TF_Tensor* tensor_1X6 = Int32Tensor({1, 6});
  CHECK_EQ(TF_OK, TF_GetCode(status_)) << TF_Message(status_);
  TF_Tensor* tensor_1X1X6 = Int32Tensor({1, 1, 6});
  CHECK_EQ(TF_OK, TF_GetCode(status_)) << TF_Message(status_);

  TFE_Op* reshape_op = TFE_NewOp(tfe_context_, "Reshape", status_);
  CHECK_EQ(TF_OK, TF_GetCode(status_)) << TF_Message(status_);
  TFE_OpSetAttrType(reshape_op, "T", TF_FLOAT);

  TFE_Context* ctx = TFE_NewContext(opts, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_DeleteContextOptions(opts);

  TFE_TensorHandle* h = TestScalarTensorHandle(ctx, 1.0f);
  TFE_TensorDebugInfo* debug_info = TFE_TensorHandleTensorDebugInfo(h, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  ASSERT_EQ(0, TFE_TensorDebugInfoOnDeviceNumDims(debug_info));

    CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
    TFE_OpAddInput(matmul, m, status);
    CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
    TFE_OpAddInput(matmul, m, status);
    CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
    TFE_Execute(matmul, &retvals[0], &num_retvals, status);
    CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* b = Placeholder(graph, s, "B");
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* two = ScalarConst(2, graph, s);
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* plus2 = Add(a, two, graph, s, "plus2");
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  TFE_OpAddInput(func, packed_handle, status);
  ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  if (remote) {
    TFE_OpSetDevice(func, task1_name, status);
    ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  }

  TFE_TensorHandle* retvals[1] = {nullptr};
  int num_retvals = 1;
  TFE_Execute(func, &retvals[0], &num_retvals, status);

    TF_Operation* add2 = Add(add1, one, params->body_graph, s_, "add2");
    ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
    params->body_outputs[0] = {add2, 0};
    params->body_outputs[1] = params->body_inputs[1];

    // Finalize while loop
    TF_FinishWhile(params.get(), s_, &outputs[0]);
    EXPECT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
  }

  // Define function, use it in graph, and run

  TF_OperationDescription* desc = TF_NewOperation(graph, "Const", name);
  TF_SetAttrTensor(desc, "value", t, s);
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
  TF_SetAttrType(desc, "dtype", TF_TensorType(t));
  *op = TF_FinishOperation(desc, s);
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
  ASSERT_NE(*op, nullptr);
}

TF_Operation* Const(TF_Tensor* t, TF_Graph* graph, TF_Status* s,
                    const char* name) {

      TF_SetStatus(first_bad_status.get(), TF_GetCode(async_wait_status.get()),
                   TF_Message(async_wait_status.get()));
    }
  }

  if (first_bad_status != nullptr) {
    TF_SetStatus(status, TF_GetCode(first_bad_status.get()),
                 TF_Message(first_bad_status.get()));
  }
}

absl::optional<std::vector<std::unique_ptr<ParallelTensor>>>
ParallelDevice::Join(

  ASSERT_NE(handle, nullptr)
      << TF_Message(status) << " (shape=[" << absl::StrJoin(shape, ",")
      << "], strides=[" << absl::StrJoin(strides, ",") << "])";

  auto* dlm_out =
      static_cast<DLManagedTensor*>(TFE_HandleToDLPack(handle, status));
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  const DLTensor* dltensor_out = &dlm_out->dl_tensor;

  if (TF_GetCode(status) != TF_OK) {
    fprintf(stderr, "TF_NewWritableFile failed: %s\n", TF_Message(status));
    return 1;
  }
  fprintf(stderr, "wrote %s\n", full_path);
  free(full_path);
  TF_CloseWritableFile(h, status);
  if (TF_GetCode(status) != TF_OK) {
    fprintf(stderr, "TF_CloseWritableFile failed: %s\n", TF_Message(status));
  }
  TF_StringStreamDone(s);

  TF_KernelBuilder* b =