since `REGISTER_OP("MatMul")` appears in ***core/math_ops.cc***, the "MatMul"
operator in the script should be in the "math" category, and it will be
generated in the output file `c/experimental/ops/math_ops.cc`.

Running this script should be a no-op, generating identical code other than

namespace tensorflow {
namespace gradients {

/* Returns numerical grad inside `dtheta_approx` given `forward` model and
 * parameter specified by `input_index`.
 *
 * I.e. if y = <output of the forward model> and w = inputs[input_index],
 * this will calculate dy/dw numerically.
 *
 * `use_function` indicates whether to use graph mode(true) or eager(false).
 *

  }
  std::string value_string;
  if (!SummarizeValue(value_string).ok()) {
    value_string = "<error computing value>";
  }
  if (value_string.length() > 100) {
    // The default NumPy-style output can be distractingly long in error
    // messages.
    value_string = absl::StrCat(value_string.substr(0, 100), " [...]");
  }
  Status s;
  const char* device_name = DeviceName(&s);
  if (!s.ok()) {

  TF_Operation* neg = Neg(feed, func_graph.get(), s.get());

  TF_Output inputs[] = {{feed, 0}};
  TF_Output outputs[] = {{neg, 0}};
  *func = TF_GraphToFunction(func_graph.get(), name, append_hash, -1,
                             /*opers=*/nullptr, 1, inputs, 1, outputs,
                             /*output_names=*/nullptr,
                             /*opts=*/nullptr, description, s.get());

  virtual bool HasCustomDeviceInput() const = 0;

  virtual const tensorflow::OpDef* OpDef() const = 0;

  virtual Status InputLength(const char* input_name, int* length) = 0;
  virtual Status OutputLength(const char* output_name, int* length) = 0;

  // Set stack trace to be used for potential async error reporting.
  virtual void SetStackTrace(ManagedStackTrace stack_trace) = 0;

          context: ./tensorflow/tools/tf_sig_build_dockerfiles
          target: devel
          build-args: |
            PYTHON_VERSION=${{ matrix.python-version }}
            CACHEBUSTER=${{ steps.date.outputs.DATE }}
          tags: |
            gcr.io/tensorflow-sigs/build:${{ github.event.number }}-${{ matrix.python-version }}
          cache-from: |
            type=registry,ref=tensorflow/build:latest-${{ matrix.python-version }}

                        TF_Bool assume_valid_feeds,
                        TF_Bool aggressive_shape_inference,
                        TF_Bool include_input_tensor_values,
                        TF_Bool include_output_tensor_values,
                        TF_Status* status) {
  TF_SetStatus(status, TF_OK, "");
  absl::Status s =
      reinterpret_cast<tensorflow::grappler::GraphProperties*>(graph_properties)

  /// Look up the entry with key `key` and copy it to `value` if found. If not
  /// found, call `compute_func`. If `compute_func` set `status` to `TF_OK`,
  /// store a copy of the output parameter in the cache, and another copy in
  /// `value`.
  void LookupOrCompute(const std::string& key, T* value,
                       const ComputeFunc& compute_func, TF_Status* status) {
    if (max_age_ == 0) {

  if (!remote && !async && !remote_func_outputs) {
    auto remote_arg =
        tensorflow::TensorHandleFromInterface(tensorflow::unwrap(h1_task2));
    // The input handles should never change since they have been mirrored.
    ASSERT_TRUE(remote_arg->HasLocalMirror(nullptr));
  }

  if (remote_func_outputs) {
    const string backing_device =

  // device if any of their inputs are on that custom device, but custom devices
  // are free to set a bad status in order to require explicit placement.
  void (*execute)(const TFE_Op* op, int* num_outputs,
                  TFE_TensorHandle** outputs, TF_Status* s, void* device_info);

  // Method to delete a device.
  void (*delete_device)(void* device_info);

  // Implements TFE_CreatePackedTensorHandle when one of `handles` is on this