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"
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
#ifdef PADDLE_WITH_DISTRIBUTE
Y
Yancey1989 已提交
49
void Executor::Complete() {
Y
Yancey1989 已提交
50 51
  ::paddle::operators::distributed::RPCClient::GetInstance<
      ::paddle::operators::distributed::GRPCClient>()
Y
Yancey1989 已提交
52
      ->SendComplete();
W
Wu Yi 已提交
53 54 55
}
#endif

Y
Stash  
Yu Yang 已提交
56
void InitializeVariable(Variable* var, proto::VarType::Type var_type) {
57
  if (var_type == proto::VarType::LOD_TENSOR) {
Q
QI JUN 已提交
58
    var->GetMutable<LoDTensor>();
59
  } else if (var_type == proto::VarType::SELECTED_ROWS) {
Q
QI JUN 已提交
60
    var->GetMutable<SelectedRows>();
61
  } else if (var_type == proto::VarType::FEED_MINIBATCH) {
Q
QI JUN 已提交
62
    var->GetMutable<FeedFetchList>();
63
  } else if (var_type == proto::VarType::FETCH_LIST) {
Q
QI JUN 已提交
64
    var->GetMutable<FeedFetchList>();
65
  } else if (var_type == proto::VarType::STEP_SCOPES) {
Y
Yu Yang 已提交
66
    var->GetMutable<std::vector<framework::Scope>>();
67
  } else if (var_type == proto::VarType::LOD_RANK_TABLE) {
Y
Yu Yang 已提交
68
    var->GetMutable<LoDRankTable>();
69
  } else if (var_type == proto::VarType::LOD_TENSOR_ARRAY) {
Y
Yu Yang 已提交
70
    var->GetMutable<LoDTensorArray>();
71
  } else if (var_type == proto::VarType::PLACE_LIST) {
Y
Yang Yu 已提交
72
    var->GetMutable<platform::PlaceList>();
73
  } else if (var_type == proto::VarType::READER) {
F
fengjiayi 已提交
74
    var->GetMutable<ReaderHolder>();
75 76
  } else if (var_type == proto::VarType::CHANNEL) {
    var->GetMutable<ChannelHolder>();
T
typhoonzero 已提交
77 78
  } else if (var_type == proto::VarType::RAW) {
    // GetMutable will be called in operator
Q
QI JUN 已提交
79 80
  } else {
    PADDLE_THROW(
Y
Yu Yang 已提交
81
        "Variable type %d is not in "
F
fengjiayi 已提交
82
        "[LOD_TENSOR, SELECTED_ROWS, FEED_MINIBATCH, FETCH_LIST, "
T
typhoonzero 已提交
83
        "LOD_RANK_TABLE, PLACE_LIST, READER, CHANNEL, RAW]",
Y
Yu Yang 已提交
84
        var_type);
Q
QI JUN 已提交
85 86 87
  }
}

L
Liu Yiqun 已提交
88 89 90
void Executor::CreateVariables(const ProgramDesc& pdesc, Scope* scope,
                               int block_id) {
  auto& global_block = pdesc.Block(block_id);
91 92 93 94 95 96 97 98 99 100 101 102 103 104

  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());
105
        InitializeVariable(ptr, var->GetType());
106 107 108 109
        VLOG(3) << "Create Variable " << var->Name()
                << " global, which pointer is " << ptr;
      } else {
        auto* ptr = scope->Var(var->Name());
110
        InitializeVariable(ptr, var->GetType());
111 112 113 114 115 116 117
        VLOG(3) << "Create Variable " << var->Name()
                << " locally, which pointer is " << ptr;
      }
    }
  } else {
    for (auto& var : global_block.AllVars()) {
      auto* ptr = scope->Var(var->Name());
118
      InitializeVariable(ptr, var->GetType());
119 120 121 122 123 124
      VLOG(3) << "Create variable " << var->Name() << ", which pointer is "
              << ptr;
    }
  }
}

Y
Yu Yang 已提交
125
void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id,
T
typhoonzero 已提交
126
                   bool create_local_scope, bool create_vars) {
X
Xin Pan 已提交
127
  platform::RecordBlock b(block_id);
128
  if (FLAGS_use_mkldnn) EnableMKLDNN(pdesc);
Q
Qiao Longfei 已提交
129 130
  auto ctx = Prepare(pdesc, block_id);
  RunPreparedContext(ctx.get(), scope, create_local_scope, create_vars);
Q
qijun 已提交
131 132
}

