Merge pull request #9911 from tonyyang-svail/unify_op_registry

Unify REGISTER_OP and REGISTER_OPERATOR

paddle/fluid/framework/grad_op_desc_maker.h

@@ -13,6 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #pragma once
+#include <algorithm>
 #include <string>
 #include <unordered_set>
 #include <vector>
@@ -69,8 +70,7 @@ class GradOpDescMakerBase {
                       " for input argument with a list of variables, "
                       " drop_empty_grad is not allowed because it makes"
                       " the correspondence bewteen a variable and its gradient"
-                      " ambiguous. Use REGISTER_OP_EX to register the op"
+                      " ambiguous."
-                      " or call InputGrad(?,false) in GradOpDescMaker."
                       " Op type %s",
                       fwd_op_.Type());
 

paddle/fluid/framework/op_registry.h

@@ -16,6 +16,8 @@ limitations under the License. */
 
 #include <algorithm>
 #include <atomic>
+#include <string>
+#include <tuple>
 #include <type_traits>
 #include <typeinfo>
 #include <unordered_map>
@@ -141,36 +143,6 @@ class OpKernelRegistrar : public Registrar {
     return 0;                                                          \
   }
 
-/**
- * Macro to register Operator. When the input is duplicable, you should
- * use REGISTER_OP_EX with drop_empty_grad=false instead.
- */
-#define REGISTER_OP(op_type, op_class, op_maker_class, grad_op_type, \
-                    grad_op_class)                                   \
-  REGISTER_OP_EX(op_type, op_class, op_maker_class, grad_op_type,    \
-                 grad_op_class, true)
-
-// When an argument is duplicable, we need to use this version.
-// Perhaps we can omit DropEmptyIG template parameter and
-// only have one version of REGISTER_OP.
-#define REGISTER_OP_EX(op_type, op_class, op_maker_class, grad_op_type,       \
-                       grad_op_class, drop_empty_grad)                        \
-  REGISTER_OPERATOR(grad_op_type, grad_op_class);                             \
-  class _GradOpDescMaker_##grad_op_type##_                                    \
-      : public ::paddle::framework::DefaultGradOpDescMaker<drop_empty_grad> { \
-    using ::paddle::framework::DefaultGradOpDescMaker<                        \
-        drop_empty_grad>::DefaultGradOpDescMaker;                             \
-                                                                              \
-   protected:                                                                 \
-    virtual std::string GradOpType() const { return #grad_op_type; }          \
-  };                                                                          \
-  REGISTER_OPERATOR(op_type, op_class, _GradOpDescMaker_##grad_op_type##_,    \
-                    op_maker_class);
-
-#define REGISTER_OP_WITH_KERNEL(op_type, ...)                         \
-  REGISTER_OPERATOR(op_type, ::paddle::framework::OperatorWithKernel, \
-                    ##__VA_ARGS__)
-
 #define REGISTER_OP_WITHOUT_GRADIENT(op_type, op_class, op_maker_class) \
   REGISTER_OPERATOR(op_type, op_class, op_maker_class)
 

paddle/fluid/operators/CMakeLists.txt

@@ -110,12 +110,12 @@ function(op_library TARGET)
     # Note that it's enough to just adding one operator to pybind in a *_op.cc file.
     # And for detail pybind information, please see generated paddle/pybind/pybind.h.
     file(READ ${TARGET}.cc TARGET_CONTENT)
-    string(REGEX MATCH "REGISTER_OP\\(.*REGISTER_OP\\(" multi_register "${TARGET_CONTENT}")
+    string(REGEX MATCH "REGISTER_OPERATOR\\(.*REGISTER_OPERATOR\\(" multi_register "${TARGET_CONTENT}")
-    string(REGEX MATCH "REGISTER_OP\\([a-z0-9_]*," one_register "${multi_register}")
+    string(REGEX MATCH "REGISTER_OPERATOR\\([a-z0-9_]*," one_register "${multi_register}")
     if (one_register STREQUAL "")
         string(REPLACE "_op" "" TARGET "${TARGET}")
     else ()
-        string(REPLACE "REGISTER_OP(" "" TARGET "${one_register}")
+        string(REPLACE "REGISTER_OPERATOR(" "" TARGET "${one_register}")
         string(REPLACE "," "" TARGET "${TARGET}")
     endif()
 

paddle/fluid/operators/activation_op.cc

@@ -558,95 +558,126 @@ $$out = \frac{x}{1 + e^{- \beta x}}$$
 
 namespace ops = paddle::operators;
 
