clip_op.h 6.6 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
W
wanghaoshuang 已提交
2

L
Luo Tao 已提交
3 4 5
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
W
wanghaoshuang 已提交
6

L
Luo Tao 已提交
7
    http://www.apache.org/licenses/LICENSE-2.0
W
wanghaoshuang 已提交
8

L
Luo Tao 已提交
9 10 11 12 13
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. */
W
wanghaoshuang 已提交
14 15 16

#pragma once

Y
Yi Wang 已提交
17 18
#include "paddle/fluid/framework/eigen.h"
#include "paddle/fluid/framework/op_registry.h"
S
sneaxiy 已提交
19
#include "paddle/fluid/operators/math/selected_rows_functor.h"
Y
Yi Wang 已提交
20
#include "paddle/fluid/platform/transform.h"
W
wanghaoshuang 已提交
21 22 23 24

namespace paddle {
namespace operators {

W
wanghaoshuang 已提交
25 26
using framework::Tensor;
using platform::Transform;
W
wanghaoshuang 已提交
27

28
#if defined(__NVCC__) || defined(__HIPCC__)
29 30 31 32 33 34 35 36 37 38
template <typename T, typename UnaryOperation>
__global__ void ClipCudaKernel(const T* input, T* out, int num,
                               UnaryOperation op) {
  int idx = threadIdx.x + blockDim.x * blockIdx.x;
  if (idx < num) {
    out[idx] = op(input[idx]);
  }
}
#endif

W
wanghaoshuang 已提交
39 40 41 42 43
template <typename T>
class ClipFunctor {
 public:
  explicit ClipFunctor(const T min, const T max) : min_(min), max_(max) {}
  HOSTDEVICE T operator()(const T& x) const {
44
    return x < min_ ? min_ : x > max_ ? max_ : x;
W
wanghaoshuang 已提交
45 46 47 48 49 50 51 52 53 54 55 56
  }

 private:
  T min_;
  T max_;
};

template <typename T>
class ClipGradFunctor {
 public:
  explicit ClipGradFunctor(const T min, const T max) : min_(min), max_(max) {}
  HOSTDEVICE T operator()(const T& x, const T& y) const {
Z
zhangbo9674 已提交
57
    return (y > min_ && y < max_) ? x : static_cast<T>(0);
W
wanghaoshuang 已提交
58
  }
W
wanghaoshuang 已提交
59

W
wanghaoshuang 已提交
60 61 62 63
 private:
  T min_;
  T max_;
};
64

Q
QI JUN 已提交
65
template <typename DeviceContext, typename T>
Y
Yu Yang 已提交
66
class ClipKernel : public framework::OpKernel<T> {
W
wanghaoshuang 已提交
67 68
 public:
  void Compute(const framework::ExecutionContext& context) const override {
Y
Yang Zhang 已提交
69
    auto max = static_cast<T>(context.Attr<float>("max"));
70 71 72 73 74 75 76 77 78 79
    Tensor max_cpu;
    if (context.HasInput("Max")) {
      auto* max_t = context.Input<Tensor>("Max");
      auto* max_data = max_t->data<T>();
      if (platform::is_gpu_place(max_t->place())) {
        TensorCopySync(*max_t, platform::CPUPlace(), &max_cpu);
        max_data = max_cpu.data<T>();
      }
      max = max_data[0];
    }
Y
Yang Zhang 已提交
80
    max = static_cast<T>(max);
81

Z
zhangbo9674 已提交
82
    auto min = static_cast<T>(context.Attr<float>("min"));
83 84 85 86 87 88 89 90 91 92
    Tensor min_cpu;
    if (context.HasInput("Min")) {
      auto* min_t = context.Input<Tensor>("Min");
      auto* min_data = min_t->data<T>();
      if (platform::is_gpu_place(min_t->place())) {
        TensorCopySync(*min_t, platform::CPUPlace(), &min_cpu);
        min_data = min_cpu.data<T>();
      }
      min = min_data[0];
    }
Y
Yang Zhang 已提交
93 94 95 96 97 98

    PADDLE_ENFORCE_LE(min, max,
                      platform::errors::InvalidArgument(
                          "max should be greater than or equal to min. "
                          "But received min = %f, max = %f",
                          min, max));
99

