conv_add_kernel.cpp 4.8 KB
Newer Older
L
liuruilong 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
/* 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 FUSION_CONVADD_OP

#include "operators/kernel/conv_add_kernel.h"

namespace paddle_mobile {
namespace operators {

template <>
bool ConvAddKernel<GPU_CL, float>::Init(FusionConvAddParam<GPU_CL> *param) {
L
liuruilong 已提交
24
  PADDLE_MOBILE_ENFORCE(
L
liuruilong 已提交
25
      param->Filter()->dims()[2] == param->Filter()->dims()[3] &&
L
liuruilong 已提交
26
          param->Paddings()[0] == param->Paddings()[1],
L
liuruilong 已提交
27
      "need equal");
L
liuruilong 已提交
28 29
  param->Bias()->InitCLImage(cl_helper_.CLContext(),
                             this->cl_helper_.CLCommandQueue());
L
liuruilong 已提交
30

L
liuruilong 已提交
31 32 33 34
  int offset = static_cast<int>(param->Filter()->dims()[2]) / 2 -
               static_cast<int>(param->Paddings()[1]);
  param->SetOffset(offset);

L
liuruilong 已提交
35 36 37 38
  if (param->Filter()->dims()[2] == 1 && param->Filter()->dims()[3] == 1) {
    param->Filter()->InitNImage(cl_helper_.CLContext(),
                                cl_helper_.CLCommandQueue());

Y
yangfei 已提交
39
    this->cl_helper_.AddKernel("conv_1x1", "conv_add_kernel.cl");
L
liuruilong 已提交
40
  } else if (param->Filter()->dims()[1] == 1) {
L
liuruilong 已提交
41 42 43
    param->Filter()->InitCLImage(cl_helper_.CLContext(),
                                 cl_helper_.CLCommandQueue());

Y
yangfei 已提交
44
    this->cl_helper_.AddKernel("depth_conv_3x3", "conv_add_kernel.cl");
L
liuruilong 已提交
45 46 47

  } else if (param->Filter()->dims()[2] == 3 &&
             param->Filter()->dims()[3] == 3) {
Z
zhaojiaying01 已提交
48
    param->Filter()->InitCLImage(cl_helper_.CLContext(),
L
liuruilong 已提交
49 50
                                 cl_helper_.CLCommandQueue());

Y
yangfei 已提交
51
    this->cl_helper_.AddKernel("conv_3x3", "conv_add_kernel.cl");
L
liuruilong 已提交
52

L
liuruilong 已提交
53 54 55 56
  } else {
    PADDLE_MOBILE_THROW_EXCEPTION(" not support ");
  }

L
liuruilong 已提交
57 58 59 60 61
  return true;
}

template <>
void ConvAddKernel<GPU_CL, float>::Compute(
L
liuruilong 已提交
62 63 64 65 66 67 68 69 70
    const FusionConvAddParam<GPU_CL> &param) {
  auto kernel = this->cl_helper_.KernelAt(0);
  auto default_work_size = this->cl_helper_.DefaultWorkSize(*param.Output());
  int c_block = default_work_size[0];
  int w = default_work_size[1];
  int nh = default_work_size[2];
  auto input = param.Input()->GetCLImage();
  auto filter = param.Filter()->GetCLImage();
  auto biase = param.Bias()->GetCLImage();
L
liuruilong 已提交
71
  auto output = param.Output()->GetCLImage();
L
liuruilong 已提交
72 73
  int stride = param.Strides()[0];
  int offset = param.Offset();
L
liuruilong 已提交
74 75 76
  int input_c = reinterpret_cast<framework::CLImageConverterFolder *>(
                    param.Input()->Converter())
                    ->GetCBlock();
L
liuruilong 已提交
77
  int dilation = param.Dilations()[0];
L
liuruilong 已提交
78 79 80 81 82

  int input_width = param.Input()->dims()[3];
  int input_height = param.Input()->dims()[2];
  int output_width = param.Output()->dims()[3];
  int output_height = param.Output()->dims()[2];
L
liuruilong 已提交
83

L
liuruilong 已提交
84 85 86
  cl_int status;

  status = clSetKernelArg(kernel, 0, sizeof(int), &c_block);
L
liuruilong 已提交
87 88
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
89
  status = clSetKernelArg(kernel, 1, sizeof(int), &w);
L
liuruilong 已提交
90 91
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
92
  status = clSetKernelArg(kernel, 2, sizeof(int), &nh);
L
liuruilong 已提交
93 94
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
95
  status = clSetKernelArg(kernel, 3, sizeof(cl_mem), &input);
L
liuruilong 已提交
96 97
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
98
  status = clSetKernelArg(kernel, 4, sizeof(cl_mem), &filter);
L
liuruilong 已提交
99 100
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
101
  status = clSetKernelArg(kernel, 5, sizeof(cl_mem), &biase);
L
liuruilong 已提交
102 103
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
104
  status = clSetKernelArg(kernel, 6, sizeof(cl_mem), &output);
L
liuruilong 已提交
105 106
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
107
  status = clSetKernelArg(kernel, 7, sizeof(int), &stride);
L
liuruilong 已提交
108 109
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
110
  status = clSetKernelArg(kernel, 8, sizeof(int), &offset);
L
liuruilong 已提交
111 112
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
113
  status = clSetKernelArg(kernel, 9, sizeof(int), &input_c);
L
liuruilong 已提交
114 115
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
116
  status = clSetKernelArg(kernel, 10, sizeof(int), &dilation);
L
liuruilong 已提交
117 118
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
119
  status = clSetKernelArg(kernel, 11, sizeof(int), &input_width);
L
liuruilong 已提交
120 121
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
122
  status = clSetKernelArg(kernel, 12, sizeof(int), &input_height);
L
liuruilong 已提交
123 124
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
125
  status = clSetKernelArg(kernel, 13, sizeof(int), &output_width);
L
liuruilong 已提交
126
  CL_CHECK_ERRORS(status);
L
liuruilong 已提交
127

L
liuruilong 已提交
128
  status = clSetKernelArg(kernel, 14, sizeof(int), &output_height);
L
liuruilong 已提交
129 130
  CL_CHECK_ERRORS(status);

L
liuruilong 已提交
131 132
  //  cl_event out_event = param.Output()->GetClEvent();
  //  cl_event wait_event = param.Input()->GetClEvent();
L
liuruilong 已提交
133

L
liuruilong 已提交
134 135 136
  status = clEnqueueNDRangeKernel(
      this->cl_helper_.CLCommandQueue(), kernel, default_work_size.size(), NULL,
      default_work_size.data(), NULL, 0, NULL, NULL);
L
liuruilong 已提交
137
  CL_CHECK_ERRORS(status);
L
liuruilong 已提交
138
}
L
liuruilong 已提交
139 140 141 142 143 144 145

template class ConvAddKernel<GPU_CL, float>;

}  // namespace operators
}  // namespace paddle_mobile

#endif