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

   1 /*
   2 // Copyright (c) 2016 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 "lrn_inst.h"
  18 #include "primitive_gpu_base.h"
  19 #include "implementation_map.h"
  20 #include "error_handler.h"
  21 #include "kernel_selector_helper.h"
  22 #include "lrn/lrn_kernel_selector.h"
  23 #include "lrn/lrn_kernel_base.h"
  24
  25 namespace cldnn { namespace gpu {
  26
  27
  28 struct lrn_gpu : typed_primitive_gpu_impl<lrn>
  29 {
  30     using parent = typed_primitive_gpu_impl<lrn>;
  31     using parent::parent;
  32
  33     static primitive_impl* create(const lrn_node& arg)
  34     {
  35         auto lrn_params = get_default_params<kernel_selector::lrn_params>(arg);
  36         auto lrn_optional_params = get_default_optional_params<kernel_selector::lrn_optional_params>(arg.get_program());
  37
  38         const auto& primitive = arg.get_primitive();
  39
  40         lrn_params.alpha      = primitive->alpha;
  41         lrn_params.beta       = primitive->beta;
  42         lrn_params.k          = primitive->k;
  43         lrn_params.localSize  = primitive->size;
  44         lrn_params.divMode    = kernel_selector::kernel_divider_mode::FIXED;
  45         lrn_params.normMode   =
  46             primitive->norm_region == cldnn_lrn_norm_region_within_channel ?
  47             kernel_selector::lrn_mode::WITHIN_CHANNEL :
  48             kernel_selector::lrn_mode::ACROSS_CHANNEL;
  49
  50
  51         auto& kernel_selector = kernel_selector::lrn_kernel_selector::Instance();
  52         auto best_kernels = kernel_selector.GetBestKernels(lrn_params, lrn_optional_params);
  53
  54         CLDNN_ERROR_BOOL(arg.id(), "Best_kernel.empty()", best_kernels.empty(), "Cannot find a proper kernel with this arguments");
  55
  56         auto lrn = new lrn_gpu(arg, best_kernels[0]);
  57
  58         return lrn;
  59     }
  60
  61 };
  62
  63 namespace {
  64     struct attach {
  65         attach() {
  66             implementation_map<lrn>::add(std::make_tuple(engine_types::ocl, data_types::f32, format::yxfb), lrn_gpu::create);
  67             implementation_map<lrn>::add(std::make_tuple(engine_types::ocl, data_types::f16, format::yxfb), lrn_gpu::create);
  68             implementation_map<lrn>::add(std::make_tuple(engine_types::ocl, data_types::f32, format::bfyx), lrn_gpu::create);
  69             implementation_map<lrn>::add(std::make_tuple(engine_types::ocl, data_types::f16, format::bfyx), lrn_gpu::create);
  70             implementation_map<lrn>::add(std::make_tuple(engine_types::ocl, data_types::f32, format::byxf), lrn_gpu::create);
  71             implementation_map<lrn>::add(std::make_tuple(engine_types::ocl, data_types::f16, format::byxf), lrn_gpu::create);
  72         }
  73         ~attach() {}
  74     };
  75     attach attach_impl;
  76 }
  77 } }