paddle_pass_builder.h 8.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
// 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.

#pragma once

#include <sstream>
#include <string>
#include <vector>
W
wanghuancoder 已提交
20

21
#include "paddle_infer_declare.h"  // NOLINT
22

23 24 25 26 27 28 29 30 31 32 33 34
///
/// \file paddle_pass_builder.h
///
/// \brief Class Paddle Passs Builder and its subclasses(pass strategies).
/// \section sec_intro Introduction
/// This class aims to build passes for paddle and define passes' strategies.
///
/// \author paddle-infer@baidu.com
/// \date 2020-3-23
/// \since 1.7

/// \namespace paddle
35
namespace paddle {
36

37 38 39 40 41 42 43 44 45 46 47
/// \class PaddlePassBuilder
/// \brief This class build passes based on vector<string> input. It is part of
/// inference API. Users can build passes, insert new passes, delete passes
/// using this class and its functions.
///
/// Example Usage:
///     Build a new pass.
/// \code{cpp}
/// const vector<string> passes(1, "conv_relu_mkldnn_fuse_pass");
/// PaddlePassBuilder builder(passes);
/// \endcode
48
class PD_INFER_DECL PaddlePassBuilder {
49
 public:
50 51
  /// \brief Constructor of the class. It stores the input passes.
  /// \param[in] passes passes' types.
52 53 54
  explicit PaddlePassBuilder(const std::vector<std::string> &passes)
      : passes_(passes) {}

55 56
  /// \brief Stores the input passes.
  /// \param[in] passes passes' types.
57 58 59 60
  void SetPasses(std::initializer_list<std::string> passes) {
    passes_ = passes;
  }

61 62
  /// \brief Append a pass to the end of the passes.
  /// \param[in] pass_type the type of the new pass.
63 64
  void AppendPass(const std::string &pass_type);

65 66 67
  /// \brief Insert a pass to a specific position.
  /// \param[in] idx the position to insert.
  /// \param[in] pass_type the type of insert pass.
68 69
  void InsertPass(size_t idx, const std::string &pass_type);

70 71
  /// \brief Delete the pass at certain position 'idx'.
  /// \param[in] idx the position to delete.
72 73
  void DeletePass(size_t idx);

74 75
  /// \brief Delete all passes that has a certain type 'pass_type'.
  /// \param[in] pass_type the certain pass type to be deleted.
76 77
  void DeletePass(const std::string &pass_type);

78
  /// \brief Delete all the passes.
79
  void ClearPasses();
80 81 82

  /// \brief Append an analysis pass.
  /// \param[in] pass the type of the new analysis pass.
Y
Yan Chunwei 已提交
83 84
  void AppendAnalysisPass(const std::string &pass);

85 86
  /// \brief Visualize the computation graph after each pass by generating a DOT
  /// language file, one can draw them with the Graphviz toolkit.
87
  void TurnOnDebug();
88
  /// \brief Human-readable information of the passes.
89 90
  std::string DebugString();

91 92
  /// \brief Get information of passes.
  /// \return Return list of the passes.
93
  const std::vector<std::string> &AllPasses() const { return passes_; }
94 95 96

  /// \brief Get information of analysis passes.
  /// \return Return list of analysis passes.
Y
Yan Chunwei 已提交
97 98 99 100 101 102 103
  std::vector<std::string> AnalysisPasses() const {
    auto passes = analysis_passes_;
    // To make sure the ir_graph_to_program should be the last pass so any
    // modication of IR will persist to the program.
    passes.push_back("ir_graph_to_program_pass");
    return passes;
  }
104 105

 protected:
106
  /// \cond Protected
Y
Yan Chunwei 已提交
107
  std::vector<std::string> analysis_passes_{
108
      {"ir_graph_build_pass", "ir_graph_clean_pass", "ir_analysis_pass",
109 110
       "ir_params_sync_among_devices_pass", "adjust_cudnn_workspace_size_pass",
       "inference_op_replace_pass"}};
111
  std::vector<std::string> passes_;
112
  /// \endcond
113 114
};

115 116 117
/// \class PassStrategy
/// \brief This class defines the pass strategies like whether to use gpu/cuDNN
/// kernel/MKLDNN.
118
class PD_INFER_DECL PassStrategy : public PaddlePassBuilder {
119
 public:
120 121
  /// \brief Constructor of PassStrategy class. It works the same as
  /// PaddlePassBuilder class. \param[in] passes passes' types.
122 123 124
  explicit PassStrategy(const std::vector<std::string> &passes)
      : PaddlePassBuilder(passes) {}

125
  /// \brief Enable the use of cuDNN kernel.
126 127
  virtual void EnableCUDNN() {}

128 129 130
  /// \brief Enable the use of MKLDNN.
  /// The MKLDNN control exists in both CPU and GPU mode, because there can
  /// still be some CPU kernels running in GPU mode.
Y
Yan Chunwei 已提交
131
  virtual void EnableMKLDNN() {}
132

133
  /// \brief Enable MKLDNN quantize optimization.
134
  virtual void EnableMkldnnQuantizer() {}
135

136 137 138
  /// \brief Enable MKLDNN bfloat16.
  virtual void EnableMkldnnBfloat16() {}

139 140
  /// \brief Check if we are using gpu.
  /// \return A bool variable implying whether we are in gpu mode.
141 142
  bool use_gpu() const { return use_gpu_; }

