register int/int64_t/float16 in pow/square kernel,test=develop (#22023)

* register int/int64_t/float16 in  pow/square kernel,test=develop

* add abs/square/exp type,test=develop

paddle/fluid/operators/activation_op.cc

@@ -1029,8 +1029,24 @@ REGISTER_OPERATOR(
     ops::ActivationOpDoubleGrad<ops::SquareGradGradFunctor<float>::FwdDeps()>,
     ops::ActivationDoubleGradOpInplaceInference);
 
-REGISTER_ACTIVATION_CPU_KERNEL(square, Square, SquareFunctor,
-                               SquareGradFunctor);
+REGISTER_OP_CPU_KERNEL(square,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::SquareFunctor<float>>,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::SquareFunctor<double>>,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::SquareFunctor<int>>,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::SquareFunctor<int64_t>>);
+REGISTER_OP_CPU_KERNEL(
+    square_grad, ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                                           ops::SquareGradFunctor<float>>,
+    ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                              ops::SquareGradFunctor<double>>,
+    ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                              ops::SquareGradFunctor<int>>,
+    ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                              ops::SquareGradFunctor<int64_t>>);
 
 REGISTER_OP_CPU_KERNEL(
     square_grad_grad,
@@ -1039,7 +1055,11 @@ REGISTER_OP_CPU_KERNEL(
     ops::SquareDoubleGradKernel<plat::CPUDeviceContext,
                                 ops::SquareGradGradFunctor<double>>,
     ops::SquareDoubleGradKernel<plat::CPUDeviceContext,
-                                ops::SquareGradGradFunctor<plat::float16>>);
+                                ops::SquareGradGradFunctor<plat::float16>>,
+    ops::SquareDoubleGradKernel<plat::CPUDeviceContext,
+                                ops::SquareGradGradFunctor<int>>,
+    ops::SquareDoubleGradKernel<plat::CPUDeviceContext,
+                                ops::SquareGradGradFunctor<int64_t>>);
 /* ========================================================================== */
 
 /* ==========================   pow register  ============================ */
@@ -1055,9 +1075,77 @@ REGISTER_OPERATOR(pow_grad, ops::PowOpGrad,
 
 REGISTER_OP_CPU_KERNEL(
     pow, ops::PowKernel<plat::CPUDeviceContext, ops::PowFunctor<float>>,
-    ops::PowKernel<plat::CPUDeviceContext, ops::PowFunctor<double>>);
+    ops::PowKernel<plat::CPUDeviceContext, ops::PowFunctor<double>>,
+    ops::PowKernel<plat::CPUDeviceContext, ops::PowFunctor<int>>,
+    ops::PowKernel<plat::CPUDeviceContext, ops::PowFunctor<int64_t>>);
 REGISTER_OP_CPU_KERNEL(
     pow_grad,
     ops::PowGradKernel<plat::CPUDeviceContext, ops::PowGradFunctor<float>>,
-    ops::PowGradKernel<plat::CPUDeviceContext, ops::PowGradFunctor<double>>);
+    ops::PowGradKernel<plat::CPUDeviceContext, ops::PowGradFunctor<double>>,
+    ops::PowGradKernel<plat::CPUDeviceContext, ops::PowGradFunctor<int>>,
+    ops::PowGradKernel<plat::CPUDeviceContext, ops::PowGradFunctor<int64_t>>);
+/* ========================================================================== */
+
+/* ==========================   exp register  ============================ */
+REGISTER_OPERATOR(
+    exp, ops::ActivationOp, ops::ExpOpMaker, ops::ActivationOpInferVarType,
+    ops::ActivationGradOpMaker<ops::ExpGradFunctor<float>::FwdDeps(),
+                               paddle::framework::OpDesc>,
+    ops::ActivationGradOpMaker<ops::ExpGradFunctor<float>::FwdDeps(),
+                               paddle::imperative::OpBase>,
+    std::conditional<ops::CanInplaceAct<ops::ExpGradFunctor<float>>(),
+                     ops::ActFwdInplaceInferer, void>::type);
+REGISTER_OPERATOR(exp_grad, ops::ActivationOpGrad,
+                  ops::ActivationGradOpInplaceInference);
+
+REGISTER_OP_CPU_KERNEL(exp,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::ExpFunctor<float>>,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::ExpFunctor<double>>,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::ExpFunctor<int>>,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::ExpFunctor<int64_t>>);
+REGISTER_OP_CPU_KERNEL(
+    exp_grad, ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                                        ops::ExpGradFunctor<float>>,
+    ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                              ops::ExpGradFunctor<double>>,
+    ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                              ops::ExpGradFunctor<int>>,
+    ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                              ops::ExpGradFunctor<int64_t>>);
+/* ========================================================================== */
+
+/* ==========================   abs register  ============================ */
+REGISTER_OPERATOR(
+    abs, ops::ActivationOp, ops::AbsOpMaker, ops::ActivationOpInferVarType,
+    ops::ActivationGradOpMaker<ops::AbsGradFunctor<float>::FwdDeps(),
+                               paddle::framework::OpDesc>,
+    ops::ActivationGradOpMaker<ops::AbsGradFunctor<float>::FwdDeps(),
+                               paddle::imperative::OpBase>,
+    std::conditional<ops::CanInplaceAct<ops::AbsGradFunctor<float>>(),
+                     ops::ActFwdInplaceInferer, void>::type);
+REGISTER_OPERATOR(abs_grad, ops::ActivationOpGrad,
+                  ops::ActivationGradOpInplaceInference);
+
+REGISTER_OP_CPU_KERNEL(abs,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::AbsFunctor<float>>,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::AbsFunctor<double>>,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::AbsFunctor<int>>,
+                       ops::ActivationKernel<paddle::platform::CPUDeviceContext,
+                                             ops::AbsFunctor<int64_t>>);
+REGISTER_OP_CPU_KERNEL(
+    abs_grad, ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                                        ops::AbsGradFunctor<float>>,
+    ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                              ops::AbsGradFunctor<double>>,
+    ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                              ops::AbsGradFunctor<int>>,
+    ops::ActivationGradKernel<paddle::platform::CPUDeviceContext,
+                              ops::AbsGradFunctor<int64_t>>);
 /* ========================================================================== */

