// Recover data format from forward pass for gradient.
    std::string data_format;
    TF_RETURN_IF_ERROR(forward_attrs_.Get("data_format", &data_format));

    // Grad for A
    grad_inputs[0] = upstream_grad;
    grad_inputs[0]->Ref();

    // Grad for bias
    std::string name = "bias_add_grad";
    TF_RETURN_IF_ERROR(BiasAddGrad(ctx, upstream_grad, &grad_inputs[1],

  }
  registry_.insert({op_name, gradient_function_factory});
  return absl::OkStatus();
}
Status GradientRegistry::Lookup(
    const ForwardOperation& op,
    std::unique_ptr<GradientFunction>* gradient_function) const {
  auto iter = registry_.find(op.op_name);
  if (iter == registry_.end()) {
    const string error_msg = "No gradient defined for op: " + op.op_name + ".";
    return errors::NotFound(error_msg);

  }

  // Both files have been opened, do the transfer.
  // Since errno would be overridden by `close` below, save it here.
  int error_code = 0;
  if (CopyFileContents(dst_fd, src_fd, size) < 0) error_code = errno;

  close(src_fd);
  close(dst_fd);
  if (error_code != 0) {
    errno = error_code;
    return -1;
  } else {
    return 0;
  }
}

int RemoveSpecialDirectoryEntries(const struct dirent* entry) {

    shape_string = shape.DebugString();
  } else {
    shape_string = "<error computing shape>";
  }
  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), " [...]");
  }

    Model model, Model grad_model, AbstractContext* ctx,
    absl::Span<AbstractTensorHandle* const> inputs, bool use_function,
    double abs_error = 1e-2);

void CheckTensorValue(AbstractTensorHandle* t, absl::Span<const float> manuals,
                      absl::Span<const int64_t> dims, double abs_error = 1e-2);

Model BuildGradModel(Model forward, GradientRegistry registry);

}  // namespace internal
}  // namespace gradients

  OwnedTFStatus c_status(TF_NewStatus());
  init_fn(&params, c_status.get());
  TF_RETURN_IF_ERROR(tsl::StatusFromTF_Status(c_status.get()));
  TF_RETURN_IF_ERROR(ValidateTPOptimizerRegistrationParams(params));
  TF_RETURN_IF_ERROR(ValidateTPOptimizer(optimizer));
  TF_RETURN_IF_ERROR(ValidateTPOptimizerConfigs(optimizer_configs));

  CGraphOptimizerRegister(

//  --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;

// As we need to test multiple URI schemes we need a parameterized test.
// Furthermore, since each test creates and deletes files, we will use the same

    for (auto input : inputs) {
      TF_RETURN_IF_ERROR(fn_op->AddInput(input));
    }
    int retvals = outputs.size() - null_indices.size();
    std::vector<AbstractTensorHandle*> fn_outputs(retvals);
    TF_RETURN_IF_ERROR(fn_op->Execute(
        absl::Span<AbstractTensorHandle*>(fn_outputs.data(), fn_outputs.size()),
        &retvals));
    int skipped_indices = 0;

    // Get transpose attrs
    bool t_a;
    TF_RETURN_IF_ERROR(forward_attrs_.Get("transpose_a", &t_a));

    bool t_b;
    TF_RETURN_IF_ERROR(forward_attrs_.Get("transpose_b", &t_b));

    // Conj each input
    AbstractTensorHandle* conj_output;
    std::string name = "Conj_A_MatMul_Grad";
    TF_RETURN_IF_ERROR(
        SafeConj(ctx, forward_inputs_[0], &conj_output, name.c_str()));

// type of files.
static Status ValidateOperations(const TF_FilesystemPluginOps* ops) {
  TF_RETURN_IF_ERROR(ValidateHelper(ops->filesystem_ops));
  TF_RETURN_IF_ERROR(ValidateHelper(ops->random_access_file_ops));
  TF_RETURN_IF_ERROR(ValidateHelper(ops->writable_file_ops));
  TF_RETURN_IF_ERROR(ValidateHelper(ops->read_only_memory_region_ops));

  if (ops->filesystem_ops->new_random_access_file != nullptr &&