executor.cc 15.6 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
Q
qijun 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14

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. */

Y
Yi Wang 已提交
15
#include "paddle/fluid/framework/executor.h"
Y
Yang Yang 已提交
16

17
#include "paddle/fluid/framework/channel.h"
Y
Yi Wang 已提交
18 19 20 21 22
#include "paddle/fluid/framework/feed_fetch_method.h"
#include "paddle/fluid/framework/lod_rank_table.h"
#include "paddle/fluid/framework/lod_tensor_array.h"
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/framework/reader.h"
G
gongweibao 已提交
23
#include "paddle/fluid/operators/detail/macros.h"
Y
Yi Wang 已提交
24
#include "paddle/fluid/platform/place.h"
X
Xin Pan 已提交
25
#include "paddle/fluid/platform/profiler.h"
Y
Yang Yu 已提交
26

D
dzhwinter 已提交
27
DECLARE_bool(benchmark);
28
DEFINE_bool(use_mkldnn, false, "Use MKLDNN to run");
Q
qijun 已提交
29 30 31

namespace paddle {
namespace framework {
X
Xin Pan 已提交
32 33 34 35 36
namespace {
// block id starts from 0. This id is used to represent the codeblock
// wrapping the first block 0.
int kProgramId = -1;
}  // namespace
Q
qijun 已提交
37

Q
Qiao Longfei 已提交
38 39 40
ExecutorPrepareContext::ExecutorPrepareContext(
    const framework::ProgramDesc& prog, size_t block_id)
    : prog_(prog), block_id_(block_id) {}
Y
Yu Yang 已提交
41

Q
Qiao Longfei 已提交
42 43 44
ExecutorPrepareContext::~ExecutorPrepareContext() {
  VLOG(5) << "destroy ExecutorPrepareContext";
}
Y
Yu Yang 已提交
45

D
dzhwinter 已提交
46
Executor::Executor(const platform::Place& place) : place_(place) {}
Q
qijun 已提交
47

W
Wu Yi 已提交
48 49
#ifdef PADDLE_WITH_DISTRIBUTE
void Executor::Complete() {
G
gongweibao 已提交
50
  ::paddle::operators::distributed::RPCClient::GetInstance<RPCCLIENT_T>()
W
Wu Yi 已提交
51 52 53 54
      ->SendComplete();
}
#endif

Y
Stash  
Yu Yang 已提交
55
void InitializeVariable(Variable* var, proto::VarType::Type var_type) {
56
  if (var_type == proto::VarType::LOD_TENSOR) {
Q
QI JUN 已提交
57
    var->GetMutable<LoDTensor>();
58
  } else if (var_type == proto::VarType::SELECTED_ROWS) {
Q
QI JUN 已提交
59
    var->GetMutable<SelectedRows>();
60
  } else if (var_type == proto::VarType::FEED_MINIBATCH) {
Q
QI JUN 已提交
61
    var->GetMutable<FeedFetchList>();
62
  } else if (var_type == proto::VarType::FETCH_LIST) {
Q
QI JUN 已提交
63
    var->GetMutable<FeedFetchList>();
64
  } else if (var_type == proto::VarType::STEP_SCOPES) {
Y
Yu Yang 已提交
65
    var->GetMutable<std::vector<framework::Scope>>();
66
  } else if (var_type == proto::VarType::LOD_RANK_TABLE) {
Y
Yu Yang 已提交
67
    var->GetMutable<LoDRankTable>();
68
  } else if (var_type == proto::VarType::LOD_TENSOR_ARRAY) {
Y
Yu Yang 已提交
69
    var->GetMutable<LoDTensorArray>();
70
  } else if (var_type == proto::VarType::PLACE_LIST) {
Y
Yang Yu 已提交
71
    var->GetMutable<platform::PlaceList>();
72
  } else if (var_type == proto::VarType::READER) {
F
fengjiayi 已提交
73
    var->GetMutable<ReaderHolder>();
74 75
  } else if (var_type == proto::VarType::CHANNEL) {
    var->GetMutable<ChannelHolder>();
T
typhoonzero 已提交
76 77
  } else if (var_type == proto::VarType::RAW) {
    // GetMutable will be called in operator
Q
QI JUN 已提交
78 79
  } else {
    PADDLE_THROW(
Y
Yu Yang 已提交
80
        "Variable type %d is not in "
F
fengjiayi 已提交
81
        "[LOD_TENSOR, SELECTED_ROWS, FEED_MINIBATCH, FETCH_LIST, "
T
typhoonzero 已提交
82
        "LOD_RANK_TABLE, PLACE_LIST, READER, CHANNEL, RAW]",
Y
Yu Yang 已提交
83
        var_type);
Q
QI JUN 已提交
84 85 86
  }
}

