MatMulOp(context, hcpu, hdevice), TFE_DeleteOp);
  TFE_OpSetDevice(matmul.get(), name, status.get());
  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  TFE_TensorHandle* retval;
  int num_retvals = 1;
  TFE_Execute(matmul.get(), &retval, &num_retvals, status.get());
  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  ASSERT_TRUE(executed);

  TFE_DeleteTensorHandle(retval);

TF_Tensor* BoolTensor(int32_t v);

// Create a tensor with values of type TF_INT8 provided by `values`.
TF_Tensor* Int8Tensor(const int64_t* dims, int num_dims, const char* values);

// Create a tensor with values of type TF_INT32 provided by `values`.
TF_Tensor* Int32Tensor(const int64_t* dims, int num_dims,
                       const int32_t* values);

// Create 1 dimensional tensor with values from `values`

typedef struct TF_FilesystemPluginOps {
  char* scheme;
  int filesystem_ops_abi;
  int filesystem_ops_api;
  size_t filesystem_ops_size;
  TF_FilesystemOps* filesystem_ops;
  int random_access_file_ops_abi;
  int random_access_file_ops_api;
  size_t random_access_file_ops_size;
  TF_RandomAccessFileOps* random_access_file_ops;
  int writable_file_ops_abi;
  int writable_file_ops_api;
  size_t writable_file_ops_size;

// `mode`. The later is only used if `dst` needs to be created.
int TransferFileContents(const char* src, const char* dst, mode_t mode,
                         off_t size);

// Returns true only if `entry` points to an entry other than `.` or `..`.
//
// This is a filter for `scandir`.
int RemoveSpecialDirectoryEntries(const struct dirent* entry);

}  // namespace tf_posix_filesystem

}

TF_Tensor* TFE_AllocateHostTensor(TFE_Context* ctx, TF_DataType dtype,
                                  const int64_t* dims, int num_dims,
                                  TF_Status* status) {
  std::vector<int64_t> dimvec(num_dims);
  for (int i = 0; i < num_dims; ++i) {
    dimvec[i] = static_cast<int64_t>(dims[i]);
  }

  if (ctx == nullptr) {

namespace internal {
namespace {

using tensorflow::TF_StatusPtr;

void CompareNumericalAndManualGradients(
    Model model, AbstractContext* ctx,
    absl::Span<AbstractTensorHandle* const> inputs, int input_index,
    float* expected_grad, int num_grad, bool use_function,
    double abs_error = 1e-2) {
  Status s;
  AbstractTensorHandlePtr numerical_grad;
  {
    AbstractTensorHandle* numerical_grad_raw;

  std::vector<TFE_TensorHandle*> handles_task0;
  if (heavy_load_on_streaming_rpc) {
    // Send 50 tensor copy requests to simulate that there have been some RPC
    // requests been enqueued.
    for (int i = 0; i < 50; ++i) {
      handles_task0.push_back(TestMatrixTensorHandle(ctx));
    }
  }
  const char task1_name[] = "/job:localhost/replica:0/task:1/device:CPU:0";

#include <memory>

#include "tensorflow/core/common_runtime/eager/eager_executor.h"

struct TFE_Executor {
  explicit TFE_Executor(bool async, bool enable_streaming_enqueue,
                        int in_flight_nodes_limit)
      : owned_executor(new tensorflow::EagerExecutor(
            async, enable_streaming_enqueue, in_flight_nodes_limit)) {}

  explicit TFE_Executor(tensorflow::EagerExecutor* executor)

TEST(CAPI_EXPERIMENTAL, IsStateful) {
  std::unique_ptr<TF_Status, decltype(&TF_DeleteStatus)> status(
      TF_NewStatus(), TF_DeleteStatus);
  int assign = TF_OpIsStateful("AssignAddVariableOp", status.get());
  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
  EXPECT_EQ(assign, 1);
  int id = TF_OpIsStateful("Identity", status.get());
  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
  EXPECT_EQ(id, 0);
}

    for (int i = 0; i < expected.size(); ++i) {
      ASSERT_EQ(expected[i], data[i]) << "Different data at index " << i;
    }
  }

  std::vector<TF_Output> ToOutput(const std::vector<TF_Operation*> ops) {
    std::vector<TF_Output> out;
    for (auto op : ops) {
      out.push_back({op, 0});
    }
    return out;
  }

  void Define(int num_opers, const std::vector<TF_Operation*>& opers,