ir_pass_manager.cc 4.7 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
// 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.

#include "paddle/fluid/inference/analysis/ir_pass_manager.h"
#include <string>
Z
zhhsplendid 已提交
17
#include <unordered_map>
L
luotao1 已提交
18
#include <vector>
Y
Yan Chunwei 已提交
19
#include "paddle/fluid/framework/ir/fuse_pass_base.h"
20 21
#include "paddle/fluid/framework/ir/graph.h"
#include "paddle/fluid/framework/scope.h"
22 23
#include "paddle/fluid/inference/analysis/argument.h"
#include "paddle/fluid/inference/analysis/ir_passes/subgraph_detector.h"
Y
Yan Chunwei 已提交
24
#include "paddle/fluid/string/pretty_log.h"
25 26 27 28

namespace paddle {
namespace inference {
namespace analysis {
Y
Yan Chunwei 已提交
29 30 31
using string::PrettyLogEndl;
using string::PrettyLog;
using string::Style;
32

33 34 35 36 37 38 39 40 41 42 43
IRPassManager::IRPassManager(Argument *argument) {
  ARGUMENT_CHECK_FIELD(argument, main_program);
  graph_ = std::unique_ptr<Graph>(new Graph(argument->main_program()));
  if (argument->Has("scope")) {
    graph_->Set(framework::ir::kParamScopeAttr,
                new framework::Scope *(
                    const_cast<framework::Scope *>(&argument->scope())));
  }

  ARGUMENT_CHECK_FIELD(argument, ir_analysis_passes);
  CreatePasses(argument, argument->ir_analysis_passes());
44 45
}

46 47
void IRPassManager::CreatePasses(Argument *argument,
                                 const std::vector<std::string> &passes) {
48
  std::string pre_pass;
L
luotao1 已提交
49
  int pass_num = 0;
50
  for (const std::string &pass_name : passes) {
51
    auto pass = framework::ir::PassRegistry::Instance().Get(pass_name);
52

53
    if (pass_name == "graph_viz_pass") {
L
luotao1 已提交
54 55 56
      std::string dot_file_path = std::to_string(pass_num) + "_ir_" +
                                  (pre_pass.empty() ? "origin" : pre_pass) +
                                  ".dot";
57
      pass->Set("graph_viz_path", new std::string(std::move(dot_file_path)));
L
luotao1 已提交
58
      pass_num++;
Z
zhhsplendid 已提交
59
    } else if (pass_name == "mkldnn_placement_pass") {
60 61 62
      pass->Set("mkldnn_enabled_op_types",
                new std::unordered_set<std::string>(
                    argument->mkldnn_enabled_op_types()));
Z
zhhsplendid 已提交
63 64 65 66
    } else if (pass_name == "cpu_quantize_pass") {
      pass->Set("quant_var_scales",
                new VarQuantScale(argument->quant_var_scales()));
    } else if (pass_name == "tensorrt_subgraph_pass") {
67 68
      pass->Set("workspace_size", new int(argument->tensorrt_workspace_size()));
      pass->Set("max_batch_size", new int(argument->tensorrt_max_batch_size()));
69 70
      pass->Set("min_subgraph_size",
                new int(argument->tensorrt_min_subgraph_size()));
N
nhzlx 已提交
71 72
      pass->Set("program",
                new framework::ProgramDesc *(&argument->main_program()));
N
nhzlx 已提交
73

N
nhzlx 已提交
74
      bool enable_int8 = argument->tensorrt_precision_mode() ==
75
                         AnalysisConfig::Precision::kInt8;
N
nhzlx 已提交
76 77

      pass->Set("enable_int8", new bool(enable_int8));
N
nhzlx 已提交
78 79 80 81 82 83 84
      std::string model_opt_cache_dir =
          argument->Has("model_dir")
              ? argument->model_dir()
              : GetDirRoot(argument->model_program_path());
      pass->Set(
          "model_opt_cache_dir",
          new std::string(GetOrCreateModelOptCacheDir(model_opt_cache_dir)));
N
nhzlx 已提交
85
      pass->Set("gpu_device_id", new int(argument->gpu_device_id()));
N
nhzlx 已提交
86 87
      pass->Set("use_static_engine",
                new bool(argument->tensorrt_use_static_engine()));
88 89
    }

90
    pre_pass = pass_name;
91 92

    passes_.emplace_back(std::move(pass));
93 94 95
  }
}

96 97 98 99 100 101 102
std::unique_ptr<Graph> IRPassManager::Apply(std::unique_ptr<Graph> graph) {
  if (passes_.empty()) {
    return graph;
  }
  PADDLE_ENFORCE(graph.get());
  // Apply all the passes
  for (const auto &pass : passes_) {
Y
Yan Chunwei 已提交
103 104 105
    if (pass->Type() != "graph_viz_pass") {
      PrettyLogEndl(Style::H2(), "--- Running IR pass [%s]", pass->Type());
    }
106 107
    graph = pass->Apply(std::move(graph));
  }
G
Gabor Buella 已提交
108
  return graph;
109 110 111
}

framework::proto::ProgramDesc IRPassManager::AcquireProgram(
N
nhzlx 已提交
112
    std::unique_ptr<Graph> *graph, ProgramDesc *program) const {
113 114 115
  auto pass =
      framework::ir::PassRegistry::Instance().Get("graph_to_program_pass");

N
nhzlx 已提交
116 117
  // Direct using ProgramDesc desc(argument->main_program()) may cause
  // incomplete copies of information.
N
nhzlx 已提交
118
  ProgramDesc desc;
N
nhzlx 已提交
119
  desc.CopyFrom(*program->Proto());
120 121 122 123 124 125
  pass->SetNotOwned("program", &desc);
  auto *the_graph = graph->release();
  *graph = pass->Apply(std::unique_ptr<Graph>(the_graph));
  return *desc.Proto();
}

126 127 128
}  // namespace analysis
}  // namespace inference
}  // namespace paddle