.NewSubScope(child_scope_name);

    if (dx != nullptr) {
      std::vector<tensorflow::Output> dx_arg = OutputsFromTFOutputs(dx, ny);
      status->status =
          AddSymbolicGradients(scope, y_arg, x_arg, dx_arg, &dy_arg);
    } else {
      status->status = AddSymbolicGradients(scope, y_arg, x_arg, &dy_arg);
    }

    // Update g->name_map with the name_map from the scope, which will contain

  if (!status->status.ok()) return nullptr;

  if (input_index >= op_def->input_arg_size() || input_index < 0) {
    status->status = tensorflow::errors::InvalidArgument(
        input_index, " out of range for ", op_name);
    return nullptr;
  }

  const tensorflow::OpDef_ArgDef& input_arg = op_def->input_arg()[input_index];

  if (input_arg.number_attr().empty()) {
    status->status = tensorflow::errors::NotFound(

  CHECK(tensorflow::protobuf::TextFormat::ParseFromString(
      "    signature {"
      "      name: 'AddVariablesFunction'"
      "      input_arg {"
      "        name: 'var'"
      "        type: DT_RESOURCE"
      "      }"
      "      output_arg {"
      "        name: 'sum'"
      "        type: DT_FLOAT"
      "      }"
      "    }"
      "    node_def {"
      "      name: 'read0'"

    ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());

    auto cpu_arg =
        tensorflow::TensorHandleFromInterface(tensorflow::unwrap(hcpu));
    auto gpu_arg =
        tensorflow::TensorHandleFromInterface(tensorflow::unwrap(hgpu));
    auto gpu_device = gpu_arg->device();
    ASSERT_FALSE(cpu_arg->HasLocalMirror(gpu_device));

    TFE_Op* matmul = MatMulOp(ctx, hcpu, hgpu);
    if (cpu_op) {

//
// Example invocation:
//  bazel test //tensorflow/c/experimental/filesystem:modular_filesystem_test \\
//  --test_arg=--dso=/path/to/one.so --test_arg=--dso=/path/to/another.so \\
//  --test_arg=--scheme= --test_arg=--scheme=file
//
// Note that to test the local filesystem we use an empty value.

namespace tensorflow {
namespace {

using ::tensorflow::error::Code;

                                                  TF_Status* status) {
  auto* cc_ctx = reinterpret_cast<tensorflow::OpKernelContext*>(ctx);
  const tensorflow::Tensor& arg_tensor = cc_ctx->input(index);
  absl::Status cc_status;
  if (arg_tensor.dtype() != tensorflow::DT_RESOURCE) {
    cc_status = absl::InvalidArgumentError(
        absl::StrCat("Trying to obtain resource handle from Input[", index,

  GeneralAssignment("AssignVariableOp", context, value, status);
}

// Passed to `TF_NewTensor` to indicate how an array of floats should be
// deleted.
static void FloatDeallocator(void* data, size_t, void* arg) {
  delete[] static_cast<float*>(data);
}

// Creates a TFE_TensorHandle with value `v`.
TensorHandlePtr FloatTensorHandle(float v, TF_Status* status) {
  const int num_bytes = sizeof(float);

                   "      input_arg {"
                   "        name: 'a'"
                   "        type: DT_FLOAT"
                   "      }"
                   "      input_arg {"
                   "        name: 'b'"
                   "        type: DT_FLOAT"
                   "      }"
                   "      output_arg {"
                   "        name: 'm'"

  // First create a simple identity function.
  TF_Graph* function_graph = TF_NewGraph();
  TF_OperationDescription* arg_descr =
      TF_NewOperation(function_graph, "Placeholder", "arg");
  TF_SetAttrType(arg_descr, "dtype", TF_INT32);
  TF_Status* status = TF_NewStatus();
  TF_Operation* arg = TF_FinishOperation(arg_descr, status);
  ASSERT_TRUE(TF_GetCode(status) == TF_OK) << TF_Message(status);
  TF_OperationDescription* id_descr =

  // Build two abstract input tensors as function arguments.
  std::vector<TF_AbstractTensor*> func_args;
  {
    TFE_Context* eager_ctx =
        TF_ExecutionContextGetTFEContext(eager_execution_ctx, s);

    // 1st Arg
    float vals1[] = {0.0f, 1.0f, 1.0f, 0.0f};
    int64_t dims[] = {2, 2};  // Matrices will be 2 x 2
    int num_dims = sizeof(dims) / sizeof(dims[0]);

    TFE_TensorHandle* input_eager =