scale_image_compute.cc 4.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
// 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>
#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),
30
                                              PRECISION(kFP16),
31 32 33 34
                                              DATALAYOUT(kImageDefault)> {
 public:
  using param_t = operators::ScaleParam;

35
  std::string doc() const override { return "Scale using cl::Image2D, kFP16"; }
36 37 38 39 40 41 42 43 44 45

  void PrepareForRun() override {
    auto& context = ctx_->As<OpenCLContext>();
    context.cl_context()->AddKernel(
        kernel_func_name_, "image/scale_kernel.cl", build_options_);
  }

  void Run() override {
    const auto& param = *param_.get_mutable<param_t>();
    const auto& in_dims = param.x->dims();
46
    auto* x_img = param.x->data<uint16_t, cl::Image2D>();
47 48 49 50 51 52 53
    const float scale = param.scale;
    const float bias = param.bias;

    LOG(INFO) << "x_image" << x_img;
    auto out_image_shape = InitImageDimInfoWith(in_dims);
    LOG(INFO) << "out_image_shape = " << out_image_shape["width"] << " "
              << out_image_shape["height"];
54
    auto* out_img = param.output->mutable_data<uint16_t, cl::Image2D>(
55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91
        out_image_shape["width"], out_image_shape["height"]);
    LOG(INFO) << "out_image" << out_img;

    auto& context = ctx_->As<OpenCLContext>();
    CHECK(context.cl_context() != nullptr);
    STL::stringstream kernel_key;
    kernel_key << kernel_func_name_ << build_options_;
    auto kernel = context.cl_context()->GetKernel(kernel_key.str());

    auto global_work_size =
        cl::NDRange{static_cast<cl::size_type>(out_image_shape["width"]),
                    static_cast<cl::size_type>(out_image_shape["height"])};

    cl_int status;
    int arg_idx = 0;
    status = kernel.setArg(arg_idx, *x_img);
    CL_CHECK_FATAL(status);
    status = kernel.setArg(++arg_idx, *out_img);
    CL_CHECK_FATAL(status);
    status = kernel.setArg(++arg_idx, scale);
    CL_CHECK_FATAL(status);
    status = kernel.setArg(++arg_idx, bias);
    CL_CHECK_FATAL(status);

    status = context.cl_context()->GetCommandQueue().enqueueNDRangeKernel(
        kernel,
        cl::NullRange,
        global_work_size,
        cl::NullRange,
        nullptr,
        event_.get());
    CL_CHECK_FATAL(status);
    context.cl_wait_list()->emplace(out_img, event_);
  }

 private:
  std::string kernel_func_name_{"scale"};
92
  std::string build_options_{"-DCL_DTYPE_half"};
93 94 95 96 97 98 99 100 101 102
  std::shared_ptr<cl::Event> event_{new cl::Event};
};

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

REGISTER_LITE_KERNEL(scale,
                     kOpenCL,
103
                     kFP16,
104 105 106 107 108
                     kImageDefault,
                     paddle::lite::kernels::opencl::ScaleComputeImage2D,
                     image2d)
    .BindInput("X",
               {LiteType::GetTensorTy(TARGET(kOpenCL),
109
                                      PRECISION(kFP16),
110 111 112
                                      DATALAYOUT(kImageDefault))})
    .BindOutput("Out",
                {LiteType::GetTensorTy(TARGET(kOpenCL),
113
                                       PRECISION(kFP16),
114 115
                                       DATALAYOUT(kImageDefault))})
    .Finalize();