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};

// `src_inputs[i]` will have that input replaced with `dst_inputs[i]`.  `prefix`
// will be prepended to copied node names. `control_deps` are nodes in
// `dst_graph` that the copied `src_graph` nodes will have control dependencies
// on. `return_nodes` are nodes in `src_graph`, and the new corresponding nodes
// in `dst_graph` will be returned. `return_nodes` must be non-null.

      plugin_memory_allocate(TF_WRITABLE_FILE_OPS_SIZE));
  ops->writable_file_ops->cleanup = tf_writable_file::Cleanup;
  ops->writable_file_ops->append = tf_writable_file::Append;
  ops->writable_file_ops->tell = tf_writable_file::Tell;
  ops->writable_file_ops->flush = tf_writable_file::Flush;
  ops->writable_file_ops->sync = tf_writable_file::Sync;
  ops->writable_file_ops->close = tf_writable_file::Close;

#include "tensorflow/core/platform/test.h"

// NOTE(allenl): These tests currently go through TFE_Execute and so are
// integration testing rather than purely testing the parallel device. They
// correspond fairly well to the implementation, but testing the C++ directly is
// another option.

namespace tensorflow {
namespace parallel_device {

Variable* Variable::Create(TFE_Context* context, TF_DataType type,

  TF_DeleteStatus(status);
  return counter;
}

void IncrementCounter0(TFE_MonitoringCounter0* counter, int64_t delta) {
  auto* cell = TFE_MonitoringGetCellCounter0(counter);
  TFE_MonitoringCounterCellIncrementBy(cell, delta);
}

TFE_MonitoringCounter1* CreateCounter1(const char* counter_name,
                                       const char* label) {
  TF_Status* status = TF_NewStatus();

#include "tensorflow/core/platform/test.h"

// NOTE(allenl): These tests currently go through TFE_Execute and so are
// integration testing rather than purely testing the parallel device. They
// correspond fairly well to the implementation, but testing the C++ directly is
// another option.

namespace tensorflow {
namespace parallel_device {

using ::testing::HasSubstr;

TEST(PARALLEL_DEVICE, TestBasicCPU) {

  length = tf_writable_file::Tell(file, status_);
  ASSERT_EQ(length, 4);
  ASSERT_TF_OK(status_);
  tf_writable_file::Flush(file, status_);
  ASSERT_TF_OK(status_);
  ASSERT_TRUE(
      CompareWithServer(filepath, 0, 4, &gcs_file->gcs_client, status_));

  tf_writable_file::Append(file, content + 4, 4, status_);
  ASSERT_TF_OK(status_);
  length = tf_writable_file::Tell(file, status_);
  ASSERT_EQ(length, 8);

  return context->GetContextId();
}

void TFE_MonitoringCounterCellIncrementBy(TFE_MonitoringCounterCell* cell,
                                          int64_t value) {
  cell->cell.IncrementBy(value);
}

int64_t TFE_MonitoringCounterCellValue(TFE_MonitoringCounterCell* cell) {
  return cell->cell.value();
}

TFE_MonitoringCounter0* TFE_MonitoringNewCounter0(const char* name,

  return result;
}

int64_t TFE_MonitoringReadCounter0(TFE_MonitoringCounterReader* cell_reader) {
  int64_t result = cell_reader->Read();

  return result;
}

int64_t TFE_MonitoringReadCounter1(TFE_MonitoringCounterReader* cell_reader,
                                   const char* label) {
  int64_t result = cell_reader->Read(label);

  return result;

  if (content_length == stream.headers().end()) {
    // When we read a file with offset that is bigger than the actual file size.
    // GCS will return an empty header (e.g no `content-length` header). In this
    // case, we will set read to `0` and continue.
    read = 0;
  } else if (!absl::SimpleAtoi(content_length->second, &read)) {
    TF_SetStatus(status, TF_UNKNOWN, "Could not get content-length header");