#include "tensorflow/c/experimental/gradients/nn_grad.h"

#include "absl/types/span.h"
#include "tensorflow/c/eager/abstract_tensor_handle.h"
#include "tensorflow/c/eager/immediate_execution_context.h"
#include "tensorflow/c/eager/immediate_execution_tensor_handle.h"
#include "tensorflow/c/experimental/ops/array_ops.h"
#include "tensorflow/c/experimental/ops/math_ops.h"
#include "tensorflow/c/experimental/ops/nn_ops.h"

void TF_SetTracingImplementation(const char* name, TF_Status* s) {
  tsl::Set_TF_Status_from_Status(s, SetDefaultTracingEngine(name));
}

// Creates a new TensorFlow function, it is an execution context attached to a
// given tracing context.
TF_ExecutionContext* TF_CreateFunction(const char* fn_name, TF_Status* s) {
  return wrap(CreateTracingExecutionContext(fn_name, s));
}

      immediate_execution_ctx_.reset(ctx_raw);
    }

    // Computing numerical gradients with TensorFloat-32 is numerically
    // unstable. Some forward pass tests also fail with TensorFloat-32 due to
    // low tolerances
    enable_tensor_float_32_execution(false);
  }

  AbstractContextPtr immediate_execution_ctx_;
  GradientRegistry registry_;
  Status status_;

#include "tensorflow/c/eager/c_api.h"
#include "tensorflow/c/eager/c_api_unified_experimental.h"
#include "tensorflow/c/eager/c_api_unified_experimental_internal.h"
#include "tensorflow/c/eager/immediate_execution_context.h"
#include "tensorflow/c/eager/immediate_execution_tensor_handle.h"
#include "tensorflow/c/eager/tfe_context_internal.h"
#include "tensorflow/c/eager/tfe_tensorhandle_internal.h"
#include "tensorflow/c/tf_status.h"

                             serialized_update.size(), status);
  EXPECT_NE(TF_OK, TF_GetCode(status)) << TF_Message(status);

  // Even after the prevoiusly failed cluster update, another update and op
  // execution should work fine as long as the provided server_def is valid.
  TFE_ContextUpdateServerDef(ctx, 0, serialized.data(), serialized.size(),
                             status);

  // CPU tensor to GPU).
  // Setting a larger thread pool does not help with the Swift caller, as we use
  // a different TFE context for each thread of execution (for running graph
  // functions, and their send/recvs corountines).
  config.set_inter_op_parallelism_threads(1);

  TF_Buffer* ret = TF_NewBuffer();
  TF_CHECK_OK(MessageToBuffer(config, ret));
  return ret;

See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/

#include "tensorflow/c/eager/immediate_execution_tensor_handle.h"

namespace tensorflow {

std::string ImmediateExecutionTensorHandle::DebugString() const {
  PartialTensorShape shape;
  std::string shape_string;
  if (Shape(&shape).ok()) {

      immediate_execution_ctx_.reset(ctx_raw);
    }

    // Computing numerical gradients with TensorFloat-32 is numerically
    // unstable. Some forward pass tests also fail with TensorFloat-32 due to
    // low tolerances
    enable_tensor_float_32_execution(false);
  }

  AbstractContextPtr immediate_execution_ctx_;
  GradientRegistry registry_;
  Status status_;

 public:

  TFE_ContextOptionsSetTfrt(opts, std::get<1>(GetParam()));
  TF_ExecutionContext* eager_execution_ctx =
      TF_NewEagerExecutionContext(opts, status.get());
  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
  TFE_DeleteContextOptions(opts);

  TF_ExecutionContextRegisterFunction(eager_execution_ctx, func, status.get());
  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());

  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  if (forward_input) {
    TFE_DeleteTensorHandle(n);
  }

  int num_retvals = 1;
  if (async) {
    // Enqueue dummy ops so we backlog async execution & actually test async.
    // This is usually unnecessary, but we've experienced the occasional test
    // failure when testing async mode with no explicit forwarding.
    for (int i = 0; i < 100000; ++i) {