prelu_op

paddle/operators/prelu_op.cc

@@ -33,20 +33,20 @@ class PreluOp : public framework::OperatorWithKernel {
   }
 };
 
-template <typename AttrType>
+// template <typename AttrType>
 class PreluOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
   PreluOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
-    AddInput("X", "The input tensor of prelu operator.").NotInGradient();
+    AddInput("X", "The input tensor of prelu operator.");
-    AddOutput("Out", "The output tensor of prelu operator.").NotInGradient();
+    AddOutput("Out", "The output tensor of prelu operator.");
     AddComment(R"DOC(Prelu operator
 
 The equation is:
 f(x) = alpha * x , for x < 0
 f(x) = x         , for x >= 0
 )DOC");
-    AddAttr<AttrType>("alpha", "The scaling factor alpha of prelu.")
+    AddAttr<float>("alpha", "The scaling factor alpha of prelu.")
         .SetDefault(0.0);
   }
 };
@@ -58,8 +58,10 @@ class PreluGradOp : public framework::OperatorWithKernel {
 
  protected:
   void InferShape(const framework::InferShapeContext &ctx) const override {
-    auto X_grad = ctx.Output<framework::LoDTensor>(framework::GradVarName("X"));
+    PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("X"), "Input(X) must not be null.");
-    auto X = ctx.Input<Tensor>("X");
+    auto *X_grad =
+        ctx.Output<framework::LoDTensor>(framework::GradVarName("X"));
+    auto *X = ctx.Input<framework::Tensor>("X");
 
     X_grad->Resize(X->dims());
   }
@@ -70,7 +72,7 @@ class PreluGradOp : public framework::OperatorWithKernel {
 
 namespace ops = paddle::operators;
 
-REGISTER_OP(prelu, ops::PreluOp, ops::PreluOpMaker<float>, prelu_grad,
+REGISTER_OP(prelu, ops::PreluOp, ops::PreluOpMaker, prelu_grad,
             ops::PreluGradOp);
 REGISTER_OP_CPU_KERNEL(prelu,
                        ops::PreluKernel<paddle::platform::CPUPlace, float>);

paddle/operators/prelu_op.cu

@@ -1,21 +0,0 @@
-/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-   http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License. */
-
-#include "paddle/operators/prelu_op.h"
-
-REGISTER_OP_GPU_KERNEL(
-    prelu, paddle::operators::PreluKernel<paddle::platform::GPUPlace, float>);
-REGISTER_OP_GPU_KERNEL(
-    prelu_grad,
-    paddle::operators::PreluGradKernel<paddle::platform::GPUPlace, float>);

paddle/operators/prelu_op.h

@@ -24,7 +24,7 @@ template <typename T, int MajorType = Eigen::RowMajor,
           typename IndexType = Eigen::DenseIndex>
 using EigenVector = framework::EigenVector<T, MajorType, IndexType>;
 
-template <typename Place, typename T, typename AttrType = T>
+template <typename Place, typename T>
 class PreluKernel : public framework::OpKernel {
  public:
   void Compute(const framework::ExecutionContext& context) const override {
@@ -33,30 +33,29 @@ class PreluKernel : public framework::OpKernel {
 
     Out->mutable_data<T>(context.GetPlace());
 
-    auto alpha = static_cast<T>(context.Attr<AttrType>("alpha"));
+    auto alpha = static_cast<T>(context.Attr<float>("alpha"));
 
     auto X_vec = EigenVector<T>::Flatten(*X);
     auto Out_vec = EigenVector<T>::Flatten(*Out);
 
-    auto place = context.GetEigenDevice<Place>();
+    // auto place = context.GetEigenDevice<Place>();
-
+    // Out_vec.device(place)
-    Out_vec.device(place) = X_vec.cwiseMax(0.f) + X_vec.cwiseMin(0.f) * alpha;
+    Out_vec = X_vec.cwiseMax(0.f) + X_vec.cwiseMin(0.f) * alpha;
   }
 };
 
-template <typename Place, typename T, typename AttrType = T>
+template <typename Place, typename T>
 class PreluGradKernel : public framework::OpKernel {
  public:
   void Compute(const framework::ExecutionContext& context) const override {
     auto* dX = context.Output<Tensor>(framework::GradVarName("X"));
     auto* dO = context.Input<Tensor>(framework::GradVarName("Out"));
 
-    auto* Out = context.Output<Tensor>("Out");
+    auto* Out = context.Input<Tensor>("Out");
 
-    auto alpha = static_cast<T>(context.Attr<AttrType>("alpha"));
+    auto alpha = static_cast<T>(context.Attr<float>("alpha"));
 
     dX->mutable_data<T>(context.GetPlace());
-
     for (int i = 0; i < dX->numel(); ++i) {
       if (Out->data<T>()[i] > 0) {
         dX->data<T>()[i] = dO->data<T>()[i];

python/paddle/v2/framework/tests/test_prelu_op.py

@@ -6,11 +6,12 @@ from op_test import OpTest
 class PreluTest(OpTest):
     def setUp(self):
         self.op_type = "prelu"
-        self.inputs = {'X': np.random.random((10, 10)).astype("float32")}
+        self.inputs = {'X': np.random.normal(size=(3, 5)).astype("float32")}
         self.attrs = {'alpha': 0.1}
         out_np = np.maximum(self.inputs['X'], 0.)
         out_np = out_np + np.minimum(self.inputs['X'], 0.) * self.attrs['alpha']
-        self.outputs = {'Out': self.inputs['X'] * self.attrs['scale']}
+        assert out_np is not self.inputs['X']
+        self.outputs = {'Out': out_np}
 
     def test_check_output(self):
         self.check_output()