2 // Copyright (c) 2016 Intel Corporation
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
8 // http://www.apache.org/licenses/LICENSE-2.0
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.
17 #include "pooling_kernel_gpu_average_opt.h"
19 namespace kernel_selector
21 ParamsKey PoolingKernelGPUAverageOpt::GetSupportedKey() const
24 k.EnableInputDataType(Datatype::F32);
25 k.EnableOutputDataType(Datatype::F32);
26 k.EnableInputLayout(DataLayout::bfyx);
27 k.EnableOutputLayout(DataLayout::bfyx);
28 k.EnablePoolType(PoolType::AVG);
29 k.EnablePoolRemainder(PoolRemainder::FLOOR);
30 k.EnablePoolRemainder(PoolRemainder::CEIL);
31 k.EnablePoolKernelDividerMode(KernelDividerMode::FIXED);
35 bool PoolingKernelGPUAverageOpt::Validate(const Params& p, const optional_params& o) const
37 if (!PoolingKernelBase::Validate(p, o))
42 const pooling_params& params = static_cast<const pooling_params&>(p);
44 if (params.activation.function != ActivationFunction::NONE)
49 if ((params.poolSize.x != 3) ||
50 (params.poolSize.y != 3) ||
51 (params.poolStride.x != 1) ||
52 (params.poolStride.y != 1) ||
53 (params.poolPad.x != 1) ||
54 (params.poolPad.y != 1) ||
55 !(params.inputs[0] == params.output) ||
56 params.inputs[0].PitchesDifferFromLogicalDims() ||
57 params.output.PitchesDifferFromLogicalDims())
65 static uSize GetTileDimentions()
67 constexpr int simdSize = 16;
69 return{ simdSize - 2, 7 };
72 PoolingKernelBase::DispatchData PoolingKernelGPUAverageOpt::SetDefault(const pooling_params& params) const
74 constexpr int simdSize = 16;
76 DispatchData runInfo = PoolingKernelBase::SetDefault(params);
78 auto tileDims = GetTileDimentions();
80 const int numTilesX = static_cast<int>(std::ceil(static_cast<float>(params.inputs[0].X().v) / static_cast<float>(tileDims.x)));
81 const int numTilesY = static_cast<int>(std::ceil(static_cast<float>(params.inputs[0].Y().v) / static_cast<float>(tileDims.y)));
83 runInfo.gws0 = numTilesX * simdSize;
84 runInfo.gws1 = numTilesY;
85 runInfo.gws2 = params.inputs[0].Feature().v;
86 runInfo.lws0 = simdSize;
93 JitConstants PoolingKernelGPUAverageOpt::GetJitConstants(const pooling_params& params, DispatchData kd) const
95 auto tileDims = GetTileDimentions();
96 auto mem_consts = PoolingKernelBase::GetJitConstants(params, kd);
98 if (tileDims.y != 0 && tileDims.x != 0)
100 mem_consts.AddConstant(MakeJitConstant("SUB_GROUP_SIZE", kd.lws0));
101 mem_consts.AddConstant(MakeJitConstant("TILE_HEIGHT", tileDims.y));
102 mem_consts.AddConstant(MakeJitConstant("TILE_WIDTH", tileDims.x));
103 mem_consts.AddConstant(MakeJitConstant("ONE_OVER_POOL_SIZE", 1.f / (params.poolSize.x * params.poolSize.y)));
109 KernelsData PoolingKernelGPUAverageOpt::GetKernelsData(const Params& params, const optional_params& options) const
111 return GetCommonKernelsData(params, options, FORCE_PRIORITY_7);