ASSERT_TRUE(TF_GetCode(status) == TF_OK) << TF_Message(status);
  TF_DeleteFunction(fn);

  for (bool async : {false, true, false}) {
    TFE_Executor* old_executor = TFE_ContextGetExecutorForThread(ctx);
    TFE_Executor* executor = TFE_NewExecutor(
        /*is_async=*/async, /*enable_streaming_enqueue=*/true,
        /*in_flight_nodes_limit=*/0);
    TFE_ContextSetExecutorForThread(ctx, executor);

  CheckTFE_TensorHandleHasFloats(retval_task0, {7, 10, 15, 22});

  TFE_DeleteTensorHandle(retval_task0);
  TFE_DeleteTensorHandle(h0_task0);
  TFE_DeleteTensorHandle(retvals[0]);

  TFE_DeleteOp(matmul);

  TFE_Executor* executor = TFE_ContextGetExecutorForThread(ctx);
  TFE_ExecutorWaitForAllPendingNodes(executor, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_DeleteExecutor(executor);
  TF_DeleteStatus(status);

    TFE_Execute(matmul, &retvals[0], &num_retvals, status);
    CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
    if (state.iterations() >= state.max_iterations && async) {
      TFE_Executor* executor = TFE_ContextGetExecutorForThread(ctx);
      TFE_ExecutorWaitForAllPendingNodes(executor, status);
      ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
      TFE_DeleteExecutor(executor);
    }