elementwise_max_op.h 2.5 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
F
wip  
fengjiayi 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

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. */

#pragma once

17
#include "paddle/fluid/operators/elementwise_op.h"
Y
Yi Wang 已提交
18
#include "paddle/fluid/operators/elementwise_op_function.h"
F
wip  
fengjiayi 已提交
19 20 21 22 23 24 25 26 27 28 29 30 31

namespace paddle {
namespace operators {

template <typename T>
struct MaxFunctor {
  inline HOSTDEVICE T operator()(T a, T b) const { return a > b ? a : b; }
};

template <typename DeviceContext, typename T>
class ElementwiseMaxKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
C
chengduoZH 已提交
32 33 34 35 36 37 38
    using Tensor = framework::Tensor;

    auto* x = ctx.Input<Tensor>("X");
    auto* y = ctx.Input<Tensor>("Y");
    auto* z = ctx.Output<Tensor>("Out");
    z->mutable_data<T>(ctx.GetPlace());
    int axis = ctx.Attr<int>("axis");
C
chengduoZH 已提交
39 40
    ElementwiseComputeEx<MaxFunctor<T>, DeviceContext, T>(ctx, x, y, axis,
                                                          MaxFunctor<T>(), z);
F
wip  
fengjiayi 已提交
41 42 43 44
  }
};

template <typename T>
C
chengduoZH 已提交
45 46 47
struct MaxGradDx {
  HOSTDEVICE T operator()(T x, T y, T out, T dout) const {
    return dout * (x > y);
F
wip  
fengjiayi 已提交
48 49 50 51
  }
};

template <typename T>
C
chengduoZH 已提交
52 53 54
struct MaxGradDy {
  HOSTDEVICE T operator()(T x, T y, T out, T dout) const {
    return dout * (x <= y);
F
wip  
fengjiayi 已提交
55 56 57
  }
};

F
fengjiayi 已提交
58
template <typename DeviceContext, typename T>
59
class ElementwiseMaxGradKernel : public ElemwiseGradKernel<T> {
F
fengjiayi 已提交
60 61
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
62
    ElemwiseGradKernel<T>::Compute(ctx);
C
chengduoZH 已提交
63 64 65 66 67 68 69 70 71
    using Tensor = framework::Tensor;

    auto* x = ctx.Input<Tensor>("X");
    auto* y = ctx.Input<Tensor>("Y");
    auto* out = ctx.Input<Tensor>("Out");
    auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
    auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
    auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
    int axis = ctx.Attr<int>("axis");
C
chengduoZH 已提交
72 73
    ElemwiseGradCompute<DeviceContext, T, MaxGradDx<T>, MaxGradDy<T>>(
        ctx, *x, *y, *out, *dout, axis, dx, dy, MaxGradDx<T>(), MaxGradDy<T>());
F
fengjiayi 已提交
74 75
  }
};
F
wip  
fengjiayi 已提交
76 77
}  // namespace operators
}  // namespace paddle