TF_Status* s, const char* name, TF_Operation** op,
                 bool check) {
  TF_OperationDescription* desc = TF_NewOperation(graph, "AddN", name);
  TF_Output add_inputs[2] = {{l, 0}, {r, 0}};
  TF_AddInputList(desc, add_inputs, 2);
  *op = TF_FinishOperation(desc, s);
  if (check) {
    ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    ASSERT_NE(*op, nullptr);
  }
}

  return parent_op_->DeviceName();
}
Status TapeOperation::SetDeviceName(const char* name) {
  return parent_op_->SetDeviceName(name);
}
Status TapeOperation::AddInput(AbstractTensorHandle* input) {
  TF_RETURN_IF_ERROR(parent_op_->AddInput(input));
  forward_op_.inputs.push_back(input);
  return absl::OkStatus();
}
Status TapeOperation::AddInputList(
    absl::Span<AbstractTensorHandle* const> inputs) {

  %bias = arith.constant dense<[7.11401462, 7.05456924]> : tensor<2xf32>

  %q_input= "quantfork.qcast"(%input) : (tensor<1x3x4x3xf32>) -> tensor<1x3x4x3x!quant.uniform<i8:f32, 0.58810077742034317:-128>>
  %dq_input= "quantfork.dcast"(%q_input) : (tensor<1x3x4x3x!quant.uniform<i8:f32, 0.58810077742034317:-128>>) -> tensor<1x3x4x3xf32>

// Records the op name in the `ForwardOperation`.
Status Reset(AbstractOperation*, const char* op, const char* raw_device_name,
             ForwardOperation*);

// Records the inputs in the `ForwardOperation`.
Status AddInput(AbstractOperation*, AbstractTensorHandle*, ForwardOperation*);
Status AddInputList(AbstractOperation*,
                    absl::Span<AbstractTensorHandle* const> inputs,
                    ForwardOperation*);

    }

    AbstractOperationPtr fn_op(ctx->CreateOperation());
    TF_RETURN_IF_ERROR(fn_op->Reset(fn_name, /*raw_device_name=*/nullptr));
    for (auto input : inputs) {
      TF_RETURN_IF_ERROR(fn_op->AddInput(input));
    }
    int retvals = outputs.size() - null_indices.size();
    std::vector<AbstractTensorHandle*> fn_outputs(retvals);
    TF_RETURN_IF_ERROR(fn_op->Execute(

  LogicalResult matchAndRewrite(stablehlo::PadOp pad_op,
                                PatternRewriter& rewriter) const override {
    Value pad_input = pad_op.getOperand();
    RankedTensorType pad_type = pad_op.getType().cast<RankedTensorType>();

    auto transpose_op = pad_input.getDefiningOp<stablehlo::TransposeOp>();
    if (!transpose_op || !transpose_op->hasOneUse()) return failure();

  auto* new_op = new EagerOperation(&ctx);
  TF_RETURN_IF_ERROR(
      new_op->Reset(fname.c_str(), orig_op->DeviceName().c_str()));
  for (auto input : orig_op->GetInputs()) {
    TF_RETURN_IF_ERROR(new_op->AddInput(input));
  }
  new_op->MutableAttrs()->CopyAttributes(orig_op->Attrs());
  out_op->reset(new_op);

  LOG_FIRST_N(INFO, 1)
      << "Finish Node Expansion Passes. Rewrite the op to call function: "
      << fname;

  SmallVector<Value, 8> operands_;
  llvm::StringMap<Attribute> attrs_;
  std::unique_ptr<OperationState> state_;
  // This is the index of the next ODS operand that will be added with AddInput
  // or AddInput;
  int current_ods_input_ = 0;
  const tensorflow::OpDef* op_def_ = nullptr;
  const char* op_name_ = nullptr;
  string tf_op_type_;
  // TODO(srbs): Use this.
  string device_name_;
};

  %bias = arith.constant dense<[7.11401462, 7.05456924]> : tensor<2xf32>

  %q_input= "quantfork.qcast"(%input) : (tensor<1x3x4x3xf32>) -> tensor<1x3x4x3x!quant.uniform<i8:f32, 0.58810077742034317:-128>>
  %dq_input= "quantfork.dcast"(%q_input) : (tensor<1x3x4x3x!quant.uniform<i8:f32, 0.58810077742034317:-128>>) -> tensor<1x3x4x3xf32>

  if (isa<TracingOperation>(check_numerics_op.get())) {
    s = dyn_cast<TracingOperation>(check_numerics_op.get())
            ->SetOpName("check_numerics");
    ASSERT_EQ(errors::OK, s.code()) << s.message();
  }
  s = AddInput(check_numerics_op.get(), t.get(), &forward_op);
  ASSERT_EQ(errors::OK, s.code()) << s.message();
  string message = "This is the way!";
  s = SetAttrString(check_numerics_op.get(), "message", message.data(),