executor.cc 15.9 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
#ifdef PADDLE_WITH_DISTRIBUTE
Y
Yancey1989 已提交
51 52 53 54 55 56 57 58 59 60 61 62 63 64
void Executor::BeginPass() {
  auto client = ::paddle::operators::distributed::RPCClient::GetInstance<
      ::paddle::operators::distributed::GRPCClient>();

  client->SendBeginPass();
  client->Wait();
}

void Executor::EndPass() {
  auto client = ::paddle::operators::distributed::RPCClient::GetInstance<
      ::paddle::operators::distributed::GRPCClient>();

  client->SendEndPass();
  client->Wait();
W
Wu Yi 已提交
65 66 67
}
#endif

Y
Stash  
Yu Yang 已提交
68
void InitializeVariable(Variable* var, proto::VarType::Type var_type) {
69
  if (var_type == proto::VarType::LOD_TENSOR) {
Q
QI JUN 已提交
70
    var->GetMutable<LoDTensor>();
71
  } else if (var_type == proto::VarType::SELECTED_ROWS) {
Q
QI JUN 已提交
72
    var->GetMutable<SelectedRows>();
73
  } else if (var_type == proto::VarType::FEED_MINIBATCH) {
Q
QI JUN 已提交
74
    var->GetMutable<FeedFetchList>();
75
  } else if (var_type == proto::VarType::FETCH_LIST) {
Q
QI JUN 已提交
76
    var->GetMutable<FeedFetchList>();
77
  } else if (var_type == proto::VarType::STEP_SCOPES) {
Y
Yu Yang 已提交
78
    var->GetMutable<std::vector<framework::Scope>>();
79
  } else if (var_type == proto::VarType::LOD_RANK_TABLE) {
Y
Yu Yang 已提交
80
    var->GetMutable<LoDRankTable>();
81
  } else if (var_type == proto::VarType::LOD_TENSOR_ARRAY) {
Y
Yu Yang 已提交
82
    var->GetMutable<LoDTensorArray>();
83
  } else if (var_type == proto::VarType::PLACE_LIST) {
Y
Yang Yu 已提交
84
    var->GetMutable<platform::PlaceList>();
85
  } else if (var_type == proto::VarType::READER) {
F
fengjiayi 已提交
86
    var->GetMutable<ReaderHolder>();
87 88
  } else if (var_type == proto::VarType::CHANNEL) {
    var->GetMutable<ChannelHolder>();
T
typhoonzero 已提交
89 90
  } else if (var_type == proto::VarType::RAW) {
    // GetMutable will be called in operator
Q
QI JUN 已提交
91 92
  } else {
    PADDLE_THROW(
Y
Yu Yang 已提交
93
        "Variable type %d is not in "
F
fengjiayi 已提交
94
        "[LOD_TENSOR, SELECTED_ROWS, FEED_MINIBATCH, FETCH_LIST, "
T
typhoonzero 已提交
95
        "LOD_RANK_TABLE, PLACE_LIST, READER, CHANNEL, RAW]",
Y
Yu Yang 已提交
96
        var_type);
Q
QI JUN 已提交
97 98 99
  }
}

L
Liu Yiqun 已提交
100 101 102
void Executor::CreateVariables(const ProgramDesc& pdesc, Scope* scope,
                               int block_id) {
  auto& global_block = pdesc.Block(block_id);
103 104 105 106 107 108 109 110 111 112 113 114 115 116

  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());
117
        InitializeVariable(ptr, var->GetType());
118 119 120 121
        VLOG(3) << "Create Variable " << var->Name()
                << " global, which pointer is " << ptr;
      } else {
        auto* ptr = scope->Var(var->Name());
122
        InitializeVariable(ptr, var->GetType());
123 124 125 126 127 128 129
        VLOG(3) << "Create Variable " << var->Name()
                << " locally, which pointer is " << ptr;
      }
    }
  } else {
    for (auto& var : global_block.AllVars()) {
      auto* ptr = scope->Var(var->Name());
130
      InitializeVariable(ptr, var->GetType());
131 132 133 134 135 136
      VLOG(3) << "Create variable " << var->Name() << ", which pointer is "
              << ptr;
    }
  }
}