-REGISTER_OP(sigmoid, ops::ActivationOp, ops::SigmoidOpMaker, sigmoid_grad,
+REGISTER_OPERATOR(sigmoid, ops::ActivationOp, ops::SigmoidOpMaker,
-            ops::ActivationOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(sigmoid_grad, ops::ActivationOpGrad)
 
-REGISTER_OP(logsigmoid, ops::ActivationOp, ops::LogSigmoidOpMaker,
+REGISTER_OPERATOR(logsigmoid, ops::ActivationOp, ops::LogSigmoidOpMaker,
-            logsigmoid_grad, ops::ActivationOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(logsigmoid_grad, ops::ActivationOpGrad)
 
-REGISTER_OP(exp, ops::ActivationOp, ops::ExpOpMaker, exp_grad,
+REGISTER_OPERATOR(exp, ops::ActivationOp, ops::ExpOpMaker,
-            ops::ActivationOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(exp_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(relu, ops::ActivationWithMKLDNNOp, ops::ReluOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(relu_grad, ops::ActivationWithMKLDNNOpGrad)
+
+REGISTER_OPERATOR(tanh, ops::ActivationWithMKLDNNOp, ops::TanhOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(tanh_grad, ops::ActivationWithMKLDNNOpGrad)
+
+REGISTER_OPERATOR(tanh_shrink, ops::ActivationOp, ops::TanhShrinkOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(tanh_shrink_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(softshrink, ops::ActivationOp, ops::SoftShrinkOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(softshrink_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(sqrt, ops::ActivationWithMKLDNNOp, ops::SqrtOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(sqrt_grad, ops::ActivationWithMKLDNNOpGrad)
+
+REGISTER_OPERATOR(abs, ops::ActivationWithMKLDNNOp, ops::AbsOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(abs_grad, ops::ActivationWithMKLDNNOpGrad)
+
+REGISTER_OPERATOR(ceil, ops::ActivationOp, ops::CeilOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(ceil_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(floor, ops::ActivationOp, ops::FloorOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(floor_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(cos, ops::ActivationOp, ops::CosOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(cos_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(sin, ops::ActivationOp, ops::SinOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(sin_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(round, ops::ActivationOp, ops::RoundOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(round_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(reciprocal, ops::ActivationOp, ops::ReciprocalOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(reciprocal_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(log, ops::ActivationOp, ops::LogOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(log_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(square, ops::ActivationOp, ops::SquareOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(square_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(softplus, ops::ActivationOp, ops::SoftplusOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(softplus_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(softsign, ops::ActivationOp, ops::SoftsignOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(softsign_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(brelu, ops::ActivationOp, ops::BReluOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(brelu_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(leaky_relu, ops::ActivationOp, ops::LeakyReluOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(leaky_relu_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(soft_relu, ops::ActivationOp, ops::SoftReluOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(soft_relu_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(elu, ops::ActivationOp, ops::ELUOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(elu_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(relu6, ops::ActivationOp, ops::Relu6OpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(relu6_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(pow, ops::ActivationOp, ops::PowOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(pow_grad, ops::ActivationOpGrad)
+
+REGISTER_OPERATOR(stanh, ops::ActivationOp, ops::STanhOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(stanh_grad, ops::ActivationOpGrad)
 
-REGISTER_OP(relu, ops::ActivationWithMKLDNNOp, ops::ReluOpMaker, relu_grad,
+REGISTER_OPERATOR(hard_shrink, ops::ActivationOp, ops::HardShrinkOpMaker,
-            ops::ActivationWithMKLDNNOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(hard_shrink_grad, ops::ActivationOpGrad)
 
