executor.cc 15.7 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"
W
Wu Yi 已提交
23
#ifdef PADDLE_WITH_DISTRIBUTE
24
#include "paddle/fluid/operators/distributed/grpc_client.h"
W
Wu Yi 已提交
25
#endif
Y
Yi Wang 已提交
26
#include "paddle/fluid/platform/place.h"
X
Xin Pan 已提交
27
#include "paddle/fluid/platform/profiler.h"
Y
Yang Yu 已提交
28

D
dzhwinter 已提交
29
DECLARE_bool(benchmark);
30
DEFINE_bool(use_mkldnn, false, "Use MKLDNN to run");
Q
qijun 已提交
31 32 33

namespace paddle {
namespace framework {
X
Xin Pan 已提交
34 35 36 37 38
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
39

Q
Qiao Longfei 已提交
40 41 42
ExecutorPrepareContext::ExecutorPrepareContext(
    const framework::ProgramDesc& prog, size_t block_id)
    : prog_(prog), block_id_(block_id) {}
Y
Yu Yang 已提交
43

Q
Qiao Longfei 已提交
44 45 46
ExecutorPrepareContext::~ExecutorPrepareContext() {
  VLOG(5) << "destroy ExecutorPrepareContext";
}
Y
Yu Yang 已提交
47

D
dzhwinter 已提交
48
Executor::Executor(const platform::Place& place) : place_(place) {}
Q
qijun 已提交
49

W
Wu Yi 已提交
50 51
#ifdef PADDLE_WITH_DISTRIBUTE
void Executor::Complete() {
52 53
  ::paddle::operators::distributed::RPCClient::GetInstance<
      ::paddle::operators::distributed::GRPCClient>()
W
Wu Yi 已提交
54 55 56 57
      ->SendComplete();
}
#endif

Y
Stash  
Yu Yang 已提交
58
void InitializeVariable(Variable* var, proto::VarType::Type var_type) {
59
  if (var_type == proto::VarType::LOD_TENSOR) {
Q
QI JUN 已提交
60
    var->GetMutable<LoDTensor>();
61
  } else if (var_type == proto::VarType::SELECTED_ROWS) {
Q
QI JUN 已提交
62
    var->GetMutable<SelectedRows>();
63
  } else if (var_type == proto::VarType::FEED_MINIBATCH) {
Q
QI JUN 已提交
64
    var->GetMutable<FeedFetchList>();
65
  } else if (var_type == proto::VarType::FETCH_LIST) {
Q
QI JUN 已提交
66
    var->GetMutable<FeedFetchList>();
67
  } else if (var_type == proto::VarType::STEP_SCOPES) {
Y
Yu Yang 已提交
68
    var->GetMutable<std::vector<framework::Scope>>();
69
  } else if (var_type == proto::VarType::LOD_RANK_TABLE) {
Y
Yu Yang 已提交
70
    var->GetMutable<LoDRankTable>();
71
  } else if (var_type == proto::VarType::LOD_TENSOR_ARRAY) {
Y
Yu Yang 已提交
72
    var->GetMutable<LoDTensorArray>();
73
  } else if (var_type == proto::VarType::PLACE_LIST) {
Y
Yang Yu 已提交
74
    var->GetMutable<platform::PlaceList>();
75
  } else if (var_type == proto::VarType::READER) {
F
fengjiayi 已提交
76
    var->GetMutable<ReaderHolder>();
77 78
  } else if (var_type == proto::VarType::CHANNEL) {
    var->GetMutable<ChannelHolder>();
T
typhoonzero 已提交
79 80
  } else if (var_type == proto::VarType::RAW) {
    // GetMutable will be called in operator
Q
QI JUN 已提交
81 82
  } else {
    PADDLE_THROW(
Y
Yu Yang 已提交
83
        "Variable type %d is not in "
F
fengjiayi 已提交
84
        "[LOD_TENSOR, SELECTED_ROWS, FEED_MINIBATCH, FETCH_LIST, "
T
typhoonzero 已提交
85
        "LOD_RANK_TABLE, PLACE_LIST, READER, CHANNEL, RAW]",
Y
Yu Yang 已提交
86
        var_type);
Q
QI JUN 已提交
87 88 89
  }
}

L
Liu Yiqun 已提交
90 91 92
void Executor::CreateVariables(const ProgramDesc& pdesc, Scope* scope,
                               int block_id) {
  auto& global_block = pdesc.Block(block_id);
93 94 95 96 97 98 99 100 101 102 103 104 105 106