paddle/fluid/operators/activation_op.cu

@@ -74,8 +74,25 @@ REGISTER_OP_CUDA_KERNEL(
 /* ========================================================================== */
 
 /* ===========================  square register  ============================ */
-REGISTER_ACTIVATION_CUDA_KERNEL(square, Square, SquareFunctor,
-                                SquareGradFunctor);
+REGISTER_OP_CUDA_KERNEL(
+    square,
+    ops::ActivationKernel<plat::CUDADeviceContext, ops::SquareFunctor<float>>,
+    ops::ActivationKernel<plat::CUDADeviceContext, ops::SquareFunctor<double>>,
+    ops::ActivationKernel<plat::CUDADeviceContext, ops::SquareFunctor<int>>,
+    ops::ActivationKernel<plat::CUDADeviceContext, ops::SquareFunctor<int64_t>>,
+    ops::ActivationKernel<plat::CUDADeviceContext,
+                          ops::SquareFunctor<plat::float16>>);
+REGISTER_OP_CUDA_KERNEL(
+    square_grad, ops::ActivationGradKernel<plat::CUDADeviceContext,
+                                           ops::SquareGradFunctor<float>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::SquareGradFunctor<double>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::SquareGradFunctor<int>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::SquareGradFunctor<int64_t>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::SquareGradFunctor<plat::float16>>);
 
 REGISTER_OP_CUDA_KERNEL(
     square_grad_grad,
@@ -84,7 +101,11 @@ REGISTER_OP_CUDA_KERNEL(
     ops::SquareDoubleGradKernel<paddle::platform::CUDADeviceContext,
                                 ops::SquareGradGradFunctor<double>>,
     ops::SquareDoubleGradKernel<plat::CUDADeviceContext,
-                                ops::SquareGradGradFunctor<plat::float16>>);
+                                ops::SquareGradGradFunctor<plat::float16>>,
+    ops::SquareDoubleGradKernel<paddle::platform::CUDADeviceContext,
+                                ops::SquareGradGradFunctor<int>>,
+    ops::SquareDoubleGradKernel<paddle::platform::CUDADeviceContext,
+                                ops::SquareGradGradFunctor<int64_t>>);
 /* ========================================================================== */
 
