inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/fully_connected/fully_connected_kernel_fb_oi_b8_ref.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 "fully_connected_kernel_fb_oi_b8_ref.h"
  18
  19 namespace kernel_selector
  20 {
  21     ParamsKey FullyConnected_fb_oi_b8_ref::GetSupportedKey() const
  22     {
  23         ParamsKey k;
  24         k.EnableInputDataType(Datatype::F32);
  25         k.EnableOutputDataType(Datatype::F32);
  26         k.EnableInputWeightsType(WeightsType::F16);
  27         k.EnableInputWeightsType(WeightsType::F32);
  28         k.EnableAllInputLayout();
  29         k.EnableOutputLayout(DataLayout::fb);
  30         k.EnableBatching();
  31         k.EnableBiasPerFeature();
  32         k.EnableNonBiasTerm();
  33         k.EnableSubGroup();
  34         return k;
  35     }
  36
  37     FullyConnected_fb_oi_b8_ref::DispatchData FullyConnected_fb_oi_b8_ref::SetDefault(const fully_connected_params& arg, int ) const
  38     {
  39         auto kd = FullyConnectedKernelBase::SetDefault(arg);
  40
  41         const auto& output = arg.output;
  42         kd.gws0 = output.Batch().v;
  43         kd.gws1 = output.LogicalSize() / kd.gws0;
  44         kd.lws0 = 8;
  45         kd.lws1 = 1;
  46
  47         return kd;
  48     }
  49
  50     bool FullyConnected_fb_oi_b8_ref::Validate(const Params& p, const optional_params& o) const
  51     {
  52         if (!FullyConnectedKernelBase::Validate(p, o))
  53         {
  54             return false;
  55         }
  56
  57         const auto& params = static_cast<const fully_connected_params&>(p);
  58
  59         if (params.inputs[0].Batch().v != 8)
  60         {
  61             return false;
  62         }
  63
  64         return true;
  65     }
  66
  67     KernelsData FullyConnected_fb_oi_b8_ref::GetKernelsData(const Params& params, const optional_params& optParams) const
  68     {
  69         KernelsData res = {};
  70         for (size_t i = 0; i < autoTuneOptions.size(); i++)
  71         {
  72             KernelsData kd = GetTunedKernelsDataByIndex(params, optParams, DataLayout::fb, { WeightsLayout::oi }, FORCE_PRIORITY_6, (int)i);
  73             if (!kd.empty())
  74             {
  75                 res.emplace_back(kd[0]);
  76             }
  77         }
  78         return res;
  79     }
  80 }