-REGISTER_OP(tanh, ops::ActivationWithMKLDNNOp, ops::TanhOpMaker, tanh_grad,
+REGISTER_OPERATOR(thresholded_relu, ops::ActivationOp,
-            ops::ActivationWithMKLDNNOpGrad);
+                  ops::ThresholdedReluOpMaker,
-
+                  paddle::framework::DefaultGradOpDescMaker<true>)
-REGISTER_OP(tanh_shrink, ops::ActivationOp, ops::TanhShrinkOpMaker,
+REGISTER_OPERATOR(thresholded_relu_grad, ops::ActivationOpGrad)
-            tanh_shrink_grad, ops::ActivationOpGrad);
+
-
+REGISTER_OPERATOR(hard_sigmoid, ops::ActivationOp, ops::HardSigmoidOpMaker,
-REGISTER_OP(softshrink, ops::ActivationOp, ops::SoftShrinkOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
-            softshrink_grad, ops::ActivationOpGrad);
+REGISTER_OPERATOR(hard_sigmoid_grad, ops::ActivationOpGrad)
-
+
-REGISTER_OP(sqrt, ops::ActivationWithMKLDNNOp, ops::SqrtOpMaker, sqrt_grad,
+REGISTER_OPERATOR(swish, ops::ActivationOp, ops::SwishOpMaker,
-            ops::ActivationWithMKLDNNOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
-
+REGISTER_OPERATOR(swish_grad, ops::ActivationOpGrad)
-REGISTER_OP(abs, ops::ActivationWithMKLDNNOp, ops::AbsOpMaker, abs_grad,
-            ops::ActivationWithMKLDNNOpGrad);
-
-REGISTER_OP(ceil, ops::ActivationOp, ops::CeilOpMaker, ceil_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(floor, ops::ActivationOp, ops::FloorOpMaker, floor_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(cos, ops::ActivationOp, ops::CosOpMaker, cos_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(sin, ops::ActivationOp, ops::SinOpMaker, sin_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(round, ops::ActivationOp, ops::RoundOpMaker, round_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(reciprocal, ops::ActivationOp, ops::ReciprocalOpMaker,
-            reciprocal_grad, ops::ActivationOpGrad);
-
-REGISTER_OP(log, ops::ActivationOp, ops::LogOpMaker, log_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(square, ops::ActivationOp, ops::SquareOpMaker, square_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(softplus, ops::ActivationOp, ops::SoftplusOpMaker, softplus_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(softsign, ops::ActivationOp, ops::SoftsignOpMaker, softsign_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(brelu, ops::ActivationOp, ops::BReluOpMaker, brelu_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(leaky_relu, ops::ActivationOp, ops::LeakyReluOpMaker,
-            leaky_relu_grad, ops::ActivationOpGrad);
-
-REGISTER_OP(soft_relu, ops::ActivationOp, ops::SoftReluOpMaker, soft_relu_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(elu, ops::ActivationOp, ops::ELUOpMaker, elu_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(relu6, ops::ActivationOp, ops::Relu6OpMaker, relu6_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(pow, ops::ActivationOp, ops::PowOpMaker, pow_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(stanh, ops::ActivationOp, ops::STanhOpMaker, stanh_grad,
-            ops::ActivationOpGrad);
-
-REGISTER_OP(hard_shrink, ops::ActivationOp, ops::HardShrinkOpMaker,
-            hard_shrink_grad, ops::ActivationOpGrad);
-
-REGISTER_OP(thresholded_relu, ops::ActivationOp, ops::ThresholdedReluOpMaker,
-            thresholded_relu_grad, ops::ActivationOpGrad);
-
-REGISTER_OP(hard_sigmoid, ops::ActivationOp, ops::HardSigmoidOpMaker,
-            hard_sigmoid_grad, ops::ActivationOpGrad);
-
-REGISTER_OP(swish, ops::ActivationOp, ops::SwishOpMaker, swish_grad,
-            ops::ActivationOpGrad);
 