133 134 135 136 137 138 139
// 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(
140
    const BlockDesc& block,
L
Liu Yiqun 已提交
141
    const std::map<std::string, const LoDTensor*>& feed_targets,
142 143
    const std::string& feed_holder_name) {
  size_t feed_count = 0;
144
  for (auto* op : block.AllOps()) {
145 146
    if (op->Type() == kFeedOpType) {
      feed_count++;
L
Liu Yiqun 已提交
147
      // The input variable's name of feed_op should be feed_holder_name.
148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
      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'");

163
    if (!feed_holder_name.empty()) {
L
Liu Yiqun 已提交
164
      // When feed operator are present, so should be feed_holder.
165 166 167 168 169 170 171
      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);
    }
172 173 174 175 176 177 178 179 180 181 182 183
  }

  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
184 185
    const BlockDesc& block,
    const std::map<std::string, LoDTensor*>& fetch_targets,
186 187
    const std::string& fetch_holder_name) {
  size_t fetch_count = 0;
188
  for (auto* op : block.AllOps()) {
189 190
    if (op->Type() == kFetchOpType) {
      fetch_count++;
L
Liu Yiqun 已提交
191
      // The output variable's name of fetch_op should be fetch_holder_name.
192 193 194 195 196 197 198 199 200 201 202 203 204 205 206
      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'");

207
    if (!fetch_holder_name.empty()) {
L
Liu Yiqun 已提交
208
      // When fetch operator are present, so should be fetch_holder.
209 210 211 212 213 214 215
      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);
    }
216 217 218 219 220 221
  }

  return fetch_count > 0;
}

void Executor::Run(const ProgramDesc& program, Scope* scope,
222 223
                   std::map<std::string, const LoDTensor*>* feed_targets,
                   std::map<std::string, LoDTensor*>* fetch_targets,
W
Wu Yi 已提交
224 225
                   bool create_local_scope, bool create_vars,
                   const std::string& feed_holder_name,
226
                   const std::string& fetch_holder_name) {
X
Xin Pan 已提交
227
  platform::RecordBlock b(kProgramId);
228
  if (FLAGS_use_mkldnn) EnableMKLDNN(program);
229
  bool has_feed_ops =
230
      has_feed_operators(program.Block(0), *feed_targets, feed_holder_name);
231
  bool has_fetch_ops =
232
      has_fetch_operators(program.Block(0), *fetch_targets, fetch_holder_name);
233 234

  ProgramDesc* copy_program = const_cast<ProgramDesc*>(&program);
S
sneaxiy 已提交
235
  std::unique_ptr<ProgramDesc> unique_ptr_of_copy_program;
236
  if (!has_feed_ops || !has_fetch_ops) {
S
sneaxiy 已提交
237 238
    unique_ptr_of_copy_program.reset(new ProgramDesc(program));
    copy_program = unique_ptr_of_copy_program.get();
239
  }
240 241
  auto* global_block = copy_program->MutableBlock(0);

242
  if (!has_feed_ops) {
243 244
    // create feed_holder variable
    auto* feed_holder = global_block->Var(feed_holder_name);
245
    feed_holder->SetType(proto::VarType::FEED_MINIBATCH);
246 247 248
    feed_holder->SetPersistable(true);

    int i = 0;
249
    for (auto& feed_target : (*feed_targets)) {
250 251 252 253 254 255 256 257 258 259 260 261 262 263 264
      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++;
    }
  }

265
  if (!has_fetch_ops) {
266 267
    // create fetch_holder variable
    auto* fetch_holder = global_block->Var(fetch_holder_name);
268
    fetch_holder->SetType(proto::VarType::FETCH_LIST);
269 270 271
    fetch_holder->SetPersistable(true);

    int i = 0;
272
    for (auto& fetch_target : (*fetch_targets)) {
273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
      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++;
    }
  }

288
  auto ctx = Prepare(*copy_program, 0);
W
Wu Yi 已提交
289 290 291
  RunPreparedContext(ctx.get(), scope, feed_targets, fetch_targets,
                     create_local_scope, create_vars, feed_holder_name,
                     fetch_holder_name);
292 293
}

