[Autogen Prim Operants] Autogen prim eager and static tensor operants

[jiahy0825]

PR types

New features

PR changes

Others

Describe

Auto-generate tensor operants codes, code example:

eager_tensor_operants.h

// Generated by paddle/fluid/prim/api/auto_code_generated/tensor_operants_gen.py

#pragma once

#include "paddle/phi/api/include/operants_base.h"
#include "paddle/phi/api/include/tensor.h"
#include "paddle/phi/core/macros.h"


namespace paddle {

namespace prim {

using Tensor = paddle::experimental::Tensor;
using TensorOperantsBase = paddle::operants::TensorOperantsBase;

class EagerTensorOperants : public TensorOperantsBase {
 private:
  DISABLE_COPY_AND_ASSIGN(EagerTensorOperants);

 public:
  EagerTensorOperants() = default;

  Tensor multiply(const Tensor& x, const Tensor& y);

};

}  // namespace prim
}  // namespace paddle

eager_tensor_operants.cc

// Generated by paddle/fluid/prim/api/auto_code_generated/tensor_operants_gen.py

#include "paddle/fluid/prim/utils/eager/eager_tensor_operants.h"

#include "paddle/fluid/eager/api/generated/eager_generated/forwards/dygraph_functions.h"


namespace paddle {

namespace prim {

Tensor EagerTensorOperants::multiply(const Tensor& x, const Tensor& y) {
  return ::multiply_ad_func(x, y);
}


}  // namespace prim
}  // namespace paddle

static_tensor_operants.h

// Generated by paddle/fluid/prim/api/auto_code_generated/tensor_operants_gen.py

#pragma once

#include "paddle/phi/api/include/operants_base.h"
#include "paddle/phi/api/include/tensor.h"
#include "paddle/phi/core/macros.h"


namespace paddle {

namespace prim {

using Tensor = paddle::experimental::Tensor;
using TensorOperantsBase = paddle::operants::TensorOperantsBase;

class StaticTensorOperants : public TensorOperantsBase {
 private:
  DISABLE_COPY_AND_ASSIGN(StaticTensorOperants);

 public:
  StaticTensorOperants() = default;

  Tensor multiply(const Tensor& x, const Tensor& y);

};

}  // namespace prim
}  // namespace paddle

static_tensor_operants.cc

// Generated by paddle/fluid/prim/api/auto_code_generated/tensor_operants_gen.py

#include "paddle/fluid/prim/utils/static/static_tensor_operants.h"

#include "paddle/fluid/prim/api/generated_prim/prim_generated_api.h"
#include "paddle/fluid/prim/utils/static/desc_tensor.h"


namespace paddle {

namespace prim {
using DescTensor = paddle::prim::DescTensor;

Tensor StaticTensorOperants::multiply(const Tensor& x, const Tensor& y) {
  return paddle::prim::multiply<DescTensor>(x, y);
}


}  // namespace prim
}  // namespace paddle

[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.

[YuanRisheng]

第一句的api_prims赋值有什么作用，另外api_prims = ('add', 'subtract', 'multiply', 'divide')这个应该会多次做出修改，是不是可以作为一个全局变量放在外边，方便修改维护

[jiahy0825]

现在只重载了四则运算，api_prims = ('add', 'subtract', 'multiply', 'divide') 只是一个短期的白名单。后续会直接将这行代码删掉，根据 api_prims = yaml.safe_load(f) 生成 prim 的 api。

api_prims = ('add', 'subtract', 'multiply', 'divide') is a temporary whitelist solution because tensor operants only overload arithmetic operators. In the future, this line of code will be deleted and generate prim api according to api_prims = yaml.safe_load(f)

[zyfncg]

2022 -> 2023

[jiahy0825]

Done, thx

[zyfncg]

从长期来看 tensor operants 和 EagerPrimAPI 是不解耦开比较好些？Operants本身和PrimAPI不是一个强绑定的关系，如果以后要新增一套支持C++组网训练的API，可能operants还需要从prim里面分出来

[jiahy0825]

Done