  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());
107
        InitializeVariable(ptr, var->GetType());
108 109 110 111
        VLOG(3) << "Create Variable " << var->Name()
                << " global, which pointer is " << ptr;
      } else {
        auto* ptr = scope->Var(var->Name());
112
        InitializeVariable(ptr, var->GetType());
113 114 115 116 117 118 119
        VLOG(3) << "Create Variable " << var->Name()
                << " locally, which pointer is " << ptr;
      }
    }
  } else {
    for (auto& var : global_block.AllVars()) {
      auto* ptr = scope->Var(var->Name());
120
      InitializeVariable(ptr, var->GetType());
121 122 123 124 125 126
      VLOG(3) << "Create variable " << var->Name() << ", which pointer is "
              << ptr;
    }
  }
}

Y
Yu Yang 已提交
127
void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id,
T
typhoonzero 已提交
128
                   bool create_local_scope, bool create_vars) {
X
Xin Pan 已提交
129
  platform::RecordBlock b(block_id);
130
  if (FLAGS_use_mkldnn) EnableMKLDNN(pdesc);
Q
Qiao Longfei 已提交
131 132
  auto ctx = Prepare(pdesc, block_id);
  RunPreparedContext(ctx.get(), scope, create_local_scope, create_vars);
Q
qijun 已提交
133 134
}

135 136 137 138 139 140 141
// 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(
142
    const BlockDesc& block,
L
Liu Yiqun 已提交
143
    const std::map<std::string, const LoDTensor*>& feed_targets,
144 145
    const std::string& feed_holder_name) {
  size_t feed_count = 0;
146
  for (auto* op : block.AllOps()) {
147 148
    if (op->Type() == kFeedOpType) {
      feed_count++;
L
Liu Yiqun 已提交
149
      // The input variable's name of feed_op should be feed_holder_name.
150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
      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'");

165
    if (!feed_holder_name.empty()) {
L
Liu Yiqun 已提交
166
      // When feed operator are present, so should be feed_holder.
167 168 169 170 171 172 173
      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);
    }
174 175 176 177 178 179 180 181 182 183 184 185
  }

  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
186 187
    const BlockDesc& block,
    const std::map<std::string, LoDTensor*>& fetch_targets,
188 189
    const std::string& fetch_holder_name) {
  size_t fetch_count = 0;
190
  for (auto* op : block.AllOps()) {
191 192
    if (op->Type() == kFetchOpType) {
      fetch_count++;
L
Liu Yiqun 已提交
193
      // The output variable's name of fetch_op should be fetch_holder_name.
194 195 196 197 198 199 200 201 202 203 204 205 206 207 208
      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'");

209
    if (!fetch_holder_name.empty()) {
L
Liu Yiqun 已提交
210
      // When fetch operator are present, so should be fetch_holder.
211 212 213 214 215 216 217
      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);
    }
218 219 220 221 222 223
  }

  return fetch_count > 0;
}

void Executor::Run(const ProgramDesc& program, Scope* scope,
224 225
                   std::map<std::string, const LoDTensor*>* feed_targets,
                   std::map<std::string, LoDTensor*>* fetch_targets,
W
Wu Yi 已提交
226 227
                   bool create_local_scope, bool create_vars,
                   const std::string& feed_holder_name,
228
                   const std::string& fetch_holder_name) {
X
Xin Pan 已提交
229
  platform::RecordBlock b(kProgramId);
230
  if (FLAGS_use_mkldnn) EnableMKLDNN(program);
231
  bool has_feed_ops =
232
      has_feed_operators(program.Block(0), *feed_targets, feed_holder_name);
233
  bool has_fetch_ops =
234
      has_fetch_operators(program.Block(0), *fetch_targets, fetch_holder_name);
235 236

  ProgramDesc* copy_program = const_cast<ProgramDesc*>(&program);
S
sneaxiy 已提交
237
  std::unique_ptr<ProgramDesc> unique_ptr_of_copy_program;
238
  if (!has_feed_ops || !has_fetch_ops) {
S
sneaxiy 已提交
239 240
    unique_ptr_of_copy_program.reset(new ProgramDesc(program));
    copy_program = unique_ptr_of_copy_program.get();
241
  }
242 243
  auto* global_block = copy_program->MutableBlock(0);

244
  if (!has_feed_ops) {
245 246
    // create feed_holder variable
    auto* feed_holder = global_block->Var(feed_holder_name);
247
    feed_holder->SetType(proto::VarType::FEED_MINIBATCH);
248 249 250
    feed_holder->SetPersistable(true);

    int i = 0;
251
    for (auto& feed_target : (*feed_targets)) {
252 253 254 255 256 257 258 259 260 261 262 263 264 265 266
      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++;
    }
  }

267
  if (!has_fetch_ops) {
268 269
    // create fetch_holder variable
    auto* fetch_holder = global_block->Var(fetch_holder_name);
270
    fetch_holder->SetType(proto::VarType::FETCH_LIST);
271 272 273
    fetch_holder->SetPersistable(true);

    int i = 0;
274
    for (auto& fetch_target : (*fetch_targets)) {
275 276 277 278 279 280 281 282 283 284 285 286 287 288 289
      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++;
    }
  }

