softmax_kernel.cpp 2.0 KB
Newer Older
H
hanbuhe 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.

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. */

#ifdef SOFTMAX_OP

#include "../softmax_kernel.h"
#include "../central-arm-func/softmax_arm_func.h"
#include "common/types.h"
H
hanbuhe 已提交
20
#include "fpga/api.h"
H
hanbuhe 已提交
21 22 23 24 25
#include "operators/math/softmax.h"
namespace paddle_mobile {
namespace operators {

template <>
N
nhzlx 已提交
26
bool SoftmaxKernel<FPGA, float>::Init(SoftmaxParam<FPGA> *param) {
Z
zhangyang 已提交
27
  auto input = const_cast<Tensor *>(param->InputX());
28
  auto input_ptr = input->data<float>();
Z
zhangyang 已提交
29 30 31
  auto float_input = new Tensor(*input);
  fpga::format_fp32_ofm(float_input);

32 33 34 35 36
  fpga::BypassArgs args;
  args.input_layout_type = fpga::LAYOUT_HWC;
  args.output_layout_type = fpga::LAYOUT_CHW;
  args.input_data_type = fpga::DATA_TYPE_FP16;
  args.output_data_type = fpga::DATA_TYPE_FP32;
Z
zhangyang 已提交
37 38 39 40 41
  args.image.address = input_ptr;
  args.image.height = 1;
  args.image.width = 1;
  args.image.channels = (uint32_t)input->dims()[1];
  args.output.address = float_input->mutable_data<float>();
42

Z
zhangyang 已提交
43
  param->SetFloatInput(float_input);
44
  param->SetFpgaArgs(args);
H
hanbuhe 已提交
45 46 47 48
  return true;
}

template <>
N
nhzlx 已提交
49 50
void SoftmaxKernel<FPGA, float>::Compute(
    const SoftmaxParam<FPGA> &param) const {
Z
zhangyang 已提交
51
  Tensor *in_x = param.FloatInput();
H
hanbuhe 已提交
52
  Tensor *out = param.Out();
Z
zhangyang 已提交
53

54
  fpga::PerformBypass(param.FpgaArgs());
Z
zhangyang 已提交
55 56 57
  fpga::fpga_invalidate(
      (void *)in_x->data<float>(),
      (size_t)fpga::get_align_image_cw((int)in_x->dims()[1]) * sizeof(float));
58

H
hanbuhe 已提交
59
  math::SoftmaxFuntor<CPU, float>()(in_x, out);
Z
zhangyang 已提交
60
  fpga::fpga_flush(out->data<float>(), out->memory_size());
H
hanbuhe 已提交
61 62 63 64 65 66
}

}  // namespace operators
}  // namespace paddle_mobile

#endif