void TestRemoteExecuteSilentCopiesOp(bool async, bool remote,
                                     bool remote_func_outputs = false) {
  return TestRemoteExecuteSilentCopies(async, remote, /*func=*/false,
                                       /*heavy_load_on_streaming_rpc=*/false,
                                       remote_func_outputs);
}

TEST(CAPI, RemoteExecuteSilentCopies) {

  TestRemoteExecuteSilentCopiesFunc(/*async=*/true, /*remote=*/false,
                                    /*heavy_load_on_streaming_rpc=*/false,
                                    /*remote_func_outputs=*/true);
}
TEST(CAPI, RemoteExecuteSilentCopiesLocalAsyncFuncOrdering) {
  // A remote input may be not ready when we start running a function. Test that
  // the function execution should wait until the remote input is ready.

A. Unique TensorFlower <******@****.***> 1682633218 -0700

  // TODO(gjn): Add support for waiting on async local mirrors
  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 =

  // Check that copying it to the old remote device (named localhost) fails.
  TFE_TensorHandleCopyToDevice(h0_task0_new, ctx, remote_device_name, status);
  EXPECT_NE(TF_OK, TF_GetCode(status)) << TF_Message(status);

  // Copying and executing on the new remote device works.
  const char new_remote_device_name[] =
      "/job:worker/replica:0/task:1/device:CPU:0";

      tensorflow::ContextFromInterface(tensorflow::unwrap(ctx));
  auto collective_executor_handle = context->GetCollectiveExecutorHandle();
  tensorflow::Notification done;
  collective_executor_handle->get()->remote_access()->CheckPeerHealth(
      task, timeout_in_ms, [&done, status](const Status& s) {
        status->status = s;
        done.Notify();
      });
  done.WaitForNotification();
}

TempFile::TempFile(TempFile&& rhs)
    : std::fstream(std::move(rhs)), name_(std::move(rhs.name_)) {}

TempFile::~TempFile() {
  std::fstream::close();
  std::remove(name_.c_str());
}

const std::string TempFile::getName() const { return name_; }

bool TempFile::truncate() {
  std::fstream::close();
  std::fstream::open(name_, std::ios::binary | std::ios::out);

namespace {

#if !defined(IS_MOBILE_PLATFORM) && !defined(IS_SLIM_BUILD)

// Creates a placeholder representing an input to the cond or body graph.
// TODO(skyewm): remove these from final graph
bool CreateInput(const TF_Output& parent_input, TF_Graph* g, const char* name,
                 TF_Output* input, TF_Status* status) {

  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  // Pack 3 variable handles into one TFE_TensorHandle.
  // When remote is false, function device is placed on task0. Handle types are
  // REMOTE, REMOTE, LOCAL on task0. When remote is true, function device is
  // placed on task1, Handle types are LOCAL, REMOTE, LOCAL on task1.
  int num_replicas = 3;
  std::vector<TFE_TensorHandle*> handles = {h0, h1, h2};

      immediate_execution_ctx_.get(), {x.get()}, UseFunction()));
}

TEST_P(CppGradients, TestMatMulGrad) {
  // TODO(vnvo2409): Figure out why `gradient_checker` does not work very
  // well with `MatMul` and remove `TestMatMul*` in
  // `mnist_gradients_test` when done.
  GTEST_SKIP();

  float A_vals[] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f};
  int64_t A_dims[] = {3, 3};
  AbstractTensorHandlePtr A;
  {