 #define REGISTER_ACTIVATION_CPU_KERNEL(act_type, functor, grad_functor)   \
   REGISTER_OP_CPU_KERNEL(                                                 \

paddle/fluid/operators/bilinear_tensor_product_op.cc

@@ -153,9 +153,11 @@ class BilinearTensorProductOpGrad : public framework::OperatorWithKernel {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(bilinear_tensor_product, ops::BilinearTensorProductOp,
+REGISTER_OPERATOR(bilinear_tensor_product, ops::BilinearTensorProductOp,
-            ops::BilinearTensorProductOpMaker, bilinear_tensor_product_grad,
+                  ops::BilinearTensorProductOpMaker,
-            ops::BilinearTensorProductOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(bilinear_tensor_product_grad,
+                  ops::BilinearTensorProductOpGrad)
 REGISTER_OP_CPU_KERNEL(
     bilinear_tensor_product,
     ops::BilinearTensorProductKernel<paddle::platform::CPUDeviceContext, float>,

paddle/fluid/operators/clip_op.cc

@@ -81,8 +81,9 @@ class ClipOpGrad : public framework::OperatorWithKernel {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(clip, ops::ClipOp, ops::ClipOpMaker<float>, clip_grad,
+REGISTER_OPERATOR(clip, ops::ClipOp, ops::ClipOpMaker<float>,
-            ops::ClipOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(clip_grad, ops::ClipOpGrad)
 REGISTER_OP_CPU_KERNEL(
     clip, ops::ClipKernel<paddle::platform::CPUDeviceContext, float>);
 REGISTER_OP_CPU_KERNEL(

paddle/fluid/operators/concat_op.cc

@@ -103,8 +103,10 @@ class ConcatOpGrad : public framework::OperatorWithKernel {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP_EX(concat, ops::ConcatOp, ops::ConcatOpMaker, concat_grad,
+REGISTER_OPERATOR(concat, ops::ConcatOp, ops::ConcatOpMaker,
-               ops::ConcatOpGrad, false)
+                  paddle::framework::DefaultGradOpDescMaker<
+                      false> /* set false to disable empty grad */)
+REGISTER_OPERATOR(concat_grad, ops::ConcatOpGrad)
 REGISTER_OP_CPU_KERNEL(
     concat, ops::ConcatKernel<paddle::platform::CPUDeviceContext, float>)
 REGISTER_OP_CPU_KERNEL(

paddle/fluid/operators/conv_op.cc

@@ -335,14 +335,17 @@ framework::OpKernelType ConvOpGrad::GetExpectedKernelType(
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(conv2d, ops::ConvOp, ops::Conv2DOpMaker, conv2d_grad,
+REGISTER_OPERATOR(conv2d, ops::ConvOp, ops::Conv2DOpMaker,
-            ops::ConvOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(conv2d_grad, ops::ConvOpGrad)
 
 // depthwise convolution op
-REGISTER_OP(depthwise_conv2d, ops::ConvOp, ops::Conv2DOpMaker,
+REGISTER_OPERATOR(depthwise_conv2d, ops::ConvOp, ops::Conv2DOpMaker,
-            depthwise_conv2d_grad, ops::ConvOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
-REGISTER_OP(conv3d, ops::ConvOp, ops::Conv3DOpMaker, conv3d_grad,
+REGISTER_OPERATOR(depthwise_conv2d_grad, ops::ConvOpGrad)
-            ops::ConvOpGrad);
+REGISTER_OPERATOR(conv3d, ops::ConvOp, ops::Conv3DOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(conv3d_grad, ops::ConvOpGrad)
 
