[PRIM][IR]support optional and invoke for more vjp code gen

[Charles-hit]

PR types

New features

PR changes

Others

Description

Pcard-66975
背景：
在PR PR56512基础上进行反向算子vjp生成扩量工作，高优支持GPT/LLama依赖的26个高优算子vjp.
PR改动：

• 支持invoke字段解析, 生成yaml中invoke算子vjp
  以cast_vjp为例:

std::vector<std::vector<paddle::Tensor>> cast_vjp(const Tensor& x, const Tensor& out_grad, const std::vector<std::vector<bool>>& stop_gradients) {
  std::vector<std::vector<paddle::Tensor>> vjp_res;
  for (auto arg: stop_gradients) {
    vjp_res.push_back(std::vector<paddle::Tensor>(arg.size()));
  }
  auto op_res = backend::cast<LazyTensor>(out_grad, x.dtype());
  vjp_res[0][0] = op_res;
  vjp_res = ConstructVjpResultByStopGradients(vjp_res, stop_gradients);
  return vjp_res;
}

• 支持optional字段解析
  以layer_norm_vjp为例：

std::vector<std::vector<paddle::Tensor>> layer_norm_vjp(const Tensor& x, const paddle::optional<Tensor>& scale, const paddle::optional<Tensor>& bias, const Tensor& mean, const Tensor& variance, const Tensor& out_grad, float epsilon, int begin_norm_axis, const std::vector<std::vector<bool>>& stop_gradients) {
  std::vector<std::vector<paddle::Tensor>> vjp_res;
  for (auto arg: stop_gradients) {
    vjp_res.push_back(std::vector<paddle::Tensor>(arg.size()));
  }
  auto op_res = backend::layer_norm_grad<LazyTensor>(x, scale, bias, mean, variance, out_grad, epsilon, begin_norm_axis);
  vjp_res[0][0] = std::get<0>(op_res);
  vjp_res[1][0] = std::get<1>(op_res);
  vjp_res[2][0] = std::get<2>(op_res);
  vjp_res = ConstructVjpResultByStopGradients(vjp_res, stop_gradients);
  return vjp_res;
}

[paddle-bot]

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.