void *, const cudnnTensorDescriptor_t, const void *,
+      const void *, const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnAddTensor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, alpha, aDesc, A, beta, cDesc, C);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateOpTensorDescriptor(cudnnOpTensorDescriptor_t *opTensorDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnOpTensorDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateOpTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(opTensorDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetOpTensorDescriptor(
+    cudnnOpTensorDescriptor_t opTensorDesc, cudnnOpTensorOp_t opTensorOp,
+    cudnnDataType_t opTensorCompType, cudnnNanPropagation_t opTensorNanOpt) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnOpTensorDescriptor_t, cudnnOpTensorOp_t,
+                                   cudnnDataType_t, cudnnNanPropagation_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetOpTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(opTensorDesc, opTensorOp, opTensorCompType, opTensorNanOpt);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetOpTensorDescriptor(
+    const cudnnOpTensorDescriptor_t opTensorDesc, cudnnOpTensorOp_t *opTensorOp,
+    cudnnDataType_t *opTensorCompType, cudnnNanPropagation_t *opTensorNanOpt) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnOpTensorDescriptor_t, cudnnOpTensorOp_t *, cudnnDataType_t *,
+      cudnnNanPropagation_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetOpTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(opTensorDesc, opTensorOp, opTensorCompType, opTensorNanOpt);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyOpTensorDescriptor(cudnnOpTensorDescriptor_t opTensorDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnOpTensorDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyOpTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(opTensorDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnOpTensor(
+    cudnnHandle_t handle, const cudnnOpTensorDescriptor_t opTensorDesc,
+    const void *alpha1, const cudnnTensorDescriptor_t aDesc, const void *A,
+    const void *alpha2, const cudnnTensorDescriptor_t bDesc, const void *B,
+    const void *beta, const cudnnTensorDescriptor_t cDesc, void *C) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnOpTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnOpTensor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opTensorDesc, alpha1, aDesc, A, alpha2, bDesc, B,
+                  beta, cDesc, C);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCreateReduceTensorDescriptor(
+    cudnnReduceTensorDescriptor_t *reduceTensorDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnReduceTensorDescriptor_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnCreateReduceTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(reduceTensorDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetReduceTensorDescriptor(
+    cudnnReduceTensorDescriptor_t reduceTensorDesc,
+    cudnnReduceTensorOp_t reduceTensorOp, cudnnDataType_t reduceTensorCompType,
+    cudnnNanPropagation_t reduceTensorNanOpt,
+    cudnnReduceTensorIndices_t reduceTensorIndices,
+    cudnnIndicesType_t reduceTensorIndicesType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnReduceTensorDescriptor_t, cudnnReduceTensorOp_t, cudnnDataType_t,
+      cudnnNanPropagation_t, cudnnReduceTensorIndices_t, cudnnIndicesType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetReduceTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(reduceTensorDesc, reduceTensorOp, reduceTensorCompType,
+                  reduceTensorNanOpt, reduceTensorIndices,
+                  reduceTensorIndicesType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetReduceTensorDescriptor(
+    const cudnnReduceTensorDescriptor_t reduceTensorDesc,
+    cudnnReduceTensorOp_t *reduceTensorOp,
+    cudnnDataType_t *reduceTensorCompType,
+    cudnnNanPropagation_t *reduceTensorNanOpt,
+    cudnnReduceTensorIndices_t *reduceTensorIndices,
+    cudnnIndicesType_t *reduceTensorIndicesType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnReduceTensorDescriptor_t, cudnnReduceTensorOp_t *,
+      cudnnDataType_t *, cudnnNanPropagation_t *, cudnnReduceTensorIndices_t *,
+      cudnnIndicesType_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetReduceTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(reduceTensorDesc, reduceTensorOp, reduceTensorCompType,
+                  reduceTensorNanOpt, reduceTensorIndices,
+                  reduceTensorIndicesType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDestroyReduceTensorDescriptor(
+    cudnnReduceTensorDescriptor_t reduceTensorDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnReduceTensorDescriptor_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnDestroyReduceTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(reduceTensorDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetReductionIndicesSize(
+    cudnnHandle_t handle, const cudnnReduceTensorDescriptor_t reduceTensorDesc,
+    const cudnnTensorDescriptor_t aDesc, const cudnnTensorDescriptor_t cDesc,
+    size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnReduceTensorDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnTensorDescriptor_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetReductionIndicesSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, reduceTensorDesc, aDesc, cDesc, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetReductionWorkspaceSize(
+    cudnnHandle_t handle, const cudnnReduceTensorDescriptor_t reduceTensorDesc,
+    const cudnnTensorDescriptor_t aDesc, const cudnnTensorDescriptor_t cDesc,
+    size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnReduceTensorDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnTensorDescriptor_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetReductionWorkspaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, reduceTensorDesc, aDesc, cDesc, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnReduceTensor(
+    cudnnHandle_t handle, const cudnnReduceTensorDescriptor_t reduceTensorDesc,
+    void *indices, size_t indicesSizeInBytes, void *workspace,
+    size_t workspaceSizeInBytes, const void *alpha,
+    const cudnnTensorDescriptor_t aDesc, const void *A, const void *beta,
+    const cudnnTensorDescriptor_t cDesc, void *C) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnReduceTensorDescriptor_t, void *, size_t,
+      void *, size_t, const void *, const cudnnTensorDescriptor_t, const void *,
+      const void *, const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnReduceTensor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, reduceTensorDesc, indices, indicesSizeInBytes,
+                  workspace, workspaceSizeInBytes, alpha, aDesc, A, beta, cDesc,
+                  C);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetTensor(cudnnHandle_t handle,
+                                         const cudnnTensorDescriptor_t yDesc,
+                                         void *y, const void *valuePtr) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t, void *, const void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetTensor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, yDesc, y, valuePtr);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnScaleTensor(cudnnHandle_t handle,
+                                           const cudnnTensorDescriptor_t yDesc,
+                                           void *y, const void *alpha) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t, void *, const void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnScaleTensor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, yDesc, y, alpha);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateFilterDescriptor(cudnnFilterDescriptor_t *filterDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnFilterDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateFilterDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(filterDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetFilter4dDescriptor(
+    cudnnFilterDescriptor_t filterDesc,
+    cudnnDataType_t dataType,          /* image data type */
+    cudnnTensorFormat_t format, int k, /* number of output feature maps */
+    int c,                             /* number of input feature maps */
+    int h,                             /* height of each input filter */
+    int w) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnFilterDescriptor_t, cudnnDataType_t,
+                                   cudnnTensorFormat_t, int, int, int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetFilter4dDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(filterDesc, dataType, format, k, c, h, w);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetFilter4dDescriptor(
+    const cudnnFilterDescriptor_t filterDesc,
+    cudnnDataType_t *dataType,           /* image data type */
+    cudnnTensorFormat_t *format, int *k, /* number of output feature maps */
+    int *c,                              /* number of input feature maps */
+    int *h,                              /* height of each input filter */
+    int *w) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnFilterDescriptor_t, cudnnDataType_t *, cudnnTensorFormat_t *,
+      int *, int *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetFilter4dDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(filterDesc, dataType, format, k, c, h, w);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetFilterNdDescriptor(
+    cudnnFilterDescriptor_t filterDesc,
+    cudnnDataType_t dataType, /* image data type */
+    cudnnTensorFormat_t format, int nbDims, const int filterDimA[]) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnFilterDescriptor_t, cudnnDataType_t,
+                                   cudnnTensorFormat_t, int, const int[]);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetFilterNdDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(filterDesc, dataType, format, nbDims, filterDimA);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetFilterNdDescriptor(
+    const cudnnFilterDescriptor_t filterDesc, int nbDimsRequested,
+    cudnnDataType_t *dataType, /* image data type */
+    cudnnTensorFormat_t *format, int *nbDims, int filterDimA[]) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnFilterDescriptor_t, int, cudnnDataType_t *,
+      cudnnTensorFormat_t *, int *, int[]);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetFilterNdDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(filterDesc, nbDimsRequested, dataType, format, nbDims,
+                  filterDimA);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetFilterSizeInBytes(
+    const cudnnFilterDescriptor_t filterDesc, size_t *size) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(const cudnnFilterDescriptor_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetFilterSizeInBytes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(filterDesc, size);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnTransformFilter(
+    cudnnHandle_t handle, const cudnnTensorTransformDescriptor_t transDesc,
+    const void *alpha, const cudnnFilterDescriptor_t srcDesc,
+    const void *srcData, const void *beta,
+    const cudnnFilterDescriptor_t destDesc, void *destData) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorTransformDescriptor_t, const void *,
+      const cudnnFilterDescriptor_t, const void *, const void *,
+      const cudnnFilterDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnTransformFilter");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transDesc, alpha, srcDesc, srcData, beta, destDesc,
+                  destData);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyFilterDescriptor(cudnnFilterDescriptor_t filterDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnFilterDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyFilterDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(filterDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSoftmaxForward(
+    cudnnHandle_t handle, cudnnSoftmaxAlgorithm_t algo, cudnnSoftmaxMode_t mode,
+    const void *alpha, const cudnnTensorDescriptor_t xDesc, const void *x,
+    const void *beta, const cudnnTensorDescriptor_t yDesc, void *y) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnSoftmaxAlgorithm_t, cudnnSoftmaxMode_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSoftmaxForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, mode, alpha, xDesc, x, beta, yDesc, y);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreatePoolingDescriptor(cudnnPoolingDescriptor_t *poolingDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnPoolingDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreatePoolingDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(poolingDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetPooling2dDescriptor(
+    cudnnPoolingDescriptor_t poolingDesc, cudnnPoolingMode_t mode,
+    cudnnNanPropagation_t maxpoolingNanOpt, int windowHeight, int windowWidth,
+    int verticalPadding, int horizontalPadding, int verticalStride,
+    int horizontalStride) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnPoolingDescriptor_t, cudnnPoolingMode_t, cudnnNanPropagation_t, int,
+      int, int, int, int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetPooling2dDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(poolingDesc, mode, maxpoolingNanOpt, windowHeight,
+                  windowWidth, verticalPadding, horizontalPadding,
+                  verticalStride, horizontalStride);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetPooling2dDescriptor(
+    const cudnnPoolingDescriptor_t poolingDesc, cudnnPoolingMode_t *mode,
+    cudnnNanPropagation_t *maxpoolingNanOpt, int *windowHeight,
+    int *windowWidth, int *verticalPadding, int *horizontalPadding,
+    int *verticalStride, int *horizontalStride) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnPoolingDescriptor_t, cudnnPoolingMode_t *,
+      cudnnNanPropagation_t *, int *, int *, int *, int *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetPooling2dDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(poolingDesc, mode, maxpoolingNanOpt, windowHeight,
+                  windowWidth, verticalPadding, horizontalPadding,
+                  verticalStride, horizontalStride);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetPoolingNdDescriptor(
+    cudnnPoolingDescriptor_t poolingDesc, const cudnnPoolingMode_t mode,
+    const cudnnNanPropagation_t maxpoolingNanOpt, int nbDims,
+    const int windowDimA[], const int paddingA[], const int strideA[]) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnPoolingDescriptor_t, const cudnnPoolingMode_t,
+      const cudnnNanPropagation_t, int, const int[], const int[], const int[]);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetPoolingNdDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(poolingDesc, mode, maxpoolingNanOpt, nbDims, windowDimA,
+                  paddingA, strideA);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetPoolingNdDescriptor(
+    const cudnnPoolingDescriptor_t poolingDesc, int nbDimsRequested,
+    cudnnPoolingMode_t *mode, cudnnNanPropagation_t *maxpoolingNanOpt,
+    int *nbDims, int windowDimA[], int paddingA[], int strideA[]) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnPoolingDescriptor_t, int, cudnnPoolingMode_t *,
+      cudnnNanPropagation_t *, int *, int[], int[], int[]);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetPoolingNdDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(poolingDesc, nbDimsRequested, mode, maxpoolingNanOpt, nbDims,
+                  windowDimA, paddingA, strideA);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnGetPoolingNdForwardOutputDim(const cudnnPoolingDescriptor_t poolingDesc,
+                                  const cudnnTensorDescriptor_t inputTensorDesc,
+                                  int nbDims, int outputTensorDimA[]) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(const cudnnPoolingDescriptor_t,
+                                   const cudnnTensorDescriptor_t, int, int[]);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetPoolingNdForwardOutputDim");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(poolingDesc, inputTensorDesc, nbDims, outputTensorDimA);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnGetPooling2dForwardOutputDim(const cudnnPoolingDescriptor_t poolingDesc,
+                                  const cudnnTensorDescriptor_t inputTensorDesc,
+                                  int *n, int *c, int *h, int *w) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(const cudnnPoolingDescriptor_t,
+                                               const cudnnTensorDescriptor_t,
+                                               int *, int *, int *, int *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetPooling2dForwardOutputDim");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(poolingDesc, inputTensorDesc, n, c, h, w);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyPoolingDescriptor(cudnnPoolingDescriptor_t poolingDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnPoolingDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyPoolingDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(poolingDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnPoolingForward(
+    cudnnHandle_t handle, const cudnnPoolingDescriptor_t poolingDesc,
+    const void *alpha, const cudnnTensorDescriptor_t xDesc, const void *x,
+    const void *beta, const cudnnTensorDescriptor_t yDesc, void *y) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnPoolingDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnPoolingForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, poolingDesc, alpha, xDesc, x, beta, yDesc, y);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateActivationDescriptor(cudnnActivationDescriptor_t *activationDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnActivationDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateActivationDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(activationDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetActivationDescriptor(
+    cudnnActivationDescriptor_t activationDesc, cudnnActivationMode_t mode,
+    cudnnNanPropagation_t reluNanOpt, double coef) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnActivationDescriptor_t,
+                                               cudnnActivationMode_t,
+                                               cudnnNanPropagation_t, double);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetActivationDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(activationDesc, mode, reluNanOpt, coef);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnGetActivationDescriptor(const cudnnActivationDescriptor_t activationDesc,
+                             cudnnActivationMode_t *mode,
+                             cudnnNanPropagation_t *reluNanOpt, double *coef) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnActivationDescriptor_t, cudnnActivationMode_t *,
+      cudnnNanPropagation_t *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetActivationDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(activationDesc, mode, reluNanOpt, coef);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyActivationDescriptor(cudnnActivationDescriptor_t activationDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnActivationDescriptor_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnDestroyActivationDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(activationDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnActivationForward(
+    cudnnHandle_t handle, cudnnActivationDescriptor_t activationDesc,
+    const void *alpha, const cudnnTensorDescriptor_t xDesc, const void *x,
+    const void *beta, const cudnnTensorDescriptor_t yDesc, void *y) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnActivationDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnActivationForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, activationDesc, alpha, xDesc, x, beta, yDesc, y);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateLRNDescriptor(cudnnLRNDescriptor_t *normDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnLRNDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateLRNDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(normDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetLRNDescriptor(cudnnLRNDescriptor_t normDesc,
+                                                unsigned lrnN, double lrnAlpha,
+                                                double lrnBeta, double lrnK) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnLRNDescriptor_t, unsigned int, double, double, double);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetLRNDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(normDesc, lrnN, lrnAlpha, lrnBeta, lrnK);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetLRNDescriptor(cudnnLRNDescriptor_t normDesc,
+                                                unsigned *lrnN,
+                                                double *lrnAlpha,
+                                                double *lrnBeta, double *lrnK) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnLRNDescriptor_t, unsigned int *, double *, double *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetLRNDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(normDesc, lrnN, lrnAlpha, lrnBeta, lrnK);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyLRNDescriptor(cudnnLRNDescriptor_t lrnDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnLRNDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyLRNDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(lrnDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnLRNCrossChannelForward(
+    cudnnHandle_t handle, cudnnLRNDescriptor_t normDesc, cudnnLRNMode_t lrnMode,
+    const void *alpha, const cudnnTensorDescriptor_t xDesc, const void *x,
+    const void *beta, const cudnnTensorDescriptor_t yDesc, void *y) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnLRNDescriptor_t, cudnnLRNMode_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnLRNCrossChannelForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, normDesc, lrnMode, alpha, xDesc, x, beta, yDesc, y);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDivisiveNormalizationForward(
+    cudnnHandle_t handle, cudnnLRNDescriptor_t normDesc,
+    cudnnDivNormMode_t mode, const void *alpha,
+    const cudnnTensorDescriptor_t xDesc, /* same desc for means, temp, temp2 */
+    const void *x,
+    const void *means, /* if NULL, means are assumed to be zero */
+    void *temp, void *temp2, const void *beta,
+    const cudnnTensorDescriptor_t yDesc, void *y) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnLRNDescriptor_t, cudnnDivNormMode_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *, void *, void *,
+      const void *, const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnDivisiveNormalizationForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, normDesc, mode, alpha, xDesc, x, means, temp, temp2,
+                  beta, yDesc, y);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDeriveBNTensorDescriptor(
+    cudnnTensorDescriptor_t derivedBnDesc, const cudnnTensorDescriptor_t xDesc,
+    cudnnBatchNormMode_t mode) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnTensorDescriptor_t,
+                                               const cudnnTensorDescriptor_t,
+                                               cudnnBatchNormMode_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDeriveBNTensorDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(derivedBnDesc, xDesc, mode);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnBatchNormalizationForwardInference(
+    cudnnHandle_t handle, cudnnBatchNormMode_t mode,
+    const void *alpha, /* alpha[0] = result blend factor */
+    const void *beta,  /* beta[0] = dest layer blend factor */
+    const cudnnTensorDescriptor_t xDesc, const void *x, /* NxCxHxW */
+    const cudnnTensorDescriptor_t yDesc, void *y,       /* NxCxHxW */
+    const cudnnTensorDescriptor_t bnScaleBiasMeanVarDesc, const void *bnScale,
+    const void *bnBias, const void *estimatedMean,
+    const void *estimatedVariance, double epsilon) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnBatchNormMode_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, void *, const cudnnTensorDescriptor_t,
+      const void *, const void *, const void *, const void *, double);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnBatchNormalizationForwardInference");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode, alpha, beta, xDesc, x, yDesc, y,
+                  bnScaleBiasMeanVarDesc, bnScale, bnBias, estimatedMean,
+                  estimatedVariance, epsilon);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCreateSpatialTransformerDescriptor(
+    cudnnSpatialTransformerDescriptor_t *stDesc) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnSpatialTransformerDescriptor_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnCreateSpatialTransformerDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetSpatialTransformerNdDescriptor(
+    cudnnSpatialTransformerDescriptor_t stDesc, cudnnSamplerType_t samplerType,
+    cudnnDataType_t dataType, const int nbDims, const int dimA[]) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnSpatialTransformerDescriptor_t, cudnnSamplerType_t, cudnnDataType_t,
+      const int, const int[]);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnSetSpatialTransformerNdDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stDesc, samplerType, dataType, nbDims, dimA);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDestroySpatialTransformerDescriptor(
+    cudnnSpatialTransformerDescriptor_t stDesc) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnSpatialTransformerDescriptor_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnDestroySpatialTransformerDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(stDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSpatialTfGridGeneratorForward(
+    cudnnHandle_t handle, const cudnnSpatialTransformerDescriptor_t stDesc,
+    const void *theta, void *grid) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnSpatialTransformerDescriptor_t, const void *,
+      void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnSpatialTfGridGeneratorForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, stDesc, theta, grid);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSpatialTfSamplerForward(
+    cudnnHandle_t handle, cudnnSpatialTransformerDescriptor_t stDesc,
+    const void *alpha, const cudnnTensorDescriptor_t xDesc, const void *x,
+    const void *grid, const void *beta, cudnnTensorDescriptor_t yDesc,
+    void *y) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnSpatialTransformerDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *, const void *,
+      cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSpatialTfSamplerForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, stDesc, alpha, xDesc, x, grid, beta, yDesc, y);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateDropoutDescriptor(cudnnDropoutDescriptor_t *dropoutDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnDropoutDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateDropoutDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dropoutDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyDropoutDescriptor(cudnnDropoutDescriptor_t dropoutDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnDropoutDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyDropoutDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dropoutDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDropoutGetStatesSize(cudnnHandle_t handle,
+                                                    size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDropoutGetStatesSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDropoutGetReserveSpaceSize(
+    cudnnTensorDescriptor_t xdesc, size_t *sizeInBytes) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnTensorDescriptor_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDropoutGetReserveSpaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(xdesc, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetDropoutDescriptor(
+    cudnnDropoutDescriptor_t dropoutDesc, cudnnHandle_t handle, float dropout,
+    void *states, size_t stateSizeInBytes, unsigned long long seed) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnDropoutDescriptor_t, cudnnHandle_t,
+                                   float, void *, size_t, unsigned long long);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetDropoutDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dropoutDesc, handle, dropout, states, stateSizeInBytes, seed);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRestoreDropoutDescriptor(
+    cudnnDropoutDescriptor_t dropoutDesc, cudnnHandle_t handle, float dropout,
+    void *states, size_t stateSizeInBytes, unsigned long long seed) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnDropoutDescriptor_t, cudnnHandle_t,
+                                   float, void *, size_t, unsigned long long);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRestoreDropoutDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dropoutDesc, handle, dropout, states, stateSizeInBytes, seed);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetDropoutDescriptor(
+    cudnnDropoutDescriptor_t dropoutDesc, cudnnHandle_t handle, float *dropout,
+    void **states, unsigned long long *seed) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnDropoutDescriptor_t, cudnnHandle_t,
+                                   float *, void **, unsigned long long *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetDropoutDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dropoutDesc, handle, dropout, states, seed);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDropoutForward(
+    cudnnHandle_t handle, const cudnnDropoutDescriptor_t dropoutDesc,
+    const cudnnTensorDescriptor_t xdesc, const void *x,
+    const cudnnTensorDescriptor_t ydesc, void *y, void *reserveSpace,
+    size_t reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnDropoutDescriptor_t,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, void *, void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDropoutForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dropoutDesc, xdesc, x, ydesc, y, reserveSpace,
+                  reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCreateAlgorithmPerformance(
+    cudnnAlgorithmPerformance_t *algoPerf, int numberToCreate) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnAlgorithmPerformance_t *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateAlgorithmPerformance");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(algoPerf, numberToCreate);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDestroyAlgorithmPerformance(
+    cudnnAlgorithmPerformance_t *algoPerf, int numberToDestroy) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnAlgorithmPerformance_t *, int);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnDestroyAlgorithmPerformance");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(algoPerf, numberToDestroy);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetCallback(unsigned mask, void *udata,
+                                           cudnnCallback_t fptr) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(unsigned int, void *, cudnnCallback_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetCallback");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mask, udata, fptr);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetCallback(unsigned *mask, void **udata,
+                                           cudnnCallback_t *fptr) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(unsigned int *, void **, cudnnCallback_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetCallback");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mask, udata, fptr);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnOpsInferVersionCheck(void) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnOpsInferVersionCheck");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateConvolutionDescriptor(cudnnConvolutionDescriptor_t *convDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnConvolutionDescriptor_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnCreateConvolutionDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyConvolutionDescriptor(cudnnConvolutionDescriptor_t convDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnConvolutionDescriptor_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnDestroyConvolutionDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetConvolutionMathType(
+    cudnnConvolutionDescriptor_t convDesc, cudnnMathType_t mathType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnConvolutionDescriptor_t,
+                                               cudnnMathType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetConvolutionMathType");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, mathType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionMathType(
+    cudnnConvolutionDescriptor_t convDesc, cudnnMathType_t *mathType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnConvolutionDescriptor_t,
+                                               cudnnMathType_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetConvolutionMathType");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, mathType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetConvolutionGroupCount(
+    cudnnConvolutionDescriptor_t convDesc, int groupCount) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnConvolutionDescriptor_t, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetConvolutionGroupCount");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, groupCount);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionGroupCount(
+    cudnnConvolutionDescriptor_t convDesc, int *groupCount) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnConvolutionDescriptor_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetConvolutionGroupCount");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, groupCount);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetConvolutionReorderType(
+    cudnnConvolutionDescriptor_t convDesc, cudnnReorderType_t reorderType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnConvolutionDescriptor_t,
+                                               cudnnReorderType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetConvolutionReorderType");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, reorderType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionReorderType(
+    cudnnConvolutionDescriptor_t convDesc, cudnnReorderType_t *reorderType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnConvolutionDescriptor_t,
+                                               cudnnReorderType_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetConvolutionReorderType");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, reorderType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetConvolution2dDescriptor(
+    cudnnConvolutionDescriptor_t convDesc, int pad_h, /* zero-padding height */
+    int pad_w,                                        /* zero-padding width */
+    int u,          /* vertical filter stride */
+    int v,          /* horizontal filter stride */
+    int dilation_h, /* filter dilation in the vertical dimension */
+    int dilation_w, /* filter dilation in the horizontal dimension */
+    cudnnConvolutionMode_t mode, cudnnDataType_t computeType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnConvolutionDescriptor_t, int, int, int, int, int, int,
+      cudnnConvolutionMode_t, cudnnDataType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetConvolution2dDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, pad_h, pad_w, u, v, dilation_h, dilation_w, mode,
+                  computeType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolution2dDescriptor(
+    const cudnnConvolutionDescriptor_t convDesc,
+    int *pad_h,      /* zero-padding height */
+    int *pad_w,      /* zero-padding width */
+    int *u,          /* vertical filter stride */
+    int *v,          /* horizontal filter stride */
+    int *dilation_h, /* filter dilation in the vertical dimension */
+    int *dilation_w, /* filter dilation in the horizontal dimension */
+    cudnnConvolutionMode_t *mode, cudnnDataType_t *computeType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnConvolutionDescriptor_t, int *, int *, int *, int *, int *,
+      int *, cudnnConvolutionMode_t *, cudnnDataType_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetConvolution2dDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, pad_h, pad_w, u, v, dilation_h, dilation_w, mode,
+                  computeType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetConvolutionNdDescriptor(
+    cudnnConvolutionDescriptor_t convDesc, int arrayLength, /* nbDims-2 size */
+    const int padA[], const int filterStrideA[], const int dilationA[],
+    cudnnConvolutionMode_t mode, cudnnDataType_t computeType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnConvolutionDescriptor_t, int, const int[], const int[], const int[],
+      cudnnConvolutionMode_t, cudnnDataType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetConvolutionNdDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, arrayLength, padA, filterStrideA, dilationA, mode,
+                  computeType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionNdDescriptor(
+    const cudnnConvolutionDescriptor_t convDesc, int arrayLengthRequested,
+    int *arrayLength, int padA[], int strideA[], int dilationA[],
+    cudnnConvolutionMode_t *mode, cudnnDataType_t *computeType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnConvolutionDescriptor_t, int, int *, int[], int[], int[],
+      cudnnConvolutionMode_t *, cudnnDataType_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetConvolutionNdDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, arrayLengthRequested, arrayLength, padA, strideA,
+                  dilationA, mode, computeType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolution2dForwardOutputDim(
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t inputTensorDesc,
+    const cudnnFilterDescriptor_t filterDesc, int *n, int *c, int *h, int *w) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnConvolutionDescriptor_t, const cudnnTensorDescriptor_t,
+      const cudnnFilterDescriptor_t, int *, int *, int *, int *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolution2dForwardOutputDim");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, inputTensorDesc, filterDesc, n, c, h, w);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionNdForwardOutputDim(
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t inputTensorDesc,
+    const cudnnFilterDescriptor_t filterDesc, int nbDims,
+    int tensorOuputDimA[]) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnConvolutionDescriptor_t, const cudnnTensorDescriptor_t,
+      const cudnnFilterDescriptor_t, int, int[]);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolutionNdForwardOutputDim");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(convDesc, inputTensorDesc, filterDesc, nbDims,
+                  tensorOuputDimA);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnGetConvolutionForwardAlgorithmMaxCount(cudnnHandle_t handle, int *count) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t, int *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolutionForwardAlgorithmMaxCount");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, count);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionForwardAlgorithm_v7(
+    cudnnHandle_t handle, const cudnnTensorDescriptor_t srcDesc,
+    const cudnnFilterDescriptor_t filterDesc,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t destDesc, const int requestedAlgoCount,
+    int *returnedAlgoCount, cudnnConvolutionFwdAlgoPerf_t *perfResults) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t,
+      const cudnnFilterDescriptor_t, const cudnnConvolutionDescriptor_t,
+      const cudnnTensorDescriptor_t, const int, int *,
+      cudnnConvolutionFwdAlgoPerf_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolutionForwardAlgorithm_v7");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, srcDesc, filterDesc, convDesc, destDesc,
+                  requestedAlgoCount, returnedAlgoCount, perfResults);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnFindConvolutionForwardAlgorithm(
+    cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
+    const cudnnFilterDescriptor_t wDesc,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t yDesc, const int requestedAlgoCount,
+    int *returnedAlgoCount, cudnnConvolutionFwdAlgoPerf_t *perfResults) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t,
+      const cudnnFilterDescriptor_t, const cudnnConvolutionDescriptor_t,
+      const cudnnTensorDescriptor_t, const int, int *,
+      cudnnConvolutionFwdAlgoPerf_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnFindConvolutionForwardAlgorithm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, xDesc, wDesc, convDesc, yDesc, requestedAlgoCount,
+                  returnedAlgoCount, perfResults);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnFindConvolutionForwardAlgorithmEx(
+    cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc, const void *x,
+    const cudnnFilterDescriptor_t wDesc, const void *w,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t yDesc, void *y, const int requestedAlgoCount,
+    int *returnedAlgoCount, cudnnConvolutionFwdAlgoPerf_t *perfResults,
+    void *workSpace, size_t workSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t, const void *,
+      const cudnnFilterDescriptor_t, const void *,
+      const cudnnConvolutionDescriptor_t, const cudnnTensorDescriptor_t, void *,
+      const int, int *, cudnnConvolutionFwdAlgoPerf_t *, void *, size_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnFindConvolutionForwardAlgorithmEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, xDesc, x, wDesc, w, convDesc, yDesc, y,
+                  requestedAlgoCount, returnedAlgoCount, perfResults, workSpace,
+                  workSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnIm2Col(cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
+            const void *x, const cudnnFilterDescriptor_t wDesc,
+            const cudnnConvolutionDescriptor_t convDesc, void *colBuffer) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t, const cudnnTensorDescriptor_t,
+                                   const void *, const cudnnFilterDescriptor_t,
+                                   const cudnnConvolutionDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnIm2Col");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, xDesc, x, wDesc, convDesc, colBuffer);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnReorderFilterAndBias(
+    cudnnHandle_t handle, const cudnnFilterDescriptor_t filterDesc,
+    cudnnReorderType_t reorderType, const void *filterData,
+    void *reorderedFilterData, int reorderBias, const void *biasData,
+    void *reorderedBiasData) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnFilterDescriptor_t, cudnnReorderType_t,
+      const void *, void *, int, const void *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnReorderFilterAndBias");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, filterDesc, reorderType, filterData,
+                  reorderedFilterData, reorderBias, biasData,
+                  reorderedBiasData);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionForwardWorkspaceSize(
+    cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
+    const cudnnFilterDescriptor_t wDesc,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t yDesc, cudnnConvolutionFwdAlgo_t algo,
+    size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t,
+      const cudnnFilterDescriptor_t, const cudnnConvolutionDescriptor_t,
+      const cudnnTensorDescriptor_t, cudnnConvolutionFwdAlgo_t, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolutionForwardWorkspaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, xDesc, wDesc, convDesc, yDesc, algo, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnConvolutionForward(
+    cudnnHandle_t handle, const void *alpha,
+    const cudnnTensorDescriptor_t xDesc, const void *x,
+    const cudnnFilterDescriptor_t wDesc, const void *w,
+    const cudnnConvolutionDescriptor_t convDesc, cudnnConvolutionFwdAlgo_t algo,
+    void *workSpace, size_t workSpaceSizeInBytes, const void *beta,
+    const cudnnTensorDescriptor_t yDesc, void *y) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const void *, const cudnnTensorDescriptor_t, const void *,
+      const cudnnFilterDescriptor_t, const void *,
+      const cudnnConvolutionDescriptor_t, cudnnConvolutionFwdAlgo_t, void *,
+      size_t, const void *, const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnConvolutionForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, alpha, xDesc, x, wDesc, w, convDesc, algo, workSpace,
+                  workSpaceSizeInBytes, beta, yDesc, y);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnConvolutionBiasActivationForward(
+    cudnnHandle_t handle, const void *alpha1,
+    const cudnnTensorDescriptor_t xDesc, const void *x,
+    const cudnnFilterDescriptor_t wDesc, const void *w,
+    const cudnnConvolutionDescriptor_t convDesc, cudnnConvolutionFwdAlgo_t algo,
+    void *workSpace, size_t workSpaceSizeInBytes, const void *alpha2,
+    const cudnnTensorDescriptor_t zDesc, const void *z,
+    const cudnnTensorDescriptor_t biasDesc, const void *bias,
+    const cudnnActivationDescriptor_t activationDesc,
+    const cudnnTensorDescriptor_t yDesc, void *y) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const void *, const cudnnTensorDescriptor_t, const void *,
+      const cudnnFilterDescriptor_t, const void *,
+      const cudnnConvolutionDescriptor_t, cudnnConvolutionFwdAlgo_t, void *,
+      size_t, const void *, const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnActivationDescriptor_t, const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnConvolutionBiasActivationForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, alpha1, xDesc, x, wDesc, w, convDesc, algo, workSpace,
+                  workSpaceSizeInBytes, alpha2, zDesc, z, biasDesc, bias,
+                  activationDesc, yDesc, y);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnBackendCreateDescriptor(cudnnBackendDescriptorType_t descriptorType,
+                             cudnnBackendDescriptor_t *descriptor) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnBackendDescriptorType_t,
+                                               cudnnBackendDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnBackendCreateDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descriptorType, descriptor);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnBackendDestroyDescriptor(cudnnBackendDescriptor_t descriptor) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnBackendDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnBackendDestroyDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descriptor);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnBackendFinalize(cudnnBackendDescriptor_t descriptor) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnBackendDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnBackendFinalize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descriptor);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnBackendSetAttribute(cudnnBackendDescriptor_t descriptor,
+                         cudnnBackendAttributeName_t attributeName,
+                         cudnnBackendAttributeType_t attributeType,
+                         int64_t elementCount, const void *arrayOfElements) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnBackendDescriptor_t, cudnnBackendAttributeName_t,
+      cudnnBackendAttributeType_t, int64_t, const void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnBackendSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descriptor, attributeName, attributeType, elementCount,
+                  arrayOfElements);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnBackendGetAttribute(
+    cudnnBackendDescriptor_t const descriptor,
+    cudnnBackendAttributeName_t attributeName,
+    cudnnBackendAttributeType_t attributeType, int64_t requestedElementCount,
+    int64_t *elementCount, void *arrayOfElements) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnBackendDescriptor_t const, cudnnBackendAttributeName_t,
+      cudnnBackendAttributeType_t, int64_t, int64_t *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnBackendGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descriptor, attributeName, attributeType,
+                  requestedElementCount, elementCount, arrayOfElements);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnBackendExecute(
+    cudnnHandle_t handle, cudnnBackendDescriptor_t executionPlan,
+    cudnnBackendDescriptor_t variantPack) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnBackendDescriptor_t, cudnnBackendDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnBackendExecute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, executionPlan, variantPack);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionBackwardDataAlgorithmMaxCount(
+    cudnnHandle_t handle, int *count) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t, int *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolutionBackwardDataAlgorithmMaxCount");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, count);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnFindConvolutionBackwardDataAlgorithm(
+    cudnnHandle_t handle, const cudnnFilterDescriptor_t wDesc,
+    const cudnnTensorDescriptor_t dyDesc,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t dxDesc, const int requestedAlgoCount,
+    int *returnedAlgoCount, cudnnConvolutionBwdDataAlgoPerf_t *perfResults) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnFilterDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnConvolutionDescriptor_t,
+      const cudnnTensorDescriptor_t, const int, int *,
+      cudnnConvolutionBwdDataAlgoPerf_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnFindConvolutionBackwardDataAlgorithm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, wDesc, dyDesc, convDesc, dxDesc, requestedAlgoCount,
+                  returnedAlgoCount, perfResults);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnFindConvolutionBackwardDataAlgorithmEx(
+    cudnnHandle_t handle, const cudnnFilterDescriptor_t wDesc, const void *w,
+    const cudnnTensorDescriptor_t dyDesc, const void *dy,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t dxDesc, void *dx,
+    const int requestedAlgoCount, int *returnedAlgoCount,
+    cudnnConvolutionBwdDataAlgoPerf_t *perfResults, void *workSpace,
+    size_t workSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnFilterDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnConvolutionDescriptor_t, const cudnnTensorDescriptor_t, void *,
+      const int, int *, cudnnConvolutionBwdDataAlgoPerf_t *, void *, size_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnFindConvolutionBackwardDataAlgorithmEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, wDesc, w, dyDesc, dy, convDesc, dxDesc, dx,
+                  requestedAlgoCount, returnedAlgoCount, perfResults, workSpace,
+                  workSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionBackwardDataAlgorithm_v7(
+    cudnnHandle_t handle, const cudnnFilterDescriptor_t filterDesc,
+    const cudnnTensorDescriptor_t diffDesc,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t gradDesc, const int requestedAlgoCount,
+    int *returnedAlgoCount, cudnnConvolutionBwdDataAlgoPerf_t *perfResults) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnFilterDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnConvolutionDescriptor_t,
+      const cudnnTensorDescriptor_t, const int, int *,
+      cudnnConvolutionBwdDataAlgoPerf_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolutionBackwardDataAlgorithm_v7");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, filterDesc, diffDesc, convDesc, gradDesc,
+                  requestedAlgoCount, returnedAlgoCount, perfResults);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionBackwardDataWorkspaceSize(
+    cudnnHandle_t handle, const cudnnFilterDescriptor_t wDesc,
+    const cudnnTensorDescriptor_t dyDesc,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t dxDesc, cudnnConvolutionBwdDataAlgo_t algo,
+    size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnFilterDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnConvolutionDescriptor_t,
+      const cudnnTensorDescriptor_t, cudnnConvolutionBwdDataAlgo_t, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolutionBackwardDataWorkspaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, wDesc, dyDesc, convDesc, dxDesc, algo, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnConvolutionBackwardData(
+    cudnnHandle_t handle, const void *alpha,
+    const cudnnFilterDescriptor_t wDesc, const void *w,
+    const cudnnTensorDescriptor_t dyDesc, const void *dy,
+    const cudnnConvolutionDescriptor_t convDesc,
+    cudnnConvolutionBwdDataAlgo_t algo, void *workSpace,
+    size_t workSpaceSizeInBytes, const void *beta,
+    const cudnnTensorDescriptor_t dxDesc, void *dx) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const void *, const cudnnFilterDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnConvolutionDescriptor_t, cudnnConvolutionBwdDataAlgo_t, void *,
+      size_t, const void *, const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnConvolutionBackwardData");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, alpha, wDesc, w, dyDesc, dy, convDesc, algo,
+                  workSpace, workSpaceSizeInBytes, beta, dxDesc, dx);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetFoldedConvBackwardDataDescriptors(
+    const cudnnHandle_t handle, const cudnnFilterDescriptor_t filterDesc,
+    const cudnnTensorDescriptor_t diffDesc,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnTensorDescriptor_t gradDesc,
+    const cudnnTensorFormat_t transformFormat,
+    cudnnFilterDescriptor_t foldedFilterDesc,
+    cudnnTensorDescriptor_t paddedDiffDesc,
+    cudnnConvolutionDescriptor_t foldedConvDesc,
+    cudnnTensorDescriptor_t foldedGradDesc,
+    cudnnTensorTransformDescriptor_t filterFoldTransDesc,
+    cudnnTensorTransformDescriptor_t diffPadTransDesc,
+    cudnnTensorTransformDescriptor_t gradFoldTransDesc,
+    cudnnTensorTransformDescriptor_t gradUnfoldTransDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnHandle_t, const cudnnFilterDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnConvolutionDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnTensorFormat_t,
+      cudnnFilterDescriptor_t, cudnnTensorDescriptor_t,
+      cudnnConvolutionDescriptor_t, cudnnTensorDescriptor_t,
+      cudnnTensorTransformDescriptor_t, cudnnTensorTransformDescriptor_t,
+      cudnnTensorTransformDescriptor_t, cudnnTensorTransformDescriptor_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetFoldedConvBackwardDataDescriptors");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, filterDesc, diffDesc, convDesc, gradDesc,
+                  transformFormat, foldedFilterDesc, paddedDiffDesc,
+                  foldedConvDesc, foldedGradDesc, filterFoldTransDesc,
+                  diffPadTransDesc, gradFoldTransDesc, gradUnfoldTransDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCreateFusedOpsConstParamPack(
+    cudnnFusedOpsConstParamPack_t *constPack, cudnnFusedOps_t ops) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnFusedOpsConstParamPack_t *,
+                                               cudnnFusedOps_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnCreateFusedOpsConstParamPack");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(constPack, ops);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyFusedOpsConstParamPack(cudnnFusedOpsConstParamPack_t constPack) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnFusedOpsConstParamPack_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnDestroyFusedOpsConstParamPack");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(constPack);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetFusedOpsConstParamPackAttribute(
+    cudnnFusedOpsConstParamPack_t constPack,
+    cudnnFusedOpsConstParamLabel_t paramLabel, const void *param) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnFusedOpsConstParamPack_t,
+                                               cudnnFusedOpsConstParamLabel_t,
+                                               const void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnSetFusedOpsConstParamPackAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(constPack, paramLabel, param);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetFusedOpsConstParamPackAttribute(
+    const cudnnFusedOpsConstParamPack_t constPack,
+    cudnnFusedOpsConstParamLabel_t paramLabel, void *param, int *isNULL) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnFusedOpsConstParamPack_t, cudnnFusedOpsConstParamLabel_t,
+      void *, int *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetFusedOpsConstParamPackAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(constPack, paramLabel, param, isNULL);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCreateFusedOpsVariantParamPack(
+    cudnnFusedOpsVariantParamPack_t *varPack, cudnnFusedOps_t ops) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnFusedOpsVariantParamPack_t *, cudnnFusedOps_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnCreateFusedOpsVariantParamPack");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(varPack, ops);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyFusedOpsVariantParamPack(cudnnFusedOpsVariantParamPack_t varPack) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnFusedOpsVariantParamPack_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnDestroyFusedOpsVariantParamPack");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(varPack);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetFusedOpsVariantParamPackAttribute(
+    cudnnFusedOpsVariantParamPack_t varPack,
+    cudnnFusedOpsVariantParamLabel_t paramLabel, void *ptr) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnFusedOpsVariantParamPack_t,
+                                   cudnnFusedOpsVariantParamLabel_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnSetFusedOpsVariantParamPackAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(varPack, paramLabel, ptr);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetFusedOpsVariantParamPackAttribute(
+    const cudnnFusedOpsVariantParamPack_t varPack,
+    cudnnFusedOpsVariantParamLabel_t paramLabel, void *ptr) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(const cudnnFusedOpsVariantParamPack_t,
+                                   cudnnFusedOpsVariantParamLabel_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetFusedOpsVariantParamPackAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(varPack, paramLabel, ptr);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCreateFusedOpsPlan(cudnnFusedOpsPlan_t *plan,
+                                                  cudnnFusedOps_t ops) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnFusedOpsPlan_t *, cudnnFusedOps_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateFusedOpsPlan");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(plan, ops);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDestroyFusedOpsPlan(cudnnFusedOpsPlan_t plan) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnFusedOpsPlan_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyFusedOpsPlan");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(plan);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnMakeFusedOpsPlan(cudnnHandle_t handle, cudnnFusedOpsPlan_t plan,
+                      const cudnnFusedOpsConstParamPack_t constPack,
+                      size_t *workspaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnFusedOpsPlan_t, const cudnnFusedOpsConstParamPack_t,
+      size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnMakeFusedOpsPlan");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, plan, constPack, workspaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnFusedOpsExecute(cudnnHandle_t handle, const cudnnFusedOpsPlan_t plan,
+                     cudnnFusedOpsVariantParamPack_t varPack) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t, const cudnnFusedOpsPlan_t,
+                                   cudnnFusedOpsVariantParamPack_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnFusedOpsExecute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, plan, varPack);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateRNNDescriptor(cudnnRNNDescriptor_t *rnnDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateRNNDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyRNNDescriptor(cudnnRNNDescriptor_t rnnDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyRNNDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetRNNDescriptor_v8(
+    cudnnRNNDescriptor_t rnnDesc, cudnnRNNAlgo_t algo, cudnnRNNMode_t cellMode,
+    cudnnRNNBiasMode_t biasMode, cudnnDirectionMode_t dirMode,
+    cudnnRNNInputMode_t inputMode, cudnnDataType_t dataType,
+    cudnnDataType_t mathPrec, cudnnMathType_t mathType, int32_t inputSize,
+    int32_t hiddenSize, int32_t projSize, int32_t numLayers,
+    cudnnDropoutDescriptor_t dropoutDesc, uint32_t auxFlags) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnRNNDescriptor_t, cudnnRNNAlgo_t, cudnnRNNMode_t, cudnnRNNBiasMode_t,
+      cudnnDirectionMode_t, cudnnRNNInputMode_t, cudnnDataType_t,
+      cudnnDataType_t, cudnnMathType_t, int32_t, int32_t, int32_t, int32_t,
+      cudnnDropoutDescriptor_t, uint32_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetRNNDescriptor_v8");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc, algo, cellMode, biasMode, dirMode, inputMode,
+                  dataType, mathPrec, mathType, inputSize, hiddenSize, projSize,
+                  numLayers, dropoutDesc, auxFlags);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNDescriptor_v8(
+    cudnnRNNDescriptor_t rnnDesc, cudnnRNNAlgo_t *algo,
+    cudnnRNNMode_t *cellMode, cudnnRNNBiasMode_t *biasMode,
+    cudnnDirectionMode_t *dirMode, cudnnRNNInputMode_t *inputMode,
+    cudnnDataType_t *dataType, cudnnDataType_t *mathPrec,
+    cudnnMathType_t *mathType, int32_t *inputSize, int32_t *hiddenSize,
+    int32_t *projSize, int32_t *numLayers,
+    cudnnDropoutDescriptor_t *dropoutDesc, uint32_t *auxFlags) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnRNNDescriptor_t, cudnnRNNAlgo_t *, cudnnRNNMode_t *,
+      cudnnRNNBiasMode_t *, cudnnDirectionMode_t *, cudnnRNNInputMode_t *,
+      cudnnDataType_t *, cudnnDataType_t *, cudnnMathType_t *, int32_t *,
+      int32_t *, int32_t *, int32_t *, cudnnDropoutDescriptor_t *, uint32_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNDescriptor_v8");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc, algo, cellMode, biasMode, dirMode, inputMode,
+                  dataType, mathPrec, mathType, inputSize, hiddenSize, projSize,
+                  numLayers, dropoutDesc, auxFlags);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetRNNDescriptor_v6(
+    cudnnHandle_t handle, cudnnRNNDescriptor_t rnnDesc, const int hiddenSize,
+    const int numLayers, cudnnDropoutDescriptor_t dropoutDesc,
+    cudnnRNNInputMode_t inputMode, cudnnDirectionMode_t direction,
+    cudnnRNNMode_t cellMode, cudnnRNNAlgo_t algo, cudnnDataType_t mathPrec) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnRNNDescriptor_t, const int, const int,
+      cudnnDropoutDescriptor_t, cudnnRNNInputMode_t, cudnnDirectionMode_t,
+      cudnnRNNMode_t, cudnnRNNAlgo_t, cudnnDataType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetRNNDescriptor_v6");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, hiddenSize, numLayers, dropoutDesc,
+                  inputMode, direction, cellMode, algo, mathPrec);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNDescriptor_v6(
+    cudnnHandle_t handle, cudnnRNNDescriptor_t rnnDesc, int *hiddenSize,
+    int *numLayers, cudnnDropoutDescriptor_t *dropoutDesc,
+    cudnnRNNInputMode_t *inputMode, cudnnDirectionMode_t *direction,
+    cudnnRNNMode_t *cellMode, cudnnRNNAlgo_t *algo, cudnnDataType_t *mathPrec) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnRNNDescriptor_t, int *, int *,
+      cudnnDropoutDescriptor_t *, cudnnRNNInputMode_t *, cudnnDirectionMode_t *,
+      cudnnRNNMode_t *, cudnnRNNAlgo_t *, cudnnDataType_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNDescriptor_v6");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, hiddenSize, numLayers, dropoutDesc,
+                  inputMode, direction, cellMode, algo, mathPrec);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnSetRNNMatrixMathType(cudnnRNNDescriptor_t rnnDesc, cudnnMathType_t mType) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDescriptor_t, cudnnMathType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetRNNMatrixMathType");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc, mType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNMatrixMathType(
+    cudnnRNNDescriptor_t rnnDesc, cudnnMathType_t *mType) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDescriptor_t, cudnnMathType_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNMatrixMathType");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc, mType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetRNNBiasMode(cudnnRNNDescriptor_t rnnDesc,
+                                              cudnnRNNBiasMode_t biasMode) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDescriptor_t, cudnnRNNBiasMode_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetRNNBiasMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc, biasMode);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNBiasMode(cudnnRNNDescriptor_t rnnDesc,
+                                              cudnnRNNBiasMode_t *biasMode) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDescriptor_t, cudnnRNNBiasMode_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNBiasMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc, biasMode);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNSetClip(cudnnHandle_t handle,
+                                          cudnnRNNDescriptor_t rnnDesc,
+                                          cudnnRNNClipMode_t clipMode,
+                                          cudnnNanPropagation_t clipNanOpt,
+                                          double lclip, double rclip) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnRNNDescriptor_t, cudnnRNNClipMode_t,
+      cudnnNanPropagation_t, double, double);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNSetClip");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, clipMode, clipNanOpt, lclip, rclip);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNGetClip(cudnnHandle_t handle,
+                                          cudnnRNNDescriptor_t rnnDesc,
+                                          cudnnRNNClipMode_t *clipMode,
+                                          cudnnNanPropagation_t *clipNanOpt,
+                                          double *lclip, double *rclip) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnRNNDescriptor_t, cudnnRNNClipMode_t *,
+      cudnnNanPropagation_t *, double *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNGetClip");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, clipMode, clipNanOpt, lclip, rclip);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnSetRNNProjectionLayers(cudnnHandle_t handle, cudnnRNNDescriptor_t rnnDesc,
+                            const int recProjSize, const int outProjSize) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnRNNDescriptor_t, const int, const int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetRNNProjectionLayers");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, recProjSize, outProjSize);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNProjectionLayers(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc, int *recProjSize,
+    int *outProjSize) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNProjectionLayers");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, recProjSize, outProjSize);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCreatePersistentRNNPlan(
+    cudnnRNNDescriptor_t rnnDesc, const int minibatch,
+    const cudnnDataType_t dataType, cudnnPersistentRNNPlan_t *plan) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDescriptor_t, const int,
+                                               const cudnnDataType_t,
+                                               cudnnPersistentRNNPlan_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreatePersistentRNNPlan");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc, minibatch, dataType, plan);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyPersistentRNNPlan(cudnnPersistentRNNPlan_t plan) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnPersistentRNNPlan_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyPersistentRNNPlan");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(plan);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetPersistentRNNPlan(
+    cudnnRNNDescriptor_t rnnDesc, cudnnPersistentRNNPlan_t plan) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDescriptor_t,
+                                               cudnnPersistentRNNPlan_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetPersistentRNNPlan");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc, plan);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnGetRNNWeightSpaceSize(cudnnHandle_t handle, cudnnRNNDescriptor_t rnnDesc,
+                           size_t *weightSpaceSize) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t,
+                                               cudnnRNNDescriptor_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNWeightSpaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, weightSpaceSize);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNWorkspaceSize(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const int seqLength, const cudnnTensorDescriptor_t *xDesc,
