fusion_conv_add_bn_op.h 3.3 KB
Newer Older
Z
zhangyang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54
/* Copyright (c) 2018 PaddlePaddle 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. */

#ifdef FUSION_CONVADDBN_OP

#pragma once

#include <string>
#include <vector>
#include "framework/operator.h"
#include "framework/program/program-optimize/fusion_op_register.h"
#include "op_param.h"
#include "operators/kernel/conv_add_bn_kernel.h"

namespace paddle_mobile {
namespace operators {
using std::string;
using std::vector;
class FusionConvAddBNMatcher : public framework::FusionOpMatcher {
 public:
  FusionConvAddBNMatcher() {
    node_ = framework::Node(G_OP_TYPE_CONV);
    node_ > std::make_shared<framework::Node>(G_OP_TYPE_ELEMENTWISE_ADD) >
        std::make_shared<framework::Node>(G_OP_TYPE_BATCHNORM);
  }

  void FolderNodes(
      framework::Node *node,
      std::vector<std::shared_ptr<framework::Node>> *removed_nodes) {
    node->Folder(node_.Depth(), Type(),
                 {{G_OP_TYPE_ELEMENTWISE_ADD, {{"Y", "Y"}}},
                  {G_OP_TYPE_BATCHNORM,
                   {{"Scale", "Scale"},
                    {"Mean", "Mean"},
                    {"Bias", "Bias"},
                    {"Variance", "Variance"}}}},
                 removed_nodes);
  }

  std::string Type() { return G_OP_TYPE_FUSION_CONV_ADD_BN; }
};

template <typename DeviceType, typename T>
Z
zhangyang 已提交
55 56 57
class FusionConvAddBNOp : public framework::OperatorWithKernel<
                              DeviceType, FusionConvAddBNParam,
                              operators::ConvAddBNKernel<DeviceType, T>> {
Z
zhangyang 已提交
58 59
 public:
  FusionConvAddBNOp(const string &type, const VariableNameMap &inputs,
Z
zhangyang 已提交
60 61 62
                    const VariableNameMap &outputs,
                    const framework::AttributeMap &attrs,
                    std::shared_ptr<framework::Scope> scope)
Z
zhangyang 已提交
63 64
      : framework::OperatorWithKernel<
            DeviceType, FusionConvAddBNParam,
Z
zhangyang 已提交
65 66
            operators::ConvAddBNKernel<DeviceType, T>>(type, inputs, outputs,
                                                       attrs, scope) {}
Z
zhangyang 已提交
67

Z
zhangyang 已提交
68
  void InferShape() const override;
Z
zhangyang 已提交
69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115

 protected:
};

#ifdef PADDLE_MOBILE_CPU

#ifndef FUSION_CONV_ADD_BN_REGISTER
static framework::FusionOpRegistrar fusion_conv_add_bn_registrar(
    new FusionConvAddBNMatcher());
#define FUSION_CONV_ADD_BN_REGISTER
#endif

#endif

#ifdef PADDLE_MOBILE_MALI_GPU

#ifndef FUSION_CONV_ADD_BN_REGISTER
static framework::FusionOpRegistrar fusion_conv_add_bn_registrar(
    new FusionConvAddBNMatcher());
#define FUSION_CONV_ADD_BN_REGISTER
#endif

#endif

#ifdef PADDLE_MOBILE_FPGA

#ifndef FUSION_CONV_ADD_BN_REGISTER
static framework::FusionOpRegistrar fusion_conv_add_bn_registrar(
    new FusionConvAddBNMatcher());
#define FUSION_CONV_ADD_BN_REGISTER
#endif

#endif

}  // namespace operators
}  // namespace paddle_mobile

#ifdef PADDLE_MOBILE_CPU
USE_OP_CPU(fusion_conv_add_bn);
#endif
#ifdef PADDLE_MOBILE_MALI_GPU
#endif
#ifdef PADDLE_MOBILE_FPGA
USE_OP_FPGA(fusion_conv_add_bn);
#endif

#endif