L
Liu Yiqun 已提交
87 88 89
void Executor::CreateVariables(const ProgramDesc& pdesc, Scope* scope,
                               int block_id) {
  auto& global_block = pdesc.Block(block_id);
90 91 92 93 94 95 96 97 98 99 100 101 102 103

  const Scope* ancestor_scope = scope;
  while (ancestor_scope->parent()) {
    ancestor_scope = ancestor_scope->parent();
  }

  if (ancestor_scope != scope) {
    for (auto& var : global_block.AllVars()) {
      if (var->Name() == framework::kEmptyVarName) {
        continue;
      }

      if (var->Persistable()) {
        auto* ptr = const_cast<Scope*>(ancestor_scope)->Var(var->Name());
104
        InitializeVariable(ptr, var->GetType());
105 106 107 108
        VLOG(3) << "Create Variable " << var->Name()
                << " global, which pointer is " << ptr;
      } else {
        auto* ptr = scope->Var(var->Name());
109
        InitializeVariable(ptr, var->GetType());
110 111 112 113 114 115 116
        VLOG(3) << "Create Variable " << var->Name()
                << " locally, which pointer is " << ptr;
      }
    }
  } else {
    for (auto& var : global_block.AllVars()) {
      auto* ptr = scope->Var(var->Name());
117
      InitializeVariable(ptr, var->GetType());
118 119 120 121 122 123
      VLOG(3) << "Create variable " << var->Name() << ", which pointer is "
              << ptr;
    }
  }
}

Y
Yu Yang 已提交
124
void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id,
T
typhoonzero 已提交
125
                   bool create_local_scope, bool create_vars) {
X
Xin Pan 已提交
126
  platform::RecordBlock b(block_id);
127
  if (FLAGS_use_mkldnn) EnableMKLDNN(pdesc);
Q
Qiao Longfei 已提交
128 129
  auto ctx = Prepare(pdesc, block_id);
  RunPreparedContext(ctx.get(), scope, create_local_scope, create_vars);
Q
qijun 已提交
130 131
}

132 133 134 135 136 137 138
// Check whether the block already has feed operators and feed_holder.
// Return false if the block does not have any feed operators.
// If some feed operators have been prepended to the block, check that
// the info contained in these feed operators matches the feed_targets
// and feed_holder_name. Raise exception when any mismatch is found.
// Return true if the block has feed operators and holder of matching info.
static bool has_feed_operators(
139
    const BlockDesc& block,
L
Liu Yiqun 已提交
140
    const std::map<std::string, const LoDTensor*>& feed_targets,
141 142
    const std::string& feed_holder_name) {
  size_t feed_count = 0;
143
  for (auto* op : block.AllOps()) {
144 145
    if (op->Type() == kFeedOpType) {
      feed_count++;
L
Liu Yiqun 已提交
146
      // The input variable's name of feed_op should be feed_holder_name.
147 148 149 150 151 152 153 154 155 156 157 158 159 160 161
      PADDLE_ENFORCE_EQ(op->Input("X")[0], feed_holder_name,
                        "Input to feed op should be '%s'", feed_holder_name);
      std::string feed_target_name = op->Output("Out")[0];
      PADDLE_ENFORCE(
          feed_targets.find(feed_target_name) != feed_targets.end(),
          "Feed operator output name '%s' cannot be found in 'feed_targets'",
          feed_target_name);
    }
  }

  if (feed_count > 0) {
    PADDLE_ENFORCE_EQ(
        feed_count, feed_targets.size(),
        "The number of feed operators should match 'feed_targets'");

