inference.cc 6.9 KB
Newer Older
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
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.

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 "inference.h"
#include <fstream>
#include "paddle/framework/executor.h"
#include "paddle/framework/feed_fetch_method.h"
#include "paddle/framework/init.h"
#include "paddle/framework/scope.h"

#ifdef PADDLE_USE_PTOOLS
#include "chooseser.h"
#endif

namespace paddle {

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
void InferenceEngine::LoadInferenceModel(const std::string& dirname) {
  std::string model_filename = dirname + "/__model__.dat";
  LOG(INFO) << "loading model from " << model_filename;
  std::ifstream inputfs(model_filename, std::ios::in | std::ios::binary);
  std::string program_desc_str;
  inputfs.seekg(0, std::ios::end);
  program_desc_str.resize(inputfs.tellg());
  inputfs.seekg(0, std::ios::beg);
  LOG(INFO) << "program_desc_str's size: " << program_desc_str.size();
  inputfs.read(&program_desc_str[0], program_desc_str.size());
  inputfs.close();

  program_ = new framework::ProgramDesc(program_desc_str);
  GenerateLoadProgram(dirname);

  framework::BlockDesc* global_block = program_->MutableBlock(0);
  feed_var_names_.clear();
  fetch_var_names_.clear();
  for (auto* op : global_block->AllOps()) {
    if (op->Type() == "feed") {
      feed_var_names_.insert(feed_var_names_.begin(), op->Output("Out")[0]);
    } else if (op->Type() == "fetch") {
      fetch_var_names_.push_back(op->Input("X")[0]);
    }
  }
}

55 56 57 58
void InferenceEngine::LoadInferenceModel(
    const std::string& dirname,
    const std::vector<std::string>& feed_var_names,
    const std::vector<std::string>& fetch_var_names) {
59
  std::string model_filename = dirname + "/__model__.dat";
60
  LOG(INFO) << "loading model from " << model_filename;
61
  std::ifstream inputfs(model_filename, std::ios::in | std::ios::binary);
62
  std::string program_desc_str;
63 64 65 66 67 68
  inputfs.seekg(0, std::ios::end);
  program_desc_str.resize(inputfs.tellg());
  inputfs.seekg(0, std::ios::beg);
  LOG(INFO) << "program_desc_str's size: " << program_desc_str.size();
  inputfs.read(&program_desc_str[0], program_desc_str.size());
  inputfs.close();
K
Kexin Zhao 已提交
69

70 71 72 73 74 75 76 77 78 79 80 81 82
  program_ = new framework::ProgramDesc(program_desc_str);
  GenerateLoadProgram(dirname);

  if (feed_var_names.empty() || fetch_var_names.empty()) {
    LOG(FATAL) << "Please specify the feed_var_names and fetch_var_names.";
  }
  feed_var_names_ = feed_var_names;
  fetch_var_names_ = fetch_var_names;
  PrependFeedOp();
  AppendFetchOp();
}

bool InferenceEngine::IsParameter(const framework::VarDesc* var) {
K
fix bug  
Kexin Zhao 已提交
83
  if (var->Persistable() && var->Name() != "feed" && var->Name() != "fetch") {
84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189
    // There are many unreachable variables in the program
    for (size_t i = 0; i < program_->Size(); ++i) {
      const framework::BlockDesc& block = program_->Block(i);
      for (auto* op : block.AllOps()) {
        for (auto input_argument_name : op->InputArgumentNames()) {
          if (input_argument_name == var->Name()) {
            return true;
          }
        }
      }
    }
  }
  return false;
}

void InferenceEngine::GenerateLoadProgram(const std::string& dirname) {
  framework::BlockDesc* global_block = program_->MutableBlock(0);

  load_program_ = new framework::ProgramDesc();
  framework::BlockDesc* load_block = load_program_->MutableBlock(0);
  for (auto* var : global_block->AllVars()) {
    if (IsParameter(var)) {
      LOG(INFO) << "parameter's name: " << var->Name();

      framework::VarDesc* new_var = load_block->Var(var->Name());
      new_var->SetShape(var->Shape());
      new_var->SetDataType(var->GetDataType());
      new_var->SetType(var->GetType());
      new_var->SetLoDLevel(var->GetLoDLevel());
      new_var->SetPersistable(true);

      // append_op
      framework::OpDesc* op = load_block->AppendOp();
      op->SetType("load");
      op->SetOutput("Out", {new_var->Name()});
      op->SetAttr("file_path", {dirname + "/" + new_var->Name()});
      op->CheckAttrs();
    }
  }
}

void InferenceEngine::PrependFeedOp() {
  if (!program_) {
    LOG(FATAL) << "Please initialize the program_ first.";
  }

  framework::BlockDesc* global_block = program_->MutableBlock(0);

  // create_var
  framework::VarDesc* feed_var = global_block->Var("feed");
  feed_var->SetType(framework::proto::VarDesc::FEED_MINIBATCH);
  feed_var->SetPersistable(true);

  // prepend feed_op
  for (size_t i = 0; i < feed_var_names_.size(); ++i) {
    std::string var_name = feed_var_names_[i];
    LOG(INFO) << "feed var's name: " << var_name;

    // prepend_op
    framework::OpDesc* op = global_block->PrependOp();
    op->SetType("feed");
    op->SetInput("X", {"feed"});
    op->SetOutput("Out", {var_name});
    op->SetAttr("col", {static_cast<int>(i)});
    op->CheckAttrs();
  }
}

void InferenceEngine::AppendFetchOp() {
  if (!program_) {
    LOG(FATAL) << "Please initialize the program_ first.";
  }

  framework::BlockDesc* global_block = program_->MutableBlock(0);

  // create_var
  framework::VarDesc* fetch_var = global_block->Var("fetch");
  fetch_var->SetType(framework::proto::VarDesc::FETCH_LIST);
  fetch_var->SetPersistable(true);

  // append fetch_op
  for (size_t i = 0; i < fetch_var_names_.size(); ++i) {
    std::string var_name = fetch_var_names_[i];
    LOG(INFO) << "fetch var's name: " << var_name;

    // append_op
    framework::OpDesc* op = global_block->AppendOp();
    op->SetType("fetch");
    op->SetInput("X", {var_name});
    op->SetOutput("Out", {"fetch"});
    op->SetAttr("col", {static_cast<int>(i)});
    op->CheckAttrs();
  }
}

void InferenceEngine::Execute(const std::vector<framework::LoDTensor>& feeds,
                              std::vector<framework::LoDTensor>& fetchs) {
  if (!program_ || !load_program_) {
    LOG(FATAL) << "Please initialize the program_ and load_program_ first.";
  }

  if (feeds.size() < feed_var_names_.size()) {
    LOG(FATAL) << "Please feed " << feed_var_names_.size() << " input Tensors.";
  }

  auto* place = new platform::CPUPlace();
190
  framework::InitDevices();
191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213
  framework::Executor* executor = new framework::Executor(*place);
  framework::Scope* scope = new framework::Scope();

  executor->Run(*load_program_, scope, 0, true, true);

  // set_feed_variable
  for (size_t i = 0; i < feed_var_names_.size(); ++i) {
    framework::SetFeedVariable(scope, feeds[i], "feed", i);
  }

  executor->Run(*program_, scope, 0, true, true);

  // get_fetch_variable
  fetchs.resize(fetch_var_names_.size());
  for (size_t i = 0; i < fetch_var_names_.size(); ++i) {
    fetchs[i] = framework::GetFetchVariable(*scope, "fetch", i);
  }

  delete place;
  delete scope;
  delete executor;
}
}  // namespace paddle