 // depthwise conv kernel
 // TODO(xingzhaolong): neon kernel for mobile

paddle/fluid/operators/conv_shift_op.cc

@@ -193,8 +193,9 @@ class ConvShiftGradKernel<platform::CPUPlace, T>
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(conv_shift, ops::ConvShiftOp, ops::ConvShiftOpMaker,
+REGISTER_OPERATOR(conv_shift, ops::ConvShiftOp, ops::ConvShiftOpMaker,
-            conv_shift_grad, ops::ConvShiftGradOp);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(conv_shift_grad, ops::ConvShiftGradOp)
 REGISTER_OP_CPU_KERNEL(conv_shift,
                        ops::ConvShiftKernel<paddle::platform::CPUPlace, float>);
 REGISTER_OP_CPU_KERNEL(

paddle/fluid/operators/conv_transpose_op.cc

@@ -298,8 +298,10 @@ framework::OpKernelType ConvTransposeOpGrad::GetExpectedKernelType(
 
 namespace ops = paddle::operators;
 
-REGISTER_OP(conv2d_transpose, ops::ConvTransposeOp, ops::Conv2DTransposeOpMaker,
+REGISTER_OPERATOR(conv2d_transpose, ops::ConvTransposeOp,
-            conv2d_transpose_grad, ops::ConvTransposeOpGrad);
+                  ops::Conv2DTransposeOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(conv2d_transpose_grad, ops::ConvTransposeOpGrad)
 
 REGISTER_OP_CPU_KERNEL(
     conv2d_transpose,
@@ -311,8 +313,10 @@ REGISTER_OP_CPU_KERNEL(
     ops::GemmConvTransposeGradKernel<paddle::platform::CPUDeviceContext,
                                      double>);
 
-REGISTER_OP(conv3d_transpose, ops::ConvTransposeOp, ops::Conv3DTransposeOpMaker,
+REGISTER_OPERATOR(conv3d_transpose, ops::ConvTransposeOp,
-            conv3d_transpose_grad, ops::ConvTransposeOpGrad);
+                  ops::Conv3DTransposeOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(conv3d_transpose_grad, ops::ConvTransposeOpGrad)
 
 REGISTER_OP_CPU_KERNEL(
     conv3d_transpose,

paddle/fluid/operators/cos_sim_op.cc

@@ -153,8 +153,9 @@ class CosSimOpGrad : public framework::OperatorWithKernel {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(cos_sim, ops::CosSimOp, ops::CosSimOpMaker, cos_sim_grad,
+REGISTER_OPERATOR(cos_sim, ops::CosSimOp, ops::CosSimOpMaker,
-            ops::CosSimOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(cos_sim_grad, ops::CosSimOpGrad)
 REGISTER_OP_CPU_KERNEL(
     cos_sim, ops::CosSimKernel<paddle::platform::CPUDeviceContext, float>);
 REGISTER_OP_CPU_KERNEL(

paddle/fluid/operators/crop_op.cc

@@ -153,7 +153,9 @@ class CropOpGrad : public framework::OperatorWithKernel {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(crop, ops::CropOp, ops::CropOpMaker, crop_grad, ops::CropOpGrad);
+REGISTER_OPERATOR(crop, ops::CropOp, ops::CropOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>);
+REGISTER_OPERATOR(crop_grad, ops::CropOpGrad);
 REGISTER_OP_CPU_KERNEL(crop, ops::CropKernel<float>);
 REGISTER_OP_CPU_KERNEL(
     crop_grad, ops::CropGradKernel<paddle::platform::CPUDeviceContext, float>);

paddle/fluid/operators/cross_entropy_op.cc

@@ -164,8 +164,9 @@ or not. But the output only shares the LoD information with input X.
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(cross_entropy, ops::CrossEntropyOp, ops::CrossEntropyOpMaker,
+REGISTER_OPERATOR(cross_entropy, ops::CrossEntropyOp, ops::CrossEntropyOpMaker,
-            cross_entropy_grad, ops::CrossEntropyGradientOp);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(cross_entropy_grad, ops::CrossEntropyGradientOp)
 REGISTER_OP_CPU_KERNEL(cross_entropy, ops::CrossEntropyOpKernel<float>,
                        ops::CrossEntropyOpKernel<double>);
 REGISTER_OP_CPU_KERNEL(cross_entropy_grad,