 /* ==========================   pow register  ============================ */
@@ -92,11 +113,59 @@ REGISTER_OP_CUDA_KERNEL(
 REGISTER_OP_CUDA_KERNEL(
     pow, ops::PowKernel<plat::CUDADeviceContext, ops::PowFunctor<float>>,
     ops::PowKernel<plat::CUDADeviceContext, ops::PowFunctor<double>>,
+    ops::PowKernel<plat::CUDADeviceContext, ops::PowFunctor<int>>,
+    ops::PowKernel<plat::CUDADeviceContext, ops::PowFunctor<int64_t>>,
     ops::PowKernel<plat::CUDADeviceContext, ops::PowFunctor<plat::float16>>);
 REGISTER_OP_CUDA_KERNEL(
     pow_grad,
     ops::PowGradKernel<plat::CUDADeviceContext, ops::PowGradFunctor<float>>,
     ops::PowGradKernel<plat::CUDADeviceContext, ops::PowGradFunctor<double>>,
+    ops::PowGradKernel<plat::CUDADeviceContext, ops::PowGradFunctor<int>>,
+    ops::PowGradKernel<plat::CUDADeviceContext, ops::PowGradFunctor<int64_t>>,
     ops::PowGradKernel<plat::CUDADeviceContext,
                        ops::PowGradFunctor<plat::float16>>);
 /* ========================================================================== */
+
+/* ==========================   exp register  ============================ */
+
+REGISTER_OP_CUDA_KERNEL(
+    exp, ops::ActivationKernel<plat::CUDADeviceContext, ops::ExpFunctor<float>>,
+    ops::ActivationKernel<plat::CUDADeviceContext, ops::ExpFunctor<double>>,
+    ops::ActivationKernel<plat::CUDADeviceContext, ops::ExpFunctor<int>>,
+    ops::ActivationKernel<plat::CUDADeviceContext, ops::ExpFunctor<int64_t>>,
+    ops::ActivationKernel<plat::CUDADeviceContext,
+                          ops::ExpFunctor<plat::float16>>);
+REGISTER_OP_CUDA_KERNEL(
+    exp_grad, ops::ActivationGradKernel<plat::CUDADeviceContext,
+                                        ops::ExpGradFunctor<float>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::ExpGradFunctor<double>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::ExpGradFunctor<int>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::ExpGradFunctor<int64_t>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::ExpGradFunctor<plat::float16>>);
+/* ========================================================================== */
+
+/* ==========================   exp register  ============================ */
+
+REGISTER_OP_CUDA_KERNEL(
+    abs, ops::ActivationKernel<plat::CUDADeviceContext, ops::AbsFunctor<float>>,
+    ops::ActivationKernel<plat::CUDADeviceContext, ops::AbsFunctor<double>>,
+    ops::ActivationKernel<plat::CUDADeviceContext, ops::AbsFunctor<int>>,
+    ops::ActivationKernel<plat::CUDADeviceContext, ops::AbsFunctor<int64_t>>,
+    ops::ActivationKernel<plat::CUDADeviceContext,
+                          ops::AbsFunctor<plat::float16>>);
+REGISTER_OP_CUDA_KERNEL(
+    abs_grad, ops::ActivationGradKernel<plat::CUDADeviceContext,
+                                        ops::AbsGradFunctor<float>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::AbsGradFunctor<double>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::AbsGradFunctor<int>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::AbsGradFunctor<int64_t>>,
+    ops::ActivationGradKernel<plat::CUDADeviceContext,
+                              ops::AbsGradFunctor<plat::float16>>);
+/* ========================================================================== */

paddle/fluid/operators/activation_op.h

@@ -1667,13 +1667,11 @@ class PowGradKernel
 #define FOR_EACH_ACTIVATION_OP(__macro)                                       \
   __macro(sigmoid, Sigmoid, SigmoidFunctor, SigmoidGradFunctor);              \
   __macro(logsigmoid, LogSigmoid, LogSigmoidFunctor, LogSigmoidGradFunctor);  \
-  __macro(exp, Exp, ExpFunctor, ExpGradFunctor);                              \
   __macro(gelu, Gelu, GeluFunctor, GeluGradFunctor);                          \
   __macro(tanh, Tanh, TanhFunctor, TanhGradFunctor);                          \
   __macro(atan, Atan, AtanFunctor, AtanGradFunctor);                          \
   __macro(softshrink, SoftShrink, SoftShrinkFunctor, SoftShrinkGradFunctor);  \
   __macro(rsqrt, Rsqrt, RsqrtFunctor, RsqrtGradFunctor);                      \
-  __macro(abs, Abs, AbsFunctor, AbsGradFunctor);                              \
   __macro(ceil, Ceil, CeilFunctor, ZeroGradFunctor);                          \
   __macro(floor, Floor, FloorFunctor, ZeroGradFunctor);                       \
   __macro(cos, Cos, CosFunctor, CosGradFunctor);                              \