feed_kernel.cpp 2.9 KB
Newer Older
Y
yangfei 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
/* 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. */

#include "operators/kernel/feed_kernel.h"
Y
yangfei 已提交
16
#include "framework/cl/cl_tensor.h"
H
hjchen2 已提交
17

Y
yangfei 已提交
18
namespace paddle_mobile {
L
liuruilong 已提交
19 20 21 22 23 24 25 26 27 28 29 30
namespace operators {

template <>
bool FeedKernel<GPU_CL, float>::Init(FeedParam<GPU_CL> *param) {
  DLOG << "Init feed";
  this->cl_helper_.AddKernel("feed", "feed_kernel.cl");
  return true;
}

template <>
void FeedKernel<GPU_CL, float>::Compute(const FeedParam<GPU_CL> &param) {
  auto kernel = this->cl_helper_.KernelAt(0);
Y
yangfei 已提交
31
  auto default_work_size = this->cl_helper_.DefaultWorkSize(*(param.Out()));
L
liuruilong 已提交
32
  cl_int status;
33 34
  param.Out()->InitEmptyImage(cl_helper_.CLContext(),
                              cl_helper_.CLCommandQueue(), param.Out()->dims());
L
liuruilong 已提交
35
  auto output = param.Out();
Y
yangfei 已提交
36
  const Tensor *input = param.InputX();
L
liuruilong 已提交
37
  //  DLOG << *input;
Y
yangfei 已提交
38 39
  const float *input_data = input->data<float>();
  int numel = input->numel();
Y
yangfei 已提交
40 41 42 43 44 45 46
  cl_mem output_image = output->GetCLImage();
  const int out_C = output->dims()[1];
  const int out_H = output->dims()[2];
  const int out_W = output->dims()[3];
  const int Stride2 = out_C * out_H * out_W;
  const int Stride1 = out_H * out_W;
  const int Stride0 = out_W;
H
hjchen2 已提交
47 48
  framework::CLTensor input_cl_tensor(this->cl_helper_.CLContext(),
                                      this->cl_helper_.CLCommandQueue());
Y
yangfei 已提交
49
  input_cl_tensor.Resize(input->dims());
L
liuruilong 已提交
50
  cl_mem inputBuffer = input_cl_tensor.mutable_with_data<float>(input_data);
L
liuruilong 已提交
51

Y
yangfei 已提交
52
  status = clSetKernelArg(kernel, 0, sizeof(cl_mem), &inputBuffer);
L
liuruilong 已提交
53
  CL_CHECK_ERRORS(status);
Y
yangfei 已提交
54
  status = clSetKernelArg(kernel, 1, sizeof(cl_mem), &output_image);
L
liuruilong 已提交
55
  CL_CHECK_ERRORS(status);
Y
yangfei 已提交
56
  status = clSetKernelArg(kernel, 2, sizeof(cl_int), &out_H);
Y
yangfei 已提交
57
  CL_CHECK_ERRORS(status);
Y
yangfei 已提交
58
  status = clSetKernelArg(kernel, 3, sizeof(cl_int), &out_W);
L
liuruilong 已提交
59
  CL_CHECK_ERRORS(status);
Y
yangfei 已提交
60 61 62 63 64 65 66
  status = clSetKernelArg(kernel, 4, sizeof(cl_int), &out_C);
  CL_CHECK_ERRORS(status);
  status = clSetKernelArg(kernel, 5, sizeof(cl_int), &Stride0);
  CL_CHECK_ERRORS(status);
  status = clSetKernelArg(kernel, 6, sizeof(cl_int), &Stride1);
  CL_CHECK_ERRORS(status);
  status = clSetKernelArg(kernel, 7, sizeof(cl_int), &Stride2);
67
  CL_CHECK_ERRORS(status);
L
liuruilong 已提交
68

Y
yangfei 已提交
69 70 71
  status = clEnqueueNDRangeKernel(
      this->cl_helper_.CLCommandQueue(), kernel, default_work_size.size(), NULL,
      default_work_size.data(), NULL, 0, NULL, NULL);
Y
yangfei 已提交
72

L
liuruilong 已提交
73
  CL_CHECK_ERRORS(status);
L
liuruilong 已提交
74 75 76 77 78
}

template class FeedKernel<GPU_CL, float>;

}  // namespace operators
Y
yangfei 已提交
79
}  // namespace paddle_mobile