290
  auto ctx = Prepare(*copy_program, 0);
W
Wu Yi 已提交
291 292 293
  RunPreparedContext(ctx.get(), scope, feed_targets, fetch_targets,
                     create_local_scope, create_vars, feed_holder_name,
                     fetch_holder_name);
294 295
}

Q
Qiao Longfei 已提交
296 297
std::unique_ptr<ExecutorPrepareContext> Executor::Prepare(
    const ProgramDesc& program, int block_id) {
Q
Qiyang Min 已提交
298 299
  std::unique_ptr<ExecutorPrepareContext> ctx(
      new ExecutorPrepareContext(program, block_id));
Y
Yu Yang 已提交
300 301 302 303 304
  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
305
  return ctx;
Y
Yu Yang 已提交
306 307
}

T
refine  
typhoonzero 已提交
308
std::vector<std::shared_ptr<ExecutorPrepareContext>> Executor::Prepare(
T
typhoonzero 已提交
309 310 311 312 313 314 315 316 317 318 319 320 321 322
    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
323
void Executor::RunPreparedContext(ExecutorPrepareContext* ctx, Scope* scope,
Q
qiaolongfei 已提交
324 325
                                  bool create_local_scope, bool create_vars,
                                  bool keep_kids) {
Y
Yu Yang 已提交
326 327 328 329
  Scope* local_scope = scope;
  if (create_vars) {
    if (create_local_scope) {
      local_scope = &scope->NewScope();
330 331
    }
    CreateVariables(ctx->prog_, local_scope, ctx->block_id_);
L
Liu Yiqun 已提交
332
  }
Y
Yu Yang 已提交
333 334

  for (auto& op : ctx->ops_) {
Q
qiaolongfei 已提交
335
    VLOG(4) << place_ << " " << op->DebugStringEx(local_scope);
336
    op->Run(*local_scope, place_);
Q
qiaolongfei 已提交
337 338 339 340
    // 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
341

Y
Yu Yang 已提交
342 343 344 345 346
    if (FLAGS_benchmark) {
      VLOG(2) << "Memory used after operator " + op->Type() + " running: "
              << memory::memory_usage(place_);
    }
  }
347
  platform::DeviceContextPool::Instance().Get(place_)->Wait();
Q
qiaolongfei 已提交
348
  if (local_scope != scope) {
Y
Yu Yang 已提交
349
    scope->DeleteScope(local_scope);
350
  } else {
Q
qiaolongfei 已提交
351 352 353 354 355
    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
356 357
      // 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
358 359
      scope->DropKids();
    }
Y
Yu Yang 已提交
360
  }
Q
qiaolongfei 已提交
361

Y
Yu Yang 已提交
362 363 364 365 366 367 368 369
  if (FLAGS_benchmark) {
    VLOG(2) << "-------------------------------------------------------";
    VLOG(2) << "Memory used after deleting local scope: "
            << memory::memory_usage(place_);
    VLOG(2) << "-------------------------------------------------------";
  }
}

370 371
void Executor::RunPreparedContext(
    ExecutorPrepareContext* ctx, Scope* scope,
372
    std::map<std::string, const LoDTensor*>* feed_targets,
W
Wu Yi 已提交
373 374 375
    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) {
376 377
  auto& global_block = ctx->prog_.Block(ctx->block_id_);

378
  PADDLE_ENFORCE(
379
      has_feed_operators(global_block, *feed_targets, feed_holder_name),
380 381
      "Program in ExecutorPrepareContext should has feed_ops.");
  PADDLE_ENFORCE(
382
      has_fetch_operators(global_block, *fetch_targets, fetch_holder_name),
383 384
      "Program in the prepared context should has fetch_ops.");

385 386 387 388 389
  // 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"));
390 391
      SetFeedVariable(scope, *(*feed_targets)[feed_target_name],
                      feed_holder_name, idx);
392 393 394
    }
  }

W
Wu Yi 已提交
395
  RunPreparedContext(ctx, scope, create_local_scope, create_vars);
396 397 398 399 400 401

  // 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"));
402
      *(*fetch_targets)[fetch_target_name] =
403 404 405 406 407
          GetFetchVariable(*scope, fetch_holder_name, idx);
    }
  }
}

408 409 410 411 412 413 414 415 416 417 418
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);
      }
    }
  }
419 420 421
#else
  LOG(WARNING)
      << "'MKLDNN' is not supported, Please re-compile with WITH_MKLDNN option";
422 423 424
#endif
}

Q
qijun 已提交
425 426
}  // namespace framework
}  // namespace paddle