143 144 145 146
  /// \brief Check if we are using xpu.
  /// \return A bool variable implying whether we are in xpu mode.
  bool use_xpu() const { return use_xpu_; }

W
Wilber 已提交
147 148 149 150
  /// \brief Check if we are using npu.
  /// \return A bool variable implying whether we are in npu mode.
  bool use_npu() const { return use_npu_; }

151
  /// \brief Default destructor.
152
  virtual ~PassStrategy() = default;
153 154

 protected:
155
  /// \cond Protected
156
  bool use_xpu_{false};
157
  bool use_gpu_{false};
W
Wilber 已提交
158
  bool use_npu_{false};
Y
Yan Chunwei 已提交
159
  bool use_mkldnn_{false};
160
  /// \endcond
161 162
};

163 164 165
/// \class CpuPassStrategy
/// \brief The CPU passes controller, it is used in AnalysisPredictor with CPU
/// mode.
166
class PD_INFER_DECL CpuPassStrategy : public PassStrategy {
167
 public:
168
  /// \brief Default constructor of CpuPassStrategy.
169
  CpuPassStrategy();
170

171 172
  /// \brief Construct by copying another CpuPassStrategy object.
  /// \param[in] other The CpuPassStrategy object we want to copy.
Y
Yan Chunwei 已提交
173
  explicit CpuPassStrategy(const CpuPassStrategy &other)
W
Wojciech Uss 已提交
174 175 176 177
      : PassStrategy(other.AllPasses()) {
    use_gpu_ = other.use_gpu_;
    use_mkldnn_ = other.use_mkldnn_;
    use_mkldnn_quantizer_ = other.use_mkldnn_quantizer_;
178
    use_mkldnn_bfloat16_ = other.use_mkldnn_bfloat16_;
W
Wojciech Uss 已提交
179
  }
180
  /// \brief Default destructor.
181 182
  virtual ~CpuPassStrategy() = default;

183
  /// \brief Enable the use of cuDNN kernel.
184
  void EnableCUDNN() override;
185 186

  /// \brief Enable the use of MKLDNN.
W
Wojciech Uss 已提交
187
  void EnableMKLDNN() override;
188 189

  /// \brief Enable MKLDNN quantize optimization.
W
Wojciech Uss 已提交
190
  void EnableMkldnnQuantizer() override;
191

192 193 194
  /// \brief Enable MKLDNN bfloat16.
  void EnableMkldnnBfloat16() override;

195
 protected:
196
  /// \cond Protected
197
  bool use_mkldnn_quantizer_{false};
198
  bool use_mkldnn_bfloat16_{false};
199
  /// \endcond
200 201
};

202 203 204
/// \class GpuPassStrategy
/// \brief The GPU passes controller, it is used in AnalysisPredictor with GPU
/// mode.
205
class PD_INFER_DECL GpuPassStrategy : public PassStrategy {
206
 public:
207
  /// \brief Default constructor of GpuPassStrategy.
208
  GpuPassStrategy();
209

210 211
  /// \brief Construct by copying another GpuPassStrategy object.
  /// \param[in] other The GpuPassStrategy object we want to copy.
Y
Yan Chunwei 已提交
212
  explicit GpuPassStrategy(const GpuPassStrategy &other)
213 214
      : PassStrategy(other.AllPasses()) {
    use_gpu_ = true;
215
    use_cudnn_ = other.use_cudnn_;
216
  }
217

218
  /// \brief Enable the use of cuDNN kernel.
219
  void EnableCUDNN() override;
220 221

  /// \brief Not supported in GPU mode yet.
222
  void EnableMKLDNN() override;
223 224

  /// \brief Not supported in GPU mode yet.
225
  void EnableMkldnnQuantizer() override;
226

227 228 229
  /// \brief Not supported in GPU mode yet.
  void EnableMkldnnBfloat16() override;

230
  /// \brief Default destructor.
231
  virtual ~GpuPassStrategy() = default;
232 233

 protected:
234
  /// \cond Protected
235
  bool use_cudnn_{false};
236
  /// \endcond
237
};
238

239 240 241 242 243
/// \class XpuPassStrategy
/// \brief The XPU passes controller, it is used in AnalysisPredictor with XPU
/// mode.
class PD_INFER_DECL XpuPassStrategy final : public PassStrategy {
 public:
W
Wilber 已提交
244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259
  XpuPassStrategy() : PassStrategy({}) { use_xpu_ = true; }
};

/// \class NpuPassStrategy
/// \brief The NPU passes controller, it is used in AnalysisPredictor with NPU
/// mode.
class PD_INFER_DECL NpuPassStrategy final : public PassStrategy {
 public:
  NpuPassStrategy() : PassStrategy({}) { use_npu_ = true; }

  /// \brief Construct by copying another NpuPassStrategy object.
  /// \param[in] other The NpuPassStrategy object we want to copy.
  explicit NpuPassStrategy(const NpuPassStrategy &other)
      : PassStrategy(other.AllPasses()) {
    use_npu_ = true;
  }
260 261
};

262
/// \brief List of tensorRT subgraph passes.
263
PD_INFER_DECL extern const std::vector<std::string> kTRTSubgraphPasses;
264

D
denglin-github 已提交
265 266 267
/// \brief List of dlnne subgraph passes.
PD_INFER_DECL extern const std::vector<std::string> kDlnneSubgraphPasses;

268
/// \brief List of lite subgraph passes.
269
PD_INFER_DECL extern const std::vector<std::string> kLiteSubgraphPasses;
270

271
}  // namespace paddle