162
    if (!feed_holder_name.empty()) {
L
Liu Yiqun 已提交
163
      // When feed operator are present, so should be feed_holder.
164 165 166 167 168 169 170
      auto var = block.FindVar(feed_holder_name);
      PADDLE_ENFORCE_NOT_NULL(var, "Block should already have a '%s' variable",
                              feed_holder_name);
      PADDLE_ENFORCE_EQ(var->GetType(), proto::VarType::FEED_MINIBATCH,
                        "'%s' variable should be 'FEED_MINIBATCH' type",
                        feed_holder_name);
    }
171 172 173 174 175 176 177 178 179 180 181 182
  }

  return feed_count > 0;
}

// Check whether the block already has fetch operators and fetch_holder.
// Return false if the block does not have any fetch operators.
// If some fetch operators have been appended to the block, check that
// the info contained in these fetch operators matches the fetch_targets
// and fetch_holder_name. Raise exception when any mismatch is found.
// Return true if the block has fetch operators and holder of matching info.
static bool has_fetch_operators(
L
Liu Yiqun 已提交
183 184
    const BlockDesc& block,
    const std::map<std::string, LoDTensor*>& fetch_targets,
185 186
    const std::string& fetch_holder_name) {
  size_t fetch_count = 0;
187
  for (auto* op : block.AllOps()) {
188 189
    if (op->Type() == kFetchOpType) {
      fetch_count++;
L
Liu Yiqun 已提交
190
      // The output variable's name of fetch_op should be fetch_holder_name.
191 192 193 194 195 196 197 198 199 200 201 202 203 204 205
      PADDLE_ENFORCE_EQ(op->Output("Out")[0], fetch_holder_name,
                        "Output of fetch op should be '%s'", fetch_holder_name);
      std::string fetch_target_name = op->Input("X")[0];
      PADDLE_ENFORCE(
          fetch_targets.find(fetch_target_name) != fetch_targets.end(),
          "Fetch operator input name '%s' cannot be found in 'fetch_targets'",
          fetch_target_name);
    }
  }

  if (fetch_count > 0) {
    PADDLE_ENFORCE_EQ(
        fetch_count, fetch_targets.size(),
        "The number of fetch operators should match 'fetch_targets'");

206
    if (!fetch_holder_name.empty()) {
L
Liu Yiqun 已提交
207
      // When fetch operator are present, so should be fetch_holder.
208 209 210 211 212 213 214
      auto var = block.FindVar(fetch_holder_name);
      PADDLE_ENFORCE_NOT_NULL(var, "Block should already have a '%s' variable",
                              fetch_holder_name);
      PADDLE_ENFORCE_EQ(var->GetType(), proto::VarType::FETCH_LIST,
                        "'%s' variable should be 'FETCH_LIST' type",
                        fetch_holder_name);
    }
215 216 217 218 219 220
  }

  return fetch_count > 0;
}

void Executor::Run(const ProgramDesc& program, Scope* scope,
221 222
                   std::map<std::string, const LoDTensor*>* feed_targets,
                   std::map<std::string, LoDTensor*>* fetch_targets,
W
Wu Yi 已提交
223 224
                   bool create_local_scope, bool create_vars,
                   const std::string& feed_holder_name,
225
                   const std::string& fetch_holder_name) {
X
Xin Pan 已提交
226
  platform::RecordBlock b(kProgramId);
227
  if (FLAGS_use_mkldnn) EnableMKLDNN(program);
228
  bool has_feed_ops =
229
      has_feed_operators(program.Block(0), *feed_targets, feed_holder_name);
230
  bool has_fetch_ops =
231
      has_fetch_operators(program.Block(0), *fetch_targets, fetch_holder_name);
232 233

  ProgramDesc* copy_program = const_cast<ProgramDesc*>(&program);
S
sneaxiy 已提交
234
  std::unique_ptr<ProgramDesc> unique_ptr_of_copy_program;
235
  if (!has_feed_ops || !has_fetch_ops) {
S
sneaxiy 已提交
236 237
    unique_ptr_of_copy_program.reset(new ProgramDesc(program));
    copy_program = unique_ptr_of_copy_program.get();
238
  }
239 240
  auto* global_block = copy_program->MutableBlock(0);

241
  if (!has_feed_ops) {
242 243
    // create feed_holder variable
    auto* feed_holder = global_block->Var(feed_holder_name);
244
    feed_holder->SetType(proto::VarType::FEED_MINIBATCH);
245 246 247
    feed_holder->SetPersistable(true);

    int i = 0;
248
    for (auto& feed_target : (*feed_targets)) {
249 250 251 252 253 254 255 256 257 258 259 260 261 262 263
      std::string var_name = feed_target.first;
      VLOG(3) << "feed target's name: " << var_name;

