Status Compute(AbstractContext* ctx,
                 absl::Span<AbstractTensorHandle* const> grad_outputs,
                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    CHECK_EQ(grad_outputs.size(), 1);
    CHECK_EQ(grad_inputs.size(), 1);
    grad_inputs[0] = grad_outputs[0];
    if (grad_inputs[0]) {
      grad_inputs[0]->Ref();
    }
    return absl::OkStatus();
  }
};

// Computes:
//
// @tf.custom_gradient

        "//tensorflow/c:tf_status",
        "@com_google_absl//absl/base:core_headers",
        "@com_google_absl//absl/synchronization",
    ],
)

tf_cc_test(
    name = "ram_file_block_cache_test",
    size = "small",
    srcs = ["ram_file_block_cache_test.cc"],
    deps = [
        ":ram_file_block_cache",
        "//tensorflow/c:tf_status_internal",
        "//tensorflow/core:lib",
        "//tensorflow/core:test",

 public:
  Status Compute(AbstractContext* ctx,
                 absl::Span<AbstractTensorHandle* const> grad_outputs,
                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    for (int i = 0; i < grad_outputs.size(); i++) {
      auto grad_input = grad_outputs[i];
      // TODO(srbs): Should we add a copy contructor to AbstractTensorHandle
      // that takes care of this similar to `Tensor`?
      if (grad_input) {

TFCI_BAZEL_TARGET_SELECTING_CONFIG_PREFIX=linux_tpu
TFCI_BUILD_PIP_PACKAGE_ARGS="--repo_env=WHEEL_NAME=tensorflow_tpu"
TFCI_LIB_SUFFIX="-tpu-linux-x86_64"
TFCI_WHL_BAZEL_TEST_ENABLE=0
TFCI_WHL_IMPORT_TEST_ENABLE=0
TFCI_WHL_SIZE_LIMIT=580M

  TFE_Context* ctx = TFE_NewContext(opts, status);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_DeleteContextOptions(opts);

  TFE_ContextSetServerDef(ctx, 0, serialized.data(), serialized.size(), status);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  TFE_TensorHandle* h0_task0 = TestMatrixTensorHandle(ctx);
  TFE_TensorHandle* h1_task0 = TestMatrixTensorHandle(ctx);

        ' binary GPU code, or as "sm_xy" to only include the binary '
        'code.\nPlease note that each additional compute capability '
        'significantly increases your build time and binary size, and that '
        'TensorFlow only supports compute capabilities >= 3.5 [Default is: '
        '%s]: ' % default_cuda_compute_capabilities)
    tf_cuda_compute_capabilities = get_from_env_or_user_or_default(

    if (read >= 0) content.resize(read);
    if (file_size != content.size())
      TF_SetStatus(
          status_, TF_DATA_LOSS,
          std::string("expected " + std::to_string(file_size) + " got " +
                      std::to_string(content.size()) + " bytes")
              .c_str());
    return content;
  }

 protected:
  TF_Filesystem* filesystem_;
  TF_Status* status_;

 private:

  for (int device_index = 0; device_index < underlying_devices_.size();
       ++device_index) {
    DeviceThread* device_thread = device_threads_[device_index].get();
    std::vector<TFE_TensorHandle*> device_inputs;
    device_inputs.reserve(inputs.size());
    for (int input_index = 0; input_index < inputs.size(); ++input_index) {
      // Parallel tensors are divided between operations by device.