cross_entropy.cu 3.5 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
2

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
6

7
    http://www.apache.org/licenses/LICENSE-2.0
8

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. */
14

S
sneaxiy 已提交
15
#include "paddle/fluid/operators/math.h"
Y
Yi Wang 已提交
16
#include "paddle/fluid/operators/math/cross_entropy.h"
17
#include "paddle/fluid/platform/cuda_device_function.h"
18
#include "paddle/fluid/platform/cuda_primitives.h"
19 20 21 22 23 24

namespace paddle {
namespace operators {
namespace math {

template <typename T>
25
__global__ void CrossEntropyKernel(T* Y, const T* X, const int64_t* label,
26 27
                                   const int N, const int D,
                                   const int ignore_index) {
28
  CUDA_KERNEL_LOOP(i, N) {
29
    PADDLE_ENFORCE(label[i] >= 0 && label[i] < D || label[i] == ignore_index,
30 31
                   "The value of label[%d] expected >= 0 and < %ld, or == %ld, "
                   "but got %ld. Please check input value.",
32
                   i, D, ignore_index, label[i]);
33
    Y[i] = ignore_index == label[i]
34 35
               ? static_cast<T>(0)
               : -math::TolerableValue<T>()(real_log(X[i * D + label[i]]));
36 37 38 39 40 41 42
  }
}

template <typename T>
__global__ void SoftCrossEntropyKernel(T* Y, const T* X, const T* label,
                                       const int class_num) {
  int tid = threadIdx.x;
43
  T val(0);
44

45 46 47
  int idx = blockIdx.x * class_num + tid;
  int end = blockIdx.x * class_num + class_num;
  for (; idx < end; idx += blockDim.x) {
48
    val += math::TolerableValue<T>()(real_log(X[idx])) * label[idx];
49 50
  }

51 52 53
  val = paddle::platform::reduceSum(val, tid, blockDim.x);
  if (threadIdx.x == 0) {
    Y[blockIdx.x] = -val;
54 55 56 57
  }
}

template <typename T>
Q
QI JUN 已提交
58
class CrossEntropyFunctor<platform::CUDADeviceContext, T> {
59
 public:
Q
QI JUN 已提交
60 61
  void operator()(const platform::CUDADeviceContext& ctx,
                  framework::Tensor* out, const framework::Tensor* prob,
62 63
                  const framework::Tensor* labels, const bool softLabel,
                  const int ignore_index, const int axis_dim) {
64 65 66 67 68 69 70 71
    const T* prob_data = prob->data<T>();
    T* loss_data = out->mutable_data<T>(ctx.GetPlace());

    int batch_size = prob->dims()[0];
    int class_num = prob->dims()[1];

    if (softLabel) {
      const T* label_data = labels->data<T>();
72 73 74
      int block = class_num > 512
                      ? 512
                      : pow(2, static_cast<int>(std::log2(class_num)));
75

76
      SoftCrossEntropyKernel<T><<<batch_size, block, 0, ctx.stream()>>>(
Q
qijun 已提交
77
          loss_data, prob_data, label_data, class_num);
78
    } else {
79
      const int64_t* label_data = labels->data<int64_t>();
80 81
      int block = 512;
      int grid = (batch_size + block - 1) / block;
Q
QI JUN 已提交
82
      CrossEntropyKernel<T><<<grid, block, 0, ctx.stream()>>>(
83 84
          loss_data, prob_data, label_data, batch_size, class_num,
          ignore_index);
85 86 87 88
    }
  }
};

Q
QI JUN 已提交
89 90
template class CrossEntropyFunctor<platform::CUDADeviceContext, float>;
template class CrossEntropyFunctor<platform::CUDADeviceContext, double>;
91 92
template class CrossEntropyFunctor<platform::CUDADeviceContext,
                                   platform::float16>;
93 94 95
}  // namespace math
}  // namespace operators
}  // namespace paddle
新手
引导
客服 返回
顶部