+    size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const int,
+      const cudnnTensorDescriptor_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNWorkspaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, seqLength, xDesc, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNTempSpaceSizes(
+    cudnnHandle_t handle, cudnnRNNDescriptor_t rnnDesc,
+    cudnnForwardMode_t fMode, cudnnRNNDataDescriptor_t xDesc,
+    size_t *workSpaceSize, size_t *reserveSpaceSize) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnRNNDescriptor_t, cudnnForwardMode_t,
+      cudnnRNNDataDescriptor_t, size_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNTempSpaceSizes");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, fMode, xDesc, workSpaceSize,
+                  reserveSpaceSize);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnGetRNNParamsSize(cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+                      const cudnnTensorDescriptor_t xDesc, size_t *sizeInBytes,
+                      cudnnDataType_t dataType) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const cudnnTensorDescriptor_t,
+      size_t *, cudnnDataType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNParamsSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, xDesc, sizeInBytes, dataType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNLinLayerMatrixParams(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const int pseudoLayer, const cudnnTensorDescriptor_t xDesc,
+    const cudnnFilterDescriptor_t wDesc, const void *w, const int linLayerID,
+    cudnnFilterDescriptor_t linLayerMatDesc, void **linLayerMat) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const int,
+      const cudnnTensorDescriptor_t, const cudnnFilterDescriptor_t,
+      const void *, const int, cudnnFilterDescriptor_t, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNLinLayerMatrixParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, pseudoLayer, xDesc, wDesc, w, linLayerID,
+                  linLayerMatDesc, linLayerMat);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNLinLayerBiasParams(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const int pseudoLayer, const cudnnTensorDescriptor_t xDesc,
+    const cudnnFilterDescriptor_t wDesc, const void *w, const int linLayerID,
+    cudnnFilterDescriptor_t linLayerBiasDesc, void **linLayerBias) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const int,
+      const cudnnTensorDescriptor_t, const cudnnFilterDescriptor_t,
+      const void *, const int, cudnnFilterDescriptor_t, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNLinLayerBiasParams");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, pseudoLayer, xDesc, wDesc, w, linLayerID,
+                  linLayerBiasDesc, linLayerBias);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNForwardInference(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const int seqLength, const cudnnTensorDescriptor_t *xDesc, const void *x,
+    const cudnnTensorDescriptor_t hxDesc, const void *hx,
+    const cudnnTensorDescriptor_t cxDesc, const void *cx,
+    const cudnnFilterDescriptor_t wDesc, const void *w,
+    const cudnnTensorDescriptor_t *yDesc, void *y,
+    const cudnnTensorDescriptor_t hyDesc, void *hy,
+    const cudnnTensorDescriptor_t cyDesc, void *cy, void *workspace,
+    size_t workSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const int,
+      const cudnnTensorDescriptor_t *, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnFilterDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t *, void *, const cudnnTensorDescriptor_t,
+      void *, const cudnnTensorDescriptor_t, void *, void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNForwardInference");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, seqLength, xDesc, x, hxDesc, hx, cxDesc, cx,
+                  wDesc, w, yDesc, y, hyDesc, hy, cyDesc, cy, workspace,
+                  workSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetRNNPaddingMode(cudnnRNNDescriptor_t rnnDesc,
+                                                 unsigned paddingMode) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDescriptor_t, unsigned int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetRNNPaddingMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc, paddingMode);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNPaddingMode(cudnnRNNDescriptor_t rnnDesc,
+                                                 unsigned *paddingMode) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDescriptor_t, unsigned int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNPaddingMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDesc, paddingMode);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateRNNDataDescriptor(cudnnRNNDataDescriptor_t *rnnDataDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDataDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateRNNDataDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDataDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyRNNDataDescriptor(cudnnRNNDataDescriptor_t rnnDataDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnRNNDataDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyRNNDataDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDataDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetRNNDataDescriptor(
+    cudnnRNNDataDescriptor_t rnnDataDesc, cudnnDataType_t dataType,
+    cudnnRNNDataLayout_t layout, int maxSeqLength, int batchSize,
+    int vectorSize,
+    const int seqLengthArray[], /* length of each sequence in the batch */
+    void *paddingFill) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnRNNDataDescriptor_t, cudnnDataType_t, cudnnRNNDataLayout_t, int, int,
+      int, const int[], void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetRNNDataDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDataDesc, dataType, layout, maxSeqLength, batchSize,
+                  vectorSize, seqLengthArray, paddingFill);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNDataDescriptor(
+    cudnnRNNDataDescriptor_t rnnDataDesc, cudnnDataType_t *dataType,
+    cudnnRNNDataLayout_t *layout, int *maxSeqLength, int *batchSize,
+    int *vectorSize, int arrayLengthRequested, int seqLengthArray[],
+    void *paddingFill) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnRNNDataDescriptor_t, cudnnDataType_t *, cudnnRNNDataLayout_t *,
+      int *, int *, int *, int, int[], void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNDataDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(rnnDataDesc, dataType, layout, maxSeqLength, batchSize,
+                  vectorSize, arrayLengthRequested, seqLengthArray,
+                  paddingFill);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNForwardInferenceEx(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const cudnnRNNDataDescriptor_t xDesc, const void *x,
+    const cudnnTensorDescriptor_t hxDesc, const void *hx,
+    const cudnnTensorDescriptor_t cxDesc, const void *cx,
+    const cudnnFilterDescriptor_t wDesc, const void *w,
+    const cudnnRNNDataDescriptor_t yDesc, void *y,
+    const cudnnTensorDescriptor_t hyDesc, void *hy,
+    const cudnnTensorDescriptor_t cyDesc, void *cy,
+    const cudnnRNNDataDescriptor_t kDesc, /* reserved, should pass NULL */
+    const void *keys,                     /* reserved, should pass NULL */
+    const cudnnRNNDataDescriptor_t cDesc, /* reserved, should pass NULL */
+    void *cAttn,                          /* reserved, should pass NULL */
+    const cudnnRNNDataDescriptor_t iDesc, /* reserved, should pass NULL */
+    void *iAttn,                          /* reserved, should pass NULL */
+    const cudnnRNNDataDescriptor_t qDesc, /* reserved, should pass NULL */
+    void *queries,                        /* reserved, should pass NULL */
+    void *workSpace, size_t workSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const cudnnRNNDataDescriptor_t,
+      const void *, const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnFilterDescriptor_t, const void *,
+      const cudnnRNNDataDescriptor_t, void *, const cudnnTensorDescriptor_t,
+      void *, const cudnnTensorDescriptor_t, void *,
+      const cudnnRNNDataDescriptor_t, const void *,
+      const cudnnRNNDataDescriptor_t, void *, const cudnnRNNDataDescriptor_t,
+      void *, const cudnnRNNDataDescriptor_t, void *, void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNForwardInferenceEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, xDesc, x, hxDesc, hx, cxDesc, cx, wDesc, w,
+                  yDesc, y, hyDesc, hy, cyDesc, cy, kDesc, keys, cDesc, cAttn,
+                  iDesc, iAttn, qDesc, queries, workSpace,
+                  workSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateSeqDataDescriptor(cudnnSeqDataDescriptor_t *seqDataDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnSeqDataDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateSeqDataDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(seqDataDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroySeqDataDescriptor(cudnnSeqDataDescriptor_t seqDataDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnSeqDataDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroySeqDataDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(seqDataDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetSeqDataDescriptor(
+    cudnnSeqDataDescriptor_t seqDataDesc, cudnnDataType_t dataType, int nbDims,
+    const int dimA[], const cudnnSeqDataAxis_t axes[],
+    size_t seqLengthArraySize, const int seqLengthArray[], void *paddingFill) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnSeqDataDescriptor_t, cudnnDataType_t, int, const int[],
+      const cudnnSeqDataAxis_t[], size_t, const int[], void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetSeqDataDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(seqDataDesc, dataType, nbDims, dimA, axes, seqLengthArraySize,
+                  seqLengthArray, paddingFill);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetSeqDataDescriptor(
+    const cudnnSeqDataDescriptor_t seqDataDesc, cudnnDataType_t *dataType,
+    int *nbDims, int nbDimsRequested, int dimA[], cudnnSeqDataAxis_t axes[],
+    size_t *seqLengthArraySize, size_t seqLengthSizeRequested,
+    int seqLengthArray[], void *paddingFill) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      const cudnnSeqDataDescriptor_t, cudnnDataType_t *, int *, int, int[],
+      cudnnSeqDataAxis_t[], size_t *, size_t, int[], void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetSeqDataDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(seqDataDesc, dataType, nbDims, nbDimsRequested, dimA, axes,
+                  seqLengthArraySize, seqLengthSizeRequested, seqLengthArray,
+                  paddingFill);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateAttnDescriptor(cudnnAttnDescriptor_t *attnDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnAttnDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateAttnDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(attnDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyAttnDescriptor(cudnnAttnDescriptor_t attnDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnAttnDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyAttnDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(attnDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetAttnDescriptor(
+    cudnnAttnDescriptor_t attnDesc, unsigned attnMode, int nHeads,
+    double smScaler, cudnnDataType_t dataType, cudnnDataType_t computePrec,
+    cudnnMathType_t mathType, cudnnDropoutDescriptor_t attnDropoutDesc,
+    cudnnDropoutDescriptor_t postDropoutDesc, int qSize, int kSize, int vSize,
+    int qProjSize, int kProjSize, int vProjSize, int oProjSize,
+    int qoMaxSeqLength, int kvMaxSeqLength, int maxBatchSize, int maxBeamSize) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnAttnDescriptor_t, unsigned int, int, double, cudnnDataType_t,
+      cudnnDataType_t, cudnnMathType_t, cudnnDropoutDescriptor_t,
+      cudnnDropoutDescriptor_t, int, int, int, int, int, int, int, int, int,
+      int, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetAttnDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(attnDesc, attnMode, nHeads, smScaler, dataType, computePrec,
+                  mathType, attnDropoutDesc, postDropoutDesc, qSize, kSize,
+                  vSize, qProjSize, kProjSize, vProjSize, oProjSize,
+                  qoMaxSeqLength, kvMaxSeqLength, maxBatchSize, maxBeamSize);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetAttnDescriptor(
+    cudnnAttnDescriptor_t attnDesc, unsigned *attnMode, int *nHeads,
+    double *smScaler, cudnnDataType_t *dataType, cudnnDataType_t *computePrec,
+    cudnnMathType_t *mathType, cudnnDropoutDescriptor_t *attnDropoutDesc,
+    cudnnDropoutDescriptor_t *postDropoutDesc, int *qSize, int *kSize,
+    int *vSize, int *qProjSize, int *kProjSize, int *vProjSize, int *oProjSize,
+    int *qoMaxSeqLength, int *kvMaxSeqLength, int *maxBatchSize,
+    int *maxBeamSize) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnAttnDescriptor_t, unsigned int *, int *, double *, cudnnDataType_t *,
+      cudnnDataType_t *, cudnnMathType_t *, cudnnDropoutDescriptor_t *,
+      cudnnDropoutDescriptor_t *, int *, int *, int *, int *, int *, int *,
+      int *, int *, int *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetAttnDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(attnDesc, attnMode, nHeads, smScaler, dataType, computePrec,
+                  mathType, attnDropoutDesc, postDropoutDesc, qSize, kSize,
+                  vSize, qProjSize, kProjSize, vProjSize, oProjSize,
+                  qoMaxSeqLength, kvMaxSeqLength, maxBatchSize, maxBeamSize);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetMultiHeadAttnBuffers(
+    cudnnHandle_t handle, const cudnnAttnDescriptor_t attnDesc,
+    size_t *weightSizeInBytes, size_t *workSpaceSizeInBytes,
+    size_t *reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnAttnDescriptor_t, size_t *, size_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetMultiHeadAttnBuffers");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, attnDesc, weightSizeInBytes, workSpaceSizeInBytes,
+                  reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetMultiHeadAttnWeights(
+    cudnnHandle_t handle, const cudnnAttnDescriptor_t attnDesc,
+    cudnnMultiHeadAttnWeightKind_t wKind, size_t weightSizeInBytes,
+    const void *weights, cudnnTensorDescriptor_t wDesc, void **wAddr) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnAttnDescriptor_t,
+      cudnnMultiHeadAttnWeightKind_t, size_t, const void *,
+      cudnnTensorDescriptor_t, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetMultiHeadAttnWeights");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, attnDesc, wKind, weightSizeInBytes, weights, wDesc,
+                  wAddr);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnMultiHeadAttnForward(
+    cudnnHandle_t handle, const cudnnAttnDescriptor_t attnDesc, int currIdx,
+    const int loWinIdx[], const int hiWinIdx[], const int devSeqLengthsQO[],
+    const int devSeqLengthsKV[], const cudnnSeqDataDescriptor_t qDesc,
+    const void *queries, const void *residuals,
+    const cudnnSeqDataDescriptor_t kDesc, const void *keys,
+    const cudnnSeqDataDescriptor_t vDesc, const void *values,
+    const cudnnSeqDataDescriptor_t oDesc, void *out, size_t weightSizeInBytes,
+    const void *weights, size_t workSpaceSizeInBytes, void *workSpace,
+    size_t reserveSpaceSizeInBytes, void *reserveSpace) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnAttnDescriptor_t, int, const int[], const int[],
+      const int[], const int[], const cudnnSeqDataDescriptor_t, const void *,
+      const void *, const cudnnSeqDataDescriptor_t, const void *,
+      const cudnnSeqDataDescriptor_t, const void *,
+      const cudnnSeqDataDescriptor_t, void *, size_t, const void *, size_t,
+      void *, size_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnMultiHeadAttnForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, attnDesc, currIdx, loWinIdx, hiWinIdx,
+                  devSeqLengthsQO, devSeqLengthsKV, qDesc, queries, residuals,
+                  kDesc, keys, vDesc, values, oDesc, out, weightSizeInBytes,
+                  weights, workSpaceSizeInBytes, workSpace,
+                  reserveSpaceSizeInBytes, reserveSpace);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnAdvInferVersionCheck(void) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnAdvInferVersionCheck");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSoftmaxBackward(
+    cudnnHandle_t handle, cudnnSoftmaxAlgorithm_t algo, cudnnSoftmaxMode_t mode,
+    const void *alpha, const cudnnTensorDescriptor_t yDesc, const void *y,
+    const cudnnTensorDescriptor_t dyDesc, const void *dy, const void *beta,
+    const cudnnTensorDescriptor_t dxDesc, void *dx) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnSoftmaxAlgorithm_t, cudnnSoftmaxMode_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSoftmaxBackward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, mode, alpha, yDesc, y, dyDesc, dy, beta, dxDesc,
+                  dx);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnPoolingBackward(
+    cudnnHandle_t handle, const cudnnPoolingDescriptor_t poolingDesc,
+    const void *alpha, const cudnnTensorDescriptor_t yDesc, const void *y,
+    const cudnnTensorDescriptor_t dyDesc, const void *dy,
+    const cudnnTensorDescriptor_t xDesc, const void *x, const void *beta,
+    const cudnnTensorDescriptor_t dxDesc, void *dx) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnPoolingDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnPoolingBackward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, poolingDesc, alpha, yDesc, y, dyDesc, dy, xDesc, x,
+                  beta, dxDesc, dx);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnActivationBackward(
+    cudnnHandle_t handle, cudnnActivationDescriptor_t activationDesc,
+    const void *alpha, const cudnnTensorDescriptor_t yDesc, const void *y,
+    const cudnnTensorDescriptor_t dyDesc, const void *dy,
+    const cudnnTensorDescriptor_t xDesc, const void *x, const void *beta,
+    const cudnnTensorDescriptor_t dxDesc, void *dx) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnActivationDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnActivationBackward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, activationDesc, alpha, yDesc, y, dyDesc, dy, xDesc, x,
+                  beta, dxDesc, dx);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnLRNCrossChannelBackward(
+    cudnnHandle_t handle, cudnnLRNDescriptor_t normDesc, cudnnLRNMode_t lrnMode,
+    const void *alpha, const cudnnTensorDescriptor_t yDesc, const void *y,
+    const cudnnTensorDescriptor_t dyDesc, const void *dy,
+    const cudnnTensorDescriptor_t xDesc, const void *x, const void *beta,
+    const cudnnTensorDescriptor_t dxDesc, void *dx) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnLRNDescriptor_t, cudnnLRNMode_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnLRNCrossChannelBackward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, normDesc, lrnMode, alpha, yDesc, y, dyDesc, dy, xDesc,
+                  x, beta, dxDesc, dx);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDivisiveNormalizationBackward(
+    cudnnHandle_t handle, cudnnLRNDescriptor_t normDesc,
+    cudnnDivNormMode_t mode, const void *alpha,
+    const cudnnTensorDescriptor_t
+        xDesc, /* same desc for x, means, dy, temp, temp2 */
+    const void *x,
+    const void *means, /* if NULL, means are assumed to be zero */
+    const void *dy, void *temp, void *temp2, const void *beta,
+    const cudnnTensorDescriptor_t dXdMeansDesc, /* same desc for dx, dMeans */
+    void *dx,                                   /* output x differential */
+    void *dMeans) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnLRNDescriptor_t, cudnnDivNormMode_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *, const void *,
+      void *, void *, const void *, const cudnnTensorDescriptor_t, void *,
+      void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnDivisiveNormalizationBackward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, normDesc, mode, alpha, xDesc, x, means, dy, temp,
+                  temp2, beta, dXdMeansDesc, dx, dMeans);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnGetBatchNormalizationForwardTrainingExWorkspaceSize(
+    cudnnHandle_t handle, cudnnBatchNormMode_t mode, cudnnBatchNormOps_t bnOps,
+    const cudnnTensorDescriptor_t xDesc, const cudnnTensorDescriptor_t zDesc,
+    const cudnnTensorDescriptor_t yDesc,
+    const cudnnTensorDescriptor_t bnScaleBiasMeanVarDesc,
+    const cudnnActivationDescriptor_t activationDesc, size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnBatchNormMode_t, cudnnBatchNormOps_t,
+      const cudnnTensorDescriptor_t, const cudnnTensorDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnTensorDescriptor_t,
+      const cudnnActivationDescriptor_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>(
+      "cudnnGetBatchNormalizationForwardTrainingExWorkspaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode, bnOps, xDesc, zDesc, yDesc,
+                  bnScaleBiasMeanVarDesc, activationDesc, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetBatchNormalizationBackwardExWorkspaceSize(
+    cudnnHandle_t handle, cudnnBatchNormMode_t mode, cudnnBatchNormOps_t bnOps,
+    const cudnnTensorDescriptor_t xDesc, const cudnnTensorDescriptor_t yDesc,
+    const cudnnTensorDescriptor_t dyDesc, const cudnnTensorDescriptor_t dzDesc,
+    const cudnnTensorDescriptor_t dxDesc,
+    const cudnnTensorDescriptor_t dBnScaleBiasDesc,
+    const cudnnActivationDescriptor_t activationDesc, size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnBatchNormMode_t, cudnnBatchNormOps_t,
+      const cudnnTensorDescriptor_t, const cudnnTensorDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnTensorDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnTensorDescriptor_t,
+      const cudnnActivationDescriptor_t, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetBatchNormalizationBackwardExWorkspaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode, bnOps, xDesc, yDesc, dyDesc, dzDesc, dxDesc,
+                  dBnScaleBiasDesc, activationDesc, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetBatchNormalizationTrainingExReserveSpaceSize(
+    cudnnHandle_t handle, cudnnBatchNormMode_t mode, cudnnBatchNormOps_t bnOps,
+    const cudnnActivationDescriptor_t activationDesc,
+    const cudnnTensorDescriptor_t xDesc, size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnBatchNormMode_t, cudnnBatchNormOps_t,
+      const cudnnActivationDescriptor_t, const cudnnTensorDescriptor_t,
+      size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>(
+      "cudnnGetBatchNormalizationTrainingExReserveSpaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode, bnOps, activationDesc, xDesc, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnBatchNormalizationForwardTraining(
+    cudnnHandle_t handle, cudnnBatchNormMode_t mode,
+
+    const void *alpha, /* alpha[0] = result blend factor */
+    const void *beta,  /* beta[0] = dest layer blend factor */
+
+    const cudnnTensorDescriptor_t xDesc, const void *x, /* NxCxHxW */
+    const cudnnTensorDescriptor_t yDesc, void *y,       /* NxCxHxW */
+
+    /* Shared desc for the next 6 tensors in the argument list.
+       Data type to be set as follows:
+       type = (typeOf(x) == double) ? double : float
+       Dimensions for this descriptor depend on normalization mode
+       - Spatial Normalization : tensors are expected to have dims 1xCx1x1
+        (normalization is performed across NxHxW)
+       - Per-Activation Normalization : tensors are expected to have dims of
+       1xCxHxW (normalization is performed across N) */
+    const cudnnTensorDescriptor_t bnScaleBiasMeanVarDesc,
+
+    /* 'Gamma' and 'Beta' respectively in Ioffe and Szegedy's paper's notation
+     */
+    const void *bnScale, const void *bnBias,
+
+    /* MUST use factor=1 in the very first call of a complete training cycle.
+       Use a factor=1/(1+n) at N-th call to the function to get
+       Cumulative Moving Average (CMA) behavior
+       CMA[n] = (x[1]+...+x[n])/n
+       Since CMA[n+1] = (n*CMA[n]+x[n+1])/(n+1) =
+       ((n+1)*CMA[n]-CMA[n])/(n+1) + x[n+1]/(n+1) =
+       CMA[n]*(1-1/(n+1)) + x[n+1]*1/(n+1) */
+    double exponentialAverageFactor,
+
+    /* Used in Training phase only.
+       runningMean = newMean*factor + runningMean*(1-factor) */
+    void *resultRunningMean,
+    /* Output in training mode, input in inference. Is the moving average
+       of  variance[x] (factor is applied in the same way as for runningMean) */
+    void *resultRunningVariance,
+
+    /* Has to be >= CUDNN_BN_MIN_EPSILON. Should be the same in forward and
+       backward functions. */
+    double epsilon,
+
+    /* Optionally save intermediate results from the forward pass here
+       - can be reused to speed up backward pass. NULL if unused */
+    void *resultSaveMean, void *resultSaveInvVariance) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnBatchNormMode_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, void *, const cudnnTensorDescriptor_t,
+      const void *, const void *, double, void *, void *, double, void *,
+      void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnBatchNormalizationForwardTraining");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(
+      handle, mode, alpha, beta, xDesc, x, yDesc, y, bnScaleBiasMeanVarDesc,
+      bnScale, bnBias, exponentialAverageFactor, resultRunningMean,
+      resultRunningVariance, epsilon, resultSaveMean, resultSaveInvVariance);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnBatchNormalizationForwardTrainingEx(
+    cudnnHandle_t handle, cudnnBatchNormMode_t mode, cudnnBatchNormOps_t bnOps,
+
+    const void *alpha, /* alpha[0] = result blend factor */
+    const void *beta,  /* beta[0] = dest layer blend factor */
+
+    const cudnnTensorDescriptor_t xDesc, const void *xData,
+    const cudnnTensorDescriptor_t zDesc, const void *zData,
+    const cudnnTensorDescriptor_t yDesc, void *yData,
+
+    const cudnnTensorDescriptor_t bnScaleBiasMeanVarDesc, const void *bnScale,
+    const void *bnBias,
+
+    double exponentialAverageFactor, void *resultRunningMean,
+    void *resultRunningVariance,
+
+    /* Has to be >= CUDNN_BN_MIN_EPSILON. Should be the same in forward and
+       backward functions. */
+    double epsilon,
+
+    /* Optionally save intermediate results from the forward pass here
+       - can be reused to speed up backward pass. NULL if unused */
+    void *resultSaveMean, void *resultSaveInvVariance,
+
+    cudnnActivationDescriptor_t activationDesc, void *workspace,
+    size_t workSpaceSizeInBytes, void *reserveSpace,
+    size_t reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnBatchNormMode_t, cudnnBatchNormOps_t, const void *,
+      const void *, const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, void *, const cudnnTensorDescriptor_t,
+      const void *, const void *, double, void *, void *, double, void *,
+      void *, cudnnActivationDescriptor_t, void *, size_t, void *, size_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnBatchNormalizationForwardTrainingEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode, bnOps, alpha, beta, xDesc, xData, zDesc, zData,
+                  yDesc, yData, bnScaleBiasMeanVarDesc, bnScale, bnBias,
+                  exponentialAverageFactor, resultRunningMean,
+                  resultRunningVariance, epsilon, resultSaveMean,
+                  resultSaveInvVariance, activationDesc, workspace,
+                  workSpaceSizeInBytes, reserveSpace, reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnBatchNormalizationBackward(
+    cudnnHandle_t handle, cudnnBatchNormMode_t mode, const void *alphaDataDiff,
+    const void *betaDataDiff, const void *alphaParamDiff,
+    const void *betaParamDiff,
+    const cudnnTensorDescriptor_t xDesc, /* same desc for x, dx, dy */
+    const void *x, const cudnnTensorDescriptor_t dyDesc, const void *dy,
+    const cudnnTensorDescriptor_t dxDesc, void *dx,
+    /* Shared tensor desc for the 4 tensors below */
+    const cudnnTensorDescriptor_t dBnScaleBiasDesc,
+    const void *bnScale, /* bnBias doesn't affect backpropagation */
+    /* scale and bias diff are not backpropagated below this layer */
+    void *dBnScaleResult, void *dBnBiasResult,
+    /* Same epsilon as forward pass */
+    double epsilon,
+
+    /* Optionally cached intermediate results from
+       forward pass */
+    const void *savedMean, const void *savedInvVariance) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnBatchNormMode_t, const void *, const void *,
+      const void *, const void *, const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, void *, const cudnnTensorDescriptor_t,
+      const void *, void *, void *, double, const void *, const void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnBatchNormalizationBackward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode, alphaDataDiff, betaDataDiff, alphaParamDiff,
+                  betaParamDiff, xDesc, x, dyDesc, dy, dxDesc, dx,
+                  dBnScaleBiasDesc, bnScale, dBnScaleResult, dBnBiasResult,
+                  epsilon, savedMean, savedInvVariance);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnBatchNormalizationBackwardEx(
+    cudnnHandle_t handle, cudnnBatchNormMode_t mode, cudnnBatchNormOps_t bnOps,
+
+    const void *alphaDataDiff, const void *betaDataDiff,
+    const void *alphaParamDiff, const void *betaParamDiff,
+    const cudnnTensorDescriptor_t xDesc, const void *xData,
+    const cudnnTensorDescriptor_t yDesc, const void *yData,
+    const cudnnTensorDescriptor_t dyDesc, const void *dyData,
+    const cudnnTensorDescriptor_t dzDesc, void *dzData,
+    const cudnnTensorDescriptor_t dxDesc, void *dxData,
+
+    /* Shared tensor desc for the 4 tensors below */
+    const cudnnTensorDescriptor_t dBnScaleBiasDesc, const void *bnScaleData,
+    const void *bnBiasData, /* needed if there is activation */
+    void *dBnScaleData, void *dBnBiasData,
+    double epsilon, /* Same epsilon as forward pass */
+
+    /* Optionally cached intermediate results from
+       forward pass */
+    const void *savedMean, const void *savedInvVariance,
+    cudnnActivationDescriptor_t activationDesc, void *workSpace,
+    size_t workSpaceSizeInBytes, void *reserveSpace,
+    size_t reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnBatchNormMode_t, cudnnBatchNormOps_t, const void *,
+      const void *, const void *, const void *, const cudnnTensorDescriptor_t,
+      const void *, const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, void *, const cudnnTensorDescriptor_t,
+      void *, const cudnnTensorDescriptor_t, const void *, const void *, void *,
+      void *, double, const void *, const void *, cudnnActivationDescriptor_t,
+      void *, size_t, void *, size_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnBatchNormalizationBackwardEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(
+      handle, mode, bnOps, alphaDataDiff, betaDataDiff, alphaParamDiff,
+      betaParamDiff, xDesc, xData, yDesc, yData, dyDesc, dyData, dzDesc, dzData,
+      dxDesc, dxData, dBnScaleBiasDesc, bnScaleData, bnBiasData, dBnScaleData,
+      dBnBiasData, epsilon, savedMean, savedInvVariance, activationDesc,
+      workSpace, workSpaceSizeInBytes, reserveSpace, reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSpatialTfGridGeneratorBackward(
+    cudnnHandle_t handle, const cudnnSpatialTransformerDescriptor_t stDesc,
+    const void *dgrid, void *dtheta) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnSpatialTransformerDescriptor_t, const void *,
+      void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnSpatialTfGridGeneratorBackward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, stDesc, dgrid, dtheta);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSpatialTfSamplerBackward(
+    cudnnHandle_t handle, cudnnSpatialTransformerDescriptor_t stDesc,
+    const void *alpha, const cudnnTensorDescriptor_t xDesc, const void *x,
+    const void *beta, const cudnnTensorDescriptor_t dxDesc, void *dx,
+    const void *alphaDgrid, const cudnnTensorDescriptor_t dyDesc,
+    const void *dy, const void *grid, const void *betaDgrid, void *dgrid) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnSpatialTransformerDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *,
+      const cudnnTensorDescriptor_t, void *, const void *,
+      const cudnnTensorDescriptor_t, const void *, const void *, const void *,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSpatialTfSamplerBackward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, stDesc, alpha, xDesc, x, beta, dxDesc, dx, alphaDgrid,
+                  dyDesc, dy, grid, betaDgrid, dgrid);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnDropoutBackward(
+    cudnnHandle_t handle, const cudnnDropoutDescriptor_t dropoutDesc,
+    const cudnnTensorDescriptor_t dydesc, const void *dy,
+    const cudnnTensorDescriptor_t dxdesc, void *dx, void *reserveSpace,
+    size_t reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnDropoutDescriptor_t,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, void *, void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDropoutBackward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dropoutDesc, dydesc, dy, dxdesc, dx, reserveSpace,
+                  reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnOpsTrainVersionCheck(void) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnOpsTrainVersionCheck");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionBackwardFilterAlgorithmMaxCount(
+    cudnnHandle_t handle, int *count) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnHandle_t, int *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolutionBackwardFilterAlgorithmMaxCount");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, count);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnFindConvolutionBackwardFilterAlgorithm(
+    cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
+    const cudnnTensorDescriptor_t dyDesc,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnFilterDescriptor_t dwDesc, const int requestedAlgoCount,
+    int *returnedAlgoCount, cudnnConvolutionBwdFilterAlgoPerf_t *perfResults) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnConvolutionDescriptor_t,
+      const cudnnFilterDescriptor_t, const int, int *,
+      cudnnConvolutionBwdFilterAlgoPerf_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnFindConvolutionBackwardFilterAlgorithm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, xDesc, dyDesc, convDesc, dwDesc, requestedAlgoCount,
+                  returnedAlgoCount, perfResults);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnFindConvolutionBackwardFilterAlgorithmEx(
+    cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc, const void *x,
+    const cudnnTensorDescriptor_t dyDesc, const void *y,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnFilterDescriptor_t dwDesc, void *dw,
+    const int requestedAlgoCount, int *returnedAlgoCount,
+    cudnnConvolutionBwdFilterAlgoPerf_t *perfResults, void *workSpace,
+    size_t workSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnConvolutionDescriptor_t, const cudnnFilterDescriptor_t, void *,
+      const int, int *, cudnnConvolutionBwdFilterAlgoPerf_t *, void *, size_t);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnFindConvolutionBackwardFilterAlgorithmEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, xDesc, x, dyDesc, y, convDesc, dwDesc, dw,
+                  requestedAlgoCount, returnedAlgoCount, perfResults, workSpace,
+                  workSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionBackwardFilterAlgorithm_v7(
+    cudnnHandle_t handle, const cudnnTensorDescriptor_t srcDesc,
+    const cudnnTensorDescriptor_t diffDesc,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnFilterDescriptor_t gradDesc, const int requestedAlgoCount,
+    int *returnedAlgoCount, cudnnConvolutionBwdFilterAlgoPerf_t *perfResults) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnConvolutionDescriptor_t,
+      const cudnnFilterDescriptor_t, const int, int *,
+      cudnnConvolutionBwdFilterAlgoPerf_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolutionBackwardFilterAlgorithm_v7");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, srcDesc, diffDesc, convDesc, gradDesc,
+                  requestedAlgoCount, returnedAlgoCount, perfResults);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetConvolutionBackwardFilterWorkspaceSize(
+    cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
+    const cudnnTensorDescriptor_t dyDesc,
+    const cudnnConvolutionDescriptor_t convDesc,
+    const cudnnFilterDescriptor_t gradDesc,
+    cudnnConvolutionBwdFilterAlgo_t algo, size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnConvolutionDescriptor_t,
+      const cudnnFilterDescriptor_t, cudnnConvolutionBwdFilterAlgo_t, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnGetConvolutionBackwardFilterWorkspaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, xDesc, dyDesc, convDesc, gradDesc, algo, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnConvolutionBackwardFilter(
+    cudnnHandle_t handle, const void *alpha,
+    const cudnnTensorDescriptor_t xDesc, const void *x,
+    const cudnnTensorDescriptor_t dyDesc, const void *dy,
+    const cudnnConvolutionDescriptor_t convDesc,
+    cudnnConvolutionBwdFilterAlgo_t algo, void *workSpace,
+    size_t workSpaceSizeInBytes, const void *beta,
+    const cudnnFilterDescriptor_t dwDesc, void *dw) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const void *, const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnConvolutionDescriptor_t, cudnnConvolutionBwdFilterAlgo_t,
+      void *, size_t, const void *, const cudnnFilterDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnConvolutionBackwardFilter");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, alpha, xDesc, x, dyDesc, dy, convDesc, algo,
+                  workSpace, workSpaceSizeInBytes, beta, dwDesc, dw);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnConvolutionBackwardBias(
+    cudnnHandle_t handle, const void *alpha,
+    const cudnnTensorDescriptor_t dyDesc, const void *dy, const void *beta,
+    const cudnnTensorDescriptor_t dbDesc, void *db) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const void *, const cudnnTensorDescriptor_t, const void *,
+      const void *, const cudnnTensorDescriptor_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnConvolutionBackwardBias");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, alpha, dyDesc, dy, beta, dbDesc, db);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetRNNTrainingReserveSize(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const int seqLength, const cudnnTensorDescriptor_t *xDesc,
+    size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const int,
+      const cudnnTensorDescriptor_t *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetRNNTrainingReserveSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, seqLength, xDesc, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNForwardTraining(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const int seqLength, const cudnnTensorDescriptor_t *xDesc, const void *x,
+    const cudnnTensorDescriptor_t hxDesc, const void *hx,
+    const cudnnTensorDescriptor_t cxDesc, const void *cx,
+    const cudnnFilterDescriptor_t wDesc, const void *w,
+    const cudnnTensorDescriptor_t *yDesc, void *y,
+    const cudnnTensorDescriptor_t hyDesc, void *hy,
+    const cudnnTensorDescriptor_t cyDesc, void *cy, void *workspace,
+    size_t workSpaceSizeInBytes, void *reserveSpace,
+    size_t reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const int,
+      const cudnnTensorDescriptor_t *, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnFilterDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t *, void *, const cudnnTensorDescriptor_t,
+      void *, const cudnnTensorDescriptor_t, void *, void *, size_t, void *,
+      size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNForwardTraining");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, seqLength, xDesc, x, hxDesc, hx, cxDesc, cx,
+                  wDesc, w, yDesc, y, hyDesc, hy, cyDesc, cy, workspace,
+                  workSpaceSizeInBytes, reserveSpace, reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnRNNBackwardData(cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+                     const int seqLength, const cudnnTensorDescriptor_t *yDesc,
+                     const void *y, const cudnnTensorDescriptor_t *dyDesc,
+                     const void *dy, const cudnnTensorDescriptor_t dhyDesc,
+                     const void *dhy, const cudnnTensorDescriptor_t dcyDesc,
+                     const void *dcy, const cudnnFilterDescriptor_t wDesc,
+                     const void *w, const cudnnTensorDescriptor_t hxDesc,
+                     const void *hx, const cudnnTensorDescriptor_t cxDesc,
+                     const void *cx, const cudnnTensorDescriptor_t *dxDesc,
+                     void *dx, const cudnnTensorDescriptor_t dhxDesc, void *dhx,
+                     const cudnnTensorDescriptor_t dcxDesc, void *dcx,
+                     void *workspace, size_t workSpaceSizeInBytes,
+                     void *reserveSpace, size_t reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const int,
+      const cudnnTensorDescriptor_t *, const void *,
+      const cudnnTensorDescriptor_t *, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnFilterDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t *, void *, const cudnnTensorDescriptor_t,
+      void *, const cudnnTensorDescriptor_t, void *, void *, size_t, void *,
+      size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNBackwardData");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, seqLength, yDesc, y, dyDesc, dy, dhyDesc,
+                  dhy, dcyDesc, dcy, wDesc, w, hxDesc, hx, cxDesc, cx, dxDesc,
+                  dx, dhxDesc, dhx, dcxDesc, dcx, workspace,
+                  workSpaceSizeInBytes, reserveSpace, reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNBackwardWeights(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const int seqLength, const cudnnTensorDescriptor_t *xDesc, const void *x,
+    const cudnnTensorDescriptor_t hxDesc, const void *hx,
+    const cudnnTensorDescriptor_t *yDesc, const void *y, const void *workspace,
+    size_t workSpaceSizeInBytes, const cudnnFilterDescriptor_t dwDesc, void *dw,
+    const void *reserveSpace, size_t reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const int,
+      const cudnnTensorDescriptor_t *, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t *, const void *, const void *, size_t,
+      const cudnnFilterDescriptor_t, void *, const void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNBackwardWeights");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, seqLength, xDesc, x, hxDesc, hx, yDesc, y,
+                  workspace, workSpaceSizeInBytes, dwDesc, dw, reserveSpace,
+                  reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNForwardTrainingEx(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const cudnnRNNDataDescriptor_t xDesc, const void *x,
+    const cudnnTensorDescriptor_t hxDesc, const void *hx,
+    const cudnnTensorDescriptor_t cxDesc, const void *cx,
+    const cudnnFilterDescriptor_t wDesc, const void *w,
+    const cudnnRNNDataDescriptor_t yDesc, void *y,
+    const cudnnTensorDescriptor_t hyDesc, void *hy,
+    const cudnnTensorDescriptor_t cyDesc, void *cy,
+    const cudnnRNNDataDescriptor_t kDesc, /* reserved, should pass NULL */
+    const void *keys,                     /* reserved, should pass NULL */
+    const cudnnRNNDataDescriptor_t cDesc, /* reserved, should pass NULL */
+    void *cAttn,                          /* reserved, should pass NULL */
+    const cudnnRNNDataDescriptor_t iDesc, /* reserved, should pass NULL */
+    void *iAttn,                          /* reserved, should pass NULL */
+    const cudnnRNNDataDescriptor_t qDesc, /* reserved, should pass NULL */
+    void *queries,                        /* reserved, should pass NULL */
+    void *workSpace, size_t workSpaceSizeInBytes, void *reserveSpace,
+    size_t reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const cudnnRNNDataDescriptor_t,
+      const void *, const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnFilterDescriptor_t, const void *,
+      const cudnnRNNDataDescriptor_t, void *, const cudnnTensorDescriptor_t,
+      void *, const cudnnTensorDescriptor_t, void *,
+      const cudnnRNNDataDescriptor_t, const void *,
+      const cudnnRNNDataDescriptor_t, void *, const cudnnRNNDataDescriptor_t,
+      void *, const cudnnRNNDataDescriptor_t, void *, void *, size_t, void *,
+      size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNForwardTrainingEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, xDesc, x, hxDesc, hx, cxDesc, cx, wDesc, w,
+                  yDesc, y, hyDesc, hy, cyDesc, cy, kDesc, keys, cDesc, cAttn,
+                  iDesc, iAttn, qDesc, queries, workSpace, workSpaceSizeInBytes,
+                  reserveSpace, reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNForward(
+    cudnnHandle_t handle, cudnnRNNDescriptor_t rnnDesc,
+    cudnnForwardMode_t fwdMode, const int32_t devSeqLengths[],
+    cudnnRNNDataDescriptor_t xDesc, const void *x,
+    cudnnRNNDataDescriptor_t yDesc, void *y, cudnnTensorDescriptor_t hDesc,
+    const void *hx, void *hy, cudnnTensorDescriptor_t cDesc, const void *cx,
+    void *cy, size_t weightSpaceSize, const void *weightSpace,
+    size_t workSpaceSize, void *workSpace, size_t reserveSpaceSize,
+    void *reserveSpace) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnRNNDescriptor_t, cudnnForwardMode_t, const int32_t[],
+      cudnnRNNDataDescriptor_t, const void *, cudnnRNNDataDescriptor_t, void *,
+      cudnnTensorDescriptor_t, const void *, void *, cudnnTensorDescriptor_t,
+      const void *, void *, size_t, const void *, size_t, void *, size_t,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNForward");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, fwdMode, devSeqLengths, xDesc, x, yDesc, y,
+                  hDesc, hx, hy, cDesc, cx, cy, weightSpaceSize, weightSpace,
+                  workSpaceSize, workSpace, reserveSpaceSize, reserveSpace);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNBackwardDataEx(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const cudnnRNNDataDescriptor_t yDesc, const void *y,
+    const cudnnRNNDataDescriptor_t dyDesc, const void *dy,
+    const cudnnRNNDataDescriptor_t dcDesc, /* reserved, should pass NULL */
+    const void *dcAttn,                    /* reserved, should pass NULL */
+    const cudnnTensorDescriptor_t dhyDesc, const void *dhy,
+    const cudnnTensorDescriptor_t dcyDesc, const void *dcy,
+    const cudnnFilterDescriptor_t wDesc, const void *w,
+    const cudnnTensorDescriptor_t hxDesc, const void *hx,
+    const cudnnTensorDescriptor_t cxDesc, const void *cx,
+    const cudnnRNNDataDescriptor_t dxDesc, void *dx,
+    const cudnnTensorDescriptor_t dhxDesc, void *dhx,
+    const cudnnTensorDescriptor_t dcxDesc, void *dcx,
+    const cudnnRNNDataDescriptor_t dkDesc, /* reserved, should pass NULL */
+    void *dkeys,                           /* reserved, should pass NULL */
+    void *workSpace, size_t workSpaceSizeInBytes, void *reserveSpace,
+    size_t reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const cudnnRNNDataDescriptor_t,
+      const void *, const cudnnRNNDataDescriptor_t, const void *,
+      const cudnnRNNDataDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnFilterDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnTensorDescriptor_t, const void *,
+      const cudnnRNNDataDescriptor_t, void *, const cudnnTensorDescriptor_t,
+      void *, const cudnnTensorDescriptor_t, void *,
+      const cudnnRNNDataDescriptor_t, void *, void *, size_t, void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNBackwardDataEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, yDesc, y, dyDesc, dy, dcDesc, dcAttn,
+                  dhyDesc, dhy, dcyDesc, dcy, wDesc, w, hxDesc, hx, cxDesc, cx,
+                  dxDesc, dx, dhxDesc, dhx, dcxDesc, dcx, dkDesc, dkeys,
+                  workSpace, workSpaceSizeInBytes, reserveSpace,
+                  reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNBackwardData_v8(
+    cudnnHandle_t handle, cudnnRNNDescriptor_t rnnDesc,
+    const int32_t devSeqLengths[], cudnnRNNDataDescriptor_t yDesc,
+    const void *y, const void *dy, cudnnRNNDataDescriptor_t xDesc, void *dx,
+    cudnnTensorDescriptor_t hDesc, const void *hx, const void *dhy, void *dhx,
+    cudnnTensorDescriptor_t cDesc, const void *cx, const void *dcy, void *dcx,
+    size_t weightSpaceSize, const void *weightSpace, size_t workSpaceSize,
+    void *workSpace, size_t reserveSpaceSize, void *reserveSpace) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnRNNDescriptor_t, const int32_t[],
+      cudnnRNNDataDescriptor_t, const void *, const void *,
+      cudnnRNNDataDescriptor_t, void *, cudnnTensorDescriptor_t, const void *,
+      const void *, void *, cudnnTensorDescriptor_t, const void *, const void *,
+      void *, size_t, const void *, size_t, void *, size_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNBackwardData_v8");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, devSeqLengths, yDesc, y, dy, xDesc, dx,
+                  hDesc, hx, dhy, dhx, cDesc, cx, dcy, dcx, weightSpaceSize,
+                  weightSpace, workSpaceSize, workSpace, reserveSpaceSize,
+                  reserveSpace);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNBackwardWeightsEx(
+    cudnnHandle_t handle, const cudnnRNNDescriptor_t rnnDesc,
+    const cudnnRNNDataDescriptor_t xDesc, const void *x,
+    const cudnnTensorDescriptor_t hxDesc, const void *hx,
+    const cudnnRNNDataDescriptor_t yDesc, const void *y, void *workSpace,
+    size_t workSpaceSizeInBytes, const cudnnFilterDescriptor_t dwDesc, void *dw,
+    void *reserveSpace, size_t reserveSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnRNNDescriptor_t, const cudnnRNNDataDescriptor_t,
+      const void *, const cudnnTensorDescriptor_t, const void *,
+      const cudnnRNNDataDescriptor_t, const void *, void *, size_t,
+      const cudnnFilterDescriptor_t, void *, void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNBackwardWeightsEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, xDesc, x, hxDesc, hx, yDesc, y, workSpace,
+                  workSpaceSizeInBytes, dwDesc, dw, reserveSpace,
+                  reserveSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnRNNBackwardWeights_v8(
+    cudnnHandle_t handle, cudnnRNNDescriptor_t rnnDesc,
+    cudnnWgradMode_t addGrad, const int32_t devSeqLengths[],
+    cudnnRNNDataDescriptor_t xDesc, const void *x,
+    cudnnTensorDescriptor_t hDesc, const void *hx,
+    cudnnRNNDataDescriptor_t yDesc, const void *y, size_t weightSpaceSize,
+    void *dweightSpace, size_t workSpaceSize, void *workSpace,
+    size_t reserveSpaceSize, void *reserveSpace) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnRNNDescriptor_t, cudnnWgradMode_t, const int32_t[],
+      cudnnRNNDataDescriptor_t, const void *, cudnnTensorDescriptor_t,
+      const void *, cudnnRNNDataDescriptor_t, const void *, size_t, void *,
+      size_t, void *, size_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnRNNBackwardWeights_v8");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, rnnDesc, addGrad, devSeqLengths, xDesc, x, hDesc, hx,
+                  yDesc, y, weightSpaceSize, dweightSpace, workSpaceSize,
+                  workSpace, reserveSpaceSize, reserveSpace);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnMultiHeadAttnBackwardData(
+    cudnnHandle_t handle, const cudnnAttnDescriptor_t attnDesc,
+    const int loWinIdx[], const int hiWinIdx[], const int devSeqLengthsDQDO[],
+    const int devSeqLengthsDKDV[], const cudnnSeqDataDescriptor_t doDesc,
+    const void *dout, const cudnnSeqDataDescriptor_t dqDesc, void *dqueries,
+    const void *queries, const cudnnSeqDataDescriptor_t dkDesc, void *dkeys,
+    const void *keys, const cudnnSeqDataDescriptor_t dvDesc, void *dvalues,
+    const void *values, size_t weightSizeInBytes, const void *weights,
+    size_t workSpaceSizeInBytes, void *workSpace,
+    size_t reserveSpaceSizeInBytes, void *reserveSpace) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnAttnDescriptor_t, const int[], const int[],
+      const int[], const int[], const cudnnSeqDataDescriptor_t, const void *,
+      const cudnnSeqDataDescriptor_t, void *, const void *,
+      const cudnnSeqDataDescriptor_t, void *, const void *,
+      const cudnnSeqDataDescriptor_t, void *, const void *, size_t,
+      const void *, size_t, void *, size_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnMultiHeadAttnBackwardData");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, attnDesc, loWinIdx, hiWinIdx, devSeqLengthsDQDO,
+                  devSeqLengthsDKDV, doDesc, dout, dqDesc, dqueries, queries,
+                  dkDesc, dkeys, keys, dvDesc, dvalues, values,
+                  weightSizeInBytes, weights, workSpaceSizeInBytes, workSpace,
+                  reserveSpaceSizeInBytes, reserveSpace);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnMultiHeadAttnBackwardWeights(
+    cudnnHandle_t handle, const cudnnAttnDescriptor_t attnDesc,
+    cudnnWgradMode_t addGrad, const cudnnSeqDataDescriptor_t qDesc,
+    const void *queries, const cudnnSeqDataDescriptor_t kDesc, const void *keys,
+    const cudnnSeqDataDescriptor_t vDesc, const void *values,
+    const cudnnSeqDataDescriptor_t doDesc, const void *dout,
+    size_t weightSizeInBytes, const void *weights, void *dweights,
+    size_t workSpaceSizeInBytes, void *workSpace,
+    size_t reserveSpaceSizeInBytes, void *reserveSpace) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnAttnDescriptor_t, cudnnWgradMode_t,
+      const cudnnSeqDataDescriptor_t, const void *,
+      const cudnnSeqDataDescriptor_t, const void *,
+      const cudnnSeqDataDescriptor_t, const void *,
+      const cudnnSeqDataDescriptor_t, const void *, size_t, const void *,
+      void *, size_t, void *, size_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cudnnMultiHeadAttnBackwardWeights");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, attnDesc, addGrad, qDesc, queries, kDesc, keys, vDesc,
+                  values, doDesc, dout, weightSizeInBytes, weights, dweights,
+                  workSpaceSizeInBytes, workSpace, reserveSpaceSizeInBytes,
+                  reserveSpace);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnCreateCTCLossDescriptor(cudnnCTCLossDescriptor_t *ctcLossDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnCTCLossDescriptor_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCreateCTCLossDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctcLossDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetCTCLossDescriptor(
+    cudnnCTCLossDescriptor_t ctcLossDesc, cudnnDataType_t compType) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnCTCLossDescriptor_t, cudnnDataType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetCTCLossDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctcLossDesc, compType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetCTCLossDescriptorEx(
+    cudnnCTCLossDescriptor_t ctcLossDesc, cudnnDataType_t compType,
+    cudnnLossNormalizationMode_t normMode, cudnnNanPropagation_t gradMode) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnCTCLossDescriptor_t, cudnnDataType_t, cudnnLossNormalizationMode_t,
+      cudnnNanPropagation_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetCTCLossDescriptorEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctcLossDesc, compType, normMode, gradMode);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnSetCTCLossDescriptor_v8(
+    cudnnCTCLossDescriptor_t ctcLossDesc, cudnnDataType_t compType,
+    cudnnLossNormalizationMode_t normMode, cudnnNanPropagation_t gradMode,
+    int maxLabelLength) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnCTCLossDescriptor_t, cudnnDataType_t, cudnnLossNormalizationMode_t,
+      cudnnNanPropagation_t, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnSetCTCLossDescriptor_v8");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctcLossDesc, compType, normMode, gradMode, maxLabelLength);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetCTCLossDescriptor(
+    cudnnCTCLossDescriptor_t ctcLossDesc, cudnnDataType_t *compType) {
+  using FuncPtr =
+      cudnnStatus_t(CUDNNWINAPI *)(cudnnCTCLossDescriptor_t, cudnnDataType_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetCTCLossDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctcLossDesc, compType);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetCTCLossDescriptorEx(
+    cudnnCTCLossDescriptor_t ctcLossDesc, cudnnDataType_t *compType,
+    cudnnLossNormalizationMode_t *normMode, cudnnNanPropagation_t *gradMode) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnCTCLossDescriptor_t, cudnnDataType_t *,
+      cudnnLossNormalizationMode_t *, cudnnNanPropagation_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetCTCLossDescriptorEx");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctcLossDesc, compType, normMode, gradMode);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetCTCLossDescriptor_v8(
+    cudnnCTCLossDescriptor_t ctcLossDesc, cudnnDataType_t *compType,
+    cudnnLossNormalizationMode_t *normMode, cudnnNanPropagation_t *gradMode,
+    int *maxLabelLength) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnCTCLossDescriptor_t, cudnnDataType_t *,
+      cudnnLossNormalizationMode_t *, cudnnNanPropagation_t *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetCTCLossDescriptor_v8");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctcLossDesc, compType, normMode, gradMode, maxLabelLength);
+}
+
+cudnnStatus_t CUDNNWINAPI
+cudnnDestroyCTCLossDescriptor(cudnnCTCLossDescriptor_t ctcLossDesc) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(cudnnCTCLossDescriptor_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnDestroyCTCLossDescriptor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(ctcLossDesc);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCTCLoss(
+    cudnnHandle_t handle,
+    const cudnnTensorDescriptor_t
+        probsDesc, /* Tensor descriptor for probabilities, the dimensions are
+                      T,N,A (T is the timing steps, N is the
+                      mini batch size, A is the alphabet size)  */
+    const void *probs,      /* probabilities after softmax, in GPU memory */
+    const int hostLabels[], /* labels, in CPU memory */
+    const int hostLabelLengths[], /* the length of each label, in CPU memory */
+    const int hostInputLengths[], /* the lengths of timing steps in each batch,
+                                     in CPU memory */
+    void *costs,                  /* the returned costs of CTC, in GPU memory */
+    const cudnnTensorDescriptor_t
+        gradientsDesc, /* Tensor descriptor for gradients, the dimensions are
+                          T,N,A */
+    void *gradients,   /* the returned CTC gradients, in GPU memory, to compute
+                          costs only, set it to NULL */
+    cudnnCTCLossAlgo_t algo, /* algorithm selected, supported now 0 and 1 */
+    cudnnCTCLossDescriptor_t ctcLossDesc,
+    void *workspace, /* pointer to the workspace, in GPU memory */
+    size_t workSpaceSizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t, const void *, const int[],
+      const int[], const int[], void *, const cudnnTensorDescriptor_t, void *,
+      cudnnCTCLossAlgo_t, cudnnCTCLossDescriptor_t, void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCTCLoss");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, probsDesc, probs, hostLabels, hostLabelLengths,
+                  hostInputLengths, costs, gradientsDesc, gradients, algo,
+                  ctcLossDesc, workspace, workSpaceSizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnCTCLoss_v8(
+    cudnnHandle_t handle,
+    cudnnCTCLossAlgo_t algo, /* algorithm selected, supported now 0 and 1 */
+    cudnnCTCLossDescriptor_t ctcLossDesc,
+    const cudnnTensorDescriptor_t
+        probsDesc, /* Tensor descriptor for probabilities, the dimensions are
+                      T,N,A (T is the timing steps, N is the
+                      mini batch size, A is the alphabet size)  */
+    const void *probs,        /* probabilities after softmax, in GPU memory */
+    const int labels[],       /* labels, in GPU memory */
+    const int labelLengths[], /* the length of each label, in GPU memory */
+    const int inputLengths[], /* the lengths of timing steps in each batch, in
+                                 GPU memory */
+    void *costs,              /* the returned costs of CTC, in GPU memory */
+    const cudnnTensorDescriptor_t
+        gradientsDesc, /* Tensor descriptor for gradients, the dimensions are
+                          T,N,A */
+    void *gradients,   /* the returned CTC gradients, in GPU memory, to compute
+                          costs only, set it to NULL */
+    size_t workSpaceSizeInBytes, /* size of the workspace */
+    void *workspace) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnCTCLossAlgo_t, cudnnCTCLossDescriptor_t,
+      const cudnnTensorDescriptor_t, const void *, const int[], const int[],
+      const int[], void *, const cudnnTensorDescriptor_t, void *, size_t,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnCTCLoss_v8");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, ctcLossDesc, probsDesc, probs, labels,
+                  labelLengths, inputLengths, costs, gradientsDesc, gradients,
+                  workSpaceSizeInBytes, workspace);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetCTCLossWorkspaceSize(
+    cudnnHandle_t handle,
+    const cudnnTensorDescriptor_t
+        probsDesc, /* Tensor descriptor for probabilities, the dimensions are
+                      T,N,A (T is the
+                      timing steps, N is the mini batch size, A is the alphabet
+                      size) */
+    const cudnnTensorDescriptor_t
+        gradientsDesc,       /* Tensor descriptor for gradients, the
+                                dimensions are T,N,A. To compute costs
+                                only, set it to NULL */
+    const int *labels,       /* labels, in CPU memory */
+    const int *labelLengths, /* the length of each label, in CPU memory */
+    const int *inputLengths, /* the lengths of timing steps in each batch, in
+                                CPU memory */
+    cudnnCTCLossAlgo_t algo, /* algorithm selected, supported now 0 and 1 */
+    cudnnCTCLossDescriptor_t ctcLossDesc, size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, const cudnnTensorDescriptor_t,
+      const cudnnTensorDescriptor_t, const int *, const int *, const int *,
+      cudnnCTCLossAlgo_t, cudnnCTCLossDescriptor_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetCTCLossWorkspaceSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, probsDesc, gradientsDesc, labels, labelLengths,
+                  inputLengths, algo, ctcLossDesc, sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnGetCTCLossWorkspaceSize_v8(
+    cudnnHandle_t handle,
+    cudnnCTCLossAlgo_t algo, /* algorithm selected, supported now 0 and 1 */
+    cudnnCTCLossDescriptor_t ctcLossDesc,
+    const cudnnTensorDescriptor_t
+        probsDesc, /* Tensor descriptor for probabilities, the dimensions are
+                      T,N,A (T is the
+                      timing steps, N is the mini batch size, A is the alphabet
+                      size) */
+    const cudnnTensorDescriptor_t
+        gradientsDesc, /* Tensor descriptor for gradients, the
+                          dimensions are T,N,A. To compute costs
+                          only, set it to NULL */
+    size_t *sizeInBytes) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)(
+      cudnnHandle_t, cudnnCTCLossAlgo_t, cudnnCTCLossDescriptor_t,
+      const cudnnTensorDescriptor_t, const cudnnTensorDescriptor_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnGetCTCLossWorkspaceSize_v8");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, ctcLossDesc, probsDesc, gradientsDesc,
+                  sizeInBytes);
+}
+
+cudnnStatus_t CUDNNWINAPI cudnnAdvTrainVersionCheck(void) {
+  using FuncPtr = cudnnStatus_t(CUDNNWINAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cudnnAdvTrainVersionCheck");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+}  // extern "C"
diff --git a/tensorflow/stream_executor/cuda/cudnn_stub.cc b/tensorflow/stream_executor/cuda/cudnn_stub.cc
index 5a05437480e..85141973ed4 100644
--- a/tensorflow/stream_executor/cuda/cudnn_stub.cc
+++ b/tensorflow/stream_executor/cuda/cudnn_stub.cc
@@ -60,5 +60,5 @@ cudnnStatus_t GetSymbolNotFoundError() { return CUDNN_STATUS_INTERNAL_ERROR; }
 #elif CUDNN_MINOR < 6
 #include "tensorflow/stream_executor/cuda/cudnn_7_4.inc"
 #else
-#include "tensorflow/stream_executor/cuda/cudnn_7_6.inc"
+#include "tensorflow/stream_executor/cuda/cudnn_8_0.inc"
 #endif
diff --git a/tensorflow/stream_executor/cuda/cusparse_12_0.inc b/tensorflow/stream_executor/cuda/cusparse_12_0.inc
new file mode 100644
index 00000000000..91641482860
--- /dev/null
+++ b/tensorflow/stream_executor/cuda/cusparse_12_0.inc
@@ -0,0 +1,6080 @@
+// Auto-generated, do not edit.
+
+extern "C" {
+cusparseStatus_t CUSPARSEAPI cusparseCreate(cusparseHandle_t *handle) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreate");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDestroy(cusparseHandle_t handle) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseGetVersion(cusparseHandle_t handle,
+                                                int *version) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetVersion");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, version);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseGetProperty(libraryPropertyType type,
+                                                 int *value) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(libraryPropertyType, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetProperty");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(type, value);
+}
+
+const char *CUSPARSEAPI cusparseGetErrorName(cusparseStatus_t status) {
+  using FuncPtr = const char *(CUSPARSEAPI *)(cusparseStatus_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetErrorName");
+  if (!func_ptr) return "cusparseGetErrorName symbol not found.";
+  return func_ptr(status);
+}
+
+const char *CUSPARSEAPI cusparseGetErrorString(cusparseStatus_t status) {
+  using FuncPtr = const char *(CUSPARSEAPI *)(cusparseStatus_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetErrorString");
+  if (!func_ptr) return "cusparseGetErrorString symbol not found.";
+  return func_ptr(status);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSetStream(cusparseHandle_t handle,
+                                               cudaStream_t streamId) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, cudaStream_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSetStream");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, streamId);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseGetStream(cusparseHandle_t handle,
+                                               cudaStream_t *streamId) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, cudaStream_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetStream");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, streamId);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseGetPointerMode(cusparseHandle_t handle, cusparsePointerMode_t *mode) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t,
+                                                  cusparsePointerMode_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetPointerMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSetPointerMode(cusparseHandle_t handle, cusparsePointerMode_t mode) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, cusparsePointerMode_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSetPointerMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mode);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseLoggerSetCallback(cusparseLoggerCallback_t callback) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseLoggerCallback_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseLoggerSetCallback");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(callback);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseLoggerSetFile(FILE *file) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(FILE *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseLoggerSetFile");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(file);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseLoggerOpenFile(const char *logFile) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(const char *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseLoggerOpenFile");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(logFile);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseLoggerSetLevel(int level) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseLoggerSetLevel");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(level);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseLoggerSetMask(int mask) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseLoggerSetMask");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(mask);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseLoggerForceDisable(void) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)();
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseLoggerForceDisable");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr();
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCreateMatDescr(cusparseMatDescr_t *descrA) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseMatDescr_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateMatDescr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descrA);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDestroyMatDescr(cusparseMatDescr_t descrA) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseMatDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyMatDescr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descrA);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSetMatType(cusparseMatDescr_t descrA,
