[NewIR]Gen most api and support gen mutable attr api

[0x45f]

PR types

Others

PR changes

Others

Description

之前的PR #56317 解决冲突时有点问题，重新开了一个PR

• 生成了绝大部分的API，因为某些op比如copy_to 因为没有infer meta，所以在pd_op中生成的Build函数是有问题的，暂时跳过了此类API的生成。

• 支持了std::vectorir::OpResult返回值类型API的生成，比如下面的rnn_grad：

std::tuple<ir::OpResult, std::vector<ir::OpResult>, std::vector<ir::OpResult>> rnn_grad(ir::OpResult x, std::vector<ir::OpResult> pre_state, std::vector<ir::OpResult> weight_list, ir::OpResult sequence_length, ir::OpResult out, ir::OpResult dropout_state_out, ir::OpResult reserve, ir::OpResult out_grad, std::vector<ir::OpResult> state_grad, float dropout_prob, bool is_bidirec, int input_size, int hidden_size, int num_layers, const std::string& mode, int seed, bool is_test){
    auto pre_state_combine_op = APIBuilder::Instance().GetBuilder()->Build<ir::CombineOp>(pre_state);
    auto weight_list_combine_op = APIBuilder::Instance().GetBuilder()->Build<ir::CombineOp>(weight_list);
    auto state_grad_combine_op = APIBuilder::Instance().GetBuilder()->Build<ir::CombineOp>(state_grad);
    paddle::dialect::RnnGradOp rnn_grad_op = APIBuilder::Instance().GetBuilder()->Build<paddle::dialect::RnnGradOp>(x, pre_state_combine_op.out(), weight_list_combine_op.out(), sequence_length, out, dropout_state_out, reserve, out_grad, state_grad_combine_op.out(), dropout_prob, is_bidirec, input_size, hidden_size, num_layers, mode, seed, is_test);
    auto pre_state_grad_split_op = APIBuilder::Instance().GetBuilder()->Build<ir::SplitOp>(rnn_grad_op.result(1));
    auto weight_list_grad_split_op = APIBuilder::Instance().GetBuilder()->Build<ir::SplitOp>(rnn_grad_op.result(2));
    return std::make_tuple(rnn_grad_op.result(0), pre_state_grad_split_op.outputs(), weight_list_grad_split_op.outputs());
}

• 针对可变attr生成了一套api，其参数顺序与op build函数的参数顺序保持了一致

// 非可变attr api
std::tuple<ir::OpResult, ir::OpResult> send_u_recv(ir::OpResult x, ir::OpResult src_index, ir::OpResult dst_index, const std::string& reduce_op = "SUM", const std::vector<int64_t>& out_size = {0});

std::tuple<ir::OpResult, ir::OpResult> send_u_recv(ir::OpResult x, ir::OpResult src_index, ir::OpResult dst_index, const std::string& reduce_op, const std::vector<int64_t>& out_size){
    paddle::dialect::SendURecvOp send_u_recv_op = APIBuilder::Instance().GetBuilder()->Build<paddle::dialect::SendURecvOp>(x, src_index, dst_index, reduce_op, out_size);
    return std::make_tuple(send_u_recv_op.result(0), send_u_recv_op.result(1));
}

// 可变attr api
std::tuple<ir::OpResult, ir::OpResult> send_u_recv(ir::OpResult x, ir::OpResult src_index, ir::OpResult dst_index, ir::OpResult out_size, const std::string& reduce_op = "SUM");

std::tuple<ir::OpResult, ir::OpResult> send_u_recv(ir::OpResult x, ir::OpResult src_index, ir::OpResult dst_index, ir::OpResult out_size, const std::string& reduce_op){
    paddle::dialect::SendURecvOp send_u_recv_op = APIBuilder::Instance().GetBuilder()->Build<paddle::dialect::SendURecvOp>(x, src_index, dst_index, out_size, reduce_op);
    return std::make_tuple(send_u_recv_op.result(0), send_u_recv_op.result(1));
}

TODO:

• 修改生成代码的格式
• 目前inplace版本的api和非inplace生成的是一样的，后续可能需要修改

Pcard-67164

[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.