“4a7cba60d2ecb6650caee7459b2fd1f54da74e7e”上不存在“static/tools/train_multi_machine.py”
one_hot_op.cu 2.9 KB
Newer Older
Y
Yang yaming 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
//   Copyright (c) 2018 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.

Y
Yi Wang 已提交
15 16 17
#include "paddle/fluid/operators/one_hot_op.h"
#include "paddle/fluid/platform/cuda_helper.h"
#include "paddle/fluid/platform/gpu_info.h"
Y
Yang yaming 已提交
18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80

namespace paddle {
namespace operators {
using platform::PADDLE_CUDA_NUM_THREADS;

template <typename InT, typename OutT>
__global__ void FillOutputKernel(const InT* p_in_data, OutT* p_out_data,
                                 const int64_t numel, const int depth) {
  int idx = blockIdx.x * blockDim.x + threadIdx.x;
  if (idx < numel) {
    *(p_out_data + (idx * depth) + p_in_data[idx]) = 1.0;
  }
}

template <typename DeviceContext, typename InT>
struct OneHotOpCUDAFunctor {
  const framework::LoDTensor* in_;
  framework::LoDTensor* out_;
  const DeviceContext& ctx_;
  int depth_;

  OneHotOpCUDAFunctor(const framework::LoDTensor* in, framework::LoDTensor* out,
                      int depth, const DeviceContext& ctx)
      : in_(in), out_(out), depth_(depth), ctx_(ctx) {}

  template <typename OutT>
  void operator()() const {
    auto* p_in_data = in_->data<InT>();
    auto numel = in_->numel();
    auto* p_out_data = out_->mutable_data<OutT>(ctx_.GetPlace());
    auto stream = ctx_.stream();
    math::set_constant(ctx_, out_, 0.0);

    FillOutputKernel<<<(numel + PADDLE_CUDA_NUM_THREADS - 1) /
                           PADDLE_CUDA_NUM_THREADS,
                       PADDLE_CUDA_NUM_THREADS, 0, stream>>>(
        p_in_data, p_out_data, numel, depth_);
  }
};

using LoDTensor = framework::LoDTensor;
template <typename DeviceContext, typename T>
class OneHotCUDAKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& context) const override {
    auto* in = context.Input<LoDTensor>("X");
    auto* out = context.Output<LoDTensor>("Out");
    int depth = context.Attr<int>("depth");

    framework::VisitDataType(
        static_cast<framework::proto::DataType>(context.Attr<int>("dtype")),
        OneHotOpCUDAFunctor<DeviceContext, T>(
            in, out, depth, context.template device_context<DeviceContext>()));
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
REGISTER_OP_CUDA_KERNEL(
    one_hot, ops::OneHotCUDAKernel<paddle::platform::CUDADeviceContext, int>,
    ops::OneHotCUDAKernel<paddle::platform::CUDADeviceContext, int64_t>);