+                                                cusparseMatrixType_t type) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseMatDescr_t, cusparseMatrixType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSetMatType");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descrA, type);
+}
+
+cusparseMatrixType_t CUSPARSEAPI
+cusparseGetMatType(const cusparseMatDescr_t descrA) {
+  using FuncPtr = cusparseMatrixType_t(CUSPARSEAPI *)(const cusparseMatDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetMatType");
+  return func_ptr(descrA);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSetMatFillMode(cusparseMatDescr_t descrA, cusparseFillMode_t fillMode) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseMatDescr_t, cusparseFillMode_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSetMatFillMode");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descrA, fillMode);
+}
+
+cusparseFillMode_t CUSPARSEAPI
+cusparseGetMatFillMode(const cusparseMatDescr_t descrA) {
+  using FuncPtr = cusparseFillMode_t(CUSPARSEAPI *)(const cusparseMatDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetMatFillMode");
+  return func_ptr(descrA);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSetMatDiagType(cusparseMatDescr_t descrA, cusparseDiagType_t diagType) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseMatDescr_t, cusparseDiagType_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSetMatDiagType");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descrA, diagType);
+}
+
+cusparseDiagType_t CUSPARSEAPI
+cusparseGetMatDiagType(const cusparseMatDescr_t descrA) {
+  using FuncPtr = cusparseDiagType_t(CUSPARSEAPI *)(const cusparseMatDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetMatDiagType");
+  return func_ptr(descrA);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSetMatIndexBase(cusparseMatDescr_t descrA,
+                                                     cusparseIndexBase_t base) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseMatDescr_t, cusparseIndexBase_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSetMatIndexBase");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descrA, base);
+}
+
+cusparseIndexBase_t CUSPARSEAPI
+cusparseGetMatIndexBase(const cusparseMatDescr_t descrA) {
+  using FuncPtr = cusparseIndexBase_t(CUSPARSEAPI *)(const cusparseMatDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetMatIndexBase");
+  return func_ptr(descrA);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateCsric02Info(csric02Info_t *info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(csric02Info_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateCsric02Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDestroyCsric02Info(csric02Info_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(csric02Info_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyCsric02Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateBsric02Info(bsric02Info_t *info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(bsric02Info_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateBsric02Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDestroyBsric02Info(bsric02Info_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(bsric02Info_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyBsric02Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateCsrilu02Info(csrilu02Info_t *info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(csrilu02Info_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateCsrilu02Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDestroyCsrilu02Info(csrilu02Info_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(csrilu02Info_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyCsrilu02Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateBsrilu02Info(bsrilu02Info_t *info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(bsrilu02Info_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateBsrilu02Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDestroyBsrilu02Info(bsrilu02Info_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(bsrilu02Info_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyBsrilu02Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateBsrsv2Info(bsrsv2Info_t *info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(bsrsv2Info_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateBsrsv2Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDestroyBsrsv2Info(bsrsv2Info_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(bsrsv2Info_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyBsrsv2Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateBsrsm2Info(bsrsm2Info_t *info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(bsrsm2Info_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateBsrsm2Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDestroyBsrsm2Info(bsrsm2Info_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(bsrsm2Info_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyBsrsm2Info");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateCsru2csrInfo(csru2csrInfo_t *info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(csru2csrInfo_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateCsru2csrInfo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDestroyCsru2csrInfo(csru2csrInfo_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(csru2csrInfo_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyCsru2csrInfo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCreateColorInfo(cusparseColorInfo_t *info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseColorInfo_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateColorInfo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDestroyColorInfo(cusparseColorInfo_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseColorInfo_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyColorInfo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSetColorAlgs(cusparseColorInfo_t info,
+                                                  cusparseColorAlg_t alg) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseColorInfo_t, cusparseColorAlg_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSetColorAlgs");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info, alg);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseGetColorAlgs(cusparseColorInfo_t info,
+                                                  cusparseColorAlg_t *alg) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseColorInfo_t,
+                                                  cusparseColorAlg_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGetColorAlgs");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info, alg);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreatePruneInfo(pruneInfo_t *info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(pruneInfo_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreatePruneInfo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDestroyPruneInfo(pruneInfo_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(pruneInfo_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyPruneInfo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(info);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSgemvi(cusparseHandle_t handle, cusparseOperation_t transA, int m,
+               int n, const float *alpha, const float *A, int lda, int nnz,
+               const float *xVal, const int *xInd, const float *beta, float *y,
+               cusparseIndexBase_t idxBase, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, int, int, const float *,
+      const float *, int, int, const float *, const int *, const float *,
+      float *, cusparseIndexBase_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgemvi");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transA, m, n, alpha, A, lda, nnz, xVal, xInd, beta, y,
+                  idxBase, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSgemvi_bufferSize(cusparseHandle_t handle, cusparseOperation_t transA,
+                          int m, int n, int nnz, int *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, int, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgemvi_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transA, m, n, nnz, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDgemvi(cusparseHandle_t handle, cusparseOperation_t transA, int m,
+               int n, const double *alpha, const double *A, int lda, int nnz,
+               const double *xVal, const int *xInd, const double *beta,
+               double *y, cusparseIndexBase_t idxBase, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, int, int, const double *,
+      const double *, int, int, const double *, const int *, const double *,
+      double *, cusparseIndexBase_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgemvi");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transA, m, n, alpha, A, lda, nnz, xVal, xInd, beta, y,
+                  idxBase, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDgemvi_bufferSize(cusparseHandle_t handle, cusparseOperation_t transA,
+                          int m, int n, int nnz, int *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, int, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgemvi_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transA, m, n, nnz, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgemvi(
+    cusparseHandle_t handle, cusparseOperation_t transA, int m, int n,
+    const cuComplex *alpha, const cuComplex *A, int lda, int nnz,
+    const cuComplex *xVal, const int *xInd, const cuComplex *beta, cuComplex *y,
+    cusparseIndexBase_t idxBase, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, int, int, const cuComplex *,
+      const cuComplex *, int, int, const cuComplex *, const int *,
+      const cuComplex *, cuComplex *, cusparseIndexBase_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgemvi");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transA, m, n, alpha, A, lda, nnz, xVal, xInd, beta, y,
+                  idxBase, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCgemvi_bufferSize(cusparseHandle_t handle, cusparseOperation_t transA,
+                          int m, int n, int nnz, int *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, int, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgemvi_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transA, m, n, nnz, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgemvi(
+    cusparseHandle_t handle, cusparseOperation_t transA, int m, int n,
+    const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, int nnz,
+    const cuDoubleComplex *xVal, const int *xInd, const cuDoubleComplex *beta,
+    cuDoubleComplex *y, cusparseIndexBase_t idxBase, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, int, int, const cuDoubleComplex *, const int *,
+      const cuDoubleComplex *, cuDoubleComplex *, cusparseIndexBase_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgemvi");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transA, m, n, alpha, A, lda, nnz, xVal, xInd, beta, y,
+                  idxBase, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseZgemvi_bufferSize(cusparseHandle_t handle, cusparseOperation_t transA,
+                          int m, int n, int nnz, int *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, int, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgemvi_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, transA, m, n, nnz, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrmv(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nb, int nnzb, const float *alpha,
+    const cusparseMatDescr_t descrA, const float *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    const float *x, const float *beta, float *y) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int, int,
+      const float *, const cusparseMatDescr_t, const float *, const int *,
+      const int *, int, const float *, const float *, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrmv");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedRowPtrA, bsrSortedColIndA, blockDim,
+                  x, beta, y);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrmv(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nb, int nnzb, const double *alpha,
+    const cusparseMatDescr_t descrA, const double *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    const double *x, const double *beta, double *y) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int, int,
+      const double *, const cusparseMatDescr_t, const double *, const int *,
+      const int *, int, const double *, const double *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrmv");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedRowPtrA, bsrSortedColIndA, blockDim,
+                  x, beta, y);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCbsrmv(cusparseHandle_t handle, cusparseDirection_t dirA,
+               cusparseOperation_t transA, int mb, int nb, int nnzb,
+               const cuComplex *alpha, const cusparseMatDescr_t descrA,
+               const cuComplex *bsrSortedValA, const int *bsrSortedRowPtrA,
+               const int *bsrSortedColIndA, int blockDim, const cuComplex *x,
+               const cuComplex *beta, cuComplex *y) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int, int,
+      const cuComplex *, const cusparseMatDescr_t, const cuComplex *,
+      const int *, const int *, int, const cuComplex *, const cuComplex *,
+      cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrmv");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedRowPtrA, bsrSortedColIndA, blockDim,
+                  x, beta, y);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrmv(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nb, int nnzb,
+    const cuDoubleComplex *alpha, const cusparseMatDescr_t descrA,
+    const cuDoubleComplex *bsrSortedValA, const int *bsrSortedRowPtrA,
+    const int *bsrSortedColIndA, int blockDim, const cuDoubleComplex *x,
+    const cuDoubleComplex *beta, cuDoubleComplex *y) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int, int,
+      const cuDoubleComplex *, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, int,
+      const cuDoubleComplex *, const cuDoubleComplex *, cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrmv");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedRowPtrA, bsrSortedColIndA, blockDim,
+                  x, beta, y);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSbsrxmv(cusparseHandle_t handle, cusparseDirection_t dirA,
+                cusparseOperation_t transA, int sizeOfMask, int mb, int nb,
+                int nnzb, const float *alpha, const cusparseMatDescr_t descrA,
+                const float *bsrSortedValA, const int *bsrSortedMaskPtrA,
+                const int *bsrSortedRowPtrA, const int *bsrSortedEndPtrA,
+                const int *bsrSortedColIndA, int blockDim, const float *x,
+                const float *beta, float *y) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int, int,
+      int, const float *, const cusparseMatDescr_t, const float *, const int *,
+      const int *, const int *, const int *, int, const float *, const float *,
+      float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrxmv");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, sizeOfMask, mb, nb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedMaskPtrA, bsrSortedRowPtrA,
+                  bsrSortedEndPtrA, bsrSortedColIndA, blockDim, x, beta, y);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDbsrxmv(cusparseHandle_t handle, cusparseDirection_t dirA,
+                cusparseOperation_t transA, int sizeOfMask, int mb, int nb,
+                int nnzb, const double *alpha, const cusparseMatDescr_t descrA,
+                const double *bsrSortedValA, const int *bsrSortedMaskPtrA,
+                const int *bsrSortedRowPtrA, const int *bsrSortedEndPtrA,
+                const int *bsrSortedColIndA, int blockDim, const double *x,
+                const double *beta, double *y) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int, int,
+      int, const double *, const cusparseMatDescr_t, const double *,
+      const int *, const int *, const int *, const int *, int, const double *,
+      const double *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrxmv");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, sizeOfMask, mb, nb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedMaskPtrA, bsrSortedRowPtrA,
+                  bsrSortedEndPtrA, bsrSortedColIndA, blockDim, x, beta, y);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrxmv(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int sizeOfMask, int mb, int nb, int nnzb,
+    const cuComplex *alpha, const cusparseMatDescr_t descrA,
+    const cuComplex *bsrSortedValA, const int *bsrSortedMaskPtrA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedEndPtrA,
+    const int *bsrSortedColIndA, int blockDim, const cuComplex *x,
+    const cuComplex *beta, cuComplex *y) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int, int,
+      int, const cuComplex *, const cusparseMatDescr_t, const cuComplex *,
+      const int *, const int *, const int *, const int *, int,
+      const cuComplex *, const cuComplex *, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrxmv");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, sizeOfMask, mb, nb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedMaskPtrA, bsrSortedRowPtrA,
+                  bsrSortedEndPtrA, bsrSortedColIndA, blockDim, x, beta, y);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrxmv(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int sizeOfMask, int mb, int nb, int nnzb,
+    const cuDoubleComplex *alpha, const cusparseMatDescr_t descrA,
+    const cuDoubleComplex *bsrSortedValA, const int *bsrSortedMaskPtrA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedEndPtrA,
+    const int *bsrSortedColIndA, int blockDim, const cuDoubleComplex *x,
+    const cuDoubleComplex *beta, cuDoubleComplex *y) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int, int,
+      int, const cuDoubleComplex *, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, const int *,
+      const int *, int, const cuDoubleComplex *, const cuDoubleComplex *,
+      cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrxmv");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, sizeOfMask, mb, nb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedMaskPtrA, bsrSortedRowPtrA,
+                  bsrSortedEndPtrA, bsrSortedColIndA, blockDim, x, beta, y);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXbsrsv2_zeroPivot(cusparseHandle_t handle,
+                                                       bsrsv2Info_t info,
+                                                       int *position) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, bsrsv2Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXbsrsv2_zeroPivot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, position);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrsv2_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, float *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, float *, const int *, const int *, int,
+      bsrsv2Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrsv2_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrsv2_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, double *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, double *, const int *, const int *, int,
+      bsrsv2Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrsv2_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrsv2_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, cuComplex *, const int *, const int *, int,
+      bsrsv2Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrsv2_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrsv2_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, cuDoubleComplex *, const int *, const int *,
+      int, bsrsv2Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrsv2_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrsv2_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, float *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockSize,
+    bsrsv2Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, float *, const int *, const int *, int,
+      bsrsv2Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrsv2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockSize, info,
+                  pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrsv2_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, double *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockSize,
+    bsrsv2Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, double *, const int *, const int *, int,
+      bsrsv2Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrsv2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockSize, info,
+                  pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrsv2_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockSize,
+    bsrsv2Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, cuComplex *, const int *, const int *, int,
+      bsrsv2Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrsv2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockSize, info,
+                  pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrsv2_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockSize,
+    bsrsv2Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, cuDoubleComplex *, const int *, const int *,
+      int, bsrsv2Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrsv2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockSize, info,
+                  pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrsv2_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, const float *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, const float *, const int *, const int *, int,
+      bsrsv2Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrsv2_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info, policy,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrsv2_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, const double *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, const double *, const int *, const int *, int,
+      bsrsv2Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrsv2_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info, policy,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrsv2_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, const cuComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, const cuComplex *, const int *, const int *,
+      int, bsrsv2Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrsv2_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info, policy,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrsv2_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cusparseMatDescr_t, const cuDoubleComplex *, const int *,
+      const int *, int, bsrsv2Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrsv2_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info, policy,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrsv2_solve(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb, const float *alpha,
+    const cusparseMatDescr_t descrA, const float *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, const float *f, float *x, cusparseSolvePolicy_t policy,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const float *, const cusparseMatDescr_t, const float *, const int *,
+      const int *, int, bsrsv2Info_t, const float *, float *,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrsv2_solve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, alpha, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info, f, x,
+                  policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrsv2_solve(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb, const double *alpha,
+    const cusparseMatDescr_t descrA, const double *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, const double *f, double *x, cusparseSolvePolicy_t policy,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const double *, const cusparseMatDescr_t, const double *, const int *,
+      const int *, int, bsrsv2Info_t, const double *, double *,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrsv2_solve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, alpha, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info, f, x,
+                  policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrsv2_solve(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb, const cuComplex *alpha,
+    const cusparseMatDescr_t descrA, const cuComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, const cuComplex *f, cuComplex *x,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cuComplex *, const cusparseMatDescr_t, const cuComplex *,
+      const int *, const int *, int, bsrsv2Info_t, const cuComplex *,
+      cuComplex *, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrsv2_solve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, alpha, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info, f, x,
+                  policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrsv2_solve(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, int mb, int nnzb, const cuDoubleComplex *alpha,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    bsrsv2Info_t info, const cuDoubleComplex *f, cuDoubleComplex *x,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t, int, int,
+      const cuDoubleComplex *, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, int, bsrsv2Info_t,
+      const cuDoubleComplex *, cuDoubleComplex *, cusparseSolvePolicy_t,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrsv2_solve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, mb, nnzb, alpha, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, blockDim, info, f, x,
+                  policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrmm(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transB, int mb, int n,
+    int kb, int nnzb, const float *alpha, const cusparseMatDescr_t descrA,
+    const float *bsrSortedValA, const int *bsrSortedRowPtrA,
+    const int *bsrSortedColIndA, const int blockSize, const float *B,
+    const int ldb, const float *beta, float *C, int ldc) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, int, const float *,
+      const cusparseMatDescr_t, const float *, const int *, const int *,
+      const int, const float *, const int, const float *, float *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrmm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transB, mb, n, kb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedRowPtrA, bsrSortedColIndA, blockSize,
+                  B, ldb, beta, C, ldc);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrmm(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transB, int mb, int n,
+    int kb, int nnzb, const double *alpha, const cusparseMatDescr_t descrA,
+    const double *bsrSortedValA, const int *bsrSortedRowPtrA,
+    const int *bsrSortedColIndA, const int blockSize, const double *B,
+    const int ldb, const double *beta, double *C, int ldc) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, int, const double *,
+      const cusparseMatDescr_t, const double *, const int *, const int *,
+      const int, const double *, const int, const double *, double *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrmm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transB, mb, n, kb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedRowPtrA, bsrSortedColIndA, blockSize,
+                  B, ldb, beta, C, ldc);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrmm(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transB, int mb, int n,
+    int kb, int nnzb, const cuComplex *alpha, const cusparseMatDescr_t descrA,
+    const cuComplex *bsrSortedValA, const int *bsrSortedRowPtrA,
+    const int *bsrSortedColIndA, const int blockSize, const cuComplex *B,
+    const int ldb, const cuComplex *beta, cuComplex *C, int ldc) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, int, const cuComplex *,
+      const cusparseMatDescr_t, const cuComplex *, const int *, const int *,
+      const int, const cuComplex *, const int, const cuComplex *, cuComplex *,
+      int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrmm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transB, mb, n, kb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedRowPtrA, bsrSortedColIndA, blockSize,
+                  B, ldb, beta, C, ldc);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrmm(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transB, int mb, int n,
+    int kb, int nnzb, const cuDoubleComplex *alpha,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA,
+    const int blockSize, const cuDoubleComplex *B, const int ldb,
+    const cuDoubleComplex *beta, cuDoubleComplex *C, int ldc) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, int, const cuDoubleComplex *,
+      const cusparseMatDescr_t, const cuDoubleComplex *, const int *,
+      const int *, const int, const cuDoubleComplex *, const int,
+      const cuDoubleComplex *, cuDoubleComplex *, int);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrmm");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transB, mb, n, kb, nnzb, alpha, descrA,
+                  bsrSortedValA, bsrSortedRowPtrA, bsrSortedColIndA, blockSize,
+                  B, ldb, beta, C, ldc);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXbsrsm2_zeroPivot(cusparseHandle_t handle,
+                                                       bsrsm2Info_t info,
+                                                       int *position) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, bsrsm2Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXbsrsm2_zeroPivot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, position);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrsm2_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t, float *,
+      const int *, const int *, int, bsrsm2Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrsm2_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrsm2_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t, double *,
+      const int *, const int *, int, bsrsm2Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrsm2_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrsm2_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t, cuComplex *,
+      const int *, const int *, int, bsrsm2Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrsm2_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrsm2_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, const int *, int, bsrsm2Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrsm2_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrsm2_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transB, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t, float *,
+      const int *, const int *, int, bsrsm2Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrsm2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transB, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrsm2_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transB, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t, double *,
+      const int *, const int *, int, bsrsm2Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrsm2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transB, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrsm2_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transB, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t, cuComplex *,
+      const int *, const int *, int, bsrsm2Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrsm2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transB, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrsm2_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transB, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, const int *, int, bsrsm2Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrsm2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transB, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrsm2_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, const float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t,
+      const float *, const int *, const int *, int, bsrsm2Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrsm2_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrsm2_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, const double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t,
+      const double *, const int *, const int *, int, bsrsm2Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrsm2_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrsm2_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA, const cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrsm2Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t,
+      const cuComplex *, const int *, const int *, int, bsrsm2Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrsm2_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrsm2_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const cusparseMatDescr_t descrA,
+    const cuDoubleComplex *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int blockSize, bsrsm2Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, int, bsrsm2Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrsm2_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrsm2_solve(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const float *alpha, const cusparseMatDescr_t descrA,
+    const float *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int blockSize, bsrsm2Info_t info,
+    const float *B, int ldb, float *X, int ldx, cusparseSolvePolicy_t policy,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const float *,
+      const cusparseMatDescr_t, const float *, const int *, const int *, int,
+      bsrsm2Info_t, const float *, int, float *, int, cusparseSolvePolicy_t,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrsm2_solve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, alpha, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, B, ldb, X, ldx, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrsm2_solve(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const double *alpha, const cusparseMatDescr_t descrA,
+    const double *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int blockSize, bsrsm2Info_t info,
+    const double *B, int ldb, double *X, int ldx, cusparseSolvePolicy_t policy,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const double *,
+      const cusparseMatDescr_t, const double *, const int *, const int *, int,
+      bsrsm2Info_t, const double *, int, double *, int, cusparseSolvePolicy_t,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrsm2_solve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, alpha, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, B, ldb, X, ldx, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrsm2_solve(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const cuComplex *alpha, const cusparseMatDescr_t descrA,
+    const cuComplex *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int blockSize, bsrsm2Info_t info,
+    const cuComplex *B, int ldb, cuComplex *X, int ldx,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cuComplex *,
+      const cusparseMatDescr_t, const cuComplex *, const int *, const int *,
+      int, bsrsm2Info_t, const cuComplex *, int, cuComplex *, int,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrsm2_solve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, alpha, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, B, ldb, X, ldx, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrsm2_solve(
+    cusparseHandle_t handle, cusparseDirection_t dirA,
+    cusparseOperation_t transA, cusparseOperation_t transXY, int mb, int n,
+    int nnzb, const cuDoubleComplex *alpha, const cusparseMatDescr_t descrA,
+    const cuDoubleComplex *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int blockSize, bsrsm2Info_t info,
+    const cuDoubleComplex *B, int ldb, cuDoubleComplex *X, int ldx,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, cusparseOperation_t,
+      cusparseOperation_t, int, int, int, const cuDoubleComplex *,
+      const cusparseMatDescr_t, const cuDoubleComplex *, const int *,
+      const int *, int, bsrsm2Info_t, const cuDoubleComplex *, int,
+      cuDoubleComplex *, int, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrsm2_solve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, transA, transXY, mb, n, nnzb, alpha, descrA,
+                  bsrSortedVal, bsrSortedRowPtr, bsrSortedColInd, blockSize,
+                  info, B, ldb, X, ldx, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsrilu02_numericBoost(
+    cusparseHandle_t handle, csrilu02Info_t info, int enable_boost, double *tol,
+    float *boost_val) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, csrilu02Info_t, int, double *, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsrilu02_numericBoost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, enable_boost, tol, boost_val);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsrilu02_numericBoost(
+    cusparseHandle_t handle, csrilu02Info_t info, int enable_boost, double *tol,
+    double *boost_val) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, csrilu02Info_t, int, double *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsrilu02_numericBoost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, enable_boost, tol, boost_val);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsrilu02_numericBoost(
+    cusparseHandle_t handle, csrilu02Info_t info, int enable_boost, double *tol,
+    cuComplex *boost_val) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, csrilu02Info_t, int, double *, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsrilu02_numericBoost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, enable_boost, tol, boost_val);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsrilu02_numericBoost(
+    cusparseHandle_t handle, csrilu02Info_t info, int enable_boost, double *tol,
+    cuDoubleComplex *boost_val) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, csrilu02Info_t, int, double *, cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsrilu02_numericBoost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, enable_boost, tol, boost_val);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcsrilu02_zeroPivot(
+    cusparseHandle_t handle, csrilu02Info_t info, int *position) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, csrilu02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcsrilu02_zeroPivot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, position);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsrilu02_bufferSize(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    float *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, float *,
+      const int *, const int *, csrilu02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsrilu02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsrilu02_bufferSize(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    double *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, double *,
+      const int *, const int *, csrilu02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsrilu02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsrilu02_bufferSize(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuComplex *,
+      const int *, const int *, csrilu02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsrilu02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsrilu02_bufferSize(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuDoubleComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuDoubleComplex *,
+      const int *, const int *, csrilu02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsrilu02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsrilu02_bufferSizeExt(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    float *csrSortedVal, const int *csrSortedRowPtr, const int *csrSortedColInd,
+    csrilu02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, float *,
+      const int *, const int *, csrilu02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsrilu02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedVal, csrSortedRowPtr,
+                  csrSortedColInd, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsrilu02_bufferSizeExt(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    double *csrSortedVal, const int *csrSortedRowPtr,
+    const int *csrSortedColInd, csrilu02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, double *,
+      const int *, const int *, csrilu02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsrilu02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedVal, csrSortedRowPtr,
+                  csrSortedColInd, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsrilu02_bufferSizeExt(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuComplex *csrSortedVal, const int *csrSortedRowPtr,
+    const int *csrSortedColInd, csrilu02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuComplex *,
+      const int *, const int *, csrilu02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsrilu02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedVal, csrSortedRowPtr,
+                  csrSortedColInd, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsrilu02_bufferSizeExt(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuDoubleComplex *csrSortedVal, const int *csrSortedRowPtr,
+    const int *csrSortedColInd, csrilu02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuDoubleComplex *,
+      const int *, const int *, csrilu02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsrilu02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedVal, csrSortedRowPtr,
+                  csrSortedColInd, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsrilu02_analysis(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const float *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const float *,
+      const int *, const int *, csrilu02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsrilu02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsrilu02_analysis(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const double *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const double *,
+      const int *, const int *, csrilu02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsrilu02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsrilu02_analysis(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const cuComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const cuComplex *,
+      const int *, const int *, csrilu02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsrilu02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsrilu02_analysis(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const cuDoubleComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, csrilu02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsrilu02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsrilu02(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    float *csrSortedValA_valM, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, float *,
+      const int *, const int *, csrilu02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsrilu02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA_valM, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsrilu02(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    double *csrSortedValA_valM, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, double *,
+      const int *, const int *, csrilu02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsrilu02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA_valM, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsrilu02(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuComplex *csrSortedValA_valM, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuComplex *,
+      const int *, const int *, csrilu02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsrilu02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA_valM, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsrilu02(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuDoubleComplex *csrSortedValA_valM, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csrilu02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuDoubleComplex *,
+      const int *, const int *, csrilu02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsrilu02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA_valM, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrilu02_numericBoost(
+    cusparseHandle_t handle, bsrilu02Info_t info, int enable_boost, double *tol,
+    float *boost_val) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, bsrilu02Info_t, int, double *, float *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrilu02_numericBoost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, enable_boost, tol, boost_val);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrilu02_numericBoost(
+    cusparseHandle_t handle, bsrilu02Info_t info, int enable_boost, double *tol,
+    double *boost_val) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, bsrilu02Info_t, int, double *, double *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrilu02_numericBoost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, enable_boost, tol, boost_val);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrilu02_numericBoost(
+    cusparseHandle_t handle, bsrilu02Info_t info, int enable_boost, double *tol,
+    cuComplex *boost_val) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, bsrilu02Info_t, int, double *, cuComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrilu02_numericBoost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, enable_boost, tol, boost_val);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrilu02_numericBoost(
+    cusparseHandle_t handle, bsrilu02Info_t info, int enable_boost, double *tol,
+    cuDoubleComplex *boost_val) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, bsrilu02Info_t, int, double *, cuDoubleComplex *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrilu02_numericBoost");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, enable_boost, tol, boost_val);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXbsrilu02_zeroPivot(
+    cusparseHandle_t handle, bsrilu02Info_t info, int *position) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, bsrilu02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXbsrilu02_zeroPivot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, position);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrilu02_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      float *, const int *, const int *, int, bsrilu02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrilu02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrilu02_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      double *, const int *, const int *, int, bsrilu02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrilu02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrilu02_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuComplex *, const int *, const int *, int, bsrilu02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrilu02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrilu02_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, const int *, int, bsrilu02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrilu02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrilu02_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrilu02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      float *, const int *, const int *, int, bsrilu02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrilu02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockSize, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrilu02_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrilu02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      double *, const int *, const int *, int, bsrilu02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrilu02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockSize, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrilu02_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrilu02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuComplex *, const int *, const int *, int, bsrilu02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrilu02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockSize, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrilu02_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsrilu02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, const int *, int, bsrilu02Info_t,
+      size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrilu02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockSize, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrilu02_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      float *, const int *, const int *, int, bsrilu02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrilu02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrilu02_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      double *, const int *, const int *, int, bsrilu02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrilu02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrilu02_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuComplex *, const int *, const int *, int, bsrilu02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrilu02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrilu02_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, const int *, int, bsrilu02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrilu02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsrilu02(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      float *, const int *, const int *, int, bsrilu02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsrilu02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsrilu02(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      double *, const int *, const int *, int, bsrilu02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsrilu02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsrilu02(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuComplex *, const int *, const int *, int, bsrilu02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsrilu02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsrilu02(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsrilu02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, const int *, int, bsrilu02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsrilu02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcsric02_zeroPivot(cusparseHandle_t handle,
+                                                        csric02Info_t info,
+                                                        int *position) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, csric02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcsric02_zeroPivot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, position);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsric02_bufferSize(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    float *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, float *,
+      const int *, const int *, csric02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsric02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsric02_bufferSize(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    double *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, double *,
+      const int *, const int *, csric02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsric02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsric02_bufferSize(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuComplex *,
+      const int *, const int *, csric02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsric02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsric02_bufferSize(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuDoubleComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuDoubleComplex *,
+      const int *, const int *, csric02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsric02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsric02_bufferSizeExt(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    float *csrSortedVal, const int *csrSortedRowPtr, const int *csrSortedColInd,
+    csric02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, float *,
+      const int *, const int *, csric02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsric02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedVal, csrSortedRowPtr,
+                  csrSortedColInd, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsric02_bufferSizeExt(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    double *csrSortedVal, const int *csrSortedRowPtr,
+    const int *csrSortedColInd, csric02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, double *,
+      const int *, const int *, csric02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsric02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedVal, csrSortedRowPtr,
+                  csrSortedColInd, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsric02_bufferSizeExt(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuComplex *csrSortedVal, const int *csrSortedRowPtr,
+    const int *csrSortedColInd, csric02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuComplex *,
+      const int *, const int *, csric02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsric02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedVal, csrSortedRowPtr,
+                  csrSortedColInd, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsric02_bufferSizeExt(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuDoubleComplex *csrSortedVal, const int *csrSortedRowPtr,
+    const int *csrSortedColInd, csric02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuDoubleComplex *,
+      const int *, const int *, csric02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsric02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedVal, csrSortedRowPtr,
+                  csrSortedColInd, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsric02_analysis(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const float *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const float *,
+      const int *, const int *, csric02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsric02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsric02_analysis(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const double *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const double *,
+      const int *, const int *, csric02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsric02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsric02_analysis(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const cuComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const cuComplex *,
+      const int *, const int *, csric02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsric02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsric02_analysis(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const cuDoubleComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, csric02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsric02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsric02(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    float *csrSortedValA_valM, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, float *,
+      const int *, const int *, csric02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsric02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA_valM, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsric02(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    double *csrSortedValA_valM, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, double *,
+      const int *, const int *, csric02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsric02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA_valM, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsric02(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuComplex *csrSortedValA_valM, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuComplex *,
+      const int *, const int *, csric02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsric02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA_valM, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsric02(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    cuDoubleComplex *csrSortedValA_valM, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, csric02Info_t info,
+    cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, cuDoubleComplex *,
+      const int *, const int *, csric02Info_t, cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsric02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA_valM, csrSortedRowPtrA,
+                  csrSortedColIndA, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXbsric02_zeroPivot(cusparseHandle_t handle,
+                                                        bsric02Info_t info,
+                                                        int *position) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, bsric02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXbsric02_zeroPivot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, info, position);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsric02_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      float *, const int *, const int *, int, bsric02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsric02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsric02_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      double *, const int *, const int *, int, bsric02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsric02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsric02_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuComplex *, const int *, const int *, int, bsric02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsric02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsric02_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, const int *, int, bsric02Info_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsric02_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsric02_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsric02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      float *, const int *, const int *, int, bsric02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsric02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockSize, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsric02_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsric02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      double *, const int *, const int *, int, bsric02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsric02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockSize, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsric02_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsric02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuComplex *, const int *, const int *, int, bsric02Info_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsric02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockSize, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsric02_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockSize,
+    bsric02Info_t info, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, const int *, int, bsric02Info_t,
+      size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsric02_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockSize, info, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsric02_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, const float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, cusparseSolvePolicy_t policy, void *pInputBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const float *, const int *, const int *, int, bsric02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsric02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pInputBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsric02_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, const double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, cusparseSolvePolicy_t policy, void *pInputBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const double *, const int *, const int *, int, bsric02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsric02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pInputBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsric02_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, const cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, cusparseSolvePolicy_t policy, void *pInputBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuComplex *, const int *, const int *, int, bsric02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsric02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pInputBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsric02_analysis(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, cusparseSolvePolicy_t policy, void *pInputBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, int, bsric02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsric02_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pInputBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsric02(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, float *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      float *, const int *, const int *, int, bsric02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsric02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsric02(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, double *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      double *, const int *, const int *, int, bsric02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsric02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsric02(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuComplex *, const int *, const int *, int, bsric02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsric02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsric02(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nnzb,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *bsrSortedVal,
+    const int *bsrSortedRowPtr, const int *bsrSortedColInd, int blockDim,
+    bsric02Info_t info, cusparseSolvePolicy_t policy, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, const int *, int, bsric02Info_t,
+      cusparseSolvePolicy_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsric02");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nnzb, descrA, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, blockDim, info, policy, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgtsv2_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const float *dl, const float *d,
+    const float *du, const float *B, int ldb, size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, const float *, const float *,
+      const float *, int, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgtsv2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgtsv2_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const double *dl, const double *d,
+    const double *du, const double *B, int ldb, size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, const double *,
+      const double *, const double *, int, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgtsv2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgtsv2_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const cuComplex *dl,
+    const cuComplex *d, const cuComplex *du, const cuComplex *B, int ldb,
+    size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuComplex *, const cuComplex *,
+      const cuComplex *, const cuComplex *, int, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgtsv2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgtsv2_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const cuDoubleComplex *dl,
+    const cuDoubleComplex *d, const cuDoubleComplex *du,
+    const cuDoubleComplex *B, int ldb, size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, const cuDoubleComplex *, const cuDoubleComplex *,
+      int, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgtsv2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgtsv2(cusparseHandle_t handle, int m,
+                                            int n, const float *dl,
+                                            const float *d, const float *du,
+                                            float *B, int ldb, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, const float *, const float *,
+      float *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgtsv2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgtsv2(cusparseHandle_t handle, int m,
+                                            int n, const double *dl,
+                                            const double *d, const double *du,
+                                            double *B, int ldb, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, const double *,
+      const double *, double *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgtsv2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgtsv2(cusparseHandle_t handle, int m,
+                                            int n, const cuComplex *dl,
+                                            const cuComplex *d,
+                                            const cuComplex *du, cuComplex *B,
+                                            int ldb, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuComplex *, const cuComplex *,
+      const cuComplex *, cuComplex *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgtsv2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgtsv2(cusparseHandle_t handle, int m,
+                                            int n, const cuDoubleComplex *dl,
+                                            const cuDoubleComplex *d,
+                                            const cuDoubleComplex *du,
+                                            cuDoubleComplex *B, int ldb,
+                                            void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, const cuDoubleComplex *, cuDoubleComplex *, int,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgtsv2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgtsv2_nopivot_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const float *dl, const float *d,
+    const float *du, const float *B, int ldb, size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, const float *, const float *,
+      const float *, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSgtsv2_nopivot_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgtsv2_nopivot_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const double *dl, const double *d,
+    const double *du, const double *B, int ldb, size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, const double *,
+      const double *, const double *, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDgtsv2_nopivot_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgtsv2_nopivot_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const cuComplex *dl,
+    const cuComplex *d, const cuComplex *du, const cuComplex *B, int ldb,
+    size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuComplex *, const cuComplex *,
+      const cuComplex *, const cuComplex *, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseCgtsv2_nopivot_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgtsv2_nopivot_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const cuDoubleComplex *dl,
+    const cuDoubleComplex *d, const cuDoubleComplex *du,
+    const cuDoubleComplex *B, int ldb, size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, const cuDoubleComplex *, const cuDoubleComplex *,
+      int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseZgtsv2_nopivot_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgtsv2_nopivot(
+    cusparseHandle_t handle, int m, int n, const float *dl, const float *d,
+    const float *du, float *B, int ldb, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, const float *, const float *,
+      float *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgtsv2_nopivot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgtsv2_nopivot(
+    cusparseHandle_t handle, int m, int n, const double *dl, const double *d,
+    const double *du, double *B, int ldb, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, const double *,
+      const double *, double *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgtsv2_nopivot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgtsv2_nopivot(
+    cusparseHandle_t handle, int m, int n, const cuComplex *dl,
+    const cuComplex *d, const cuComplex *du, cuComplex *B, int ldb,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuComplex *, const cuComplex *,
+      const cuComplex *, cuComplex *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgtsv2_nopivot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgtsv2_nopivot(
+    cusparseHandle_t handle, int m, int n, const cuDoubleComplex *dl,
+    const cuDoubleComplex *d, const cuDoubleComplex *du, cuDoubleComplex *B,
+    int ldb, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, const cuDoubleComplex *, cuDoubleComplex *, int,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgtsv2_nopivot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, dl, d, du, B, ldb, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgtsv2StridedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int m, const float *dl, const float *d,
+    const float *du, const float *x, int batchCount, int batchStride,
+    size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const float *, const float *, const float *,
+      const float *, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSgtsv2StridedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, dl, d, du, x, batchCount, batchStride,
+                  bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgtsv2StridedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int m, const double *dl, const double *d,
