opr_impl.h 3.5 KB
Newer Older
1 2 3 4 5 6 7 8
/**
 * \file dnn/src/cuda/conv_bias/opr_impl.h
 * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
 *
 * Copyright (c) 2014-2020 Megvii Inc. All rights reserved.
 *
 * Unless required by applicable law or agreed to in writing,
 * software distributed under the License is distributed on an
9 10
 * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or
 * implied.
11 12 13 14 15 16 17 18 19 20 21 22 23
 */
#pragma once
#include "../elemwise/opr_impl.h"
#include "megdnn/oprs.h"

namespace megdnn {
namespace cuda {

class ConvBiasForwardImpl : public ConvBiasForward {
public:
    using ConvBiasForward::ConvBiasForward;
    void exec(_megdnn_tensor_in src, _megdnn_tensor_in filter,
              _megdnn_tensor_in bias, _megdnn_tensor_in z,
24 25 26
              _megdnn_tensor_out dst,
              const PreprocessedFilter* preprocessed_filter,
              _megdnn_workspace workspace) override;
27 28 29 30 31 32 33 34 35 36 37 38 39
    std::vector<Algorithm*> get_all_algorithms(
            const TensorLayout& src, const TensorLayout& filter,
            const TensorLayout& bias, const TensorLayout& z,
            const TensorLayout& dst) override;
    Algorithm* get_algorithm_heuristic(const TensorLayout& src,
                                       const TensorLayout& filter,
                                       const TensorLayout& bias,
                                       const TensorLayout& z,
                                       const TensorLayout& dst,
                                       size_t workspace_limit_in_bytes,
                                       bool reproducible) override;
    size_t get_workspace_in_bytes(const TensorLayout&, const TensorLayout&,
                                  const TensorLayout&, const TensorLayout&,
40 41 42 43 44 45 46
                                  const TensorLayout&,
                                  const PreprocessedFilter*) override;

    size_t get_preprocess_workspace_in_bytes(const TensorLayout&,
                                             const TensorLayout&,
                                             const TensorLayout&,
                                             const TensorLayout&,
M
Megvii Engine Team 已提交
47
                                             const TensorLayout&) override;
48 49
    SmallVector<TensorLayout> deduce_preprocessed_filter_layout(
            const TensorLayout&, const TensorLayout&, const TensorLayout&,
M
Megvii Engine Team 已提交
50
            const TensorLayout&, const TensorLayout&) override;
51 52 53
    void exec_preprocess(const TensorLayout&, _megdnn_tensor_in,
                         const TensorLayout&, const TensorLayout&,
                         const TensorLayout&, PreprocessedFilter*,
M
Megvii Engine Team 已提交
54
                         _megdnn_workspace) override;
55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75
    const char* get_algorithm_set_name() const override;

    class AlgoBase;
    class AlgoCUDNNConvBiasActivation;
    class AlgoChanwise;
    class AlgoChanwiseSmall;
    class AlgoChanwise8x8x32;
    class AlgoCUDNNConv;
    class AlgoInplaceMatmul;
    class AlgoMatmul;
    class AlgoMatmul8x8x32;
    class Algo1x1;
    class AlgoBatchedMatmul;
    class AlgoGroupConvGeneral;
    class AlgoQUInt4x4x32WMMA;
    class AlgoInt8CHWN4DotProdImplicitGemm;
    class AlgoInt8NCHW4DotProdImplicitGemm;
    class AlgoInt8CHWN4IMMAImplicitGemm;
    class AlgoInt8NCHW4IMMAImplicitGemm;
    class AlgoInt8CHWN4IMMAImplicitGemmReorderFilter;
    class AlgoInt8CHWN4IMMAImplicitGemmUnrollWidth;
76
    class AlgoInt8NCHW32IMMAImplicitGemm;
77
    class AlgoBFloat16;
78 79 80 81 82 83 84 85 86 87 88

    class AlgoPack;

    static const AlgoPack& algo_pack() { return sm_algo_pack; }

private:
    static AlgoPack sm_algo_pack;
};

}  // namespace cuda
}  // namespace megdnn
89

90
// vim: syntax=cpp.doxygen