accuracy_op.cu 3.5 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
武毅 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14

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. */

15 16
#include <thrust/execution_policy.h>
#include <thrust/reduce.h>
W
Wu Yi 已提交
17
#include "paddle/fluid/operators/metrics/accuracy_op.h"
D
dzhwinter 已提交
18
#include "paddle/fluid/platform/cuda_primitives.h"
Y
Yi Wang 已提交
19
#include "paddle/fluid/platform/gpu_info.h"
武毅 已提交
20 21 22

namespace paddle {
namespace operators {
23
using platform::PADDLE_CUDA_NUM_THREADS;
武毅 已提交
24

武毅 已提交
25 26 27
template <int BlockSize>
__global__ void AccuracyCudaKernel(const int N, const int D,
                                   const int64_t* Xdata,
D
Dong Zhihong 已提交
28
                                   const int64_t* labeldata, int* correct_data,
C
chengduo 已提交
29
                                   float* accuracy, int* total_data) {
30 31 32 33 34 35 36 37
  int count = 0;
  __shared__ int total[BlockSize];

  // support only 1 block
  for (int i = threadIdx.x; i < (N); i += BlockSize) {
    for (int j = 0; j < D; ++j) {
      if (Xdata[i * D + j] == labeldata[i]) {
        ++count;
武毅 已提交
38 39 40 41
        break;
      }
    }
  }
42 43 44 45 46 47
  total[threadIdx.x] = count;
  __syncthreads();

  // reduce the count with init value 0, and output accuracy.
  int result = thrust::reduce(thrust::device, total, total + BlockSize, 0);
  if (threadIdx.x == 0) {
D
Dong Zhihong 已提交
48
    *correct_data = result;
49
    *accuracy = static_cast<float>(result) / static_cast<float>(N);
C
chengduo 已提交
50
    *total_data = N;
51
  }
武毅 已提交
52 53 54
}

template <typename T>
Y
Yu Yang 已提交
55
class AccuracyOpCUDAKernel : public framework::OpKernel<T> {
武毅 已提交
56 57 58
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
    PADDLE_ENFORCE(platform::is_gpu_place(ctx.GetPlace()),
D
dzhwinter 已提交
59
                   "It must use CUDAPlace.");
武毅 已提交
60 61
    auto* inference = ctx.Input<Tensor>("Out");
    auto* indices = ctx.Input<Tensor>("Indices");
武毅 已提交
62
    auto* label = ctx.Input<Tensor>("Label");
D
Dong Zhihong 已提交
63

武毅 已提交
64
    auto* accuracy = ctx.Output<Tensor>("Accuracy");
D
Dong Zhihong 已提交
65 66
    auto* correct = ctx.Output<Tensor>("Correct");
    auto* total = ctx.Output<Tensor>("Total");
武毅 已提交
67 68
    // FIXME(typhoonzero): only support indices currently
    // if add support for output values, how to detect the data type?
武毅 已提交
69 70
    const int64_t* indices_data = indices->data<int64_t>();
    const int64_t* label_data = label->data<int64_t>();
D
Dong Zhihong 已提交
71 72 73

    int* correct_data = correct->mutable_data<int>(ctx.GetPlace());
    int* total_data = total->mutable_data<int>(ctx.GetPlace());
武毅 已提交
74 75
    float* accuracy_data = accuracy->mutable_data<float>(ctx.GetPlace());

D
Dong Zhihong 已提交
76
    int num_samples = static_cast<int>(inference->dims()[0]);
武毅 已提交
77
    size_t infer_width = inference->dims()[1];
D
dzhwinter 已提交
78 79
    auto stream = ctx.cuda_device_context().stream();
    platform::GpuMemsetAsync(accuracy_data, 0, sizeof(float), stream);
武毅 已提交
80 81 82 83 84

    if (num_samples == 0) {
      return;
    }

D
dzhwinter 已提交
85 86
    AccuracyCudaKernel<
        PADDLE_CUDA_NUM_THREADS><<<1, PADDLE_CUDA_NUM_THREADS, 0, stream>>>(
D
Dong Zhihong 已提交
87
        num_samples, infer_width, indices_data, label_data, correct_data,
C
chengduo 已提交
88
        accuracy_data, total_data);
武毅 已提交
89 90 91 92 93 94
  }
};

}  // namespace operators
}  // namespace paddle

D
Dong Zhihong 已提交
95 96
// FIXME(typhoonzero): types of T is for inference data.
// label data is always int64
Q
QI JUN 已提交
97 98 99
REGISTER_OP_CUDA_KERNEL(accuracy,
                        paddle::operators::AccuracyOpCUDAKernel<float>,
                        paddle::operators::AccuracyOpCUDAKernel<double>);