+    const double *du, const double *x, int batchCount, int batchStride,
+    size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const double *, const double *, const double *,
+      const double *, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDgtsv2StridedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, dl, d, du, x, batchCount, batchStride,
+                  bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgtsv2StridedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int m, const cuComplex *dl, const cuComplex *d,
+    const cuComplex *du, const cuComplex *x, int batchCount, int batchStride,
+    size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const cuComplex *, const cuComplex *,
+      const cuComplex *, const cuComplex *, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseCgtsv2StridedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, dl, d, du, x, batchCount, batchStride,
+                  bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgtsv2StridedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int m, const cuDoubleComplex *dl,
+    const cuDoubleComplex *d, const cuDoubleComplex *du,
+    const cuDoubleComplex *x, int batchCount, int batchStride,
+    size_t *bufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const cuDoubleComplex *, const cuDoubleComplex *,
+      const cuDoubleComplex *, const cuDoubleComplex *, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseZgtsv2StridedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, dl, d, du, x, batchCount, batchStride,
+                  bufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgtsv2StridedBatch(
+    cusparseHandle_t handle, int m, const float *dl, const float *d,
+    const float *du, float *x, int batchCount, int batchStride, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const float *, const float *, const float *,
+      float *, int, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgtsv2StridedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, dl, d, du, x, batchCount, batchStride, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDgtsv2StridedBatch(cusparseHandle_t handle, int m, const double *dl,
+                           const double *d, const double *du, double *x,
+                           int batchCount, int batchStride, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const double *, const double *, const double *,
+      double *, int, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgtsv2StridedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, dl, d, du, x, batchCount, batchStride, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgtsv2StridedBatch(
+    cusparseHandle_t handle, int m, const cuComplex *dl, const cuComplex *d,
+    const cuComplex *du, cuComplex *x, int batchCount, int batchStride,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const cuComplex *, const cuComplex *,
+      const cuComplex *, cuComplex *, int, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgtsv2StridedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, dl, d, du, x, batchCount, batchStride, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgtsv2StridedBatch(
+    cusparseHandle_t handle, int m, const cuDoubleComplex *dl,
+    const cuDoubleComplex *d, const cuDoubleComplex *du, cuDoubleComplex *x,
+    int batchCount, int batchStride, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const cuDoubleComplex *, const cuDoubleComplex *,
+      const cuDoubleComplex *, cuDoubleComplex *, int, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgtsv2StridedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, dl, d, du, x, batchCount, batchStride, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgtsvInterleavedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int algo, int m, const float *dl, const float *d,
+    const float *du, const float *x, int batchCount,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, const float *, const float *,
+      const float *, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSgtsvInterleavedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, dl, d, du, x, batchCount,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgtsvInterleavedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int algo, int m, const double *dl, const double *d,
+    const double *du, const double *x, int batchCount,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, const double *,
+      const double *, const double *, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDgtsvInterleavedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, dl, d, du, x, batchCount,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgtsvInterleavedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int algo, int m, const cuComplex *dl,
+    const cuComplex *d, const cuComplex *du, const cuComplex *x, int batchCount,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuComplex *, const cuComplex *,
+      const cuComplex *, const cuComplex *, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseCgtsvInterleavedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, dl, d, du, x, batchCount,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgtsvInterleavedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int algo, int m, const cuDoubleComplex *dl,
+    const cuDoubleComplex *d, const cuDoubleComplex *du,
+    const cuDoubleComplex *x, int batchCount, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, const cuDoubleComplex *, const cuDoubleComplex *,
+      int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseZgtsvInterleavedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, dl, d, du, x, batchCount,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgtsvInterleavedBatch(
+    cusparseHandle_t handle, int algo, int m, float *dl, float *d, float *du,
+    float *x, int batchCount, void *pBuffer) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, int, int, float *,
+                                      float *, float *, float *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgtsvInterleavedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, dl, d, du, x, batchCount, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgtsvInterleavedBatch(
+    cusparseHandle_t handle, int algo, int m, double *dl, double *d, double *du,
+    double *x, int batchCount, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, int, int,
+                                                  double *, double *, double *,
+                                                  double *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgtsvInterleavedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, dl, d, du, x, batchCount, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgtsvInterleavedBatch(
+    cusparseHandle_t handle, int algo, int m, cuComplex *dl, cuComplex *d,
+    cuComplex *du, cuComplex *x, int batchCount, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, cuComplex *, cuComplex *, cuComplex *,
+      cuComplex *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgtsvInterleavedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, dl, d, du, x, batchCount, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgtsvInterleavedBatch(
+    cusparseHandle_t handle, int algo, int m, cuDoubleComplex *dl,
+    cuDoubleComplex *d, cuDoubleComplex *du, cuDoubleComplex *x, int batchCount,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, cuDoubleComplex *, cuDoubleComplex *,
+      cuDoubleComplex *, cuDoubleComplex *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgtsvInterleavedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, dl, d, du, x, batchCount, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgpsvInterleavedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int algo, int m, const float *ds, const float *dl,
+    const float *d, const float *du, const float *dw, const float *x,
+    int batchCount, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, const float *, const float *,
+      const float *, const float *, const float *, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSgpsvInterleavedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, ds, dl, d, du, dw, x, batchCount,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgpsvInterleavedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int algo, int m, const double *ds,
+    const double *dl, const double *d, const double *du, const double *dw,
+    const double *x, int batchCount, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, const double *,
+      const double *, const double *, const double *, const double *, int,
+      size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDgpsvInterleavedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, ds, dl, d, du, dw, x, batchCount,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgpsvInterleavedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int algo, int m, const cuComplex *ds,
+    const cuComplex *dl, const cuComplex *d, const cuComplex *du,
+    const cuComplex *dw, const cuComplex *x, int batchCount,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuComplex *, const cuComplex *,
+      const cuComplex *, const cuComplex *, const cuComplex *,
+      const cuComplex *, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseCgpsvInterleavedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, ds, dl, d, du, dw, x, batchCount,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgpsvInterleavedBatch_bufferSizeExt(
+    cusparseHandle_t handle, int algo, int m, const cuDoubleComplex *ds,
+    const cuDoubleComplex *dl, const cuDoubleComplex *d,
+    const cuDoubleComplex *du, const cuDoubleComplex *dw,
+    const cuDoubleComplex *x, int batchCount, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuDoubleComplex *,
+      const cuDoubleComplex *, const cuDoubleComplex *, const cuDoubleComplex *,
+      const cuDoubleComplex *, const cuDoubleComplex *, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseZgpsvInterleavedBatch_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, ds, dl, d, du, dw, x, batchCount,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgpsvInterleavedBatch(
+    cusparseHandle_t handle, int algo, int m, float *ds, float *dl, float *d,
+    float *du, float *dw, float *x, int batchCount, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, float *, float *, float *, float *, float *,
+      float *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgpsvInterleavedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, ds, dl, d, du, dw, x, batchCount, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgpsvInterleavedBatch(
+    cusparseHandle_t handle, int algo, int m, double *ds, double *dl, double *d,
+    double *du, double *dw, double *x, int batchCount, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, double *, double *, double *, double *,
+      double *, double *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgpsvInterleavedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, ds, dl, d, du, dw, x, batchCount, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgpsvInterleavedBatch(
+    cusparseHandle_t handle, int algo, int m, cuComplex *ds, cuComplex *dl,
+    cuComplex *d, cuComplex *du, cuComplex *dw, cuComplex *x, int batchCount,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, cuComplex *, cuComplex *, cuComplex *,
+      cuComplex *, cuComplex *, cuComplex *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgpsvInterleavedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, ds, dl, d, du, dw, x, batchCount, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgpsvInterleavedBatch(
+    cusparseHandle_t handle, int algo, int m, cuDoubleComplex *ds,
+    cuDoubleComplex *dl, cuDoubleComplex *d, cuDoubleComplex *du,
+    cuDoubleComplex *dw, cuDoubleComplex *x, int batchCount, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, cuDoubleComplex *, cuDoubleComplex *,
+      cuDoubleComplex *, cuDoubleComplex *, cuDoubleComplex *,
+      cuDoubleComplex *, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgpsvInterleavedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, algo, m, ds, dl, d, du, dw, x, batchCount, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsrgeam2_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const float *alpha,
+    const cusparseMatDescr_t descrA, int nnzA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, const float *beta,
+    const cusparseMatDescr_t descrB, int nnzB, const float *csrSortedValB,
+    const int *csrSortedRowPtrB, const int *csrSortedColIndB,
+    const cusparseMatDescr_t descrC, const float *csrSortedValC,
+    const int *csrSortedRowPtrC, const int *csrSortedColIndC,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, const cusparseMatDescr_t, int,
+      const float *, const int *, const int *, const float *,
+      const cusparseMatDescr_t, int, const float *, const int *, const int *,
+      const cusparseMatDescr_t, const float *, const int *, const int *,
+      size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsrgeam2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, descrA, nnzA, csrSortedValA,
+                  csrSortedRowPtrA, csrSortedColIndA, beta, descrB, nnzB,
+                  csrSortedValB, csrSortedRowPtrB, csrSortedColIndB, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsrgeam2_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const double *alpha,
+    const cusparseMatDescr_t descrA, int nnzA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const double *beta, const cusparseMatDescr_t descrB, int nnzB,
+    const double *csrSortedValB, const int *csrSortedRowPtrB,
+    const int *csrSortedColIndB, const cusparseMatDescr_t descrC,
+    const double *csrSortedValC, const int *csrSortedRowPtrC,
+    const int *csrSortedColIndC, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, const cusparseMatDescr_t, int,
+      const double *, const int *, const int *, const double *,
+      const cusparseMatDescr_t, int, const double *, const int *, const int *,
+      const cusparseMatDescr_t, const double *, const int *, const int *,
+      size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsrgeam2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, descrA, nnzA, csrSortedValA,
+                  csrSortedRowPtrA, csrSortedColIndA, beta, descrB, nnzB,
+                  csrSortedValB, csrSortedRowPtrB, csrSortedColIndB, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsrgeam2_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const cuComplex *alpha,
+    const cusparseMatDescr_t descrA, int nnzA, const cuComplex *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const cuComplex *beta, const cusparseMatDescr_t descrB, int nnzB,
+    const cuComplex *csrSortedValB, const int *csrSortedRowPtrB,
+    const int *csrSortedColIndB, const cusparseMatDescr_t descrC,
+    const cuComplex *csrSortedValC, const int *csrSortedRowPtrC,
+    const int *csrSortedColIndC, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuComplex *, const cusparseMatDescr_t,
+      int, const cuComplex *, const int *, const int *, const cuComplex *,
+      const cusparseMatDescr_t, int, const cuComplex *, const int *,
+      const int *, const cusparseMatDescr_t, const cuComplex *, const int *,
+      const int *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsrgeam2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, descrA, nnzA, csrSortedValA,
+                  csrSortedRowPtrA, csrSortedColIndA, beta, descrB, nnzB,
+                  csrSortedValB, csrSortedRowPtrB, csrSortedColIndB, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsrgeam2_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const cuDoubleComplex *alpha,
+    const cusparseMatDescr_t descrA, int nnzA,
+    const cuDoubleComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, const cuDoubleComplex *beta,
+    const cusparseMatDescr_t descrB, int nnzB,
+    const cuDoubleComplex *csrSortedValB, const int *csrSortedRowPtrB,
+    const int *csrSortedColIndB, const cusparseMatDescr_t descrC,
+    const cuDoubleComplex *csrSortedValC, const int *csrSortedRowPtrC,
+    const int *csrSortedColIndC, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuDoubleComplex *,
+      const cusparseMatDescr_t, int, const cuDoubleComplex *, const int *,
+      const int *, const cuDoubleComplex *, const cusparseMatDescr_t, int,
+      const cuDoubleComplex *, const int *, const int *,
+      const cusparseMatDescr_t, const cuDoubleComplex *, const int *,
+      const int *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsrgeam2_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, descrA, nnzA, csrSortedValA,
+                  csrSortedRowPtrA, csrSortedColIndA, beta, descrB, nnzB,
+                  csrSortedValB, csrSortedRowPtrB, csrSortedColIndB, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcsrgeam2Nnz(
+    cusparseHandle_t handle, int m, int n, const cusparseMatDescr_t descrA,
+    int nnzA, const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const cusparseMatDescr_t descrB, int nnzB, const int *csrSortedRowPtrB,
+    const int *csrSortedColIndB, const cusparseMatDescr_t descrC,
+    int *csrSortedRowPtrC, int *nnzTotalDevHostPtr, void *workspace) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, int, const int *,
+      const int *, const cusparseMatDescr_t, int, const int *, const int *,
+      const cusparseMatDescr_t, int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcsrgeam2Nnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, descrA, nnzA, csrSortedRowPtrA,
+                  csrSortedColIndA, descrB, nnzB, csrSortedRowPtrB,
+                  csrSortedColIndB, descrC, csrSortedRowPtrC,
+                  nnzTotalDevHostPtr, workspace);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsrgeam2(
+    cusparseHandle_t handle, int m, int n, const float *alpha,
+    const cusparseMatDescr_t descrA, int nnzA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, const float *beta,
+    const cusparseMatDescr_t descrB, int nnzB, const float *csrSortedValB,
+    const int *csrSortedRowPtrB, const int *csrSortedColIndB,
+    const cusparseMatDescr_t descrC, float *csrSortedValC,
+    int *csrSortedRowPtrC, int *csrSortedColIndC, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, const cusparseMatDescr_t, int,
+      const float *, const int *, const int *, const float *,
+      const cusparseMatDescr_t, int, const float *, const int *, const int *,
+      const cusparseMatDescr_t, float *, int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsrgeam2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, descrA, nnzA, csrSortedValA,
+                  csrSortedRowPtrA, csrSortedColIndA, beta, descrB, nnzB,
+                  csrSortedValB, csrSortedRowPtrB, csrSortedColIndB, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsrgeam2(
+    cusparseHandle_t handle, int m, int n, const double *alpha,
+    const cusparseMatDescr_t descrA, int nnzA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const double *beta, const cusparseMatDescr_t descrB, int nnzB,
+    const double *csrSortedValB, const int *csrSortedRowPtrB,
+    const int *csrSortedColIndB, const cusparseMatDescr_t descrC,
+    double *csrSortedValC, int *csrSortedRowPtrC, int *csrSortedColIndC,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, const cusparseMatDescr_t, int,
+      const double *, const int *, const int *, const double *,
+      const cusparseMatDescr_t, int, const double *, const int *, const int *,
+      const cusparseMatDescr_t, double *, int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsrgeam2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, descrA, nnzA, csrSortedValA,
+                  csrSortedRowPtrA, csrSortedColIndA, beta, descrB, nnzB,
+                  csrSortedValB, csrSortedRowPtrB, csrSortedColIndB, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsrgeam2(
+    cusparseHandle_t handle, int m, int n, const cuComplex *alpha,
+    const cusparseMatDescr_t descrA, int nnzA, const cuComplex *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const cuComplex *beta, const cusparseMatDescr_t descrB, int nnzB,
+    const cuComplex *csrSortedValB, const int *csrSortedRowPtrB,
+    const int *csrSortedColIndB, const cusparseMatDescr_t descrC,
+    cuComplex *csrSortedValC, int *csrSortedRowPtrC, int *csrSortedColIndC,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuComplex *, const cusparseMatDescr_t,
+      int, const cuComplex *, const int *, const int *, const cuComplex *,
+      const cusparseMatDescr_t, int, const cuComplex *, const int *,
+      const int *, const cusparseMatDescr_t, cuComplex *, int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsrgeam2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, descrA, nnzA, csrSortedValA,
+                  csrSortedRowPtrA, csrSortedColIndA, beta, descrB, nnzB,
+                  csrSortedValB, csrSortedRowPtrB, csrSortedColIndB, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsrgeam2(
+    cusparseHandle_t handle, int m, int n, const cuDoubleComplex *alpha,
+    const cusparseMatDescr_t descrA, int nnzA,
+    const cuDoubleComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, const cuDoubleComplex *beta,
+    const cusparseMatDescr_t descrB, int nnzB,
+    const cuDoubleComplex *csrSortedValB, const int *csrSortedRowPtrB,
+    const int *csrSortedColIndB, const cusparseMatDescr_t descrC,
+    cuDoubleComplex *csrSortedValC, int *csrSortedRowPtrC,
+    int *csrSortedColIndC, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cuDoubleComplex *,
+      const cusparseMatDescr_t, int, const cuDoubleComplex *, const int *,
+      const int *, const cuDoubleComplex *, const cusparseMatDescr_t, int,
+      const cuDoubleComplex *, const int *, const int *,
+      const cusparseMatDescr_t, cuDoubleComplex *, int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsrgeam2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, alpha, descrA, nnzA, csrSortedValA,
+                  csrSortedRowPtrA, csrSortedColIndA, beta, descrB, nnzB,
+                  csrSortedValB, csrSortedRowPtrB, csrSortedColIndB, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsrcolor(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const float *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, const float *fractionToColor, int *ncolors,
+    int *coloring, int *reordering, const cusparseColorInfo_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const float *,
+      const int *, const int *, const float *, int *, int *, int *,
+      const cusparseColorInfo_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsrcolor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, fractionToColor, ncolors, coloring,
+                  reordering, info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsrcolor(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const double *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, const double *fractionToColor, int *ncolors,
+    int *coloring, int *reordering, const cusparseColorInfo_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const double *,
+      const int *, const int *, const double *, int *, int *, int *,
+      const cusparseColorInfo_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsrcolor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, fractionToColor, ncolors, coloring,
+                  reordering, info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsrcolor(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const cuComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, const float *fractionToColor, int *ncolors,
+    int *coloring, int *reordering, const cusparseColorInfo_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const cuComplex *,
+      const int *, const int *, const float *, int *, int *, int *,
+      const cusparseColorInfo_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsrcolor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, fractionToColor, ncolors, coloring,
+                  reordering, info);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsrcolor(
+    cusparseHandle_t handle, int m, int nnz, const cusparseMatDescr_t descrA,
+    const cuDoubleComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, const double *fractionToColor, int *ncolors,
+    int *coloring, int *reordering, const cusparseColorInfo_t info) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, const double *, int *,
+      int *, int *, const cusparseColorInfo_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsrcolor");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, nnz, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, fractionToColor, ncolors, coloring,
+                  reordering, info);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSnnz(cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+             const cusparseMatDescr_t descrA, const float *A, int lda,
+             int *nnzPerRowCol, int *nnzTotalDevHostPtr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const float *, int, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSnnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, A, lda, nnzPerRowCol,
+                  nnzTotalDevHostPtr);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDnnz(cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+             const cusparseMatDescr_t descrA, const double *A, int lda,
+             int *nnzPerRowCol, int *nnzTotalDevHostPtr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const double *, int, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDnnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, A, lda, nnzPerRowCol,
+                  nnzTotalDevHostPtr);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCnnz(cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+             const cusparseMatDescr_t descrA, const cuComplex *A, int lda,
+             int *nnzPerRowCol, int *nnzTotalDevHostPtr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuComplex *, int, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCnnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, A, lda, nnzPerRowCol,
+                  nnzTotalDevHostPtr);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseZnnz(cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+             const cusparseMatDescr_t descrA, const cuDoubleComplex *A, int lda,
+             int *nnzPerRowCol, int *nnzTotalDevHostPtr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, int, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZnnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, A, lda, nnzPerRowCol,
+                  nnzTotalDevHostPtr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSnnz_compress(
+    cusparseHandle_t handle, int m, const cusparseMatDescr_t descr,
+    const float *csrSortedValA, const int *csrSortedRowPtrA, int *nnzPerRow,
+    int *nnzC, float tol) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const cusparseMatDescr_t, const float *,
+      const int *, int *, int *, float);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSnnz_compress");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, descr, csrSortedValA, csrSortedRowPtrA, nnzPerRow,
+                  nnzC, tol);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDnnz_compress(
+    cusparseHandle_t handle, int m, const cusparseMatDescr_t descr,
+    const double *csrSortedValA, const int *csrSortedRowPtrA, int *nnzPerRow,
+    int *nnzC, double tol) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const cusparseMatDescr_t, const double *,
+      const int *, int *, int *, double);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDnnz_compress");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, descr, csrSortedValA, csrSortedRowPtrA, nnzPerRow,
+                  nnzC, tol);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCnnz_compress(
+    cusparseHandle_t handle, int m, const cusparseMatDescr_t descr,
+    const cuComplex *csrSortedValA, const int *csrSortedRowPtrA, int *nnzPerRow,
+    int *nnzC, cuComplex tol) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const cusparseMatDescr_t, const cuComplex *,
+      const int *, int *, int *, cuComplex);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCnnz_compress");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, descr, csrSortedValA, csrSortedRowPtrA, nnzPerRow,
+                  nnzC, tol);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZnnz_compress(
+    cusparseHandle_t handle, int m, const cusparseMatDescr_t descr,
+    const cuDoubleComplex *csrSortedValA, const int *csrSortedRowPtrA,
+    int *nnzPerRow, int *nnzC, cuDoubleComplex tol) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, const cusparseMatDescr_t, const cuDoubleComplex *,
+      const int *, int *, int *, cuDoubleComplex);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZnnz_compress");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, descr, csrSortedValA, csrSortedRowPtrA, nnzPerRow,
+                  nnzC, tol);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsr2csr_compress(
+    cusparseHandle_t handle, int m, int n, const cusparseMatDescr_t descrA,
+    const float *csrSortedValA, const int *csrSortedColIndA,
+    const int *csrSortedRowPtrA, int nnzA, const int *nnzPerRow,
+    float *csrSortedValC, int *csrSortedColIndC, int *csrSortedRowPtrC,
+    float tol) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const float *,
+      const int *, const int *, int, const int *, float *, int *, int *, float);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsr2csr_compress");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, descrA, csrSortedValA, csrSortedColIndA,
+                  csrSortedRowPtrA, nnzA, nnzPerRow, csrSortedValC,
+                  csrSortedColIndC, csrSortedRowPtrC, tol);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsr2csr_compress(
+    cusparseHandle_t handle, int m, int n, const cusparseMatDescr_t descrA,
+    const double *csrSortedValA, const int *csrSortedColIndA,
+    const int *csrSortedRowPtrA, int nnzA, const int *nnzPerRow,
+    double *csrSortedValC, int *csrSortedColIndC, int *csrSortedRowPtrC,
+    double tol) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const double *,
+      const int *, const int *, int, const int *, double *, int *, int *,
+      double);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsr2csr_compress");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, descrA, csrSortedValA, csrSortedColIndA,
+                  csrSortedRowPtrA, nnzA, nnzPerRow, csrSortedValC,
+                  csrSortedColIndC, csrSortedRowPtrC, tol);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsr2csr_compress(
+    cusparseHandle_t handle, int m, int n, const cusparseMatDescr_t descrA,
+    const cuComplex *csrSortedValA, const int *csrSortedColIndA,
+    const int *csrSortedRowPtrA, int nnzA, const int *nnzPerRow,
+    cuComplex *csrSortedValC, int *csrSortedColIndC, int *csrSortedRowPtrC,
+    cuComplex tol) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t, const cuComplex *,
+      const int *, const int *, int, const int *, cuComplex *, int *, int *,
+      cuComplex);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsr2csr_compress");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, descrA, csrSortedValA, csrSortedColIndA,
+                  csrSortedRowPtrA, nnzA, nnzPerRow, csrSortedValC,
+                  csrSortedColIndC, csrSortedRowPtrC, tol);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsr2csr_compress(
+    cusparseHandle_t handle, int m, int n, const cusparseMatDescr_t descrA,
+    const cuDoubleComplex *csrSortedValA, const int *csrSortedColIndA,
+    const int *csrSortedRowPtrA, int nnzA, const int *nnzPerRow,
+    cuDoubleComplex *csrSortedValC, int *csrSortedColIndC,
+    int *csrSortedRowPtrC, cuDoubleComplex tol) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, int, const int *,
+      cuDoubleComplex *, int *, int *, cuDoubleComplex);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsr2csr_compress");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, descrA, csrSortedValA, csrSortedColIndA,
+                  csrSortedRowPtrA, nnzA, nnzPerRow, csrSortedValC,
+                  csrSortedColIndC, csrSortedRowPtrC, tol);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcoo2csr(cusparseHandle_t handle,
+                                              const int *cooRowInd, int nnz,
+                                              int m, int *csrSortedRowPtr,
+                                              cusparseIndexBase_t idxBase) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, const int *, int, int, int *, cusparseIndexBase_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcoo2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, cooRowInd, nnz, m, csrSortedRowPtr, idxBase);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcsr2coo(cusparseHandle_t handle,
+                                              const int *csrSortedRowPtr,
+                                              int nnz, int m, int *cooRowInd,
+                                              cusparseIndexBase_t idxBase) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, const int *, int, int, int *, cusparseIndexBase_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcsr2coo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, csrSortedRowPtr, nnz, m, cooRowInd, idxBase);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcsr2bsrNnz(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, int blockDim, const cusparseMatDescr_t descrC,
+    int *bsrSortedRowPtrC, int *nnzTotalDevHostPtr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const int *, const int *, int, const cusparseMatDescr_t, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcsr2bsrNnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedRowPtrA,
+                  csrSortedColIndA, blockDim, descrC, bsrSortedRowPtrC,
+                  nnzTotalDevHostPtr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsr2bsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int blockDim,
+    const cusparseMatDescr_t descrC, float *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const float *, const int *, const int *, int, const cusparseMatDescr_t,
+      float *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsr2bsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, blockDim, descrC, bsrSortedValC,
+                  bsrSortedRowPtrC, bsrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsr2bsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int blockDim,
+    const cusparseMatDescr_t descrC, double *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const double *, const int *, const int *, int, const cusparseMatDescr_t,
+      double *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsr2bsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, blockDim, descrC, bsrSortedValC,
+                  bsrSortedRowPtrC, bsrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsr2bsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const cuComplex *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int blockDim,
+    const cusparseMatDescr_t descrC, cuComplex *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuComplex *, const int *, const int *, int,
+      const cusparseMatDescr_t, cuComplex *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsr2bsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, blockDim, descrC, bsrSortedValC,
+                  bsrSortedRowPtrC, bsrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsr2bsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int blockDim,
+    const cusparseMatDescr_t descrC, cuDoubleComplex *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, int,
+      const cusparseMatDescr_t, cuDoubleComplex *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsr2bsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, blockDim, descrC, bsrSortedValC,
+                  bsrSortedRowPtrC, bsrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSbsr2csr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb,
+    const cusparseMatDescr_t descrA, const float *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    const cusparseMatDescr_t descrC, float *csrSortedValC,
+    int *csrSortedRowPtrC, int *csrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const float *, const int *, const int *, int, const cusparseMatDescr_t,
+      float *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSbsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, descrA, bsrSortedValA, bsrSortedRowPtrA,
+                  bsrSortedColIndA, blockDim, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDbsr2csr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb,
+    const cusparseMatDescr_t descrA, const double *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    const cusparseMatDescr_t descrC, double *csrSortedValC,
+    int *csrSortedRowPtrC, int *csrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const double *, const int *, const int *, int, const cusparseMatDescr_t,
+      double *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDbsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, descrA, bsrSortedValA, bsrSortedRowPtrA,
+                  bsrSortedColIndA, blockDim, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCbsr2csr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb,
+    const cusparseMatDescr_t descrA, const cuComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    const cusparseMatDescr_t descrC, cuComplex *csrSortedValC,
+    int *csrSortedRowPtrC, int *csrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuComplex *, const int *, const int *, int,
+      const cusparseMatDescr_t, cuComplex *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCbsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, descrA, bsrSortedValA, bsrSortedRowPtrA,
+                  bsrSortedColIndA, blockDim, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZbsr2csr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int blockDim,
+    const cusparseMatDescr_t descrC, cuDoubleComplex *csrSortedValC,
+    int *csrSortedRowPtrC, int *csrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, int,
+      const cusparseMatDescr_t, cuDoubleComplex *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZbsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, descrA, bsrSortedValA, bsrSortedRowPtrA,
+                  bsrSortedColIndA, blockDim, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgebsr2gebsc_bufferSize(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const float *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const float *, const int *, const int *,
+      int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgebsr2gebsc_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgebsr2gebsc_bufferSize(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const double *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const double *, const int *, const int *,
+      int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgebsr2gebsc_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgebsr2gebsc_bufferSize(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const cuComplex *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cuComplex *, const int *,
+      const int *, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgebsr2gebsc_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgebsr2gebsc_bufferSize(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const cuDoubleComplex *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cuDoubleComplex *, const int *,
+      const int *, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgebsr2gebsc_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgebsr2gebsc_bufferSizeExt(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const float *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const float *, const int *, const int *,
+      int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSgebsr2gebsc_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgebsr2gebsc_bufferSizeExt(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const double *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const double *, const int *, const int *,
+      int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDgebsr2gebsc_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgebsr2gebsc_bufferSizeExt(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const cuComplex *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cuComplex *, const int *,
+      const int *, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseCgebsr2gebsc_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgebsr2gebsc_bufferSizeExt(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const cuDoubleComplex *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cuDoubleComplex *, const int *,
+      const int *, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseZgebsr2gebsc_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgebsr2gebsc(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const float *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim, float *bscVal,
+    int *bscRowInd, int *bscColPtr, cusparseAction_t copyValues,
+    cusparseIndexBase_t idxBase, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const float *, const int *, const int *,
+      int, int, float *, int *, int *, cusparseAction_t, cusparseIndexBase_t,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgebsr2gebsc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim, bscVal, bscRowInd,
+                  bscColPtr, copyValues, idxBase, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgebsr2gebsc(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const double *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    double *bscVal, int *bscRowInd, int *bscColPtr, cusparseAction_t copyValues,
+    cusparseIndexBase_t idxBase, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const double *, const int *, const int *,
+      int, int, double *, int *, int *, cusparseAction_t, cusparseIndexBase_t,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgebsr2gebsc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim, bscVal, bscRowInd,
+                  bscColPtr, copyValues, idxBase, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgebsr2gebsc(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const cuComplex *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    cuComplex *bscVal, int *bscRowInd, int *bscColPtr,
+    cusparseAction_t copyValues, cusparseIndexBase_t idxBase, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cuComplex *, const int *,
+      const int *, int, int, cuComplex *, int *, int *, cusparseAction_t,
+      cusparseIndexBase_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgebsr2gebsc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim, bscVal, bscRowInd,
+                  bscColPtr, copyValues, idxBase, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgebsr2gebsc(
+    cusparseHandle_t handle, int mb, int nb, int nnzb,
+    const cuDoubleComplex *bsrSortedVal, const int *bsrSortedRowPtr,
+    const int *bsrSortedColInd, int rowBlockDim, int colBlockDim,
+    cuDoubleComplex *bscVal, int *bscRowInd, int *bscColPtr,
+    cusparseAction_t copyValues, cusparseIndexBase_t idxBase, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cuDoubleComplex *, const int *,
+      const int *, int, int, cuDoubleComplex *, int *, int *, cusparseAction_t,
+      cusparseIndexBase_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgebsr2gebsc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, mb, nb, nnzb, bsrSortedVal, bsrSortedRowPtr,
+                  bsrSortedColInd, rowBlockDim, colBlockDim, bscVal, bscRowInd,
+                  bscColPtr, copyValues, idxBase, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXgebsr2csr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb,
+    const cusparseMatDescr_t descrA, const int *bsrSortedRowPtrA,
+    const int *bsrSortedColIndA, int rowBlockDim, int colBlockDim,
+    const cusparseMatDescr_t descrC, int *csrSortedRowPtrC,
+    int *csrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const int *, const int *, int, int, const cusparseMatDescr_t, int *,
+      int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXgebsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, descrA, bsrSortedRowPtrA,
+                  bsrSortedColIndA, rowBlockDim, colBlockDim, descrC,
+                  csrSortedRowPtrC, csrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgebsr2csr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb,
+    const cusparseMatDescr_t descrA, const float *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDim,
+    int colBlockDim, const cusparseMatDescr_t descrC, float *csrSortedValC,
+    int *csrSortedRowPtrC, int *csrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const float *, const int *, const int *, int, int,
+      const cusparseMatDescr_t, float *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgebsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, descrA, bsrSortedValA, bsrSortedRowPtrA,
+                  bsrSortedColIndA, rowBlockDim, colBlockDim, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgebsr2csr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb,
+    const cusparseMatDescr_t descrA, const double *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDim,
+    int colBlockDim, const cusparseMatDescr_t descrC, double *csrSortedValC,
+    int *csrSortedRowPtrC, int *csrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const double *, const int *, const int *, int, int,
+      const cusparseMatDescr_t, double *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgebsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, descrA, bsrSortedValA, bsrSortedRowPtrA,
+                  bsrSortedColIndA, rowBlockDim, colBlockDim, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgebsr2csr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb,
+    const cusparseMatDescr_t descrA, const cuComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDim,
+    int colBlockDim, const cusparseMatDescr_t descrC, cuComplex *csrSortedValC,
+    int *csrSortedRowPtrC, int *csrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuComplex *, const int *, const int *, int, int,
+      const cusparseMatDescr_t, cuComplex *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgebsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, descrA, bsrSortedValA, bsrSortedRowPtrA,
+                  bsrSortedColIndA, rowBlockDim, colBlockDim, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgebsr2csr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDim,
+    int colBlockDim, const cusparseMatDescr_t descrC,
+    cuDoubleComplex *csrSortedValC, int *csrSortedRowPtrC,
+    int *csrSortedColIndC) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, int, int,
+      const cusparseMatDescr_t, cuDoubleComplex *, int *, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgebsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, descrA, bsrSortedValA, bsrSortedRowPtrA,
+                  bsrSortedColIndA, rowBlockDim, colBlockDim, descrC,
+                  csrSortedValC, csrSortedRowPtrC, csrSortedColIndC);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsr2gebsr_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int rowBlockDim,
+    int colBlockDim, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const float *, const int *, const int *, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsr2gebsr_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, rowBlockDim, colBlockDim,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsr2gebsr_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int rowBlockDim,
+    int colBlockDim, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const double *, const int *, const int *, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsr2gebsr_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, rowBlockDim, colBlockDim,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsr2gebsr_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const cuComplex *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int rowBlockDim,
+    int colBlockDim, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuComplex *, const int *, const int *, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsr2gebsr_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, rowBlockDim, colBlockDim,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsr2gebsr_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int rowBlockDim,
+    int colBlockDim, int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsr2gebsr_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, rowBlockDim, colBlockDim,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsr2gebsr_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int rowBlockDim,
+    int colBlockDim, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const float *, const int *, const int *, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseScsr2gebsr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, rowBlockDim, colBlockDim, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsr2gebsr_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int rowBlockDim,
+    int colBlockDim, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const double *, const int *, const int *, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDcsr2gebsr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, rowBlockDim, colBlockDim, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsr2gebsr_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const cuComplex *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int rowBlockDim,
+    int colBlockDim, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuComplex *, const int *, const int *, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseCcsr2gebsr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, rowBlockDim, colBlockDim, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsr2gebsr_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, int rowBlockDim,
+    int colBlockDim, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseZcsr2gebsr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, rowBlockDim, colBlockDim, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcsr2gebsrNnz(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const int *csrSortedRowPtrA,
+    const int *csrSortedColIndA, const cusparseMatDescr_t descrC,
+    int *bsrSortedRowPtrC, int rowBlockDim, int colBlockDim,
+    int *nnzTotalDevHostPtr, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const int *, const int *, const cusparseMatDescr_t, int *, int, int,
+      int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcsr2gebsrNnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedRowPtrA,
+                  csrSortedColIndA, descrC, bsrSortedRowPtrC, rowBlockDim,
+                  colBlockDim, nnzTotalDevHostPtr, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsr2gebsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const cusparseMatDescr_t descrC, float *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC, int rowBlockDim,
+    int colBlockDim, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const float *, const int *, const int *, const cusparseMatDescr_t,
+      float *, int *, int *, int, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsr2gebsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, descrC, bsrSortedValC, bsrSortedRowPtrC,
+                  bsrSortedColIndC, rowBlockDim, colBlockDim, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsr2gebsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const cusparseMatDescr_t descrC, double *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC, int rowBlockDim,
+    int colBlockDim, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const double *, const int *, const int *, const cusparseMatDescr_t,
+      double *, int *, int *, int, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsr2gebsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, descrC, bsrSortedValC, bsrSortedRowPtrC,
+                  bsrSortedColIndC, rowBlockDim, colBlockDim, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsr2gebsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const cuComplex *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const cusparseMatDescr_t descrC, cuComplex *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC, int rowBlockDim,
+    int colBlockDim, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuComplex *, const int *, const int *, const cusparseMatDescr_t,
+      cuComplex *, int *, int *, int, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsr2gebsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, descrC, bsrSortedValC, bsrSortedRowPtrC,
+                  bsrSortedColIndC, rowBlockDim, colBlockDim, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsr2gebsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int m, int n,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const cusparseMatDescr_t descrC, cuDoubleComplex *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC, int rowBlockDim,
+    int colBlockDim, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, const cusparseMatDescr_t,
+      const cuDoubleComplex *, const int *, const int *,
+      const cusparseMatDescr_t, cuDoubleComplex *, int *, int *, int, int,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsr2gebsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, m, n, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, descrC, bsrSortedValC, bsrSortedRowPtrC,
+                  bsrSortedColIndC, rowBlockDim, colBlockDim, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgebsr2gebsr_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const float *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, int rowBlockDimC, int colBlockDimC,
+    int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const float *, const int *, const int *, int,
+      int, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgebsr2gebsr_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, rowBlockDimC, colBlockDimC, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgebsr2gebsr_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const double *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, int rowBlockDimC, int colBlockDimC,
+    int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const double *, const int *, const int *, int,
+      int, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgebsr2gebsr_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, rowBlockDimC, colBlockDimC, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgebsr2gebsr_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const cuComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, int rowBlockDimC, int colBlockDimC,
+    int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const cuComplex *, const int *, const int *,
+      int, int, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgebsr2gebsr_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, rowBlockDimC, colBlockDimC, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgebsr2gebsr_bufferSize(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, int rowBlockDimC, int colBlockDimC,
+    int *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const cuDoubleComplex *, const int *,
+      const int *, int, int, int, int, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgebsr2gebsr_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, rowBlockDimC, colBlockDimC, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgebsr2gebsr_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const float *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, int rowBlockDimC, int colBlockDimC, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const float *, const int *, const int *, int,
+      int, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSgebsr2gebsr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, rowBlockDimC, colBlockDimC, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgebsr2gebsr_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const double *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, int rowBlockDimC, int colBlockDimC, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const double *, const int *, const int *, int,
+      int, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDgebsr2gebsr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, rowBlockDimC, colBlockDimC, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgebsr2gebsr_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const cuComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, int rowBlockDimC, int colBlockDimC, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const cuComplex *, const int *, const int *,
+      int, int, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseCgebsr2gebsr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, rowBlockDimC, colBlockDimC, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgebsr2gebsr_bufferSizeExt(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, int rowBlockDimC, int colBlockDimC, size_t *pBufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const cuDoubleComplex *, const int *,
+      const int *, int, int, int, int, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseZgebsr2gebsr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, rowBlockDimC, colBlockDimC, pBufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXgebsr2gebsrNnz(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const int *bsrSortedRowPtrA,
+    const int *bsrSortedColIndA, int rowBlockDimA, int colBlockDimA,
+    const cusparseMatDescr_t descrC, int *bsrSortedRowPtrC, int rowBlockDimC,
+    int colBlockDimC, int *nnzTotalDevHostPtr, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const int *, const int *, int, int,
+      const cusparseMatDescr_t, int *, int, int, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXgebsr2gebsrNnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedRowPtrA,
+                  bsrSortedColIndA, rowBlockDimA, colBlockDimA, descrC,
+                  bsrSortedRowPtrC, rowBlockDimC, colBlockDimC,
+                  nnzTotalDevHostPtr, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSgebsr2gebsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const float *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, const cusparseMatDescr_t descrC, float *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC, int rowBlockDimC,
+    int colBlockDimC, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const float *, const int *, const int *, int,
+      int, const cusparseMatDescr_t, float *, int *, int *, int, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSgebsr2gebsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, descrC, bsrSortedValC, bsrSortedRowPtrC,
+                  bsrSortedColIndC, rowBlockDimC, colBlockDimC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDgebsr2gebsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const double *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, const cusparseMatDescr_t descrC, double *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC, int rowBlockDimC,
+    int colBlockDimC, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const double *, const int *, const int *, int,
+      int, const cusparseMatDescr_t, double *, int *, int *, int, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDgebsr2gebsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, descrC, bsrSortedValC, bsrSortedRowPtrC,
+                  bsrSortedColIndC, rowBlockDimC, colBlockDimC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCgebsr2gebsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const cuComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, const cusparseMatDescr_t descrC, cuComplex *bsrSortedValC,
+    int *bsrSortedRowPtrC, int *bsrSortedColIndC, int rowBlockDimC,
+    int colBlockDimC, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const cuComplex *, const int *, const int *,
+      int, int, const cusparseMatDescr_t, cuComplex *, int *, int *, int, int,
+      void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCgebsr2gebsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, descrC, bsrSortedValC, bsrSortedRowPtrC,
+                  bsrSortedColIndC, rowBlockDimC, colBlockDimC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZgebsr2gebsr(
+    cusparseHandle_t handle, cusparseDirection_t dirA, int mb, int nb, int nnzb,
+    const cusparseMatDescr_t descrA, const cuDoubleComplex *bsrSortedValA,
+    const int *bsrSortedRowPtrA, const int *bsrSortedColIndA, int rowBlockDimA,
+    int colBlockDimA, const cusparseMatDescr_t descrC,
+    cuDoubleComplex *bsrSortedValC, int *bsrSortedRowPtrC,
+    int *bsrSortedColIndC, int rowBlockDimC, int colBlockDimC, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseDirection_t, int, int, int,
+      const cusparseMatDescr_t, const cuDoubleComplex *, const int *,
+      const int *, int, int, const cusparseMatDescr_t, cuDoubleComplex *, int *,
+      int *, int, int, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZgebsr2gebsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, dirA, mb, nb, nnzb, descrA, bsrSortedValA,
+                  bsrSortedRowPtrA, bsrSortedColIndA, rowBlockDimA,
+                  colBlockDimA, descrC, bsrSortedValC, bsrSortedRowPtrC,
+                  bsrSortedColIndC, rowBlockDimC, colBlockDimC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCreateIdentityPermutation(cusparseHandle_t handle, int n, int *p) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseHandle_t, int, int *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseCreateIdentityPermutation");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, n, p);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcoosort_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnz, const int *cooRowsA,
+    const int *cooColsA, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const int *, const int *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcoosort_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, cooRowsA, cooColsA, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcoosortByRow(cusparseHandle_t handle,
+                                                   int m, int n, int nnz,
+                                                   int *cooRowsA, int *cooColsA,
+                                                   int *P, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, int *, int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcoosortByRow");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, cooRowsA, cooColsA, P, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcoosortByColumn(cusparseHandle_t handle,
+                                                      int m, int n, int nnz,
+                                                      int *cooRowsA,
+                                                      int *cooColsA, int *P,
+                                                      void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, int *, int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcoosortByColumn");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, cooRowsA, cooColsA, P, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcsrsort_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnz, const int *csrRowPtrA,
+    const int *csrColIndA, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const int *, const int *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcsrsort_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, csrRowPtrA, csrColIndA,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcsrsort(cusparseHandle_t handle, int m,
+                                              int n, int nnz,
+                                              const cusparseMatDescr_t descrA,
+                                              const int *csrRowPtrA,
+                                              int *csrColIndA, int *P,
+                                              void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const int *,
+      int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcsrsort");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, descrA, csrRowPtrA, csrColIndA, P,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcscsort_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnz, const int *cscColPtrA,
+    const int *cscRowIndA, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const int *, const int *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcscsort_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, cscColPtrA, cscRowIndA,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseXcscsort(cusparseHandle_t handle, int m,
+                                              int n, int nnz,
+                                              const cusparseMatDescr_t descrA,
+                                              const int *cscColPtrA,
+                                              int *cscRowIndA, int *P,
+                                              void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const int *,
+      int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseXcscsort");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, descrA, cscColPtrA, cscRowIndA, P,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsru2csr_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnz, float *csrVal,
+    const int *csrRowPtr, int *csrColInd, csru2csrInfo_t info,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, float *, const int *, int *,
+      csru2csrInfo_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsru2csr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, csrVal, csrRowPtr, csrColInd, info,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsru2csr_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnz, double *csrVal,
+    const int *csrRowPtr, int *csrColInd, csru2csrInfo_t info,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, double *, const int *, int *,
+      csru2csrInfo_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsru2csr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, csrVal, csrRowPtr, csrColInd, info,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsru2csr_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnz, cuComplex *csrVal,
+    const int *csrRowPtr, int *csrColInd, csru2csrInfo_t info,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, cuComplex *, const int *, int *,
+      csru2csrInfo_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsru2csr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, csrVal, csrRowPtr, csrColInd, info,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsru2csr_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnz, cuDoubleComplex *csrVal,
+    const int *csrRowPtr, int *csrColInd, csru2csrInfo_t info,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, cuDoubleComplex *, const int *, int *,
+      csru2csrInfo_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsru2csr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, csrVal, csrRowPtr, csrColInd, info,
+                  pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsru2csr(
+    cusparseHandle_t handle, int m, int n, int nnz,
+    const cusparseMatDescr_t descrA, float *csrVal, const int *csrRowPtr,
+    int *csrColInd, csru2csrInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, float *,
+      const int *, int *, csru2csrInfo_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsru2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, descrA, csrVal, csrRowPtr, csrColInd, info,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsru2csr(
+    cusparseHandle_t handle, int m, int n, int nnz,
+    const cusparseMatDescr_t descrA, double *csrVal, const int *csrRowPtr,
+    int *csrColInd, csru2csrInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, double *,
+      const int *, int *, csru2csrInfo_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsru2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, descrA, csrVal, csrRowPtr, csrColInd, info,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsru2csr(
+    cusparseHandle_t handle, int m, int n, int nnz,
+    const cusparseMatDescr_t descrA, cuComplex *csrVal, const int *csrRowPtr,
+    int *csrColInd, csru2csrInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, cuComplex *,
+      const int *, int *, csru2csrInfo_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsru2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, descrA, csrVal, csrRowPtr, csrColInd, info,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsru2csr(
+    cusparseHandle_t handle, int m, int n, int nnz,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *csrVal,
+    const int *csrRowPtr, int *csrColInd, csru2csrInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, int *, csru2csrInfo_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsru2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, descrA, csrVal, csrRowPtr, csrColInd, info,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScsr2csru(
+    cusparseHandle_t handle, int m, int n, int nnz,
+    const cusparseMatDescr_t descrA, float *csrVal, const int *csrRowPtr,
+    int *csrColInd, csru2csrInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, float *,
+      const int *, int *, csru2csrInfo_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScsr2csru");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, descrA, csrVal, csrRowPtr, csrColInd, info,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDcsr2csru(
+    cusparseHandle_t handle, int m, int n, int nnz,
+    const cusparseMatDescr_t descrA, double *csrVal, const int *csrRowPtr,
+    int *csrColInd, csru2csrInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, double *,
+      const int *, int *, csru2csrInfo_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDcsr2csru");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, descrA, csrVal, csrRowPtr, csrColInd, info,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCcsr2csru(
+    cusparseHandle_t handle, int m, int n, int nnz,
+    const cusparseMatDescr_t descrA, cuComplex *csrVal, const int *csrRowPtr,
+    int *csrColInd, csru2csrInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, cuComplex *,