      // prepend feed op
      auto* op = global_block->PrependOp();
      op->SetType(kFeedOpType);
      op->SetInput("X", {feed_holder_name});
      op->SetOutput("Out", {var_name});
      op->SetAttr("col", {static_cast<int>(i)});
      op->CheckAttrs();

      i++;
    }
  }

264
  if (!has_fetch_ops) {
265 266
    // create fetch_holder variable
    auto* fetch_holder = global_block->Var(fetch_holder_name);
267
    fetch_holder->SetType(proto::VarType::FETCH_LIST);
268 269 270
    fetch_holder->SetPersistable(true);

    int i = 0;
271
    for (auto& fetch_target : (*fetch_targets)) {
272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
      std::string var_name = fetch_target.first;
      VLOG(3) << "fetch target's name: " << var_name;

      // append fetch op
      auto* op = global_block->AppendOp();
      op->SetType(kFetchOpType);
      op->SetInput("X", {var_name});
      op->SetOutput("Out", {fetch_holder_name});
      op->SetAttr("col", {static_cast<int>(i)});
      op->CheckAttrs();

      i++;
    }
  }

287
  auto ctx = Prepare(*copy_program, 0);
W
Wu Yi 已提交
288 289 290
  RunPreparedContext(ctx.get(), scope, feed_targets, fetch_targets,
                     create_local_scope, create_vars, feed_holder_name,
                     fetch_holder_name);
291 292
}

Q
Qiao Longfei 已提交
293 294
std::unique_ptr<ExecutorPrepareContext> Executor::Prepare(
    const ProgramDesc& program, int block_id) {
Q
Qiyang Min 已提交
295 296
  std::unique_ptr<ExecutorPrepareContext> ctx(
      new ExecutorPrepareContext(program, block_id));
Y
Yu Yang 已提交
297 298 299 300 301
  PADDLE_ENFORCE_LT(static_cast<size_t>(block_id), program.Size());
  auto& block = program.Block(block_id);
  for (auto& op_desc : block.AllOps()) {
    ctx->ops_.push_back(OpRegistry::CreateOp(*op_desc));
  }
Q
Qiyang Min 已提交
302
  return ctx;
Y
Yu Yang 已提交
303 304
}

T
refine  
typhoonzero 已提交
305
std::vector<std::shared_ptr<ExecutorPrepareContext>> Executor::Prepare(
T
typhoonzero 已提交
306 307 308 309 310 311 312 313 314 315 316 317 318 319
    const ProgramDesc& program, const std::vector<int>& block_ids) {
  std::vector<std::shared_ptr<ExecutorPrepareContext>> result;
  for (auto& bid : block_ids) {
    auto* ctx = new ExecutorPrepareContext(program, bid);
    PADDLE_ENFORCE_LT(static_cast<size_t>(bid), program.Size());
    auto& block = program.Block(bid);
    for (auto& op_desc : block.AllOps()) {
      ctx->ops_.push_back(OpRegistry::CreateOp(*op_desc));
    }
    result.push_back(std::shared_ptr<ExecutorPrepareContext>(ctx));
  }
  return result;
}

Y
Yu Yang 已提交
320
void Executor::RunPreparedContext(ExecutorPrepareContext* ctx, Scope* scope,
Q
qiaolongfei 已提交
321 322
                                  bool create_local_scope, bool create_vars,
                                  bool keep_kids) {
Y
Yu Yang 已提交
323 324 325 326
  Scope* local_scope = scope;
  if (create_vars) {
    if (create_local_scope) {
      local_scope = &scope->NewScope();
327 328
    }
    CreateVariables(ctx->prog_, local_scope, ctx->block_id_);
L
Liu Yiqun 已提交
329
  }
Y
Yu Yang 已提交
330 331

  for (auto& op : ctx->ops_) {
Q
qiaolongfei 已提交
332
    VLOG(4) << place_ << " " << op->DebugStringEx(local_scope);
333
    op->Run(*local_scope, place_);
Q
qiaolongfei 已提交
334 335 336 337
    // NOTE! Please do not delete this line, it's usefull because the debug
    // string before and after op.run are different, after run the output
    // will have right shape which is usefull for debug.
    VLOG(3) << place_ << " " << op->DebugStringEx(local_scope);
Y
Yang Yang 已提交
338