paddle/fluid/operators/dropout_op.cc

@@ -101,8 +101,9 @@ class DropoutOpGrad : public framework::OperatorWithKernel {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(dropout, ops::DropoutOp, ops::DropoutOpMaker, dropout_grad,
+REGISTER_OPERATOR(dropout, ops::DropoutOp, ops::DropoutOpMaker,
-            ops::DropoutOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(dropout_grad, ops::DropoutOpGrad)
 REGISTER_OP_CPU_KERNEL(
     dropout, ops::CPUDropoutKernel<paddle::platform::CPUDeviceContext, float>);
 REGISTER_OP_CPU_KERNEL(

paddle/fluid/operators/elementwise_div_op.cc

@@ -30,8 +30,10 @@ class ElementwiseDivOpMaker : public ElementwiseOpMaker {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(elementwise_div, ops::ElementwiseOp, ops::ElementwiseDivOpMaker,
+REGISTER_OPERATOR(elementwise_div, ops::ElementwiseOp,
-            elementwise_div_grad, ops::ElementwiseOpGrad);
+                  ops::ElementwiseDivOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(elementwise_div_grad, ops::ElementwiseOpGrad)
 REGISTER_OP_CPU_KERNEL(
     elementwise_div,
     ops::ElementwiseDivKernel<paddle::platform::CPUDeviceContext, float>,

paddle/fluid/operators/elementwise_max_op.cc

@@ -29,8 +29,10 @@ class ElementwiseMaxOpMaker : public ElementwiseOpMaker {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(elementwise_max, ops::ElementwiseOp, ops::ElementwiseMaxOpMaker,
+REGISTER_OPERATOR(elementwise_max, ops::ElementwiseOp,
-            elementwise_max_grad, ops::ElementwiseOpGrad);
+                  ops::ElementwiseMaxOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(elementwise_max_grad, ops::ElementwiseOpGrad)
 REGISTER_OP_CPU_KERNEL(
     elementwise_max,
     ops::ElementwiseMaxKernel<paddle::platform::CPUDeviceContext, float>,

paddle/fluid/operators/elementwise_min_op.cc

@@ -29,8 +29,10 @@ class ElementwiseMinOpMaker : public ElementwiseOpMaker {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(elementwise_min, ops::ElementwiseOp, ops::ElementwiseMinOpMaker,
+REGISTER_OPERATOR(elementwise_min, ops::ElementwiseOp,
-            elementwise_min_grad, ops::ElementwiseOpGrad);
+                  ops::ElementwiseMinOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(elementwise_min_grad, ops::ElementwiseOpGrad)
 REGISTER_OP_CPU_KERNEL(
     elementwise_min,
     ops::ElementwiseMinKernel<paddle::platform::CPUDeviceContext, float>,

paddle/fluid/operators/elementwise_mul_op.cc

@@ -31,8 +31,10 @@ class ElementwiseMulOpMaker : public ElementwiseOpMaker {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(elementwise_mul, ops::ElementwiseOp, ops::ElementwiseMulOpMaker,
+REGISTER_OPERATOR(elementwise_mul, ops::ElementwiseOp,
-            elementwise_mul_grad, ops::ElementwiseOpGrad);
+                  ops::ElementwiseMulOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(elementwise_mul_grad, ops::ElementwiseOpGrad)
 REGISTER_OP_CPU_KERNEL(
     elementwise_mul,
     ops::ElementwiseMulKernel<paddle::platform::CPUDeviceContext, float>,

paddle/fluid/operators/elementwise_sub_op.cc

@@ -29,8 +29,10 @@ class ElementwiseSubOpMaker : public ElementwiseOpMaker {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(elementwise_sub, ops::ElementwiseOp, ops::ElementwiseSubOpMaker,
+REGISTER_OPERATOR(elementwise_sub, ops::ElementwiseOp,
-            elementwise_sub_grad, ops::ElementwiseOpGrad);
+                  ops::ElementwiseSubOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(elementwise_sub_grad, ops::ElementwiseOpGrad)
 REGISTER_OP_CPU_KERNEL(
     elementwise_sub,
     ops::ElementwiseSubKernel<paddle::platform::CPUDeviceContext, float>,

paddle/fluid/operators/expand_op.cc

@@ -14,6 +14,8 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/expand_op.h"
 