+      const int *, int *, csru2csrInfo_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCcsr2csru");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, descrA, csrVal, csrRowPtr, csrColInd, info,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseZcsr2csru(
+    cusparseHandle_t handle, int m, int n, int nnz,
+    const cusparseMatDescr_t descrA, cuDoubleComplex *csrVal,
+    const int *csrRowPtr, int *csrColInd, csru2csrInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t,
+      cuDoubleComplex *, const int *, int *, csru2csrInfo_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseZcsr2csru");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, descrA, csrVal, csrRowPtr, csrColInd, info,
+                  pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneDense2csr_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const float *A, int lda,
+    const float *threshold, const cusparseMatDescr_t descrC,
+    const float *csrSortedValC, const int *csrSortedRowPtrC,
+    const int *csrSortedColIndC, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, int, const float *,
+      const cusparseMatDescr_t, const float *, const int *, const int *,
+      size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSpruneDense2csr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, threshold, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneDense2csr_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const double *A, int lda,
+    const double *threshold, const cusparseMatDescr_t descrC,
+    const double *csrSortedValC, const int *csrSortedRowPtrC,
+    const int *csrSortedColIndC, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, int, const double *,
+      const cusparseMatDescr_t, const double *, const int *, const int *,
+      size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDpruneDense2csr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, threshold, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneDense2csrNnz(
+    cusparseHandle_t handle, int m, int n, const float *A, int lda,
+    const float *threshold, const cusparseMatDescr_t descrC, int *csrRowPtrC,
+    int *nnzTotalDevHostPtr, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, int, const float *,
+      const cusparseMatDescr_t, int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpruneDense2csrNnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, threshold, descrC, csrRowPtrC,
+                  nnzTotalDevHostPtr, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneDense2csrNnz(
+    cusparseHandle_t handle, int m, int n, const double *A, int lda,
+    const double *threshold, const cusparseMatDescr_t descrC,
+    int *csrSortedRowPtrC, int *nnzTotalDevHostPtr, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, int, const double *,
+      const cusparseMatDescr_t, int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDpruneDense2csrNnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, threshold, descrC, csrSortedRowPtrC,
+                  nnzTotalDevHostPtr, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneDense2csr(
+    cusparseHandle_t handle, int m, int n, const float *A, int lda,
+    const float *threshold, const cusparseMatDescr_t descrC,
+    float *csrSortedValC, const int *csrSortedRowPtrC, int *csrSortedColIndC,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, int, const float *,
+      const cusparseMatDescr_t, float *, const int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpruneDense2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, threshold, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneDense2csr(
+    cusparseHandle_t handle, int m, int n, const double *A, int lda,
+    const double *threshold, const cusparseMatDescr_t descrC,
+    double *csrSortedValC, const int *csrSortedRowPtrC, int *csrSortedColIndC,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, int, const double *,
+      const cusparseMatDescr_t, double *, const int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDpruneDense2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, threshold, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneCsr2csr_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const float *threshold, const cusparseMatDescr_t descrC,
+    const float *csrSortedValC, const int *csrSortedRowPtrC,
+    const int *csrSortedColIndC, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const float *,
+      const int *, const int *, const float *, const cusparseMatDescr_t,
+      const float *, const int *, const int *, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSpruneCsr2csr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, threshold, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneCsr2csr_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const double *threshold, const cusparseMatDescr_t descrC,
+    const double *csrSortedValC, const int *csrSortedRowPtrC,
+    const int *csrSortedColIndC, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const double *,
+      const int *, const int *, const double *, const cusparseMatDescr_t,
+      const double *, const int *, const int *, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDpruneCsr2csr_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, threshold, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneCsr2csrNnz(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const float *threshold, const cusparseMatDescr_t descrC,
+    int *csrSortedRowPtrC, int *nnzTotalDevHostPtr, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const float *,
+      const int *, const int *, const float *, const cusparseMatDescr_t, int *,
+      int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpruneCsr2csrNnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, threshold, descrC, csrSortedRowPtrC,
+                  nnzTotalDevHostPtr, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneCsr2csrNnz(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const double *threshold, const cusparseMatDescr_t descrC,
+    int *csrSortedRowPtrC, int *nnzTotalDevHostPtr, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const double *,
+      const int *, const int *, const double *, const cusparseMatDescr_t, int *,
+      int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDpruneCsr2csrNnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, threshold, descrC, csrSortedRowPtrC,
+                  nnzTotalDevHostPtr, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneCsr2csr(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const float *threshold, const cusparseMatDescr_t descrC,
+    float *csrSortedValC, const int *csrSortedRowPtrC, int *csrSortedColIndC,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const float *,
+      const int *, const int *, const float *, const cusparseMatDescr_t,
+      float *, const int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpruneCsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, threshold, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneCsr2csr(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA,
+    const double *threshold, const cusparseMatDescr_t descrC,
+    double *csrSortedValC, const int *csrSortedRowPtrC, int *csrSortedColIndC,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const double *,
+      const int *, const int *, const double *, const cusparseMatDescr_t,
+      double *, const int *, int *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDpruneCsr2csr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, threshold, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneDense2csrByPercentage_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const float *A, int lda,
+    float percentage, const cusparseMatDescr_t descrC,
+    const float *csrSortedValC, const int *csrSortedRowPtrC,
+    const int *csrSortedColIndC, pruneInfo_t info, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, int, float,
+      const cusparseMatDescr_t, const float *, const int *, const int *,
+      pruneInfo_t, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSpruneDense2csrByPercentage_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, percentage, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneDense2csrByPercentage_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, const double *A, int lda,
+    float percentage, const cusparseMatDescr_t descrC,
+    const double *csrSortedValC, const int *csrSortedRowPtrC,
+    const int *csrSortedColIndC, pruneInfo_t info, size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, int, float,
+      const cusparseMatDescr_t, const double *, const int *, const int *,
+      pruneInfo_t, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDpruneDense2csrByPercentage_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, percentage, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneDense2csrNnzByPercentage(
+    cusparseHandle_t handle, int m, int n, const float *A, int lda,
+    float percentage, const cusparseMatDescr_t descrC, int *csrRowPtrC,
+    int *nnzTotalDevHostPtr, pruneInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, int, float,
+      const cusparseMatDescr_t, int *, int *, pruneInfo_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSpruneDense2csrNnzByPercentage");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, percentage, descrC, csrRowPtrC,
+                  nnzTotalDevHostPtr, info, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneDense2csrNnzByPercentage(
+    cusparseHandle_t handle, int m, int n, const double *A, int lda,
+    float percentage, const cusparseMatDescr_t descrC, int *csrRowPtrC,
+    int *nnzTotalDevHostPtr, pruneInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, int, float,
+      const cusparseMatDescr_t, int *, int *, pruneInfo_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDpruneDense2csrNnzByPercentage");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, percentage, descrC, csrRowPtrC,
+                  nnzTotalDevHostPtr, info, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneDense2csrByPercentage(
+    cusparseHandle_t handle, int m, int n, const float *A, int lda,
+    float percentage, const cusparseMatDescr_t descrC, float *csrSortedValC,
+    const int *csrSortedRowPtrC, int *csrSortedColIndC, pruneInfo_t info,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const float *, int, float,
+      const cusparseMatDescr_t, float *, const int *, int *, pruneInfo_t,
+      void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSpruneDense2csrByPercentage");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, percentage, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, info, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneDense2csrByPercentage(
+    cusparseHandle_t handle, int m, int n, const double *A, int lda,
+    float percentage, const cusparseMatDescr_t descrC, double *csrSortedValC,
+    const int *csrSortedRowPtrC, int *csrSortedColIndC, pruneInfo_t info,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, const double *, int, float,
+      const cusparseMatDescr_t, double *, const int *, int *, pruneInfo_t,
+      void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDpruneDense2csrByPercentage");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, A, lda, percentage, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, info, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneCsr2csrByPercentage_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, float percentage,
+    const cusparseMatDescr_t descrC, const float *csrSortedValC,
+    const int *csrSortedRowPtrC, const int *csrSortedColIndC, pruneInfo_t info,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const float *,
+      const int *, const int *, float, const cusparseMatDescr_t, const float *,
+      const int *, const int *, pruneInfo_t, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSpruneCsr2csrByPercentage_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, percentage, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneCsr2csrByPercentage_bufferSizeExt(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, float percentage,
+    const cusparseMatDescr_t descrC, const double *csrSortedValC,
+    const int *csrSortedRowPtrC, const int *csrSortedColIndC, pruneInfo_t info,
+    size_t *pBufferSizeInBytes) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const double *,
+      const int *, const int *, float, const cusparseMatDescr_t, const double *,
+      const int *, const int *, pruneInfo_t, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDpruneCsr2csrByPercentage_bufferSizeExt");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, percentage, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, info, pBufferSizeInBytes);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneCsr2csrNnzByPercentage(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, float percentage,
+    const cusparseMatDescr_t descrC, int *csrSortedRowPtrC,
+    int *nnzTotalDevHostPtr, pruneInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const float *,
+      const int *, const int *, float, const cusparseMatDescr_t, int *, int *,
+      pruneInfo_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSpruneCsr2csrNnzByPercentage");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, percentage, descrC, csrSortedRowPtrC,
+                  nnzTotalDevHostPtr, info, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneCsr2csrNnzByPercentage(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, float percentage,
+    const cusparseMatDescr_t descrC, int *csrSortedRowPtrC,
+    int *nnzTotalDevHostPtr, pruneInfo_t info, void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const double *,
+      const int *, const int *, float, const cusparseMatDescr_t, int *, int *,
+      pruneInfo_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDpruneCsr2csrNnzByPercentage");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, percentage, descrC, csrSortedRowPtrC,
+                  nnzTotalDevHostPtr, info, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpruneCsr2csrByPercentage(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const float *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, float percentage,
+    const cusparseMatDescr_t descrC, float *csrSortedValC,
+    const int *csrSortedRowPtrC, int *csrSortedColIndC, pruneInfo_t info,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const float *,
+      const int *, const int *, float, const cusparseMatDescr_t, float *,
+      const int *, int *, pruneInfo_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSpruneCsr2csrByPercentage");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, percentage, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, info, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDpruneCsr2csrByPercentage(
+    cusparseHandle_t handle, int m, int n, int nnzA,
+    const cusparseMatDescr_t descrA, const double *csrSortedValA,
+    const int *csrSortedRowPtrA, const int *csrSortedColIndA, float percentage,
+    const cusparseMatDescr_t descrC, double *csrSortedValC,
+    const int *csrSortedRowPtrC, int *csrSortedColIndC, pruneInfo_t info,
+    void *pBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const cusparseMatDescr_t, const double *,
+      const int *, const int *, float, const cusparseMatDescr_t, double *,
+      const int *, int *, pruneInfo_t, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDpruneCsr2csrByPercentage");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnzA, descrA, csrSortedValA, csrSortedRowPtrA,
+                  csrSortedColIndA, percentage, descrC, csrSortedValC,
+                  csrSortedRowPtrC, csrSortedColIndC, info, pBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCsr2cscEx2(
+    cusparseHandle_t handle, int m, int n, int nnz, const void *csrVal,
+    const int *csrRowPtr, const int *csrColInd, void *cscVal, int *cscColPtr,
+    int *cscRowInd, cudaDataType valType, cusparseAction_t copyValues,
+    cusparseIndexBase_t idxBase, cusparseCsr2CscAlg_t alg, void *buffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const void *, const int *, const int *,
+      void *, int *, int *, cudaDataType, cusparseAction_t, cusparseIndexBase_t,
+      cusparseCsr2CscAlg_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCsr2cscEx2");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, csrVal, csrRowPtr, csrColInd, cscVal,
+                  cscColPtr, cscRowInd, valType, copyValues, idxBase, alg,
+                  buffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCsr2cscEx2_bufferSize(
+    cusparseHandle_t handle, int m, int n, int nnz, const void *csrVal,
+    const int *csrRowPtr, const int *csrColInd, void *cscVal, int *cscColPtr,
+    int *cscRowInd, cudaDataType valType, cusparseAction_t copyValues,
+    cusparseIndexBase_t idxBase, cusparseCsr2CscAlg_t alg, size_t *bufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, int, int, int, const void *, const int *, const int *,
+      void *, int *, int *, cudaDataType, cusparseAction_t, cusparseIndexBase_t,
+      cusparseCsr2CscAlg_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCsr2cscEx2_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, m, n, nnz, csrVal, csrRowPtr, csrColInd, cscVal,
+                  cscColPtr, cscRowInd, valType, copyValues, idxBase, alg,
+                  bufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCreateSpVec(cusparseSpVecDescr_t *spVecDescr, int64_t size, int64_t nnz,
+                    void *indices, void *values, cusparseIndexType_t idxType,
+                    cusparseIndexBase_t idxBase, cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpVecDescr_t *, int64_t, int64_t, void *, void *,
+      cusparseIndexType_t, cusparseIndexBase_t, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateSpVec");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spVecDescr, size, nnz, indices, values, idxType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateConstSpVec(
+    cusparseConstSpVecDescr_t *spVecDescr, int64_t size, int64_t nnz,
+    const void *indices, const void *values, cusparseIndexType_t idxType,
+    cusparseIndexBase_t idxBase, cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpVecDescr_t *, int64_t, int64_t, const void *, const void *,
+      cusparseIndexType_t, cusparseIndexBase_t, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateConstSpVec");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spVecDescr, size, nnz, indices, values, idxType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDestroySpVec(cusparseConstSpVecDescr_t spVecDescr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseConstSpVecDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroySpVec");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spVecDescr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpVecGet(cusparseSpVecDescr_t spVecDescr,
+                                              int64_t *size, int64_t *nnz,
+                                              void **indices, void **values,
+                                              cusparseIndexType_t *idxType,
+                                              cusparseIndexBase_t *idxBase,
+                                              cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpVecDescr_t, int64_t *, int64_t *, void **, void **,
+      cusparseIndexType_t *, cusparseIndexBase_t *, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpVecGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spVecDescr, size, nnz, indices, values, idxType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseConstSpVecGet(
+    cusparseConstSpVecDescr_t spVecDescr, int64_t *size, int64_t *nnz,
+    const void **indices, const void **values, cusparseIndexType_t *idxType,
+    cusparseIndexBase_t *idxBase, cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpVecDescr_t, int64_t *, int64_t *, const void **,
+      const void **, cusparseIndexType_t *, cusparseIndexBase_t *,
+      cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstSpVecGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spVecDescr, size, nnz, indices, values, idxType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpVecGetIndexBase(
+    cusparseConstSpVecDescr_t spVecDescr, cusparseIndexBase_t *idxBase) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseConstSpVecDescr_t,
+                                                  cusparseIndexBase_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpVecGetIndexBase");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spVecDescr, idxBase);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpVecGetValues(cusparseSpVecDescr_t spVecDescr, void **values) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseSpVecDescr_t, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpVecGetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spVecDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseConstSpVecGetValues(
+    cusparseConstSpVecDescr_t spVecDescr, const void **values) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseConstSpVecDescr_t, const void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstSpVecGetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spVecDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpVecSetValues(cusparseSpVecDescr_t spVecDescr, void *values) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpVecDescr_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpVecSetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spVecDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCreateDnVec(cusparseDnVecDescr_t *dnVecDescr, int64_t size,
+                    void *values, cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseDnVecDescr_t *, int64_t, void *, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateDnVec");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnVecDescr, size, values, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCreateConstDnVec(cusparseConstDnVecDescr_t *dnVecDescr, int64_t size,
+                         const void *values, cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstDnVecDescr_t *, int64_t, const void *, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateConstDnVec");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnVecDescr, size, values, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDestroyDnVec(cusparseConstDnVecDescr_t dnVecDescr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseConstDnVecDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyDnVec");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnVecDescr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDnVecGet(cusparseDnVecDescr_t dnVecDescr,
+                                              int64_t *size, void **values,
+                                              cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseDnVecDescr_t, int64_t *, void **, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDnVecGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnVecDescr, size, values, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseConstDnVecGet(cusparseConstDnVecDescr_t dnVecDescr, int64_t *size,
+                      const void **values, cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstDnVecDescr_t, int64_t *, const void **, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstDnVecGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnVecDescr, size, values, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDnVecGetValues(cusparseDnVecDescr_t dnVecDescr, void **values) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseDnVecDescr_t, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDnVecGetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnVecDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseConstDnVecGetValues(
+    cusparseConstDnVecDescr_t dnVecDescr, const void **values) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseConstDnVecDescr_t, const void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstDnVecGetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnVecDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDnVecSetValues(cusparseDnVecDescr_t dnVecDescr, void *values) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseDnVecDescr_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDnVecSetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnVecDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDestroySpMat(cusparseConstSpMatDescr_t spMatDescr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseConstSpMatDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroySpMat");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMatGetFormat(
+    cusparseConstSpMatDescr_t spMatDescr, cusparseFormat_t *format) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseConstSpMatDescr_t,
+                                                  cusparseFormat_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMatGetFormat");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, format);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMatGetIndexBase(
+    cusparseConstSpMatDescr_t spMatDescr, cusparseIndexBase_t *idxBase) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseConstSpMatDescr_t,
+                                                  cusparseIndexBase_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMatGetIndexBase");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, idxBase);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpMatGetValues(cusparseSpMatDescr_t spMatDescr, void **values) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseSpMatDescr_t, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMatGetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseConstSpMatGetValues(
+    cusparseConstSpMatDescr_t spMatDescr, const void **values) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseConstSpMatDescr_t, const void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstSpMatGetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpMatSetValues(cusparseSpMatDescr_t spMatDescr, void *values) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpMatDescr_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMatSetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpMatGetSize(cusparseConstSpMatDescr_t spMatDescr, int64_t *rows,
+                     int64_t *cols, int64_t *nnz) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpMatDescr_t, int64_t *, int64_t *, int64_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMatGetSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMatGetStridedBatch(
+    cusparseConstSpMatDescr_t spMatDescr, int *batchCount) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseConstSpMatDescr_t, int *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMatGetStridedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, batchCount);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCooSetStridedBatch(
+    cusparseSpMatDescr_t spMatDescr, int batchCount, int64_t batchStride) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseSpMatDescr_t, int, int64_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCooSetStridedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, batchCount, batchStride);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCsrSetStridedBatch(
+    cusparseSpMatDescr_t spMatDescr, int batchCount, int64_t offsetsBatchStride,
+    int64_t columnsValuesBatchStride) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpMatDescr_t, int,
+                                                  int64_t, int64_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCsrSetStridedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, batchCount, offsetsBatchStride,
+                  columnsValuesBatchStride);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMatGetAttribute(
+    cusparseConstSpMatDescr_t spMatDescr, cusparseSpMatAttribute_t attribute,
+    void *data, size_t dataSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpMatDescr_t, cusparseSpMatAttribute_t, void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMatGetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, attribute, data, dataSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMatSetAttribute(
+    cusparseSpMatDescr_t spMatDescr, cusparseSpMatAttribute_t attribute,
+    void *data, size_t dataSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpMatDescr_t, cusparseSpMatAttribute_t, void *, size_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMatSetAttribute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, attribute, data, dataSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateCsr(
+    cusparseSpMatDescr_t *spMatDescr, int64_t rows, int64_t cols, int64_t nnz,
+    void *csrRowOffsets, void *csrColInd, void *csrValues,
+    cusparseIndexType_t csrRowOffsetsType, cusparseIndexType_t csrColIndType,
+    cusparseIndexBase_t idxBase, cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpMatDescr_t *, int64_t, int64_t, int64_t, void *, void *, void *,
+      cusparseIndexType_t, cusparseIndexType_t, cusparseIndexBase_t,
+      cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateCsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, csrRowOffsets, csrColInd,
+                  csrValues, csrRowOffsetsType, csrColIndType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateConstCsr(
+    cusparseConstSpMatDescr_t *spMatDescr, int64_t rows, int64_t cols,
+    int64_t nnz, const void *csrRowOffsets, const void *csrColInd,
+    const void *csrValues, cusparseIndexType_t csrRowOffsetsType,
+    cusparseIndexType_t csrColIndType, cusparseIndexBase_t idxBase,
+    cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpMatDescr_t *, int64_t, int64_t, int64_t, const void *,
+      const void *, const void *, cusparseIndexType_t, cusparseIndexType_t,
+      cusparseIndexBase_t, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateConstCsr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, csrRowOffsets, csrColInd,
+                  csrValues, csrRowOffsetsType, csrColIndType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateCsc(
+    cusparseSpMatDescr_t *spMatDescr, int64_t rows, int64_t cols, int64_t nnz,
+    void *cscColOffsets, void *cscRowInd, void *cscValues,
+    cusparseIndexType_t cscColOffsetsType, cusparseIndexType_t cscRowIndType,
+    cusparseIndexBase_t idxBase, cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpMatDescr_t *, int64_t, int64_t, int64_t, void *, void *, void *,
+      cusparseIndexType_t, cusparseIndexType_t, cusparseIndexBase_t,
+      cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateCsc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, cscColOffsets, cscRowInd,
+                  cscValues, cscColOffsetsType, cscRowIndType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateConstCsc(
+    cusparseConstSpMatDescr_t *spMatDescr, int64_t rows, int64_t cols,
+    int64_t nnz, const void *cscColOffsets, const void *cscRowInd,
+    const void *cscValues, cusparseIndexType_t cscColOffsetsType,
+    cusparseIndexType_t cscRowIndType, cusparseIndexBase_t idxBase,
+    cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpMatDescr_t *, int64_t, int64_t, int64_t, const void *,
+      const void *, const void *, cusparseIndexType_t, cusparseIndexType_t,
+      cusparseIndexBase_t, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateConstCsc");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, cscColOffsets, cscRowInd,
+                  cscValues, cscColOffsetsType, cscRowIndType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCsrGet(
+    cusparseSpMatDescr_t spMatDescr, int64_t *rows, int64_t *cols, int64_t *nnz,
+    void **csrRowOffsets, void **csrColInd, void **csrValues,
+    cusparseIndexType_t *csrRowOffsetsType, cusparseIndexType_t *csrColIndType,
+    cusparseIndexBase_t *idxBase, cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpMatDescr_t, int64_t *, int64_t *, int64_t *, void **, void **,
+      void **, cusparseIndexType_t *, cusparseIndexType_t *,
+      cusparseIndexBase_t *, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCsrGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, csrRowOffsets, csrColInd,
+                  csrValues, csrRowOffsetsType, csrColIndType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseConstCsrGet(
+    cusparseConstSpMatDescr_t spMatDescr, int64_t *rows, int64_t *cols,
+    int64_t *nnz, const void **csrRowOffsets, const void **csrColInd,
+    const void **csrValues, cusparseIndexType_t *csrRowOffsetsType,
+    cusparseIndexType_t *csrColIndType, cusparseIndexBase_t *idxBase,
+    cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpMatDescr_t, int64_t *, int64_t *, int64_t *, const void **,
+      const void **, const void **, cusparseIndexType_t *,
+      cusparseIndexType_t *, cusparseIndexBase_t *, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstCsrGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, csrRowOffsets, csrColInd,
+                  csrValues, csrRowOffsetsType, csrColIndType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCscGet(
+    cusparseSpMatDescr_t spMatDescr, int64_t *rows, int64_t *cols, int64_t *nnz,
+    void **cscColOffsets, void **cscRowInd, void **cscValues,
+    cusparseIndexType_t *cscColOffsetsType, cusparseIndexType_t *cscRowIndType,
+    cusparseIndexBase_t *idxBase, cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpMatDescr_t, int64_t *, int64_t *, int64_t *, void **, void **,
+      void **, cusparseIndexType_t *, cusparseIndexType_t *,
+      cusparseIndexBase_t *, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCscGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, cscColOffsets, cscRowInd,
+                  cscValues, cscColOffsetsType, cscRowIndType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseConstCscGet(
+    cusparseConstSpMatDescr_t spMatDescr, int64_t *rows, int64_t *cols,
+    int64_t *nnz, const void **cscColOffsets, const void **cscRowInd,
+    const void **cscValues, cusparseIndexType_t *cscColOffsetsType,
+    cusparseIndexType_t *cscRowIndType, cusparseIndexBase_t *idxBase,
+    cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpMatDescr_t, int64_t *, int64_t *, int64_t *, const void **,
+      const void **, const void **, cusparseIndexType_t *,
+      cusparseIndexType_t *, cusparseIndexBase_t *, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstCscGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, cscColOffsets, cscRowInd,
+                  cscValues, cscColOffsetsType, cscRowIndType, idxBase,
+                  valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCsrSetPointers(cusparseSpMatDescr_t spMatDescr, void *csrRowOffsets,
+                       void *csrColInd, void *csrValues) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpMatDescr_t, void *,
+                                                  void *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCsrSetPointers");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, csrRowOffsets, csrColInd, csrValues);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCscSetPointers(cusparseSpMatDescr_t spMatDescr, void *cscColOffsets,
+                       void *cscRowInd, void *cscValues) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpMatDescr_t, void *,
+                                                  void *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCscSetPointers");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, cscColOffsets, cscRowInd, cscValues);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateCoo(cusparseSpMatDescr_t *spMatDescr,
+                                               int64_t rows, int64_t cols,
+                                               int64_t nnz, void *cooRowInd,
+                                               void *cooColInd, void *cooValues,
+                                               cusparseIndexType_t cooIdxType,
+                                               cusparseIndexBase_t idxBase,
+                                               cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpMatDescr_t *, int64_t, int64_t, int64_t, void *, void *, void *,
+      cusparseIndexType_t, cusparseIndexBase_t, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateCoo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, cooRowInd, cooColInd, cooValues,
+                  cooIdxType, idxBase, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateConstCoo(
+    cusparseConstSpMatDescr_t *spMatDescr, int64_t rows, int64_t cols,
+    int64_t nnz, const void *cooRowInd, const void *cooColInd,
+    const void *cooValues, cusparseIndexType_t cooIdxType,
+    cusparseIndexBase_t idxBase, cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpMatDescr_t *, int64_t, int64_t, int64_t, const void *,
+      const void *, const void *, cusparseIndexType_t, cusparseIndexBase_t,
+      cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateConstCoo");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, cooRowInd, cooColInd, cooValues,
+                  cooIdxType, idxBase, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCooGet(
+    cusparseSpMatDescr_t spMatDescr, int64_t *rows, int64_t *cols, int64_t *nnz,
+    void **cooRowInd,  // COO row indices
+    void **cooColInd,  // COO column indices
+    void **cooValues,  // COO values
+    cusparseIndexType_t *idxType, cusparseIndexBase_t *idxBase,
+    cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpMatDescr_t, int64_t *, int64_t *, int64_t *,
+      void **,  // COO row indices
+      void **,  // COO column indices
+      void **,  // COO values
+      cusparseIndexType_t *, cusparseIndexBase_t *, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCooGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, cooRowInd, cooColInd, cooValues,
+                  idxType, idxBase, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseConstCooGet(cusparseConstSpMatDescr_t spMatDescr, int64_t *rows,
+                    int64_t *cols, int64_t *nnz,
+                    const void **cooRowInd,  // COO row indices
+                    const void **cooColInd,  // COO column indices
+                    const void **cooValues,  // COO values
+                    cusparseIndexType_t *idxType, cusparseIndexBase_t *idxBase,
+                    cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpMatDescr_t, int64_t *, int64_t *, int64_t *,
+      const void **,  // COO row indices
+      const void **,  // COO column indices
+      const void **,  // COO values
+      cusparseIndexType_t *, cusparseIndexBase_t *, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstCooGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, nnz, cooRowInd, cooColInd, cooValues,
+                  idxType, idxBase, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCooSetPointers(cusparseSpMatDescr_t spMatDescr, void *cooRows,
+                       void *cooColumns, void *cooValues) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpMatDescr_t, void *,
+                                                  void *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCooSetPointers");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, cooRows, cooColumns, cooValues);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateBlockedEll(
+    cusparseSpMatDescr_t *spMatDescr, int64_t rows, int64_t cols,
+    int64_t ellBlockSize, int64_t ellCols, void *ellColInd, void *ellValue,
+    cusparseIndexType_t ellIdxType, cusparseIndexBase_t idxBase,
+    cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpMatDescr_t *, int64_t, int64_t, int64_t, int64_t, void *,
+      void *, cusparseIndexType_t, cusparseIndexBase_t, cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateBlockedEll");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, ellBlockSize, ellCols, ellColInd,
+                  ellValue, ellIdxType, idxBase, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateConstBlockedEll(
+    cusparseConstSpMatDescr_t *spMatDescr, int64_t rows, int64_t cols,
+    int64_t ellBlockSize, int64_t ellCols, const void *ellColInd,
+    const void *ellValue, cusparseIndexType_t ellIdxType,
+    cusparseIndexBase_t idxBase, cudaDataType valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpMatDescr_t *, int64_t, int64_t, int64_t, int64_t,
+      const void *, const void *, cusparseIndexType_t, cusparseIndexBase_t,
+      cudaDataType);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateConstBlockedEll");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, ellBlockSize, ellCols, ellColInd,
+                  ellValue, ellIdxType, idxBase, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseBlockedEllGet(
+    cusparseSpMatDescr_t spMatDescr, int64_t *rows, int64_t *cols,
+    int64_t *ellBlockSize, int64_t *ellCols, void **ellColInd, void **ellValue,
+    cusparseIndexType_t *ellIdxType, cusparseIndexBase_t *idxBase,
+    cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseSpMatDescr_t, int64_t *, int64_t *, int64_t *, int64_t *, void **,
+      void **, cusparseIndexType_t *, cusparseIndexBase_t *, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseBlockedEllGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, ellBlockSize, ellCols, ellColInd,
+                  ellValue, ellIdxType, idxBase, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseConstBlockedEllGet(
+    cusparseConstSpMatDescr_t spMatDescr, int64_t *rows, int64_t *cols,
+    int64_t *ellBlockSize, int64_t *ellCols, const void **ellColInd,
+    const void **ellValue, cusparseIndexType_t *ellIdxType,
+    cusparseIndexBase_t *idxBase, cudaDataType *valueType) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstSpMatDescr_t, int64_t *, int64_t *, int64_t *, int64_t *,
+      const void **, const void **, cusparseIndexType_t *,
+      cusparseIndexBase_t *, cudaDataType *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstBlockedEllGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spMatDescr, rows, cols, ellBlockSize, ellCols, ellColInd,
+                  ellValue, ellIdxType, idxBase, valueType);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseCreateDnMat(
+    cusparseDnMatDescr_t *dnMatDescr, int64_t rows, int64_t cols, int64_t ld,
+    void *values, cudaDataType valueType, cusparseOrder_t order) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseDnMatDescr_t *, int64_t, int64_t, int64_t, void *, cudaDataType,
+      cusparseOrder_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateDnMat");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnMatDescr, rows, cols, ld, values, valueType, order);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseCreateConstDnMat(cusparseConstDnMatDescr_t *dnMatDescr, int64_t rows,
+                         int64_t cols, int64_t ld, const void *values,
+                         cudaDataType valueType, cusparseOrder_t order) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstDnMatDescr_t *, int64_t, int64_t, int64_t, const void *,
+      cudaDataType, cusparseOrder_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseCreateConstDnMat");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnMatDescr, rows, cols, ld, values, valueType, order);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDestroyDnMat(cusparseConstDnMatDescr_t dnMatDescr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseConstDnMatDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDestroyDnMat");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnMatDescr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDnMatGet(cusparseDnMatDescr_t dnMatDescr,
+                                              int64_t *rows, int64_t *cols,
+                                              int64_t *ld, void **values,
+                                              cudaDataType *type,
+                                              cusparseOrder_t *order) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseDnMatDescr_t, int64_t *, int64_t *, int64_t *, void **,
+      cudaDataType *, cusparseOrder_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDnMatGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnMatDescr, rows, cols, ld, values, type, order);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseConstDnMatGet(cusparseConstDnMatDescr_t dnMatDescr, int64_t *rows,
+                      int64_t *cols, int64_t *ld, const void **values,
+                      cudaDataType *type, cusparseOrder_t *order) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseConstDnMatDescr_t, int64_t *, int64_t *, int64_t *, const void **,
+      cudaDataType *, cusparseOrder_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstDnMatGet");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnMatDescr, rows, cols, ld, values, type, order);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDnMatGetValues(cusparseDnMatDescr_t dnMatDescr, void **values) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseDnMatDescr_t, void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDnMatGetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnMatDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseConstDnMatGetValues(
+    cusparseConstDnMatDescr_t dnMatDescr, const void **values) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseConstDnMatDescr_t, const void **);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseConstDnMatGetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnMatDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDnMatSetValues(cusparseDnMatDescr_t dnMatDescr, void *values) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseDnMatDescr_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDnMatSetValues");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnMatDescr, values);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDnMatSetStridedBatch(
+    cusparseDnMatDescr_t dnMatDescr, int batchCount, int64_t batchStride) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseDnMatDescr_t, int, int64_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDnMatSetStridedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnMatDescr, batchCount, batchStride);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseDnMatGetStridedBatch(cusparseConstDnMatDescr_t dnMatDescr,
+                             int *batchCount, int64_t *batchStride) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseConstDnMatDescr_t,
+                                                  int *, int64_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDnMatGetStridedBatch");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(dnMatDescr, batchCount, batchStride);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseAxpby(cusparseHandle_t handle,
+                                           const void *alpha,
+                                           cusparseConstSpVecDescr_t vecX,
+                                           const void *beta,
+                                           cusparseDnVecDescr_t vecY) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, const void *, cusparseConstSpVecDescr_t, const void *,
+      cusparseDnVecDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseAxpby");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, alpha, vecX, beta, vecY);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseGather(cusparseHandle_t handle,
+                                            cusparseConstDnVecDescr_t vecY,
+                                            cusparseSpVecDescr_t vecX) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseConstDnVecDescr_t, cusparseSpVecDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseGather");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, vecY, vecX);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseScatter(cusparseHandle_t handle,
+                                             cusparseConstSpVecDescr_t vecX,
+                                             cusparseDnVecDescr_t vecY) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseConstSpVecDescr_t, cusparseDnVecDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseScatter");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, vecX, vecY);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseRot(cusparseHandle_t handle,
+                                         const void *c_coeff,
+                                         const void *s_coeff,
+                                         cusparseSpVecDescr_t vecX,
+                                         cusparseDnVecDescr_t vecY) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, const void *, const void *, cusparseSpVecDescr_t,
+      cusparseDnVecDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseRot");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, c_coeff, s_coeff, vecX, vecY);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpVV_bufferSize(
+    cusparseHandle_t handle, cusparseOperation_t opX,
+    cusparseConstSpVecDescr_t vecX, cusparseConstDnVecDescr_t vecY,
+    const void *result, cudaDataType computeType, size_t *bufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseConstSpVecDescr_t,
+      cusparseConstDnVecDescr_t, const void *, cudaDataType, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpVV_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opX, vecX, vecY, result, computeType, bufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpVV(cusparseHandle_t handle, cusparseOperation_t opX,
+             cusparseConstSpVecDescr_t vecX, cusparseConstDnVecDescr_t vecY,
+             void *result, cudaDataType computeType, void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseConstSpVecDescr_t,
+      cusparseConstDnVecDescr_t, void *, cudaDataType, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpVV");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opX, vecX, vecY, result, computeType, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSparseToDense_bufferSize(
+    cusparseHandle_t handle, cusparseConstSpMatDescr_t matA,
+    cusparseDnMatDescr_t matB, cusparseSparseToDenseAlg_t alg,
+    size_t *bufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseConstSpMatDescr_t, cusparseDnMatDescr_t,
+      cusparseSparseToDenseAlg_t, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSparseToDense_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, matA, matB, alg, bufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSparseToDense(cusparseHandle_t handle, cusparseConstSpMatDescr_t matA,
+                      cusparseDnMatDescr_t matB, cusparseSparseToDenseAlg_t alg,
+                      void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseConstSpMatDescr_t, cusparseDnMatDescr_t,
+      cusparseSparseToDenseAlg_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSparseToDense");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, matA, matB, alg, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDenseToSparse_bufferSize(
+    cusparseHandle_t handle, cusparseConstDnMatDescr_t matA,
+    cusparseSpMatDescr_t matB, cusparseDenseToSparseAlg_t alg,
+    size_t *bufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseConstDnMatDescr_t, cusparseSpMatDescr_t,
+      cusparseDenseToSparseAlg_t, size_t *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseDenseToSparse_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, matA, matB, alg, bufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDenseToSparse_analysis(
+    cusparseHandle_t handle, cusparseConstDnMatDescr_t matA,
+    cusparseSpMatDescr_t matB, cusparseDenseToSparseAlg_t alg,
+    void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseConstDnMatDescr_t, cusparseSpMatDescr_t,
+      cusparseDenseToSparseAlg_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDenseToSparse_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, matA, matB, alg, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseDenseToSparse_convert(
+    cusparseHandle_t handle, cusparseConstDnMatDescr_t matA,
+    cusparseSpMatDescr_t matB, cusparseDenseToSparseAlg_t alg,
+    void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseConstDnMatDescr_t, cusparseSpMatDescr_t,
+      cusparseDenseToSparseAlg_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseDenseToSparse_convert");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, matA, matB, alg, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMV(
+    cusparseHandle_t handle, cusparseOperation_t opA, const void *alpha,
+    cusparseConstSpMatDescr_t matA, cusparseConstDnVecDescr_t vecX,
+    const void *beta, cusparseDnVecDescr_t vecY, cudaDataType computeType,
+    cusparseSpMVAlg_t alg, void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnVecDescr_t, const void *,
+      cusparseDnVecDescr_t, cudaDataType, cusparseSpMVAlg_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMV");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, alpha, matA, vecX, beta, vecY, computeType, alg,
+                  externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMV_bufferSize(
+    cusparseHandle_t handle, cusparseOperation_t opA, const void *alpha,
+    cusparseConstSpMatDescr_t matA, cusparseConstDnVecDescr_t vecX,
+    const void *beta, cusparseDnVecDescr_t vecY, cudaDataType computeType,
+    cusparseSpMVAlg_t alg, size_t *bufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnVecDescr_t, const void *,
+      cusparseDnVecDescr_t, cudaDataType, cusparseSpMVAlg_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMV_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, alpha, matA, vecX, beta, vecY, computeType, alg,
+                  bufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpSV_createDescr(cusparseSpSVDescr_t *descr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpSVDescr_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpSV_createDescr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descr);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpSV_destroyDescr(cusparseSpSVDescr_t descr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpSVDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpSV_destroyDescr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpSV_bufferSize(
+    cusparseHandle_t handle, cusparseOperation_t opA, const void *alpha,
+    cusparseConstSpMatDescr_t matA, cusparseConstDnVecDescr_t vecX,
+    cusparseDnVecDescr_t vecY, cudaDataType computeType, cusparseSpSVAlg_t alg,
+    cusparseSpSVDescr_t spsvDescr, size_t *bufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnVecDescr_t,
+      cusparseDnVecDescr_t, cudaDataType, cusparseSpSVAlg_t,
+      cusparseSpSVDescr_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpSV_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, alpha, matA, vecX, vecY, computeType, alg,
+                  spsvDescr, bufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpSV_analysis(
+    cusparseHandle_t handle, cusparseOperation_t opA, const void *alpha,
+    cusparseConstSpMatDescr_t matA, cusparseConstDnVecDescr_t vecX,
+    cusparseDnVecDescr_t vecY, cudaDataType computeType, cusparseSpSVAlg_t alg,
+    cusparseSpSVDescr_t spsvDescr, void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnVecDescr_t,
+      cusparseDnVecDescr_t, cudaDataType, cusparseSpSVAlg_t,
+      cusparseSpSVDescr_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpSV_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, alpha, matA, vecX, vecY, computeType, alg,
+                  spsvDescr, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpSV_solve(
+    cusparseHandle_t handle, cusparseOperation_t opA, const void *alpha,
+    cusparseConstSpMatDescr_t matA, cusparseConstDnVecDescr_t vecX,
+    cusparseDnVecDescr_t vecY, cudaDataType computeType, cusparseSpSVAlg_t alg,
+    cusparseSpSVDescr_t spsvDescr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnVecDescr_t,
+      cusparseDnVecDescr_t, cudaDataType, cusparseSpSVAlg_t,
+      cusparseSpSVDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpSV_solve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, alpha, matA, vecX, vecY, computeType, alg,
+                  spsvDescr);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpSM_createDescr(cusparseSpSMDescr_t *descr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpSMDescr_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpSM_createDescr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descr);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpSM_destroyDescr(cusparseSpSMDescr_t descr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpSMDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpSM_destroyDescr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpSM_bufferSize(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstDnMatDescr_t matB, cusparseDnMatDescr_t matC,
+    cudaDataType computeType, cusparseSpSMAlg_t alg,
+    cusparseSpSMDescr_t spsmDescr, size_t *bufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnMatDescr_t,
+      cusparseDnMatDescr_t, cudaDataType, cusparseSpSMAlg_t,
+      cusparseSpSMDescr_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpSM_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, matC, computeType, alg,
+                  spsmDescr, bufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpSM_analysis(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstDnMatDescr_t matB, cusparseDnMatDescr_t matC,
+    cudaDataType computeType, cusparseSpSMAlg_t alg,
+    cusparseSpSMDescr_t spsmDescr, void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnMatDescr_t,
+      cusparseDnMatDescr_t, cudaDataType, cusparseSpSMAlg_t,
+      cusparseSpSMDescr_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpSM_analysis");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, matC, computeType, alg,
+                  spsmDescr, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpSM_solve(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstDnMatDescr_t matB, cusparseDnMatDescr_t matC,
+    cudaDataType computeType, cusparseSpSMAlg_t alg,
+    cusparseSpSMDescr_t spsmDescr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnMatDescr_t,
+      cusparseDnMatDescr_t, cudaDataType, cusparseSpSMAlg_t,
+      cusparseSpSMDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpSM_solve");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, matC, computeType, alg,
+                  spsmDescr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMM_bufferSize(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstDnMatDescr_t matB, const void *beta, cusparseDnMatDescr_t matC,
+    cudaDataType computeType, cusparseSpMMAlg_t alg, size_t *bufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnMatDescr_t, const void *,
+      cusparseDnMatDescr_t, cudaDataType, cusparseSpMMAlg_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMM_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, bufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMM_preprocess(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstDnMatDescr_t matB, const void *beta, cusparseDnMatDescr_t matC,
+    cudaDataType computeType, cusparseSpMMAlg_t alg, void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnMatDescr_t, const void *,
+      cusparseDnMatDescr_t, cudaDataType, cusparseSpMMAlg_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMM_preprocess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMM(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstDnMatDescr_t matB, const void *beta, cusparseDnMatDescr_t matC,
+    cudaDataType computeType, cusparseSpMMAlg_t alg, void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstDnMatDescr_t, const void *,
+      cusparseDnMatDescr_t, cudaDataType, cusparseSpMMAlg_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMM");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpGEMM_createDescr(cusparseSpGEMMDescr_t *descr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpGEMMDescr_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpGEMM_createDescr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descr);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpGEMM_destroyDescr(cusparseSpGEMMDescr_t descr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpGEMMDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpGEMM_destroyDescr");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(descr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpGEMM_workEstimation(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstSpMatDescr_t matB, const void *beta, cusparseSpMatDescr_t matC,
+    cudaDataType computeType, cusparseSpGEMMAlg_t alg,
+    cusparseSpGEMMDescr_t spgemmDescr, size_t *bufferSize1,
+    void *externalBuffer1) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstSpMatDescr_t, const void *,
+      cusparseSpMatDescr_t, cudaDataType, cusparseSpGEMMAlg_t,
+      cusparseSpGEMMDescr_t, size_t *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpGEMM_workEstimation");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, spgemmDescr, bufferSize1, externalBuffer1);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpGEMM_getNumProducts(
+    cusparseSpGEMMDescr_t spgemmDescr, int64_t *num_prods) {
+  using FuncPtr =
+      cusparseStatus_t(CUSPARSEAPI *)(cusparseSpGEMMDescr_t, int64_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpGEMM_getNumProducts");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(spgemmDescr, num_prods);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpGEMM_estimateMemory(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstSpMatDescr_t matB, const void *beta, cusparseSpMatDescr_t matC,
+    cudaDataType computeType, cusparseSpGEMMAlg_t alg,
+    cusparseSpGEMMDescr_t spgemmDescr, float chunk_fraction,
+    size_t *bufferSize3, void *externalBuffer3, size_t *bufferSize2) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstSpMatDescr_t, const void *,
+      cusparseSpMatDescr_t, cudaDataType, cusparseSpGEMMAlg_t,
+      cusparseSpGEMMDescr_t, float, size_t *, void *, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpGEMM_estimateMemory");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, spgemmDescr, chunk_fraction, bufferSize3,
+                  externalBuffer3, bufferSize2);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpGEMM_compute(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstSpMatDescr_t matB, const void *beta, cusparseSpMatDescr_t matC,
+    cudaDataType computeType, cusparseSpGEMMAlg_t alg,
+    cusparseSpGEMMDescr_t spgemmDescr, size_t *bufferSize2,
+    void *externalBuffer2) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstSpMatDescr_t, const void *,
+      cusparseSpMatDescr_t, cudaDataType, cusparseSpGEMMAlg_t,
+      cusparseSpGEMMDescr_t, size_t *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpGEMM_compute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, spgemmDescr, bufferSize2, externalBuffer2);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpGEMM_copy(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstSpMatDescr_t matB, const void *beta, cusparseSpMatDescr_t matC,
+    cudaDataType computeType, cusparseSpGEMMAlg_t alg,
+    cusparseSpGEMMDescr_t spgemmDescr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstSpMatDescr_t, const void *,
+      cusparseSpMatDescr_t, cudaDataType, cusparseSpGEMMAlg_t,
+      cusparseSpGEMMDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpGEMM_copy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, spgemmDescr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpGEMMreuse_workEstimation(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    cusparseConstSpMatDescr_t matA, cusparseConstSpMatDescr_t matB,
+    cusparseSpMatDescr_t matC, cusparseSpGEMMAlg_t alg,
+    cusparseSpGEMMDescr_t spgemmDescr, size_t *bufferSize1,
+    void *externalBuffer1) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t,
+      cusparseConstSpMatDescr_t, cusparseConstSpMatDescr_t,
+      cusparseSpMatDescr_t, cusparseSpGEMMAlg_t, cusparseSpGEMMDescr_t,
+      size_t *, void *);
+  static auto func_ptr =
+      LoadSymbol<FuncPtr>("cusparseSpGEMMreuse_workEstimation");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, matA, matB, matC, alg, spgemmDescr,
+                  bufferSize1, externalBuffer1);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpGEMMreuse_nnz(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    cusparseConstSpMatDescr_t matA, cusparseConstSpMatDescr_t matB,
+    cusparseSpMatDescr_t matC, cusparseSpGEMMAlg_t alg,
+    cusparseSpGEMMDescr_t spgemmDescr, size_t *bufferSize2,
+    void *externalBuffer2, size_t *bufferSize3, void *externalBuffer3,
+    size_t *bufferSize4, void *externalBuffer4) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t,
+      cusparseConstSpMatDescr_t, cusparseConstSpMatDescr_t,
+      cusparseSpMatDescr_t, cusparseSpGEMMAlg_t, cusparseSpGEMMDescr_t,
+      size_t *, void *, size_t *, void *, size_t *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpGEMMreuse_nnz");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, matA, matB, matC, alg, spgemmDescr,
+                  bufferSize2, externalBuffer2, bufferSize3, externalBuffer3,
+                  bufferSize4, externalBuffer4);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpGEMMreuse_copy(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    cusparseConstSpMatDescr_t matA, cusparseConstSpMatDescr_t matB,
+    cusparseSpMatDescr_t matC, cusparseSpGEMMAlg_t alg,
+    cusparseSpGEMMDescr_t spgemmDescr, size_t *bufferSize5,
+    void *externalBuffer5) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t,
+      cusparseConstSpMatDescr_t, cusparseConstSpMatDescr_t,
+      cusparseSpMatDescr_t, cusparseSpGEMMAlg_t, cusparseSpGEMMDescr_t,
+      size_t *, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpGEMMreuse_copy");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, matA, matB, matC, alg, spgemmDescr,
+                  bufferSize5, externalBuffer5);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpGEMMreuse_compute(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstSpMatDescr_t matA,
+    cusparseConstSpMatDescr_t matB, const void *beta, cusparseSpMatDescr_t matC,
+    cudaDataType computeType, cusparseSpGEMMAlg_t alg,
+    cusparseSpGEMMDescr_t spgemmDescr) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstSpMatDescr_t, cusparseConstSpMatDescr_t, const void *,
+      cusparseSpMatDescr_t, cudaDataType, cusparseSpGEMMAlg_t,
+      cusparseSpGEMMDescr_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpGEMMreuse_compute");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, spgemmDescr);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSDDMM_bufferSize(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstDnMatDescr_t matA,
+    cusparseConstDnMatDescr_t matB, const void *beta, cusparseSpMatDescr_t matC,
+    cudaDataType computeType, cusparseSDDMMAlg_t alg, size_t *bufferSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstDnMatDescr_t, cusparseConstDnMatDescr_t, const void *,
+      cusparseSpMatDescr_t, cudaDataType, cusparseSDDMMAlg_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSDDMM_bufferSize");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, bufferSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSDDMM_preprocess(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstDnMatDescr_t matA,
+    cusparseConstDnMatDescr_t matB, const void *beta, cusparseSpMatDescr_t matC,
+    cudaDataType computeType, cusparseSDDMMAlg_t alg, void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstDnMatDescr_t, cusparseConstDnMatDescr_t, const void *,
+      cusparseSpMatDescr_t, cudaDataType, cusparseSDDMMAlg_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSDDMM_preprocess");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSDDMM(
+    cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB,
+    const void *alpha, cusparseConstDnMatDescr_t matA,
+    cusparseConstDnMatDescr_t matB, const void *beta, cusparseSpMatDescr_t matC,
+    cudaDataType computeType, cusparseSDDMMAlg_t alg, void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseOperation_t, cusparseOperation_t, const void *,
+      cusparseConstDnMatDescr_t, cusparseConstDnMatDescr_t, const void *,
+      cusparseSpMatDescr_t, cudaDataType, cusparseSDDMMAlg_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSDDMM");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, opA, opB, alpha, matA, matB, beta, matC, computeType,
+                  alg, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMMOp_createPlan(
+    cusparseHandle_t handle, cusparseSpMMOpPlan_t *plan,
+    cusparseOperation_t opA, cusparseOperation_t opB,
+    cusparseConstSpMatDescr_t matA, cusparseConstDnMatDescr_t matB,
+    cusparseDnMatDescr_t matC, cudaDataType computeType,
+    cusparseSpMMOpAlg_t alg, const void *addOperationNvvmBuffer,
+    size_t addOperationBufferSize, const void *mulOperationNvvmBuffer,
+    size_t mulOperationBufferSize, const void *epilogueNvvmBuffer,
+    size_t epilogueBufferSize, size_t *SpMMWorkspaceSize) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(
+      cusparseHandle_t, cusparseSpMMOpPlan_t *, cusparseOperation_t,
+      cusparseOperation_t, cusparseConstSpMatDescr_t, cusparseConstDnMatDescr_t,
+      cusparseDnMatDescr_t, cudaDataType, cusparseSpMMOpAlg_t, const void *,
+      size_t, const void *, size_t, const void *, size_t, size_t *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMMOp_createPlan");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(handle, plan, opA, opB, matA, matB, matC, computeType, alg,
+                  addOperationNvvmBuffer, addOperationBufferSize,
+                  mulOperationNvvmBuffer, mulOperationBufferSize,
+                  epilogueNvvmBuffer, epilogueBufferSize, SpMMWorkspaceSize);
+}
+
+cusparseStatus_t CUSPARSEAPI cusparseSpMMOp(cusparseSpMMOpPlan_t plan,
+                                            void *externalBuffer) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpMMOpPlan_t, void *);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMMOp");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(plan, externalBuffer);
+}
+
+cusparseStatus_t CUSPARSEAPI
+cusparseSpMMOp_destroyPlan(cusparseSpMMOpPlan_t plan) {
+  using FuncPtr = cusparseStatus_t(CUSPARSEAPI *)(cusparseSpMMOpPlan_t);
+  static auto func_ptr = LoadSymbol<FuncPtr>("cusparseSpMMOp_destroyPlan");
+  if (!func_ptr) return GetSymbolNotFoundError();
+  return func_ptr(plan);
+}
+
+}  // extern "C"
diff --git a/tensorflow/stream_executor/cuda/cusparse_stub.cc b/tensorflow/stream_executor/cuda/cusparse_stub.cc
index 439de5eb83a..751ecb838eb 100644
--- a/tensorflow/stream_executor/cuda/cusparse_stub.cc
+++ b/tensorflow/stream_executor/cuda/cusparse_stub.cc
@@ -50,8 +50,4 @@ cusparseStatus_t GetSymbolNotFoundError() {
 }
 }  // namespace
 