Y
Yu Yang 已提交
137
void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id,
T
typhoonzero 已提交
138
                   bool create_local_scope, bool create_vars) {
X
Xin Pan 已提交
139
  platform::RecordBlock b(block_id);
140
  if (FLAGS_use_mkldnn) EnableMKLDNN(pdesc);
Q
Qiao Longfei 已提交
141 142
  auto ctx = Prepare(pdesc, block_id);
  RunPreparedContext(ctx.get(), scope, create_local_scope, create_vars);
Q
qijun 已提交
143 144
}

145 146 147 148 149 150 151
// 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(
152
    const BlockDesc& block,
L
Liu Yiqun 已提交
153
    const std::map<std::string, const LoDTensor*>& feed_targets,
154 155
    const std::string& feed_holder_name) {
  size_t feed_count = 0;
156
  for (auto* op : block.AllOps()) {
157 158
    if (op->Type() == kFeedOpType) {
      feed_count++;
L
Liu Yiqun 已提交
159
      // The input variable's name of feed_op should be feed_holder_name.
160 161 162 163 164 165 166 167 168 169 170 171 172 173 174
      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'");

175
    if (!feed_holder_name.empty()) {
L
Liu Yiqun 已提交
176
      // When feed operator are present, so should be feed_holder.
177 178 179 180 181 182 183
      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);
    }
184 185 186 187 188 189 190 191 192 193 194 195
  }

  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
196 197
    const BlockDesc& block,
    const std::map<std::string, LoDTensor*>& fetch_targets,
198 199
    const std::string& fetch_holder_name) {
  size_t fetch_count = 0;
200
  for (auto* op : block.AllOps()) {
201 202
    if (op->Type() == kFetchOpType) {
      fetch_count++;
L
Liu Yiqun 已提交
203
      // The output variable's name of fetch_op should be fetch_holder_name.
204 205 206 207 208 209 210 211 212 213 214 215 216 217 218
      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'");

219
    if (!fetch_holder_name.empty()) {
L
Liu Yiqun 已提交
220
      // When fetch operator are present, so should be fetch_holder.
221 222 223 224 225 226 227
      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);
    }
228 229 230 231 232 233
  }

  return fetch_count > 0;
}

void Executor::Run(const ProgramDesc& program, Scope* scope,
234 235
                   std::map<std::string, const LoDTensor*>* feed_targets,
                   std::map<std::string, LoDTensor*>* fetch_targets,
W
Wu Yi 已提交
236 237
                   bool create_local_scope, bool create_vars,
                   const std::string& feed_holder_name,
238
                   const std::string& fetch_holder_name) {
X
Xin Pan 已提交
239
  platform::RecordBlock b(kProgramId);
240
  if (FLAGS_use_mkldnn) EnableMKLDNN(program);
241
  bool has_feed_ops =
242
      has_feed_operators(program.Block(0), *feed_targets, feed_holder_name);
243
  bool has_fetch_ops =
244
      has_fetch_operators(program.Block(0), *fetch_targets, fetch_holder_name);
245 246

  ProgramDesc* copy_program = const_cast<ProgramDesc*>(&program);
S
sneaxiy 已提交
247
  std::unique_ptr<ProgramDesc> unique_ptr_of_copy_program;
248
  if (!has_feed_ops || !has_fetch_ops) {
S
sneaxiy 已提交
249 250
    unique_ptr_of_copy_program.reset(new ProgramDesc(program));
    copy_program = unique_ptr_of_copy_program.get();
251
  }
252 253
  auto* global_block = copy_program->MutableBlock(0);

254
  if (!has_feed_ops) {
255 256
    // create feed_holder variable
    auto* feed_holder = global_block->Var(feed_holder_name);
257
    feed_holder->SetType(proto::VarType::FEED_MINIBATCH);
258 259 260
    feed_holder->SetPersistable(true);

    int i = 0;
261
    for (auto& feed_target : (*feed_targets)) {
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276
      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++;
    }
  }

277
  if (!has_fetch_ops) {
278 279
    // create fetch_holder variable
    auto* fetch_holder = global_block->Var(fetch_holder_name);
280
    fetch_holder->SetType(proto::VarType::FETCH_LIST);
281 282 283
    fetch_holder->SetPersistable(true);

    int i = 0;
284
    for (auto& fetch_target : (*fetch_targets)) {
285 286 287 288 289 290 291 292 293 294 295 296 297 298 299
      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++;
    }
  }

300
  auto ctx = Prepare(*copy_program, 0);
W
Wu Yi 已提交
301 302 303
  RunPreparedContext(ctx.get(), scope, feed_targets, fetch_targets,
                     create_local_scope, create_vars, feed_holder_name,
                     fetch_holder_name);
304 305
}

