softmax_kernel.cpp 2.1 KB
Newer Older
H
hanbuhe 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/* 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

Z
zhangyang 已提交
17 18 19
#include "operators/kernel/softmax_kernel.h"
#include "operators/kernel/central-arm-func/softmax_arm_func.h"

H
hanbuhe 已提交
20 21 22 23
namespace paddle_mobile {
namespace operators {

template <>
N
nhzlx 已提交
24
bool SoftmaxKernel<FPGA, float>::Init(SoftmaxParam<FPGA> *param) {
Z
zhangyang 已提交
25
  auto input = const_cast<Tensor *>(param->InputX());
26
  auto input_ptr = input->data<float>();
Z
zhangyang 已提交
27
  auto float_input = new Tensor;
28
  float_input->mutable_data<float>({1, input->dims()[1]});
Z
zhangyang 已提交
29 30
  fpga::format_fp32_ofm(float_input);

Z
zhangyang 已提交
31
  fpga::BypassArgs args = {fpga::DATA_TYPE_FP16};
32 33 34 35
  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
36 37 38 39
  args.image.address = input_ptr;
  args.image.height = 1;
  args.image.width = 1;
  args.image.channels = (uint32_t)input->dims()[1];
Z
zhangyang 已提交
40 41
  args.output.address = float_input->data<float>();
  args.output.scale_address = float_input->scale;
Z
zhangyang 已提交
42
  param->SetFloatInput(float_input);
43
  param->SetFpgaArgs(args);
H
hanbuhe 已提交
44 45 46 47
  return true;
}

template <>
48
void SoftmaxKernel<FPGA, float>::Compute(const SoftmaxParam<FPGA> &param) {
49 50 51 52 53 54 55 56 57
  Tensor *in_x = param.FloatInput();
  Tensor *out = param.Out();

  fpga::PerformBypass(param.FpgaArgs());
  fpga::fpga_invalidate(
      (void *)in_x->data<float>(),  // NOLINT
      fpga::get_align_image_cw(in_x->dims()[1]) * sizeof(float));
  math::SoftmaxFuntor<CPU, float>()(in_x, out);
  fpga::fpga_flush(out->data<float>(), out->memory_size());
H
hanbuhe 已提交
58 59 60 61 62 63
}

}  // namespace operators
}  // namespace paddle_mobile

#endif