"//tensorflow/core/common_runtime/next_pluggable_device/c:tf_rendezvous_c_api",
        "@local_tsl//tsl/platform",
    ],
)

tf_cuda_library(
    name = "kernels",
    srcs = [
        "kernels.cc",
    ],
    hdrs = [
        "kernels.h",
    ],
    copts = tf_copts(),
    visibility = ["//visibility:public"],
    deps = [
        ":c_api_internal",
        ":c_api_macros_hdrs",

        "//conditions:default": ["@com_github_grpc_grpc//:grpc++"],
    }),
)
# copybara:comment_end

# A shared object which includes registration mechanisms for ops and
# kernels. Does not include the implementations of any ops or kernels. Instead,
# the library which loads libtensorflow_framework.so
# (e.g. _pywrap_tensorflow_internal.so for Python, libtensorflow.so for the C

  # TODO(gunan): Add sanity checks to loaded modules here.

  # Load first party dynamic kernels.
  _tf_dir = _os.path.dirname(_current_file_location)
  _kernel_dir = _os.path.join(_tf_dir, "core", "kernels")
  if _os.path.exists(_kernel_dir):
    _ll.load_library(_kernel_dir)

  # Load third party dynamic kernels.
  for _s in _site_packages_dirs:
    _plugin_dir = _os.path.join(_s, "tensorflow-plugins")

# Only include debug info for files under tensorflow/, excluding kernels, to
# reduce the size of the debug info in the binary. This is because if the debug
# sections in the ELF binary are too large, errors can occur. See
# https://github.com/tensorflow/tensorflow/issues/48919.
# Users can still include debug info for a specific kernel, e.g. with:
#     --config=dbg --per_file_copt=+tensorflow/core/kernels/identity_op.*@-g

  # TODO(gunan): Add sanity checks to loaded modules here.

  # Load first party dynamic kernels.
  _tf_dir = _os.path.dirname(_current_file_location)
  _kernel_dir = _os.path.join(_tf_dir, "core", "kernels")
  if _os.path.exists(_kernel_dir):
    _ll.load_library(_kernel_dir)

  # Load third party dynamic kernels.
  for _s in _site_packages_dirs:
    _plugin_dir = _os.path.join(_s, "tensorflow-plugins")

```

TensorFlow kernels and TensorFlow's dependencies are still not built with
debugging information with `--config=dbg`, as issues occur on Linux if
there is too much debug info (see [this GitHub
issue](https://github.com/tensorflow/tensorflow/issues/48919) for context). If
you want to debug a kernel, you can compile specific files with `-g` using the

        mean, var = tf.nn.moments(self.kernel, axes=[0, 1, 2], keepdims=True)
        return self.convolution_op(inputs, (self.kernel - mean) / tf.sqrt(var +
        1e-10))` Alternatively, you can override `convolution_op`: `python class
        StandardizedConv2D(tf.keras.Layer): def convolution_op(self, inputs,
        kernel): mean, var = tf.nn.moments(kernel, axes=[0, 1, 2],

// function. Otherwise return nullptr in result and error Status from the
// function.
//
// If `library_filename` has already been loaded, we return a cached handle.
// Device and Kernels/Ops are registered as globals when a library is loaded
// for the first time.
TF_Library* TF_LoadPluggableDeviceLibrary(const char* library_filename,
                                          TF_Status* status) {

//tensorflow/cc/saved_model:saved_model_half_plus_two
//tensorflow:no_tensorflow_py_deps
//tensorflow/tools/pip_package:win_pip_package_marker
//tensorflow/core:image_testdata
//tensorflow/core/kernels/cloud:bigquery_reader_ops
//tensorflow/python:extra_py_tests_deps
//tensorflow/python:mixed_precision
//tensorflow/python:tf_optimizer
//tensorflow/python:compare_test_proto_py

  config_info_line('monolithic', 'Config for mostly static monolithic build.')
  config_info_line('numa', 'Build with NUMA support.')
  config_info_line(
      'dynamic_kernels',
      '(Experimental) Build kernels into separate shared objects.')
  config_info_line('v1', 'Build with TensorFlow 1 API instead of TF 2 API.')

  print('Preconfigured Bazel build configs to DISABLE default on features:')