// 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],

  }

  // 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) {

      "' (available: ");
  // Ensure deterministic (sorted) order in the error message
  std::set<string> factories_sorted;
  for (const auto& factory : GetFactories())
    factories_sorted.insert(factory.first);
  const char* comma = "";
  for (const string& factory : factories_sorted) {
    msg += comma + factory;
    comma = ", ";
  }
  msg += ")";

  return errors::InvalidArgument(msg.c_str());
}

using tensorflow::string;
using tensorflow::Tensor;
using tensorflow::TensorId;
using tensorflow::TensorShapeProto;
using tensorflow::VersionDef;
using tensorflow::errors::FailedPrecondition;
using tensorflow::errors::InvalidArgument;
using tensorflow::errors::OutOfRange;
using tensorflow::gtl::ArraySlice;
using tensorflow::strings::StrCat;

extern "C" {

  return parent_op_->SetDeviceName(name);
}
Status TapeOperation::AddInput(AbstractTensorHandle* input) {
  TF_RETURN_IF_ERROR(parent_op_->AddInput(input));
  forward_op_.inputs.push_back(input);
  return OkStatus();
}
Status TapeOperation::AddInputList(
    absl::Span<AbstractTensorHandle* const> inputs) {
  TF_RETURN_IF_ERROR(parent_op_->AddInputList(inputs));
  for (auto input : inputs) {
    forward_op_.inputs.push_back(input);
  }

      ASSERT_NEAR(dnumerical[j], danalytical[j], abs_error);
    }
    TF_DeleteTensor(analytical_tensor);
    TF_DeleteTensor(numerical_tensor);
    delete[] danalytical;
    delete[] dnumerical;
    outputs[i]->Unref();
  }
}

void CheckTensorValue(AbstractTensorHandle* t, absl::Span<const float> manuals,
                      absl::Span<const int64_t> dims, double abs_error) {
  TF_Tensor* analytical_tensor;

    status_ = StatusFromTF_Status(status.get());
    ASSERT_EQ(errors::OK, status_.code()) << status_.message();

    {
      AbstractContext* ctx_raw = nullptr;
      status_ =
          BuildImmediateExecutionContext(std::get<1>(GetParam()), &ctx_raw);
      ASSERT_EQ(errors::OK, status_.code()) << status_.message();
      immediate_execution_ctx_.reset(ctx_raw);
    }

      }
    }
  });
  // Loop until either block content is successfully fetched, or our request
  // encounters an error.
  absl::MutexLock l(&block->mu);
  TF_SetStatus(status, TF_OK, "");
  while (true) {
    switch (block->state) {
      case FetchState::ERROR:
        // TF_FALLTHROUGH_INTENDED
      case FetchState::CREATED:
        block->state = FetchState::FETCHING;

  EXPECT_EQ(TF_ALREADY_EXISTS,
            static_cast<TF_Code>(tensorflow::error::ALREADY_EXISTS));
  EXPECT_EQ(TF_PERMISSION_DENIED,
            static_cast<TF_Code>(tensorflow::error::PERMISSION_DENIED));
  EXPECT_EQ(TF_UNAUTHENTICATED,
            static_cast<TF_Code>(tensorflow::error::UNAUTHENTICATED));
  EXPECT_EQ(TF_RESOURCE_EXHAUSTED,

  tape.Watch(inputs[0]);  // Watch x.
  AbstractTensorHandle* exp_output;
  TF_RETURN_IF_ERROR(ops::Exp(ctx, inputs[0], &exp_output, "Exp"));
  std::unique_ptr<GradientFunction> gradient_function(
      new PassThroughGradientFunction);
  tape.RecordOperation(inputs, {exp_output}, gradient_function.release());
  TF_RETURN_IF_ERROR(tape.ComputeGradient(ctx,
                                          /*targets*/ {exp_output},