S
sneaxiy 已提交
100 101 102 103 104 105 106
    auto* x_var = context.InputVar("X");
    if (x_var->IsType<framework::LoDTensor>()) {
      auto* x = context.Input<framework::LoDTensor>("X");
      auto* out = context.Output<framework::LoDTensor>("Out");
      T* out_data = out->mutable_data<T>(context.GetPlace());
      const T* x_data = x->data<T>();
      int64_t numel = x->numel();
107
      if (platform::is_gpu_place(context.GetPlace())) {
108
#if defined(__NVCC__) || defined(__HIPCC__)
109 110 111 112 113 114 115 116 117 118 119 120
        int threads = 256;
        int blocks = (numel + threads - 1) / threads;
        ClipCudaKernel<T, ClipFunctor<T>><<<
            blocks, threads, 0,
            context.template device_context<platform::CUDADeviceContext>()
                .stream()>>>(x_data, out_data, numel, ClipFunctor<T>(min, max));
#endif
      } else {
        Transform<DeviceContext> trans;
        trans(context.template device_context<DeviceContext>(), x_data,
              x_data + numel, out_data, ClipFunctor<T>(min, max));
      }
S
sneaxiy 已提交
121 122 123
    } else if (x_var->IsType<framework::SelectedRows>()) {
      auto* x = context.Input<framework::SelectedRows>("X");
      auto* out = context.Output<framework::SelectedRows>("Out");
124 125 126
      PADDLE_ENFORCE_NE(x, out, platform::errors::InvalidArgument(
                                    "Inplace clip is not allowed "
                                    "when x is SelectedRows"));
S
sneaxiy 已提交
127 128 129 130 131 132 133 134 135
      math::scatter::MergeAdd<DeviceContext, T> merge_func;
      merge_func(context.template device_context<DeviceContext>(), *x, out);
      auto* out_tensor = out->mutable_value();
      auto* out_data = out_tensor->data<T>();
      int64_t numel = out_tensor->numel();
      Transform<DeviceContext> trans;
      trans(context.template device_context<DeviceContext>(), out_data,
            out_data + numel, out_data, ClipFunctor<T>(min, max));
    } else {
136 137
      PADDLE_THROW(platform::errors::Unavailable(
          "ClipOp only supports LoDTensor and SelectedRows."));
S
sneaxiy 已提交
138
    }
W
wanghaoshuang 已提交
139 140 141
  }
};

Q
QI JUN 已提交
142
template <typename DeviceContext, typename T>
Y
Yu Yang 已提交
143
class ClipGradKernel : public framework::OpKernel<T> {
W
wanghaoshuang 已提交
144 145
 public:
  void Compute(const framework::ExecutionContext& context) const override {
Y
Yang Zhang 已提交
146
    auto max = static_cast<T>(context.Attr<float>("max"));
147 148 149 150 151 152 153 154 155 156
    Tensor max_cpu;
    if (context.HasInput("Max")) {
      auto* max_t = context.Input<Tensor>("Max");
      auto* max_data = max_t->data<T>();
      if (platform::is_gpu_place(max_t->place())) {
        TensorCopySync(*max_t, platform::CPUPlace(), &max_cpu);
        max_data = max_cpu.data<T>();
      }
      max = max_data[0];
    }
Y
Yang Zhang 已提交
157
    max = static_cast<T>(max);
158

Z
zhangbo9674 已提交
159
    auto min = static_cast<T>(context.Attr<float>("min"));
160 161 162 163 164 165 166 167 168 169
    Tensor min_cpu;
    if (context.HasInput("Min")) {
      auto* min_t = context.Input<Tensor>("Min");
      auto* min_data = min_t->data<T>();
      if (platform::is_gpu_place(min_t->place())) {
        TensorCopySync(*min_t, platform::CPUPlace(), &min_cpu);
        min_data = min_cpu.data<T>();
      }
      min = min_data[0];
    }
Y
Yang Zhang 已提交
170
    min = static_cast<T>(min);
171

S
sneaxiy 已提交
172 173 174 175
    auto* d_out =
        context.Input<framework::LoDTensor>(framework::GradVarName("Out"));
    auto* d_x =
        context.Output<framework::LoDTensor>(framework::GradVarName("X"));
W
wanghaoshuang 已提交
176
    if (d_x != nullptr) {
S
sneaxiy 已提交
177
      auto* x = context.Input<framework::LoDTensor>("X");
W
wanghaoshuang 已提交
178
      int64_t numel = d_out->numel();
W
wanghaoshuang 已提交
179
      auto* d_x_data = d_x->mutable_data<T>(context.GetPlace());
W
wanghaoshuang 已提交
180 181
      const T* d_out_data = d_out->data<T>();
      const T* x_data = x->data<T>();
Q
QI JUN 已提交
182 183 184
      Transform<DeviceContext> trans;
      trans(context.template device_context<DeviceContext>(), d_out_data,
            d_out_data + numel, x_data, d_x_data, ClipGradFunctor<T>(min, max));
W
wanghaoshuang 已提交
185 186 187 188 189 190
    }
  }
};

}  // namespace operators
}  // namespace paddle