+#include <vector>
+
 namespace paddle {
 namespace operators {
 
@@ -128,8 +130,9 @@ class ExpandGradOp : public framework::OperatorWithKernel {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(expand, ops::ExpandOp, ops::ExpandOpMaker, expand_grad,
+REGISTER_OPERATOR(expand, ops::ExpandOp, ops::ExpandOpMaker,
-            ops::ExpandGradOp);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(expand_grad, ops::ExpandGradOp)
 REGISTER_OP_CPU_KERNEL(
     expand, ops::ExpandKernel<paddle::platform::CPUDeviceContext, float>);
 REGISTER_OP_CPU_KERNEL(

paddle/fluid/operators/fc_op.cc

@@ -98,5 +98,6 @@ FCOpMaker::FCOpMaker(OpProto* proto, OpAttrChecker* op_checker)
 }  // namespace operators
 }  // namespace paddle
 
-REGISTER_OP(fc, paddle::operators::FCOp, paddle::operators::FCOpMaker, fc_grad,
+REGISTER_OPERATOR(fc, paddle::operators::FCOp, paddle::operators::FCOpMaker,
-            paddle::operators::FCOpGrad);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(fc_grad, paddle::operators::FCOpGrad)

paddle/fluid/operators/gather_op.cc

@@ -100,7 +100,8 @@ Out = [[3, 4],
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(gather, ops::GatherOp, ops::GatherOpMaker, gather_grad,
+REGISTER_OPERATOR(gather, ops::GatherOp, ops::GatherOpMaker,
-            ops::GatherGradOp);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(gather_grad, ops::GatherGradOp)
 REGISTER_OP_CPU_KERNEL(gather, ops::GatherOpKernel<float>);
 REGISTER_OP_CPU_KERNEL(gather_grad, ops::GatherGradientOpKernel<float>);

paddle/fluid/operators/gru_op.cc

@@ -216,7 +216,9 @@ class GRUGradOp : public framework::OperatorWithKernel {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(gru, ops::GRUOp, ops::GRUOpMaker, gru_grad, ops::GRUGradOp);
+REGISTER_OPERATOR(gru, ops::GRUOp, ops::GRUOpMaker,
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(gru_grad, ops::GRUGradOp)
 REGISTER_OP_CPU_KERNEL(
     gru, ops::GRUKernel<paddle::platform::CPUDeviceContext, float>,
     ops::GRUKernel<paddle::platform::CPUDeviceContext, double>);

paddle/fluid/operators/gru_unit_op.cc

@@ -198,8 +198,9 @@ class GRUUnitGradOp : public framework::OperatorWithKernel {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(gru_unit, ops::GRUUnitOp, ops::GRUUnitOpMaker, gru_unit_grad,
+REGISTER_OPERATOR(gru_unit, ops::GRUUnitOp, ops::GRUUnitOpMaker,
-            ops::GRUUnitGradOp);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(gru_unit_grad, ops::GRUUnitGradOp)
 REGISTER_OP_CPU_KERNEL(
     gru_unit, ops::GRUUnitKernel<paddle::platform::CPUDeviceContext, float>,
     ops::GRUUnitKernel<paddle::platform::CPUDeviceContext, double>);

paddle/fluid/operators/hinge_loss_op.cc

@@ -103,8 +103,9 @@ class HingeLossGradOp : public framework::OperatorWithKernel {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP(hinge_loss, ops::HingeLossOp, ops::HingeLossOpMaker<float>,
+REGISTER_OPERATOR(hinge_loss, ops::HingeLossOp, ops::HingeLossOpMaker<float>,
-            hinge_loss_grad, ops::HingeLossGradOp);
+                  paddle::framework::DefaultGradOpDescMaker<true>)
+REGISTER_OPERATOR(hinge_loss_grad, ops::HingeLossGradOp)
 REGISTER_OP_CPU_KERNEL(
     hinge_loss,
     ops::HingeLossKernel<paddle::platform::CPUDeviceContext, float>);