Y
Yu Yang 已提交
339 340 341 342 343
    if (FLAGS_benchmark) {
      VLOG(2) << "Memory used after operator " + op->Type() + " running: "
              << memory::memory_usage(place_);
    }
  }
344
  platform::DeviceContextPool::Instance().Get(place_)->Wait();
Q
qiaolongfei 已提交
345
  if (local_scope != scope) {
Y
Yu Yang 已提交
346
    scope->DeleteScope(local_scope);
347
  } else {
Q
qiaolongfei 已提交
348 349 350 351 352
    if (!keep_kids) {
      // By default, we should delete all kid scopes after run executor because
      // some operators may create local scope when running, such as while_op.
      // But when while_op also create a local executor to run it's sub block,
      // the sub scopes it created should not be dropped immediately, because
Q
qiaolongfei 已提交
353 354
      // while_grad_op will use some variables created during while_op run, so
      // we need to keep the kids and wait for the outer executor to drop them.
Q
qiaolongfei 已提交
355 356
      scope->DropKids();
    }
Y
Yu Yang 已提交
357
  }
Q
qiaolongfei 已提交
358

Y
Yu Yang 已提交
359 360 361 362 363 364 365 366
  if (FLAGS_benchmark) {
    VLOG(2) << "-------------------------------------------------------";
    VLOG(2) << "Memory used after deleting local scope: "
            << memory::memory_usage(place_);
    VLOG(2) << "-------------------------------------------------------";
  }
}

367 368
void Executor::RunPreparedContext(
    ExecutorPrepareContext* ctx, Scope* scope,
369
    std::map<std::string, const LoDTensor*>* feed_targets,
W
Wu Yi 已提交
370 371 372
    std::map<std::string, LoDTensor*>* fetch_targets, bool create_local_scope,
    bool create_vars, const std::string& feed_holder_name,
    const std::string& fetch_holder_name) {
373 374
  auto& global_block = ctx->prog_.Block(ctx->block_id_);

375
  PADDLE_ENFORCE(
376
      has_feed_operators(global_block, *feed_targets, feed_holder_name),
377 378
      "Program in ExecutorPrepareContext should has feed_ops.");
  PADDLE_ENFORCE(
379
      has_fetch_operators(global_block, *fetch_targets, fetch_holder_name),
380 381
      "Program in the prepared context should has fetch_ops.");

382 383 384 385 386
  // map the data of feed_targets to feed_holder
  for (auto* op : global_block.AllOps()) {
    if (op->Type() == kFeedOpType) {
      std::string feed_target_name = op->Output("Out")[0];
      int idx = boost::get<int>(op->GetAttr("col"));
387 388
      SetFeedVariable(scope, *(*feed_targets)[feed_target_name],
                      feed_holder_name, idx);
389 390 391
    }
  }

W
Wu Yi 已提交
392
  RunPreparedContext(ctx, scope, create_local_scope, create_vars);
393 394 395 396 397 398

  // obtain the data of fetch_targets from fetch_holder
  for (auto* op : global_block.AllOps()) {
    if (op->Type() == kFetchOpType) {
      std::string fetch_target_name = op->Input("X")[0];
      int idx = boost::get<int>(op->GetAttr("col"));
399
      *(*fetch_targets)[fetch_target_name] =
400 401 402 403 404
          GetFetchVariable(*scope, fetch_holder_name, idx);
    }
  }
}

405 406 407 408 409 410 411 412 413 414 415
void Executor::EnableMKLDNN(const ProgramDesc& program) {
#ifdef PADDLE_WITH_MKLDNN
  VLOG(3) << "use_mkldnn=True";
  for (size_t bid = 0; bid < program.Size(); ++bid) {
    auto* block = const_cast<ProgramDesc&>(program).MutableBlock(bid);
    for (auto* op : block->AllOps()) {
      if (op->HasAttr("use_mkldnn")) {
        op->SetAttr("use_mkldnn", true);
      }
    }
  }
416 417 418
#else
  LOG(WARNING)
      << "'MKLDNN' is not supported, Please re-compile with WITH_MKLDNN option";
419 420 421
#endif
}

Q
qijun 已提交
422 423
}  // namespace framework
}  // namespace paddle