-#if CUDA_VERSION < 9020
-#include "tensorflow/stream_executor/cuda/cusparse_9_0.inc"
-#else
-#include "tensorflow/stream_executor/cuda/cusparse_10_0.inc"
-#endif
+#include "tensorflow/stream_executor/cuda/cusparse_12_0.inc"
diff --git a/tensorflow/stream_executor/platform/default/dso_loader.cc b/tensorflow/stream_executor/platform/default/dso_loader.cc
index 9ae8b41ccf4..8f4d9904322 100644
--- a/tensorflow/stream_executor/platform/default/dso_loader.cc
+++ b/tensorflow/stream_executor/platform/default/dso_loader.cc
@@ -73,7 +73,7 @@ port::StatusOr<void*> GetCudaDriverDsoHandle() {
 }
 
 port::StatusOr<void*> GetCudaRuntimeDsoHandle() {
-  return GetDsoHandle("cudart", GetCudaVersion());
+  return GetDsoHandle("cudart", GetCudaLibVersion());
 }
 
 port::StatusOr<void*> GetCublasDsoHandle() {
@@ -81,11 +81,11 @@ port::StatusOr<void*> GetCublasDsoHandle() {
 }
 
 port::StatusOr<void*> GetCufftDsoHandle() {
-  return GetDsoHandle("cufft", GetCudaLibVersion());
+  return GetDsoHandle("cufft", "11");
 }
 
