inference-engine/thirdparty/clDNN/src/gpu/shuffle_channels_gpu.cpp

   1 /*
   2 // Copyright (c) 2019 Intel Corporation
   3 //
   4 // Licensed under the Apache License, Version 2.0 (the "License");
   5 // you may not use this file except in compliance with the License.
   6 // You may obtain a copy of the License at
   7 //
   8 //      http://www.apache.org/licenses/LICENSE-2.0
   9 //
  10 // Unless required by applicable law or agreed to in writing, software
  11 // distributed under the License is distributed on an "AS IS" BASIS,
  12 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13 // See the License for the specific language governing permissions and
  14 // limitations under the License.
  15 */
  16
  17 #include "shuffle_channels_inst.h"
  18 #include "primitive_gpu_base.h"
  19 #include "implementation_map.h"
  20 #include "kernel_selector_helper.h"
  21 #include "shuffle_channels/shuffle_channels_kernel_selector.h"
  22 #include "shuffle_channels/shuffle_channels_kernel_ref.h"
  23 #include "error_handler.h"
  24
  25 using namespace cldnn;
  26
  27 namespace cldnn { namespace gpu {
  28
  29 struct shuffle_channels_gpu : typed_primitive_gpu_impl<shuffle_channels>
  30 {
  31     using parent = typed_primitive_gpu_impl<shuffle_channels>;
  32     using parent::parent;
  33
  34 public:
  35
  36     static primitive_impl* create(const shuffle_channels_node& arg)
  37     {
  38         auto shuffle_channels_params = get_default_params<kernel_selector::shuffle_channels_params>(arg);
  39         auto shuffle_channels_optional_params = get_default_optional_params<kernel_selector::shuffle_channels_optional_params>(arg.get_program());
  40
  41         const int32_t number_of_dims = 4;
  42         int32_t axis = arg.get_primitive()->axis;
  43
  44         if (axis < 0)
  45             axis += number_of_dims;
  46
  47         shuffle_channels_params.group = arg.get_primitive()->group;
  48         shuffle_channels_params.axis = axis;
  49
  50         auto& kernel_selector = kernel_selector::shuffle_channels_kernel_selector::Instance();
  51         auto best_kernels = kernel_selector.GetBestKernels(shuffle_channels_params, shuffle_channels_optional_params);
  52
  53         CLDNN_ERROR_BOOL(arg.id(), "Best_kernel.empty()", best_kernels.empty(), "Cannot find a proper kernel with this arguments");
  54
  55         auto shuffle_channels = new shuffle_channels_gpu(arg, best_kernels[0]);
  56
  57         return shuffle_channels;
  58     }
  59 };
  60
  61 namespace
  62 {
  63     struct attach
  64     {
  65         attach()
  66         {
  67             auto val_fw = shuffle_channels_gpu::create;
  68             implementation_map<shuffle_channels>::add(std::make_tuple(engine_types::ocl, data_types::f32, format::bfyx), val_fw);
  69             implementation_map<shuffle_channels>::add(std::make_tuple(engine_types::ocl, data_types::f16, format::bfyx), val_fw);
  70         }
  71         ~attach() = default;
  72     };
  73     attach attach_impl;
  74 }
  75 } }