Q
Qiao Longfei 已提交
294 295
std::unique_ptr<ExecutorPrepareContext> Executor::Prepare(
    const ProgramDesc& program, int block_id) {
Q
Qiyang Min 已提交
296 297
  std::unique_ptr<ExecutorPrepareContext> ctx(
      new ExecutorPrepareContext(program, block_id));
Y
Yu Yang 已提交
298 299 300 301 302
  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
303
  return ctx;
Y
Yu Yang 已提交
304 305
}

T
refine  
typhoonzero 已提交
306
std::vector<std::shared_ptr<ExecutorPrepareContext>> Executor::Prepare(
T
typhoonzero 已提交
307 308 309 310 311 312 313 314 315 316 317 318 319 320
    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
321
void Executor::RunPreparedContext(ExecutorPrepareContext* ctx, Scope* scope,
Q
qiaolongfei 已提交
322 323
                                  bool create_local_scope, bool create_vars,
                                  bool keep_kids) {
Y
Yu Yang 已提交
324 325 326 327
  Scope* local_scope = scope;
  if (create_vars) {
    if (create_local_scope) {
      local_scope = &scope->NewScope();
328 329
    }
    CreateVariables(ctx->prog_, local_scope, ctx->block_id_);
L
Liu Yiqun 已提交
330
  }
Y
Yu Yang 已提交
331 332

  for (auto& op : ctx->ops_) {
Q
qiaolongfei 已提交
333
    VLOG(4) << place_ << " " << op->DebugStringEx(local_scope);
334
    op->Run(*local_scope, place_);
Q
qiaolongfei 已提交
335 336 337 338
    // 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
339

Y
Yu Yang 已提交
340 341 342 343 344
    if (FLAGS_benchmark) {
      VLOG(2) << "Memory used after operator " + op->Type() + " running: "
              << memory::memory_usage(place_);
    }
  }
345
  platform::DeviceContextPool::Instance().Get(place_)->Wait();
Q
qiaolongfei 已提交
346
  if (local_scope != scope) {
Y
Yu Yang 已提交
347
    scope->DeleteScope(local_scope);
348
  } else {
Q
qiaolongfei 已提交
349 350 351 352 353
    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
354 355
      // 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
356 357
      scope->DropKids();
    }
Y
Yu Yang 已提交
358
  }
Q
qiaolongfei 已提交
359

Y
Yu Yang 已提交
360 361 362 363 364 365 366 367
  if (FLAGS_benchmark) {
    VLOG(2) << "-------------------------------------------------------";
    VLOG(2) << "Memory used after deleting local scope: "
            << memory::memory_usage(place_);
    VLOG(2) << "-------------------------------------------------------";
  }
}

368 369
void Executor::RunPreparedContext(
    ExecutorPrepareContext* ctx, Scope* scope,
370
    std::map<std::string, const LoDTensor*>* feed_targets,
W
Wu Yi 已提交
371 372 373
    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) {
374 375
  auto& global_block = ctx->prog_.Block(ctx->block_id_);

376
  PADDLE_ENFORCE(
377
      has_feed_operators(global_block, *feed_targets, feed_holder_name),
378 379
      "Program in ExecutorPrepareContext should has feed_ops.");
  PADDLE_ENFORCE(
380
      has_fetch_operators(global_block, *fetch_targets, fetch_holder_name),
381 382
      "Program in the prepared context should has fetch_ops.");

383 384 385 386 387
  // 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"));
388 389
      SetFeedVariable(scope, *(*feed_targets)[feed_target_name],
                      feed_holder_name, idx);
390 391 392
    }
  }

W
Wu Yi 已提交
393
  RunPreparedContext(ctx, scope, create_local_scope, create_vars);
394 395 396 397 398 399

  // 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"));
400
      *(*fetch_targets)[fetch_target_name] =
401 402 403 404 405
          GetFetchVariable(*scope, fetch_holder_name, idx);
    }
  }
}

406 407 408 409 410 411 412 413 414 415 416
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);
      }
    }
  }
417 418 419
#else
  LOG(WARNING)
      << "'MKLDNN' is not supported, Please re-compile with WITH_MKLDNN option";
420 421 422
#endif
}

Q
qijun 已提交
423 424
}  // namespace framework
}  // namespace paddle