scale_image_compute.cc 4.9 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
// Copyright (c) 2019 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 <vector>
16
#include "lite/backends/opencl/cl_half.h"
17 18 19 20 21 22 23 24 25 26 27 28 29 30
#include "lite/backends/opencl/cl_include.h"
#include "lite/core/kernel.h"
#include "lite/core/op_registry.h"
#include "lite/kernels/opencl/image_helper.h"
#include "lite/operators/op_params.h"
#include "lite/utils/replace_stl/stream.h"
#include "lite/utils/string.h"

namespace paddle {
namespace lite {
namespace kernels {
namespace opencl {

class ScaleComputeImage2D : public KernelLite<TARGET(kOpenCL),
31
                                              PRECISION(kFP16),
32 33 34 35
                                              DATALAYOUT(kImageDefault)> {
 public:
  using param_t = operators::ScaleParam;

36
  std::string doc() const override { return "Scale using cl::Image2D, kFP16"; }
37 38 39

  void PrepareForRun() override {
    auto& context = ctx_->As<OpenCLContext>();
40 41 42 43
    context.cl_context()->AddKernel(kernel_func_name_,
                                    "image/scale_kernel.cl",
                                    build_options_,
                                    time_stamp_);
44 45 46
    VLOG(1) << "kernel_func_name_:" << kernel_func_name_;

    STL::stringstream kernel_key;
47
    kernel_key << kernel_func_name_ << build_options_ << time_stamp_;
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
    kernel_ = context.cl_context()->GetKernel(kernel_key.str());
  }

  void ReInitWhenNeeded() override {
    scale_param_ = param_.get_mutable<param_t>();
    auto x_dims = scale_param_->x->dims();
    if ((!first_epoch_for_reinit_ && x_dims != last_x_dims_) ||
        first_epoch_for_reinit_) {
      last_x_dims_ = x_dims;
      first_epoch_for_reinit_ = false;

      // compute image shape
      paddle::lite::CLImageConverterDefault default_convertor;
      out_img_shape_ =
          default_convertor.InitImageDimInfoWith(scale_param_->output->dims());

      // compute global work size
      GetGlobalWorkSize();
    }
  }

  void GetGlobalWorkSize() {
    global_work_size_ =
        cl::NDRange{static_cast<cl::size_type>(out_img_shape_[0]),
                    static_cast<cl::size_type>(out_img_shape_[1])};
73 74 75
  }

  void Run() override {
76 77 78 79 80
    auto* x_img = scale_param_->x->data<half_t, cl::Image2D>();
    auto* out_img = scale_param_->output->mutable_data<half_t, cl::Image2D>(
        out_img_shape_[0], out_img_shape_[1]);
    const float scale = scale_param_->scale;
    const float bias = scale_param_->bias;
81 82 83 84

    auto& context = ctx_->As<OpenCLContext>();
    CHECK(context.cl_context() != nullptr);

85
    auto kernel = kernel_;
86
    cl_int status;
87
    status = kernel.setArg(0, *x_img);
88
    CL_CHECK_FATAL(status);
89
    status = kernel.setArg(1, *out_img);
90
    CL_CHECK_FATAL(status);
91
    status = kernel.setArg(2, scale);
92
    CL_CHECK_FATAL(status);
93
    status = kernel.setArg(3, bias);
94 95 96 97 98
    CL_CHECK_FATAL(status);

    status = context.cl_context()->GetCommandQueue().enqueueNDRangeKernel(
        kernel,
        cl::NullRange,
99
        global_work_size_,
100 101 102 103 104 105 106 107 108
        cl::NullRange,
        nullptr,
        event_.get());
    CL_CHECK_FATAL(status);
    context.cl_wait_list()->emplace(out_img, event_);
  }

 private:
  std::string kernel_func_name_{"scale"};
109
  std::string build_options_{"-DCL_DTYPE_half"};
110
  std::string time_stamp_{GetTimeStamp()};
111
  std::shared_ptr<cl::Event> event_{new cl::Event};
112 113 114 115 116 117 118 119 120

  param_t* scale_param_{nullptr};
  cl::Kernel kernel_;
  bool first_epoch_for_reinit_{true};
  DDim last_x_dims_;
  DDim out_img_shape_ = DDim(std::vector<DDim::value_type>(
      {static_cast<DDim::value_type>(1), static_cast<DDim::value_type>(1)}));
  cl::NDRange global_work_size_ = cl::NDRange{
      static_cast<size_t>(1), static_cast<size_t>(1), static_cast<size_t>(1)};
121 122 123 124 125 126 127 128 129
};

}  // namespace opencl
}  // namespace kernels
}  // namespace lite
}  // namespace paddle

REGISTER_LITE_KERNEL(scale,
                     kOpenCL,
130
                     kFP16,
131 132 133 134 135
                     kImageDefault,
                     paddle::lite::kernels::opencl::ScaleComputeImage2D,
                     image2d)
    .BindInput("X",
               {LiteType::GetTensorTy(TARGET(kOpenCL),
136
                                      PRECISION(kFP16),
137 138 139
                                      DATALAYOUT(kImageDefault))})
    .BindOutput("Out",
                {LiteType::GetTensorTy(TARGET(kOpenCL),
140
                                       PRECISION(kFP16),
141 142
                                       DATALAYOUT(kImageDefault))})
    .Finalize();