 port::StatusOr<void*> GetCusolverDsoHandle() {
-  return GetDsoHandle("cusolver", GetCudaLibVersion());
+  return GetDsoHandle("cusolver", "11");
 }
 
 port::StatusOr<void*> GetCusparseDsoHandle() {
@@ -93,7 +93,7 @@ port::StatusOr<void*> GetCusparseDsoHandle() {
 }
 
 port::StatusOr<void*> GetCurandDsoHandle() {
-  return GetDsoHandle("curand", GetCudaLibVersion());
+  return GetDsoHandle("curand", "10");
 }
 
 port::StatusOr<void*> GetCuptiDsoHandle() {
diff --git a/tensorflow/stream_executor/stream_executor_pimpl.h b/tensorflow/stream_executor/stream_executor_pimpl.h
index efa4034c88a..aae01ce94a7 100644
--- a/tensorflow/stream_executor/stream_executor_pimpl.h
+++ b/tensorflow/stream_executor/stream_executor_pimpl.h
@@ -772,7 +772,7 @@ inline port::StatusOr<std::unique_ptr<TypedKernel<Args...>>>
 StreamExecutor::CreateTypedKernel(absl::string_view kernel_name,
                                   absl::string_view ptx,
                                   absl::Span<const uint8> cubin_data) {
-  auto kernel_base = absl::make_unique<TypedKernel<Args...>>(this);
+  auto kernel_base = std::make_unique<TypedKernel<Args...>>(this);
   MultiKernelLoaderSpec loader_spec(kernel_base->kNumberOfParameters);
   loader_spec.AddCudaPtxInMemory(ptx, kernel_name);
 
diff --git a/tensorflow/tools/lib_package/BUILD b/tensorflow/tools/lib_package/BUILD
index f7b9255092f..c6cbb2cdb62 100644
--- a/tensorflow/tools/lib_package/BUILD
+++ b/tensorflow/tools/lib_package/BUILD
@@ -187,7 +187,6 @@ genrule(
         ],
         "//conditions:default": [],
     }) + if_cuda([
-        "@cub_archive//:LICENSE.TXT",
         "@local_config_nccl//:LICENSE",
     ]) + if_mkl([
         "//third_party/mkl:LICENSE",
@@ -261,7 +260,6 @@ genrule(
         ],
         "//conditions:default": [],
     }) + if_cuda([
-        "@cub_archive//:LICENSE.TXT",
         "@local_config_nccl//:LICENSE",
     ]) + if_mkl([
         "//third_party/mkl:LICENSE",
diff --git a/tensorflow/tools/pip_package/BUILD b/tensorflow/tools/pip_package/BUILD
index 1573618f7c3..ccc5bf97ada 100644
--- a/tensorflow/tools/pip_package/BUILD
+++ b/tensorflow/tools/pip_package/BUILD
@@ -156,7 +156,7 @@ filegroup(
         "@nsync//:LICENSE",
         "@opt_einsum_archive//:LICENSE",
         "@pasta//:LICENSE",
-        "@pcre//:LICENCE",
+        "@pcre2//:LICENCE.md",
         "@png_archive//:LICENSE",
         "@com_google_protobuf//:LICENSE",
         "@six_archive//:LICENSE",
@@ -199,7 +199,6 @@ filegroup(
         ],
         "//conditions:default": [],
     }) + if_cuda([
-        "@cub_archive//:LICENSE.TXT",
         "@local_config_nccl//:LICENSE",
     ]) + if_mkl([
         "//third_party/mkl:LICENSE",
diff --git a/tensorflow/tools/pip_package/setup.py b/tensorflow/tools/pip_package/setup.py
index f47878fdd7f..546e381c0df 100644
--- a/tensorflow/tools/pip_package/setup.py
+++ b/tensorflow/tools/pip_package/setup.py
@@ -62,11 +62,11 @@ REQUIRED_PACKAGES = [
     'keras_preprocessing >= 1.0.5',
     # mock comes with unittest.mock for python3, need to install for python2
     'mock >= 2.0.0;python_version<"3"',
-    'numpy >= 1.16.0, < 1.19.0',
+    'numpy >= 1.16.0, < 2',
     'opt_einsum >= 2.3.2',
     'six >= 1.10.0',
-    'protobuf >= 3.6.1',
-    'tensorboard >= 1.15.0, < 1.16.0',
+    'protobuf >= 3.6.1, < 4',
+    'tensorboard >= 1.15.0',
     'tensorflow-estimator == 1.15.1',
     'termcolor >= 1.1.0',
     # python3 requires wheel 0.26
@@ -74,7 +74,7 @@ REQUIRED_PACKAGES = [
     'wheel;python_version<"3"',
     'wrapt >= 1.11.1',
     # Pin h5py to at most 2.10.0 as newer versions break old keras tests
-    'h5py <= 2.10.0',
+    'h5py',
 ]
 
 if sys.byteorder == 'little':
diff --git a/tensorflow/workspace.bzl b/tensorflow/workspace.bzl
index 0616aa91fa6..a258abb3948 100755
--- a/tensorflow/workspace.bzl
+++ b/tensorflow/workspace.bzl
@@ -157,6 +157,7 @@ def tf_repositories(path_prefix = "", tf_repo_name = ""):
     tf_http_archive(
         name = "com_google_absl",
         build_file = clean_dep("//third_party:com_google_absl.BUILD"),
+        patch_file = clean_dep("//third_party:com_google_absl.patch"),
         sha256 = "acd93f6baaedc4414ebd08b33bebca7c7a46888916101d8c0b8083573526d070",
         strip_prefix = "abseil-cpp-43ef2148c0936ebf7cb4be6b19927a9d9d145b8f",
         urls = [
@@ -215,6 +216,7 @@ def tf_repositories(path_prefix = "", tf_repo_name = ""):
         sha256 = "fd0c3e3b50f32af332b53857f8cd1bfa009e33d1eeecabc5c79a4825d906a90c",
         strip_prefix = "google-cloud-cpp-0.10.0",
         system_build_file = clean_dep("//third_party/systemlibs:google_cloud_cpp.BUILD"),
+	patch_file = "//third_party:google_cloud_cpp.patch",
         system_link_files = {
             "//third_party/systemlibs:google_cloud_cpp.google.cloud.bigtable.BUILD": "google/cloud/bigtable/BUILD",
         },
@@ -370,7 +372,8 @@ def tf_repositories(path_prefix = "", tf_repo_name = ""):
         name = "absl_py",
         sha256 = "3d0f39e0920379ff1393de04b573bca3484d82a5f8b939e9e83b20b6106c9bbe",
         strip_prefix = "abseil-py-pypi-v0.7.1",
-        system_build_file = clean_dep("//third_party/systemlibs:absl_py.BUILD"),
+        patch_file = clean_dep("//third_party/systemlibs:absl_py.patch"),
+	system_build_file = clean_dep("//third_party/systemlibs:absl_py.BUILD"),
         system_link_files = {
             "//third_party/systemlibs:absl_py.absl.BUILD": "absl/BUILD",
             "//third_party/systemlibs:absl_py.absl.flags.BUILD": "absl/flags/BUILD",
@@ -434,7 +437,7 @@ def tf_repositories(path_prefix = "", tf_repo_name = ""):
         sha256 = PROTOBUF_SHA256,
         strip_prefix = PROTOBUF_STRIP_PREFIX,
         system_build_file = clean_dep("//third_party/systemlibs:protobuf.BUILD"),
-        system_link_files = {
+	system_link_files = {
             "//third_party/systemlibs:protobuf.bzl": "protobuf.bzl",
         },
         urls = PROTOBUF_URLS,
@@ -472,27 +475,27 @@ def tf_repositories(path_prefix = "", tf_repo_name = ""):
     )
 
     tf_http_archive(
-        name = "pcre",
-        build_file = clean_dep("//third_party:pcre.BUILD"),
-        sha256 = "aecafd4af3bd0f3935721af77b889d9024b2e01d96b58471bd91a3063fb47728",
-        strip_prefix = "pcre-8.44",
-        system_build_file = clean_dep("//third_party/systemlibs:pcre.BUILD"),
+        name = "pcre2",
+        build_file = clean_dep("//third_party:pcre2.BUILD"),
+        sha256 = "409c443549b13b216da40049850a32f3e6c57d4224ab11553ab5a786878a158e",
+        strip_prefix = "pcre2-pcre2-10.47",
+        system_build_file = clean_dep("//third_party/systemlibs:pcre2.BUILD"),
         urls = [
-            "https://storage.googleapis.com/mirror.tensorflow.org/ftp.exim.org/pub/pcre/pcre-8.44.tar.gz",
-            "https://ftp.exim.org/pub/pcre/pcre-8.44.tar.gz",
+            "https://storage.googleapis.com/mirror.tensorflow.org/github.com/PCRE2Project/pcre2/archive//pcre2-10.47.tar.gz",
+	    "https://github.com/PCRE2Project/pcre2/archive//pcre2-10.47.tar.gz"
         ],
     )
 
     tf_http_archive(
         name = "swig",
         build_file = clean_dep("//third_party:swig.BUILD"),
-        sha256 = "58a475dbbd4a4d7075e5fe86d4e54c9edde39847cdb96a3053d87cb64a23a453",
-        strip_prefix = "swig-3.0.8",
+        sha256 = "40162a706c56f7592d08fd52ef5511cb7ac191f3593cf07306a0a554c6281fcf",
+        strip_prefix = "swig-4.4.1",
         system_build_file = clean_dep("//third_party/systemlibs:swig.BUILD"),
         urls = [
-            "https://storage.googleapis.com/mirror.tensorflow.org/ufpr.dl.sourceforge.net/project/swig/swig/swig-3.0.8/swig-3.0.8.tar.gz",
-            "http://ufpr.dl.sourceforge.net/project/swig/swig/swig-3.0.8/swig-3.0.8.tar.gz",
-            "http://pilotfiber.dl.sourceforge.net/project/swig/swig/swig-3.0.8/swig-3.0.8.tar.gz",
+            "https://storage.googleapis.com/mirror.tensorflow.org/ufpr.dl.sourceforge.net/project/swig/swig/swig-4.4.1/swig-4.4.1.tar.gz",
+            "http://ufpr.dl.sourceforge.net/project/swig/swig/swig-4.4.1/swig-4.4.1.tar.gz",
+            "http://pilotfiber.dl.sourceforge.net/project/swig/swig/swig-4.4.1/swig-4.4.1.tar.gz",
         ],
     )
 
@@ -514,6 +517,7 @@ def tf_repositories(path_prefix = "", tf_repo_name = ""):
         sha256 = "67a6c26db56f345f7cee846e681db2c23f919eba46dd639b09462d1b6203d28c",
         strip_prefix = "grpc-4566c2a29ebec0835643b972eb99f4306c4234a3",
         system_build_file = clean_dep("//third_party/systemlibs:grpc.BUILD"),
+	patch_file = clean_dep("//third_party/systemlibs:grpc.patch"),
         urls = [
             "https://storage.googleapis.com/mirror.tensorflow.org/github.com/grpc/grpc/archive/4566c2a29ebec0835643b972eb99f4306c4234a3.tar.gz",
             "https://github.com/grpc/grpc/archive/4566c2a29ebec0835643b972eb99f4306c4234a3.tar.gz",
@@ -584,6 +588,7 @@ def tf_repositories(path_prefix = "", tf_repo_name = ""):
         sha256 = "1188e29000013ed6517168600fc35a010d58c5d321846d6a6dfee74e4c788b45",
         strip_prefix = "boringssl-7f634429a04abc48e2eb041c81c5235816c96514",
         system_build_file = clean_dep("//third_party/systemlibs:boringssl.BUILD"),
+        patch_file = clean_dep("//third_party/boringssl:patch"),
         urls = [
             "https://storage.googleapis.com/mirror.tensorflow.org/github.com/google/boringssl/archive/7f634429a04abc48e2eb041c81c5235816c96514.tar.gz",
             "https://github.com/google/boringssl/archive/7f634429a04abc48e2eb041c81c5235816c96514.tar.gz",
@@ -728,17 +733,6 @@ def tf_repositories(path_prefix = "", tf_repo_name = ""):
         ],
     )
 
-    tf_http_archive(
-        name = "cub_archive",
-        build_file = clean_dep("//third_party:cub.BUILD"),
-        sha256 = "6bfa06ab52a650ae7ee6963143a0bbc667d6504822cbd9670369b598f18c58c3",
-        strip_prefix = "cub-1.8.0",
-        urls = [
-            "https://storage.googleapis.com/mirror.tensorflow.org/github.com/NVlabs/cub/archive/1.8.0.zip",
-            "https://github.com/NVlabs/cub/archive/1.8.0.zip",
-        ],
-    )
-
     tf_http_archive(
         name = "rocprim_archive",
         build_file = clean_dep("//third_party:rocprim.BUILD"),
@@ -754,12 +748,12 @@ def tf_repositories(path_prefix = "", tf_repo_name = ""):
         name = "cython",
         build_file = clean_dep("//third_party:cython.BUILD"),
         delete = ["BUILD.bazel"],
-        sha256 = "bccc9aa050ea02595b2440188813b936eaf345e85fb9692790cecfe095cf91aa",
-        strip_prefix = "cython-0.28.4",
+        sha256 = "824eb14045d85c5af677536134199dd6709db8fb0835452fd2d54bc3c8df8887",
+        strip_prefix = "cython-0.29.37",
         system_build_file = clean_dep("//third_party/systemlibs:cython.BUILD"),
         urls = [
-            "https://storage.googleapis.com/mirror.tensorflow.org/github.com/cython/cython/archive/0.28.4.tar.gz",
-            "https://github.com/cython/cython/archive/0.28.4.tar.gz",
+            "https://mirror.bazel.build/github.com/cython/cython/archive/0.29.37.tar.gz",
+            "https://github.com/cython/cython/archive/0.29.37.tar.gz",
         ],
     )
 
@@ -925,23 +919,23 @@ def tf_repositories(path_prefix = "", tf_repo_name = ""):
     tf_http_archive(
         name = "pybind11",
         urls = [
-            "https://mirror.bazel.build/github.com/pybind/pybind11/archive/v2.3.0.tar.gz",
-            "https://github.com/pybind/pybind11/archive/v2.3.0.tar.gz",
+            "https://mirror.bazel.build/github.com/pybind/pybind11/archive/v2.13.6.tar.gz",
+            "https://github.com/pybind/pybind11/archive/v2.13.6.tar.gz",
         ],
-        sha256 = "0f34838f2c8024a6765168227ba587b3687729ebf03dc912f88ff75c7aa9cfe8",
-        strip_prefix = "pybind11-2.3.0",
+        sha256 = "e08cb87f4773da97fa7b5f035de8763abc656d87d5773e62f6da0587d1f0ec20",
+        strip_prefix = "pybind11-2.13.6",
         build_file = clean_dep("//third_party:pybind11.BUILD"),
     )
 
     tf_http_archive(
         name = "wrapt",
         build_file = clean_dep("//third_party:wrapt.BUILD"),
-        sha256 = "8a6fb40e8f8b6a66b4ba81a4044c68e6a7b1782f21cfabc06fb765332b4c3e51",
-        strip_prefix = "wrapt-1.11.1/src/wrapt",
+        sha256 = "9f289d4a27cb94eaa4ecf91cdcdb2508ba38db655ba3f43e018f88e1750b8915",
+        strip_prefix = "wrapt-2.1.2/src/wrapt",
         system_build_file = clean_dep("//third_party/systemlibs:wrapt.BUILD"),
         urls = [
-            "https://storage.googleapis.com/mirror.tensorflow.org/github.com/GrahamDumpleton/wrapt/archive/1.11.1.tar.gz",
-            "https://github.com/GrahamDumpleton/wrapt/archive/1.11.1.tar.gz",
+            "https://storage.googleapis.com/mirror.tensorflow.org/github.com/GrahamDumpleton/wrapt/archive/2.1.2.tar.gz",
+            "https://github.com/GrahamDumpleton/wrapt/archive/2.1.2.tar.gz",
         ],
     )
 
diff --git a/third_party/aws/aws.patch b/third_party/aws/aws.patch
new file mode 100644
index 00000000000..192a4f6a376
--- /dev/null
+++ b/third_party/aws/aws.patch
@@ -0,0 +1,10 @@
+--- ./aws-cpp-sdk-core/include/aws/core/utils/crypto/Hash.org.h	2026-04-28 16:22:19.764053626 +0000
++++ ./aws-cpp-sdk-core/include/aws/core/utils/crypto/Hash.h	2026-04-28 16:22:47.655858646 +0000
+@@ -15,6 +15,7 @@
+ 
+ #pragma once
+ 
++#include <cstdint>
+ #include <aws/core/Core_EXPORTS.h>
+ #include <aws/core/utils/crypto/HashResult.h>
+ #include <aws/core/utils/memory/stl/AWSStreamFwd.h>
diff --git a/third_party/aws/workspace.bzl b/third_party/aws/workspace.bzl
index f37699e34c5..811ca23d719 100644
--- a/third_party/aws/workspace.bzl
+++ b/third_party/aws/workspace.bzl
@@ -15,4 +15,5 @@ def repo():
         sha256 = "89905075fe50aa13e0337ff905c2e8c1ce9caf77a3504484a7cda39179120ffc",
         strip_prefix = "aws-sdk-cpp-1.5.8",
         build_file = "//third_party/aws:BUILD.bazel",
+	patch_file = "//third_party/aws:aws.patch",
     )
diff --git a/third_party/boringssl/patch b/third_party/boringssl/patch
new file mode 100644
index 00000000000..6adfab1e4be
--- /dev/null
+++ b/third_party/boringssl/patch
@@ -0,0 +1,61 @@
+--- boringssl/src/crypto/x509/t_x509.org.c	2026-04-28 16:32:42.664037294 +0000
++++ boringssl/src/crypto/x509/t_x509.c	2026-04-28 16:34:56.391899068 +0000
+@@ -500,9 +500,7 @@
+ int X509_NAME_print(BIO *bp, X509_NAME *name, int obase)
+ {
+     char *s, *c, *b;
+-    int ret = 0, l, i;
+-
+-    l = 80 - 2 - obase;
++    int ret = 0, i;
+ 
+     b = X509_NAME_oneline(name, NULL, 0);
+     if (!b)
+@@ -529,12 +527,10 @@
+                 if (BIO_write(bp, ", ", 2) != 2)
+                     goto err;
+             }
+-            l--;
+         }
+         if (*s == '\0')
+             break;
+         s++;
+-        l--;
+     }
+ 
+     ret = 1;
+--- boringssl/src/third_party/fiat/curve25519.org.c	2026-04-28 18:08:53.752703283 +0000
++++ boringssl/src/third_party/fiat/curve25519.c	2026-04-28 18:11:07.433554030 +0000
+@@ -1902,7 +1902,7 @@
+ //
+ // Preconditions:
+ //   a[31] <= 127
+-void x25519_ge_scalarmult_base(ge_p3 *h, const uint8_t *a) {
++void x25519_ge_scalarmult_base(ge_p3 *h, const uint8_t a[32]) {
+   signed char e[64];
+   signed char carry;
+   ge_p1p1 r;
+--- boringssl/src/crypto/fipsmodule/bn/internal.org.h	2026-04-29 06:30:13.414347370 +0000
++++ boringssl/src/crypto/fipsmodule/bn/internal.h	2026-04-29 06:30:31.495865508 +0000
+@@ -285,7 +285,7 @@
+ void bn_mul_comba8(BN_ULONG r[16], const BN_ULONG a[8], const BN_ULONG b[8]);
+ 
+ // bn_sqr_comba8 sets |r| to |a|^2.
+-void bn_sqr_comba8(BN_ULONG r[16], const BN_ULONG a[4]);
++void bn_sqr_comba8(BN_ULONG r[16], const BN_ULONG a[8]);
+ 
+ // bn_sqr_comba4 sets |r| to |a|^2.
+ void bn_sqr_comba4(BN_ULONG r[8], const BN_ULONG a[4]);
+--- boringssl/src/crypto/bytestring/cbb.org.c	2018-08-11 01:30:56.000000000 +0900
++++ boringssl/src/crypto/bytestring/cbb.c	2026-05-03 21:58:22.586746285 +0900
+@@ -23,6 +23,9 @@
+ 
+ #include "../internal.h"
+ 
++#if defined(__GNUC__) && (__GNUC__ >= 12)
++    #pragma GCC diagnostic ignored "-Wdangling-pointer"
++#endif
+ 
+ void CBB_zero(CBB *cbb) {
+   OPENSSL_memset(cbb, 0, sizeof(CBB));
+
diff --git a/third_party/com_google_absl.patch b/third_party/com_google_absl.patch
new file mode 100644
index 00000000000..3c09ff0d5c9
--- /dev/null
+++ b/third_party/com_google_absl.patch
@@ -0,0 +1,20 @@
+--- com_google_absl/absl/synchronization/internal/graphcycles_org.cc	2019-06-18 20:10:39.000000000 +0000
++++ com_google_absl/absl/synchronization/internal/graphcycles.cc	2026-04-28 04:49:36.608653670 +0000
+@@ -37,6 +37,7 @@
+ 
+ #include <algorithm>
+ #include <array>
++#include <limits>
+ #include "absl/base/internal/hide_ptr.h"
+ #include "absl/base/internal/raw_logging.h"
+ #include "absl/base/internal/spinlock.h"
+--- com_google_absl/absl/strings/internal/str_format/extension.org.h	2026-04-28 16:09:20.784594158 +0000
++++ com_google_absl/absl/strings/internal/str_format/extension.h	2026-04-28 16:10:04.683857902 +0000
+@@ -20,6 +20,7 @@
+ #include <cstddef>
+ #include <cstring>
+ #include <ostream>
++#include <cstdint>
+ 
+ #include "absl/base/port.h"
+ #include "absl/strings/internal/str_format/output.h"
diff --git a/third_party/cython.patch b/third_party/cython.patch
new file mode 100644
index 00000000000..f62cc7034ab
--- /dev/null
+++ b/third_party/cython.patch
@@ -0,0 +1,27 @@
+--- a/Cython/Includes/cpython/longintrepr.pxd	2018-07-08 16:44:43.000000000 +0900
++++ b/Cython/Includes/cpython/longintrepr.pxd	2026-05-03 22:20:02.056018537 +0900
+@@ -1,7 +1,7 @@
+ # Internals of the "long" type (Python 2) or "int" type (Python 3).
+ # This is not part of Python's published API.
+ 
+-cdef extern from "longintrepr.h":
++cdef extern from "Python.h":
+     # Add explicit cast to avoid compiler warnings
+     cdef _PyLong_New "(PyObject*)_PyLong_New"(Py_ssize_t s)
+ 
+--- a/Cython/Utility/ModuleSetupCode.c	2018-07-08 16:44:43.000000000 +0900
++++ b/Cython/Utility/ModuleSetupCode.c	2026-05-03 23:03:53.521631265 +0900
+@@ -188,7 +188,11 @@
+ #endif
+ 
+ #if CYTHON_USE_PYLONG_INTERNALS
+-  #include "longintrepr.h"
++  #if PY_VERSION_HEX >= 0x03000000
++    #include "Python.h"
++  #else
++    #include "longintrepr.h"
++  #endif
+   /* These short defines can easily conflict with other code */
+   #undef SHIFT
+   #undef BASE
+
diff --git a/third_party/google_cloud_cpp.patch b/third_party/google_cloud_cpp.patch
new file mode 100644
index 00000000000..fdb96ee7acf
--- /dev/null
+++ b/third_party/google_cloud_cpp.patch
@@ -0,0 +1,11 @@
+--- com_github_googlecloudplatform_google_cloud_cpp/google/cloud/iam_policy.org.h	2026-04-28 23:45:05.612601190 +0000
++++ com_github_googlecloudplatform_google_cloud_cpp/google/cloud/iam_policy.h	2026-04-28 23:45:33.705603481 +0000
+@@ -16,6 +16,7 @@
+ #define GOOGLE_CLOUD_CPP_GOOGLE_CLOUD_IAM_POLICY_H_
+ 
+ #include "google/cloud/iam_bindings.h"
++#include <cstdint>
+ 
+ namespace google {
+ namespace cloud {
+
diff --git a/third_party/gpus/crosstool/cc_toolchain_config.bzl.tpl b/third_party/gpus/crosstool/cc_toolchain_config.bzl.tpl
index 282ba08cda5..1ca6ba12908 100644
--- a/third_party/gpus/crosstool/cc_toolchain_config.bzl.tpl
+++ b/third_party/gpus/crosstool/cc_toolchain_config.bzl.tpl
@@ -1248,12 +1248,12 @@ def _impl(ctx):
         ],
     )
 