Q
Qiao Longfei 已提交
306 307
std::unique_ptr<ExecutorPrepareContext> Executor::Prepare(
    const ProgramDesc& program, int block_id) {
Q
Qiyang Min 已提交
308 309
  std::unique_ptr<ExecutorPrepareContext> ctx(
      new ExecutorPrepareContext(program, block_id));
Y
Yu Yang 已提交
310 311 312 313 314
  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
315
  return ctx;
Y
Yu Yang 已提交
316 317
}

T
refine  
typhoonzero 已提交
318
std::vector<std::shared_ptr<ExecutorPrepareContext>> Executor::Prepare(
T
typhoonzero 已提交
319 320 321 322 323 324 325 326 327 328 329 330 331 332
    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
333
void Executor::RunPreparedContext(ExecutorPrepareContext* ctx, Scope* scope,
Q
qiaolongfei 已提交
334 335
                                  bool create_local_scope, bool create_vars,
                                  bool keep_kids) {
Y
Yu Yang 已提交
336 337 338 339
  Scope* local_scope = scope;
  if (create_vars) {
    if (create_local_scope) {
      local_scope = &scope->NewScope();
340 341
    }
    CreateVariables(ctx->prog_, local_scope, ctx->block_id_);
L
Liu Yiqun 已提交
342
  }
Y
Yu Yang 已提交
343 344

  for (auto& op : ctx->ops_) {
Q
qiaolongfei 已提交
345
    VLOG(4) << place_ << " " << op->DebugStringEx(local_scope);
346
    op->Run(*local_scope, place_);
Q
qiaolongfei 已提交
347 348 349 350
    // 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
351

Y
Yu Yang 已提交
352 353 354 355 356
    if (FLAGS_benchmark) {
      VLOG(2) << "Memory used after operator " + op->Type() + " running: "
              << memory::memory_usage(place_);
    }
  }
357
  platform::DeviceContextPool::Instance().Get(place_)->Wait();
Q
qiaolongfei 已提交
358
  if (local_scope != scope) {
Y
Yu Yang 已提交
359
    scope->DeleteScope(local_scope);
360
  } else {
Q
qiaolongfei 已提交
361 362 363 364 365
    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
366 367
      // 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
368 369
      scope->DropKids();
    }
Y
Yu Yang 已提交
370
  }
Q
qiaolongfei 已提交
371

Y
Yu Yang 已提交
372 373 374 375 376 377 378 379
  if (FLAGS_benchmark) {
    VLOG(2) << "-------------------------------------------------------";
    VLOG(2) << "Memory used after deleting local scope: "
            << memory::memory_usage(place_);
    VLOG(2) << "-------------------------------------------------------";
  }
}

380 381
void Executor::RunPreparedContext(
    ExecutorPrepareContext* ctx, Scope* scope,
382
    std::map<std::string, const LoDTensor*>* feed_targets,
W
Wu Yi 已提交
383 384 385
    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) {
386 387
  auto& global_block = ctx->prog_.Block(ctx->block_id_);

388
  PADDLE_ENFORCE(
389
      has_feed_operators(global_block, *feed_targets, feed_holder_name),
390 391
      "Program in ExecutorPrepareContext should has feed_ops.");
  PADDLE_ENFORCE(
392
      has_fetch_operators(global_block, *fetch_targets, fetch_holder_name),
393 394
      "Program in the prepared context should has fetch_ops.");

395 396 397 398 399
  // 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"));
400 401
      SetFeedVariable(scope, *(*feed_targets)[feed_target_name],
                      feed_holder_name, idx);
402 403 404
    }
  }

W
Wu Yi 已提交
405
  RunPreparedContext(ctx, scope, create_local_scope, create_vars);
406 407 408 409 410 411

  // 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"));
412
      *(*fetch_targets)[fetch_target_name] =
413 414 415 416 417
          GetFetchVariable(*scope, fetch_holder_name, idx);
    }
  }
}

418 419 420 421 422 423 424 425 426 427 428
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);
      }
    }
  }
429 430 431
#else
  LOG(WARNING)
      << "'MKLDNN' is not supported, Please re-compile with WITH_MKLDNN option";
432 433 434
#endif
}

Q
qijun 已提交
435 436
}  // namespace framework
}  // namespace paddle