chromium/third_party/gemmlowp/src/internal/dispatch_gemm_shape.h

// Copyright 2017 The Gemmlowp 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.

// dispatch_gemm_shape.h: dispatch GEMM calls according to their shape

#ifndef GEMMLOWP_INTERNAL_DISPATCH_GEMM_SHAPE_H_
#define GEMMLOWP_INTERNAL_DISPATCH_GEMM_SHAPE_H_

#include "../internal/kernel_default.h"
#include "../public/map.h"
#include "../public/output_stages.h"
#include "multi_thread_gemm.h"

namespace gemmlowp {

template <typename T>
struct TransposeImpl { … };

TransposeType;

template <typename T>
TransposeType<T> Transpose(const T& t) { … }

template <MapOrder Order>
struct TransposeMapOrder { … };

template <VectorShape Shape>
struct TransposeVectorShape { … };

TransposeImpl<VectorMap<Scalar, Shape>>;

TransposeImpl<MatrixMap<Scalar, Order>>;

TransposeImpl<OutputStageQuantizeDownInt32ToUint8ScalePC<Shape>>;

TransposeImpl<OutputStageScaleInt32ByFixedPointAndExponentPC<Shape>>;

TransposeImpl<OutputStageBiasAddition<VectorMapType>>;

// TODO(benoitjacob) - does anyone understand C++ variadic templates?
// How to use them to implement TransposeTuple? Note: there are lots
// of answers on StackOverflow but they seem to all involve either
// C++14/C++17 (we can only use C++11) or lots of abstract nonsense.
inline std::tuple<> TransposeTuple(const std::tuple<>& t) { … }

template <typename T0>
std::tuple<TransposeType<T0>> TransposeTuple(const std::tuple<T0>& t) { … }

template <typename T0, typename T1>
std::tuple<TransposeType<T0>, TransposeType<T1>> TransposeTuple(
    const std::tuple<T0, T1>& t) { … }

template <typename T0, typename T1, typename T2>
std::tuple<TransposeType<T0>, TransposeType<T1>, TransposeType<T2>>
TransposeTuple(const std::tuple<T0, T1, T2>& t) { … }

template <typename T0, typename T1, typename T2, typename T3>
std::tuple<TransposeType<T0>, TransposeType<T1>, TransposeType<T2>,
           TransposeType<T3>>
TransposeTuple(const std::tuple<T0, T1, T2, T3>& t) { … }

template <typename T0, typename T1, typename T2, typename T3, typename T4>
std::tuple<TransposeType<T0>, TransposeType<T1>, TransposeType<T2>,
           TransposeType<T3>, TransposeType<T4>>
TransposeTuple(const std::tuple<T0, T1, T2, T3, T4>& t) { … }

template <typename T0, typename T1, typename T2, typename T3, typename T4,
          typename T5>
std::tuple<TransposeType<T0>, TransposeType<T1>, TransposeType<T2>,
           TransposeType<T3>, TransposeType<T4>, TransposeType<T5>>
TransposeTuple(const std::tuple<T0, T1, T2, T3, T4, T5>& t) { … }

template <typename InputScalar, typename OutputScalar, typename BitDepthParams,
          MapOrder LhsOrder, MapOrder RhsOrder, MapOrder ResultOrder,
          typename LhsOffset, typename RhsOffset, typename OutputPipelineType,
          typename GemmContextType>
void DispatchGemmShape(GemmContextType* context,
                       const MatrixMap<const InputScalar, LhsOrder>& lhs,
                       const MatrixMap<const InputScalar, RhsOrder>& rhs,
                       MatrixMap<OutputScalar, ResultOrder>* result,
                       const LhsOffset& lhs_offset, const RhsOffset& rhs_offset,
                       const OutputPipelineType& output_pipeline) { … }

}  // end namespace gemmlowp

#endif  // GEMMLOWP_INTERNAL_DISPATCH_GEMM_SHAPE_H_