-    cpp11_feature = feature(
-        name = "c++11",
+    cpp14_feature = feature(
+        name = "c++14",
         flag_sets = [
             flag_set(
                 actions = [ACTION_NAMES.cpp_compile],
-                flag_groups = [flag_group(flags = ["-std=c++11"])],
+                flag_groups = [flag_group(flags = ["-std=c++14"])],
             ),
         ],
     )
@@ -1263,7 +1263,7 @@ def _impl(ctx):
             name = "common",
             implies = [
                 "stdlib",
-                "c++11",
+                "c++14",
                 "determinism",
                 "alwayslink",
                 "hardening",
@@ -1279,7 +1279,7 @@ def _impl(ctx):
             name = "common",
             implies = [
                 "stdlib",
-                "c++11",
+                "c++14",
                 "determinism",
                 "hardening",
                 "warnings",
@@ -1294,7 +1294,7 @@ def _impl(ctx):
 
     if (ctx.attr.cpu == "local"):
         features = [
-            cpp11_feature,
+            cpp14_feature,
             stdlib_feature,
             determinism_feature,
             alwayslink_feature,
@@ -1315,7 +1315,7 @@ def _impl(ctx):
         ]
     elif (ctx.attr.cpu == "darwin"):
         features = [
-            cpp11_feature,
+            cpp14_feature,
             stdlib_feature,
             determinism_feature,
             pic_feature,
diff --git a/third_party/gpus/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc.tpl b/third_party/gpus/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc.tpl
index a69be47945b..9548b756211 100755
--- a/third_party/gpus/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc.tpl
+++ b/third_party/gpus/crosstool/clang/bin/crosstool_wrapper_driver_is_not_gcc.tpl
@@ -43,7 +43,7 @@ import os
 import subprocess
 import re
 import sys
-import pipes
+import shlex
 
 # Template values set by cuda_autoconf.
 CPU_COMPILER = ('%{cpu_compiler}')
@@ -170,7 +170,7 @@ def InvokeNvcc(argv, log=False):
   undefines = ''.join([' -U' + define for define in undefines])
   std_options = GetOptionValue(argv, 'std')
   # currently only c++11 is supported by Cuda 7.0 std argument
-  nvcc_allowed_std_options = ["c++11"]
+  nvcc_allowed_std_options = ["c++14"]
   std_options = ''.join([' -std=' + define
       for define in std_options if define in nvcc_allowed_std_options])
 
@@ -249,7 +249,7 @@ def main():
 
   if args.x and args.x[0] == 'cuda':
     if args.cuda_log: Log('-x cuda')
-    leftover = [pipes.quote(s) for s in leftover]
+    leftover = [shlex.quote(s) for s in leftover]
     if args.cuda_log: Log('using nvcc')
     return InvokeNvcc(leftover, log=args.cuda_log)
 
diff --git a/third_party/gpus/cuda_configure.bzl b/third_party/gpus/cuda_configure.bzl
index cf63adcbaa2..dfb23daa442 100644
--- a/third_party/gpus/cuda_configure.bzl
+++ b/third_party/gpus/cuda_configure.bzl
@@ -617,14 +617,14 @@ def _find_libs(repository_ctx, cuda_config):
             repository_ctx,
             cpu_value,
             cuda_config.config["cuda_library_dir"],
-            cuda_config.cuda_version,
+            None,
         ),
         "cudart_static": _find_cuda_lib(
             "cudart_static",
             repository_ctx,
             cpu_value,
             cuda_config.config["cuda_library_dir"],
-            cuda_config.cuda_version,
+            None,
             static = True,
         ),
         "cublas": _find_cuda_lib(
@@ -632,49 +632,49 @@ def _find_libs(repository_ctx, cuda_config):
             repository_ctx,
             cpu_value,
             cuda_config.config["cublas_library_dir"],
-            cuda_config.cuda_lib_version,
+            None,
         ),
         "cusolver": _find_cuda_lib(
             "cusolver",
             repository_ctx,
             cpu_value,
             cuda_config.config["cuda_library_dir"],
-            cuda_config.cuda_lib_version,
+            None,
         ),
         "curand": _find_cuda_lib(
             "curand",
             repository_ctx,
             cpu_value,
             cuda_config.config["cuda_library_dir"],
-            cuda_config.cuda_lib_version,
+            None,
         ),
         "cufft": _find_cuda_lib(
             "cufft",
             repository_ctx,
             cpu_value,
             cuda_config.config["cuda_library_dir"],
-            cuda_config.cuda_lib_version,
+            None,
         ),
         "cudnn": _find_cuda_lib(
             "cudnn",
             repository_ctx,
             cpu_value,
             cuda_config.config["cudnn_library_dir"],
-            cuda_config.cudnn_version,
+            None,
         ),
         "cupti": _find_cuda_lib(
             "cupti",
             repository_ctx,
             cpu_value,
             cuda_config.config["cupti_library_dir"],
-            cuda_config.cuda_version,
+            None,
         ),
         "cusparse": _find_cuda_lib(
             "cusparse",
             repository_ctx,
             cpu_value,
             cuda_config.config["cuda_library_dir"],
-            cuda_config.cuda_lib_version,
+            None,
         ),
     }
 
@@ -1053,8 +1053,26 @@ def _create_local_cuda_repository(repository_ctx):
     copy_rules.append(make_copy_files_rule(
         repository_ctx,
         name = "cudnn-include",
-        srcs = [cudnn_header_dir + "/cudnn.h"],
-        outs = ["cudnn/include/cudnn.h"],
+        srcs = [
+			cudnn_header_dir + "/cudnn.h",
+			cudnn_header_dir + "/cudnn_ops_infer.h",
+			cudnn_header_dir + "/cudnn_ops_train.h",
+			cudnn_header_dir + "/cudnn_adv_infer.h",
+			cudnn_header_dir + "/cudnn_adv_train.h",
+			cudnn_header_dir + "/cudnn_cnn_infer.h",
+			cudnn_header_dir + "/cudnn_cnn_train.h",
+			cudnn_header_dir + "/cudnn_backend.h",
+		],
+        outs = [
+			"cudnn/include/cudnn.h",
+			"cudnn/include/cudnn_ops_infer.h",
+			"cudnn/include/cudnn_ops_train.h",
+			"cudnn/include/cudnn_adv_infer.h",
+			"cudnn/include/cudnn_adv_train.h",
+			"cudnn/include/cudnn_cnn_infer.h",
+			"cudnn/include/cudnn_cnn_train.h",
+			"cudnn/include/cudnn_backend.h",
+		],
     ))
 
     # Set up BUILD file for cuda/
@@ -1159,7 +1177,7 @@ def _create_local_cuda_repository(repository_ctx):
         # .d file - given that includes that are prefixed with "../" multiple
         # time quickly grow longer than the root of the tree, this can lead to
         # bazel's header check failing.
-        cuda_defines["%{extra_no_canonical_prefixes_flags}"] = "\"-fno-canonical-system-headers\""
+        #cuda_defines["%{extra_no_canonical_prefixes_flags}"] = "\"-fno-canonical-system-headers\""
 
         nvcc_path = str(
             repository_ctx.path("%s/nvcc%s" % (
diff --git a/third_party/gpus/find_cuda_config.py b/third_party/gpus/find_cuda_config.py
index be59515661a..2bae2d1d2f7 100644
--- a/third_party/gpus/find_cuda_config.py
+++ b/third_party/gpus/find_cuda_config.py
@@ -344,7 +344,7 @@ def _find_cudnn_config(base_paths, required_version):
         for name in ("CUDNN_MAJOR", "CUDNN_MINOR", "CUDNN_PATCHLEVEL"))
     return ".".join(version)
 
-  header_path, header_version = _find_header(base_paths, "cudnn.h",
+  header_path, header_version = _find_header(base_paths, "cudnn_version.h",
                                              required_version,
                                              get_header_version)
   cudnn_version = header_version.split(".")[0]
diff --git a/third_party/hwloc/patch b/third_party/hwloc/patch
new file mode 100644
index 00000000000..15d146472db
--- /dev/null
+++ b/third_party/hwloc/patch
@@ -0,0 +1,15 @@
+--- hwloc/hwloc/topology.org.c	2026-05-02 16:04:16.438684110 +0000
++++ hwloc/hwloc/topology.c	2026-05-02 16:05:55.729487835 +0000
+@@ -42,8 +42,12 @@
+ #endif
+ 
+ #ifdef HAVE_SYS_SYSCTL_H
++#if __GLIBC_PREREQ(2, 32)
++#include <linux/sysctl.h>
++#else
+ #include <sys/sysctl.h>
+ #endif
++#endif
+ 
+ #ifdef HWLOC_WIN_SYS
+ #include <windows.h>
diff --git a/third_party/hwloc/workspace.bzl b/third_party/hwloc/workspace.bzl
index dc8e1579e9c..2501baecf88 100644
--- a/third_party/hwloc/workspace.bzl
+++ b/third_party/hwloc/workspace.bzl
@@ -12,5 +12,6 @@ def repo():
         sha256 = "64def246aaa5b3a6e411ce10932a22e2146c3031b735c8f94739534f06ad071c",
         strip_prefix = "hwloc-2.0.3",
         build_file = "//third_party/hwloc:BUILD.bazel",
+	patch_file = "//third_party/hwloc:patch",
         system_build_file = "//third_party/hwloc:BUILD.system",
     )
diff --git a/third_party/mlir/include/mlir/Support/STLExtras.h b/third_party/mlir/include/mlir/Support/STLExtras.h
index 3448b080d03..64ac556d61e 100644
--- a/third_party/mlir/include/mlir/Support/STLExtras.h
+++ b/third_party/mlir/include/mlir/Support/STLExtras.h
@@ -26,6 +26,7 @@
 #include "mlir/Support/LLVM.h"
 #include "llvm/ADT/iterator.h"
 #include <tuple>
+#include <cstdint>
 
 namespace mlir {
 
diff --git a/third_party/nccl/build_defs.bzl.tpl b/third_party/nccl/build_defs.bzl.tpl
index 57191398553..5f5c3a1008a 100644
--- a/third_party/nccl/build_defs.bzl.tpl
+++ b/third_party/nccl/build_defs.bzl.tpl
@@ -113,7 +113,6 @@ def _device_link_impl(ctx):
             "--cmdline=--compile-only",
             "--link",
             "--compress-all",
-            "--bin2c-path=%s" % bin2c.dirname,
             "--create=%s" % tmp_fatbin.path,
             "--embedded-fatbin=%s" % fatbin_h.path,
         ] + images,
diff --git a/third_party/pcre2.BUILD b/third_party/pcre2.BUILD
index 3a8e7a10b43..1760ffa7245 100644
--- a/third_party/pcre2.BUILD
+++ b/third_party/pcre2.BUILD
@@ -1,38 +1,19 @@
 licenses(["notice"])  # BSD
 
-exports_files(["LICENCE"])
+exports_files(["LICENCE.md"])
 
 cc_library(
-    name = "pcre",
-    srcs = [
-        "pcre_byte_order.c",
-        "pcre_chartables.c",
-        "pcre_compile.c",
-        "pcre_config.c",
-        "pcre_dfa_exec.c",
-        "pcre_exec.c",
-        "pcre_fullinfo.c",
-        "pcre_get.c",
-        "pcre_globals.c",
-        "pcre_internal.h",
-        "pcre_jit_compile.c",
-        "pcre_maketables.c",
-        "pcre_newline.c",
-        "pcre_ord2utf8.c",
-        "pcre_refcount.c",
-        "pcre_string_utils.c",
-        "pcre_study.c",
-        "pcre_tables.c",
-        "pcre_ucd.c",
-        "pcre_valid_utf8.c",
-        "pcre_version.c",
-        "pcre_xclass.c",
-        "ucp.h",
-    ],
-    hdrs = [
-        "pcre.h",
-        "pcreposix.h",
-    ],
+    name = "pcre2",
+    srcs = glob([
+        "src/*.c",
+    ], exclude=[
+	"src/pcre2posix_test.c",
+	"src/pcre2test.c",
+	"src/pcre2_jit_test.c",
+	"src/pcre2_dftables.c",
+	"src/pcre2demo.c",
+	"src/pcre2grep.c",
+    ]) + ["src/pcre2_chartables.c"],
     copts = [
         "-DHAVE_BCOPY=1",
         "-DHAVE_INTTYPES_H=1",
@@ -44,37 +25,58 @@ cc_library(
         "-DHAVE_UNISTD_H=1",
         "-DLINK_SIZE=2",
         "-DMATCH_LIMIT=10000000",
+	"-DMATCH_LIMIT_DEPTH=10000000",
         "-DMATCH_LIMIT_RECURSION=1000",
         "-DMAX_NAME_COUNT=10000",
         "-DMAX_NAME_SIZE=32",
         "-DNEWLINE=10",
+	"-DNEWLINE_DEFAULT=2",
         "-DNO_RECURSE",
         "-DPARENS_NEST_LIMIT=50",
         "-DPOSIX_MALLOC_THRESHOLD=10",
+	"-DHEAP_LIMIT=20000000",
+	"-DMAX_VARLOOKBEHIND=255",
+	"-DPCRE2GREP_BUFSIZE=20480",
+        "-DPCRE2GREP_MAX_BUFSIZE=1048576",
         "-DSTDC_HEADERS=1",
         "-DSUPPORT_UCP",
         "-DSUPPORT_UTF",
+	"-DSUPPORT_PCRE2_8",
+        "-DSUPPORT_UNICODE",
+    ],
+    defines = [
+    	"PCRE2_STATIC=1",
+	"PCRE2_CODE_UNIT_WIDTH=8",
+	"PCRE2_EXPORT='__attribute__ ((visibility (\"default\")))'",
     ],
-    defines = ["PCRE_STATIC=1"],
-    includes = ["."],
+    deps = [":config"],
+    includes = ["src"],
     visibility = ["@swig//:__pkg__"],  # Please use RE2
     alwayslink = 1,
 )
 
+cc_library(
+    name = "config",
+    hdrs = ["src/pcre2.h"],
+    srcs = [":pcre_h"],
+    includes = ["src"],
+    visibility = ["//visibility:public"],
+)
+
 genrule(
     name = "pcre_h",
-    srcs = ["pcre.h.in"],
-    outs = ["pcre.h"],
-    cmd = "sed -e s/@PCRE_MAJOR@/8/" +
-          "    -e s/@PCRE_MINOR@/39/" +
-          "    -e s/@PCRE_PRERELEASE@//" +
-          "    -e s/@PCRE_DATE@/redacted/" +
+    srcs = ["src/pcre2.h.in"],
+    outs = ["src/pcre2.h"],
+    cmd = "sed -e s/@PCRE2_MAJOR@/10/" +
+          "    -e s/@PCRE2_MINOR@/47/" +
+          "    -e s/@PCRE2_PRERELEASE@//" +
+          "    -e s/@PCRE2_DATE@/redacted/" +
           "    $< >$@",
 )
 
 genrule(
     name = "pcre_chartables_c",
-    srcs = ["pcre_chartables.c.dist"],
-    outs = ["pcre_chartables.c"],
+    srcs = ["src/pcre2_chartables.c.dist"],
+    outs = ["src/pcre2_chartables.c"],
     cmd = "cp $< $@",
 )
diff --git a/third_party/protobuf/protobuf.patch b/third_party/protobuf/protobuf.patch
index df0648563dc..626c153e18e 100644
--- a/third_party/protobuf/protobuf.patch
+++ b/third_party/protobuf/protobuf.patch
@@ -20,3 +20,82 @@ index 2fb26050..c2744d5b 100644
      includes = ["src/"],
      visibility = ["//visibility:public"],
  )
+--- com_google_protobuf/python/google/protobuf/pyext/message.org.cc	2026-05-03 03:30:31.941531635 +0000
++++ com_google_protobuf/python/google/protobuf/pyext/message.cc	2026-05-03 03:29:24.583572174 +0000
+@@ -2992,10 +2992,17 @@
+             &RepeatedCompositeContainer_Type));
+ 
+     // Register them as collections.Sequence
++#if PY_MINOR_VERSION >= 10
++    ScopedPyObjectPtr collections(PyImport_ImportModule("collections.abc"));
++    if (collections == NULL) {
++      return false;
++    }
++#else
+     ScopedPyObjectPtr collections(PyImport_ImportModule("collections"));
+     if (collections == NULL) {
+       return false;
+     }
++#endif
+     ScopedPyObjectPtr mutable_sequence(
+         PyObject_GetAttrString(collections.get(), "MutableSequence"));
+     if (mutable_sequence == NULL) {
+--- com_google_protobuf/python/google/protobuf/pyext/descriptor.org.cc	2019-06-13 02:00:36.000000000 +0900
++++ com_google_protobuf/python/google/protobuf/pyext/descriptor.cc	2026-05-03 22:51:39.710739108 +0900
+@@ -104,6 +104,46 @@
+   if (frame == NULL) {
+     return false;
+   }
++#if PY_MINOR_VERSION >= 11
++  while (stacklevel-- > 0) {
++    frame = PyFrame_GetBack(frame);
++    if (frame == NULL) {
++      return false;
++    }
++  }
++
++  PyCodeObject *f_code = PyFrame_GetCode(frame);
++  if (f_code->co_filename == NULL) {
++    return false;
++  }
++  char* filename;
++  Py_ssize_t filename_size;
++  if (PyString_AsStringAndSize(f_code->co_filename,
++                               &filename, &filename_size) < 0) {
++    // filename is not a string.
++    PyErr_Clear();
++    return false;
++  }
++  if ((filename_size < 3) ||
++      (strcmp(&filename[filename_size - 3], ".py") != 0)) {
++    // Cython's stack does not have .py file name and is not at global module
++    // scope.
++    return true;
++  }
++  if (filename_size < 7) {
++    // filename is too short.
++    return false;
++  }
++  if (strcmp(&filename[filename_size - 7], "_pb2.py") != 0) {
++    // Filename is not ending with _pb2.
++    return false;
++  }
++
++  if(PyFrame_GetGlobals(frame) != PyFrame_GetLocals(frame)) {
++    // Not at global module scope
++    return false;
++  }
++#else
+   while (stacklevel-- > 0) {
+     frame = frame->f_back;
+     if (frame == NULL) {
+@@ -141,7 +181,8 @@
+     // Not at global module scope
+     return false;
+   }
+-#endif
++#endif // #if PY_MINOR_VERSION >= 11
++#endif // #ifndef PYPY_VERSION
+   return true;
+ }
+ 
diff --git a/third_party/py/python_configure.bzl b/third_party/py/python_configure.bzl
index d1b1e594921..788cc4d3b42 100644
--- a/third_party/py/python_configure.bzl
+++ b/third_party/py/python_configure.bzl
@@ -244,8 +244,8 @@ def _get_python_include(repository_ctx, python_bin):
             python_bin,
             "-c",
             "from __future__ import print_function;" +
-            "from distutils import sysconfig;" +
-            "print(sysconfig.get_python_inc())",
+            "import os, sysconfig;" +
+            "print(os.path.dirname(sysconfig.get_config_h_filename()))",
         ],
         error_msg = "Problem getting python include path.",
         error_details = ("Is the Python binary path set up right? " +
diff --git a/third_party/swig.BUILD b/third_party/swig.BUILD
index 59a3d9e6714..67a34e354c4 100644
--- a/third_party/swig.BUILD
+++ b/third_party/swig.BUILD
@@ -4,71 +4,22 @@ exports_files(["LICENSE"])
 
 cc_binary(
     name = "swig",
-    srcs = [
-        "Source/CParse/cparse.h",
-        "Source/CParse/cscanner.c",
-        "Source/CParse/parser.c",
-        "Source/CParse/parser.h",
-        "Source/CParse/templ.c",
-        "Source/CParse/util.c",
-        "Source/DOH/base.c",
-        "Source/DOH/doh.h",
-        "Source/DOH/dohint.h",
-        "Source/DOH/file.c",
-        "Source/DOH/fio.c",
-        "Source/DOH/hash.c",
-        "Source/DOH/list.c",
-        "Source/DOH/memory.c",
-        "Source/DOH/string.c",
-        "Source/DOH/void.c",
+    srcs = glob([
+        "Source/CParse/*.h",
+        "Source/CParse/*.c",
+        "Source/DOH/*.h",
+        "Source/DOH/*.c",
+        "Source/Doxygen/*.h",
+        "Source/Doxygen/*.cxx",
         "Source/Include/swigconfig.h",
         "Source/Include/swigwarn.h",
-        "Source/Modules/allocate.cxx",
-        "Source/Modules/browser.cxx",
-        "Source/Modules/contract.cxx",
-        "Source/Modules/directors.cxx",
-        "Source/Modules/emit.cxx",
-        "Source/Modules/lang.cxx",
-        "Source/Modules/main.cxx",
-        "Source/Modules/module.cxx",
-        "Source/Modules/nested.cxx",
-        "Source/Modules/overload.cxx",
-        "Source/Modules/python.cxx",
-        "Source/Modules/swigmain-lite.cxx",
-        "Source/Modules/swigmod.h",
-        "Source/Modules/typepass.cxx",
-        "Source/Modules/uffi.cxx",
-        "Source/Modules/utils.cxx",
-        "Source/Modules/xml.cxx",
-        "Source/Preprocessor/cpp.c",
-        "Source/Preprocessor/expr.c",
-        "Source/Preprocessor/preprocessor.h",
-        "Source/Swig/cwrap.c",
-        "Source/Swig/deprecate.c",
-        "Source/Swig/error.c",
-        "Source/Swig/extend.c",
-        "Source/Swig/fragment.c",
-        "Source/Swig/getopt.c",
-        "Source/Swig/include.c",
-        "Source/Swig/misc.c",
-        "Source/Swig/naming.c",
-        "Source/Swig/parms.c",
-        "Source/Swig/scanner.c",
-        "Source/Swig/stype.c",
-        "Source/Swig/swig.h",
-        "Source/Swig/swigfile.h",
-        "Source/Swig/swigopt.h",
-        "Source/Swig/swigparm.h",
-        "Source/Swig/swigscan.h",
-        "Source/Swig/swigtree.h",
-        "Source/Swig/swigwrap.h",
-        "Source/Swig/symbol.c",
-        "Source/Swig/tree.c",
-        "Source/Swig/typemap.c",
-        "Source/Swig/typeobj.c",
-        "Source/Swig/typesys.c",
-        "Source/Swig/wrapfunc.c",
-    ],
+        "Source/Modules/*.h",
+        "Source/Modules/*.cxx",
+        "Source/Preprocessor/*.h",
+        "Source/Preprocessor/*.c",
+        "Source/Swig/*.h",
+        "Source/Swig/*.c",
+    ]),
     copts = ["$(STACK_FRAME_UNLIMITED)"] + select({
         ":windows": [],
         "//conditions:default": [
@@ -81,6 +32,7 @@ cc_binary(
     includes = [
         "Source/CParse",
         "Source/DOH",
+        "Source/Doxygen",
         "Source/Include",
         "Source/Modules",
         "Source/Preprocessor",
@@ -88,211 +40,31 @@ cc_binary(
     ],
     output_licenses = ["unencumbered"],
     visibility = ["//visibility:public"],
-    deps = ["@pcre"],
+    deps = [
+      "@pcre2",
+      ":swigconfig",
+    ],
 )
 
 filegroup(
     name = "templates",
-    srcs = [
-        "Lib/allkw.swg",
-        "Lib/attribute.i",
-        "Lib/carrays.i",
-        "Lib/cdata.i",
-        "Lib/cffi/cffi.swg",
-        "Lib/cmalloc.i",
-        "Lib/constraints.i",
-        "Lib/cpointer.i",
-        "Lib/cstring.i",
-        "Lib/cwstring.i",
-        "Lib/exception.i",
-        "Lib/intrusive_ptr.i",
-        "Lib/inttypes.i",
-        "Lib/linkruntime.c",
-        "Lib/math.i",
-        "Lib/pointer.i",
-        "Lib/python/argcargv.i",
-        "Lib/python/attribute.i",
-        "Lib/python/boost_shared_ptr.i",
-        "Lib/python/builtin.swg",
-        "Lib/python/carrays.i",
-        "Lib/python/ccomplex.i",
-        "Lib/python/cdata.i",
-        "Lib/python/cmalloc.i",
-        "Lib/python/cni.i",
-        "Lib/python/complex.i",
-        "Lib/python/cpointer.i",
-        "Lib/python/cstring.i",
-        "Lib/python/cwstring.i",
-        "Lib/python/defarg.swg",
-        "Lib/python/director.swg",
-        "Lib/python/embed.i",
-        "Lib/python/embed15.i",
-        "Lib/python/exception.i",
-        "Lib/python/factory.i",
-        "Lib/python/file.i",
-        "Lib/python/implicit.i",
-        "Lib/python/jstring.i",
-        "Lib/python/pyabc.i",
-        "Lib/python/pyapi.swg",
-        "Lib/python/pybackward.swg",
-        "Lib/python/pybuffer.i",
-        "Lib/python/pyclasses.swg",
-        "Lib/python/pycomplex.swg",
-        "Lib/python/pycontainer.swg",
-        "Lib/python/pydocs.swg",
-        "Lib/python/pyerrors.swg",
-        "Lib/python/pyfragments.swg",
-        "Lib/python/pyhead.swg",
-        "Lib/python/pyinit.swg",
-        "Lib/python/pyiterators.swg",
-        "Lib/python/pymacros.swg",
-        "Lib/python/pyname_compat.i",
-        "Lib/python/pyopers.swg",
-        "Lib/python/pyprimtypes.swg",
-        "Lib/python/pyrun.swg",
-        "Lib/python/pyruntime.swg",
-        "Lib/python/pystdcommon.swg",
-        "Lib/python/pystrings.swg",
-        "Lib/python/python.swg",
-        "Lib/python/pythonkw.swg",
-        "Lib/python/pythreads.swg",
-        "Lib/python/pytuplehlp.swg",
-        "Lib/python/pytypemaps.swg",
-        "Lib/python/pyuserdir.swg",
-        "Lib/python/pywstrings.swg",
-        "Lib/python/std_alloc.i",
-        "Lib/python/std_auto_ptr.i",
-        "Lib/python/std_basic_string.i",
-        "Lib/python/std_carray.i",
-        "Lib/python/std_char_traits.i",
-        "Lib/python/std_common.i",
-        "Lib/python/std_complex.i",
-        "Lib/python/std_container.i",
-        "Lib/python/std_deque.i",
-        "Lib/python/std_except.i",
-        "Lib/python/std_ios.i",
-        "Lib/python/std_iostream.i",
-        "Lib/python/std_list.i",
-        "Lib/python/std_map.i",
-        "Lib/python/std_multimap.i",
-        "Lib/python/std_multiset.i",
-        "Lib/python/std_pair.i",
-        "Lib/python/std_set.i",
-        "Lib/python/std_shared_ptr.i",
-        "Lib/python/std_sstream.i",
-        "Lib/python/std_streambuf.i",
-        "Lib/python/std_string.i",
-        "Lib/python/std_unordered_map.i",
-        "Lib/python/std_unordered_multimap.i",
-        "Lib/python/std_unordered_multiset.i",
-        "Lib/python/std_unordered_set.i",
-        "Lib/python/std_vector.i",
-        "Lib/python/std_vectora.i",
-        "Lib/python/std_wios.i",
-        "Lib/python/std_wiostream.i",
-        "Lib/python/std_wsstream.i",
-        "Lib/python/std_wstreambuf.i",
-        "Lib/python/std_wstring.i",
-        "Lib/python/stl.i",
-        "Lib/python/typemaps.i",
-        "Lib/python/wchar.i",
-        "Lib/runtime.swg",
-        "Lib/shared_ptr.i",
-        "Lib/std/_std_deque.i",
-        "Lib/std/std_alloc.i",
-        "Lib/std/std_basic_string.i",
-        "Lib/std/std_carray.swg",
-        "Lib/std/std_char_traits.i",
-        "Lib/std/std_common.i",
-        "Lib/std/std_container.i",
-        "Lib/std/std_deque.i",
-        "Lib/std/std_except.i",
-        "Lib/std/std_ios.i",
-        "Lib/std/std_iostream.i",
-        "Lib/std/std_list.i",
-        "Lib/std/std_map.i",
-        "Lib/std/std_multimap.i",
-        "Lib/std/std_multiset.i",
-        "Lib/std/std_pair.i",
-        "Lib/std/std_queue.i",
-        "Lib/std/std_set.i",
-        "Lib/std/std_sstream.i",
-        "Lib/std/std_stack.i",
-        "Lib/std/std_streambuf.i",
-        "Lib/std/std_string.i",
-        "Lib/std/std_unordered_map.i",
-        "Lib/std/std_unordered_multimap.i",
-        "Lib/std/std_unordered_multiset.i",
-        "Lib/std/std_unordered_set.i",
-        "Lib/std/std_vector.i",
-        "Lib/std/std_vectora.i",
-        "Lib/std/std_wios.i",
-        "Lib/std/std_wiostream.i",
-        "Lib/std/std_wsstream.i",
-        "Lib/std/std_wstreambuf.i",
-        "Lib/std/std_wstring.i",
-        "Lib/std_except.i",
-        "Lib/stdint.i",
-        "Lib/stl.i",
-        "Lib/swig.swg",
-        "Lib/swigarch.i",
-        "Lib/swigerrors.swg",
-        "Lib/swiginit.swg",
-        "Lib/swiglabels.swg",
-        "Lib/swigrun.i",
-        "Lib/swigrun.swg",
-        "Lib/swigwarn.swg",
-        "Lib/swigwarnings.swg",
-        "Lib/typemaps/attribute.swg",
-        "Lib/typemaps/carrays.swg",
-        "Lib/typemaps/cdata.swg",
-        "Lib/typemaps/cmalloc.swg",
-        "Lib/typemaps/cpointer.swg",
-        "Lib/typemaps/cstring.swg",
-        "Lib/typemaps/cstrings.swg",
-        "Lib/typemaps/cwstring.swg",
-        "Lib/typemaps/enumint.swg",
-        "Lib/typemaps/exception.swg",
-        "Lib/typemaps/factory.swg",
-        "Lib/typemaps/fragments.swg",
-        "Lib/typemaps/implicit.swg",
-        "Lib/typemaps/inoutlist.swg",
-        "Lib/typemaps/misctypes.swg",
-        "Lib/typemaps/primtypes.swg",
-        "Lib/typemaps/ptrtypes.swg",
-        "Lib/typemaps/std_except.swg",
-        "Lib/typemaps/std_string.swg",
-        "Lib/typemaps/std_strings.swg",
-        "Lib/typemaps/std_wstring.swg",
-        "Lib/typemaps/string.swg",
-        "Lib/typemaps/strings.swg",
-        "Lib/typemaps/swigmacros.swg",
-        "Lib/typemaps/swigobject.swg",
-        "Lib/typemaps/swigtype.swg",
-        "Lib/typemaps/swigtypemaps.swg",
-        "Lib/typemaps/traits.swg",
-        "Lib/typemaps/typemaps.swg",
-        "Lib/typemaps/valtypes.swg",
-        "Lib/typemaps/void.swg",
-        "Lib/typemaps/wstring.swg",
-        "Lib/wchar.i",
-        "Lib/windows.i",
-    ],
+    srcs = glob([
+      "Lib/*.*",
+      "Lib/python/*.*",
+    ]),
     licenses = ["notice"],  # simple notice license for Lib/
     path = "Lib",
     visibility = ["//visibility:public"],
 )
 
 genrule(
-    name = "swigconfig",
+    name = "swigconfig_gen",
     outs = ["Source/Include/swigconfig.h"],
     cmd = "cat <<EOF >$@\n" +
-          "#define HAVE_BOOL\n" +
-          "#define HAVE_PCRE\n" +
-          "#define HAVE_POPEN\n" +
+          "#define HAVE_CXX14 1\n" +
+          "#define HAVE_PCRE 1\n" +
           "#define PACKAGE_BUGREPORT \"http://www.swig.org\"\n" +
-          "#define PACKAGE_VERSION \"3.0.8\"\n" +
-          "#define STDC_HEADERS\n" +
+          "#define PACKAGE_VERSION \"4.2.1\"\n" +
           "#define SWIG_CXX \"bazel4lyfe\"\n" +
           "#define SWIG_LIB \"external/swig/Lib\"\n" +
           "#define SWIG_LIB_WIN_UNIX \"\"\n" +
@@ -300,34 +72,12 @@ genrule(
           "EOF",
 )
 
-genrule(
-    name = "get_rid_of_stuff_we_dont_need_yet",
-    srcs = ["Source/Modules/swigmain.cxx"],
-    outs = ["Source/Modules/swigmain-lite.cxx"],
-    cmd = "sed -e '/swig_allegrocl/d'" +
-          "    -e '/swig_cffi/d'" +
-          "    -e '/swig_chicken/d'" +
-          "    -e '/swig_clisp/d'" +
-          "    -e '/swig_csharp/d'" +
-          "    -e '/swig_d/d'" +
-          "    -e '/swig_go/d'" +
-          "    -e '/swig_guile/d'" +
-          "    -e '/swig_java/d'" +
-          "    -e '/swig_lua/d'" +
-          "    -e '/swig_modula3/d'" +
-          "    -e '/swig_mzscheme/d'" +
-          "    -e '/swig_ocaml/d'" +
-          "    -e '/swig_octave/d'" +
-          "    -e '/swig_perl/d'" +
-          "    -e '/swig_php/d'" +
-          "    -e '/swig_pike/d'" +
-          "    -e '/swig_r/d'" +
-          "    -e '/swig_ruby/d'" +
-          "    -e '/swig_scilab/d'" +
-          "    -e '/swig_sexp/d'" +
-          "    -e '/swig_tcl/d'" +
-          "    -e '/swig_uffi/d'" +
-          "    $< >$@",
+cc_library(
+    name = "swigconfig",
+    srcs = [":swigconfig_gen"], 
+    hdrs = ["Source/Include/swigconfig.h"],
+    includes = ["Source/Include"],
+    visibility = ["//visibility:public"],
 )
 
 config_setting(
diff --git a/third_party/systemlibs/absl_py.patch b/third_party/systemlibs/absl_py.patch
new file mode 100644
index 00000000000..ca69feacf1b
--- /dev/null
+++ b/third_party/systemlibs/absl_py.patch
@@ -0,0 +1,58 @@
+--- absl_py/absl/flags/_helpers.org.py	2019-03-13 04:54:28.000000000 +0900
++++ absl_py/absl/flags/_helpers.py	2026-05-04 15:27:45.723802783 +0900
+@@ -35,7 +35,6 @@
+   termios = None
+ 
+ import six
+-from six.moves import range  # pylint: disable=redefined-builtin
+ 
+ 
+ _DEFAULT_HELP_WIDTH = 80  # Default width of help output.
+--- absl_py/absl/testing/absltest.org.py	2019-03-13 04:54:28.000000000 +0900
++++ absl_py/absl/testing/absltest.py	2026-05-04 18:07:47.637444351 +0900
+@@ -61,8 +61,7 @@
+ from absl.testing import xml_reporter
+ from absl.third_party import unittest3_backport
+ import six
+-from six.moves import urllib
+-from six.moves import xrange  # pylint: disable=redefined-builtin
++import urllib
+ 
+ 
+ # Make typing an optional import to avoid it being a required dependency
+@@ -1306,7 +1305,7 @@
+     subsequence = list(subsequence)
+     longest_match = 0
+ 
+-    for start in xrange(1 + len(container) - len(subsequence)):
++    for start in range(1 + len(container) - len(subsequence)):
+       if longest_match == len(subsequence):
+         break
+       index = 0
+@@ -1727,14 +1726,14 @@
+   elif (isinstance(a, collections.Sequence) and
+         not isinstance(a, _TEXT_OR_BINARY_TYPES)):
+     minlen = min(len(a), len(b))
+-    for i in xrange(minlen):
++    for i in range(minlen):
+       _walk_structure_for_problems(
+           a[i], b[i], '%s[%d]' % (aname, i), '%s[%d]' % (bname, i),
+           problem_list)
+-    for i in xrange(minlen, len(a)):
++    for i in range(minlen, len(a)):
+       problem_list.append('%s has [%i] with value %r but %s does not' %
+                           (aname, i, a[i], bname))
+-    for i in xrange(minlen, len(b)):
++    for i in range(minlen, len(b)):
+       problem_list.append('%s lacks [%i] but %s has it with value %r' %
+                           (aname, i, bname, b[i]))
+ 
+@@ -2114,7 +2113,7 @@
+   # the test case names for this shard.
+   delegate_get_names = base_loader.getTestCaseNames
+ 
+-  bucket_iterator = itertools.cycle(xrange(total_shards))
++  bucket_iterator = itertools.cycle(range(total_shards))
+ 
+   def getShardedTestCaseNames(testCaseClass):
+     filtered_names = []
diff --git a/third_party/systemlibs/grpc.patch b/third_party/systemlibs/grpc.patch
new file mode 100644
index 00000000000..0867dd7323c
--- /dev/null
+++ b/third_party/systemlibs/grpc.patch
@@ -0,0 +1,12 @@
+--- grpc/src/core/lib/gpr/log_linux.org.cc	2026-05-02 14:09:32.079548021 +0900
++++ grpc/src/core/lib/gpr/log_linux.cc	2026-05-02 14:07:02.037469605 +0900
+@@ -40,7 +40,9 @@
+ #include <time.h>
+ #include <unistd.h>
+ 
++#if !__GLIBC_PREREQ(2, 30)
+ static long gettid(void) { return syscall(__NR_gettid); }
++#endif
+ 
+ void gpr_log(const char* file, int line, gpr_log_severity severity,
+              const char* format, ...) {
diff --git a/third_party/systemlibs/pcre2.BUILD b/third_party/systemlibs/pcre2.BUILD
index df742388474..2245993ea90 100644
--- a/third_party/systemlibs/pcre2.BUILD
+++ b/third_party/systemlibs/pcre2.BUILD
@@ -1,12 +1,12 @@
 licenses(["notice"])  # BSD
 
 filegroup(
-    name = "LICENCE",
+    name = "LICENCE.md",
     visibility = ["//visibility:public"],
 )
 
 cc_library(
-    name = "pcre",
-    linkopts = ["-lpcre"],
+    name = "pcre2",
+    linkopts = ["-lpcre2"],
     visibility = ["//visibility:public"],
 )
diff --git a/third_party/wrapt.BUILD b/third_party/wrapt.BUILD
index bed5332edaf..c90d0f1cee1 100644
--- a/third_party/wrapt.BUILD
+++ b/third_party/wrapt.BUILD
@@ -2,9 +2,14 @@ py_library(
     name = "wrapt",
     srcs = [
         "__init__.py",
-        "decorators.py",
-        "importer.py",
-        "wrappers.py",
+	"__wrapt__.py",
+	"arguments.py",
+	"decorators.py",
+	"importer.py",
+	"patches.py",
+	"proxies.py",
+	"weakrefs.py",
+	"wrappers.py",
     ],
     visibility = ["//visibility:public"],
 )