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"
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 50 51 52 53 54 55 56 57 58 59 60 61 62
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
63 64 65
}
#endif

Y
Stash  
Yu Yang 已提交
66
void InitializeVariable(Variable* var, proto::VarType::Type var_type) {
67
  if (var_type == proto::VarType::LOD_TENSOR) {
Q
QI JUN 已提交
68
    var->GetMutable<LoDTensor>();
69
  } else if (var_type == proto::VarType::SELECTED_ROWS) {
Q
QI JUN 已提交
70
    var->GetMutable<SelectedRows>();
71
  } else if (var_type == proto::VarType::FEED_MINIBATCH) {
Q
QI JUN 已提交
72
    var->GetMutable<FeedFetchList>();
73
  } else if (var_type == proto::VarType::FETCH_LIST) {
Q
QI JUN 已提交
74
    var->GetMutable<FeedFetchList>();
75
  } else if (var_type == proto::VarType::STEP_SCOPES) {
Y
Yu Yang 已提交
76
    var->GetMutable<std::vector<framework::Scope>>();
77
  } else if (var_type == proto::VarType::LOD_RANK_TABLE) {
Y
Yu Yang 已提交
78
    var->GetMutable<LoDRankTable>();
79
  } else if (var_type == proto::VarType::LOD_TENSOR_ARRAY) {
Y
Yu Yang 已提交
80
    var->GetMutable<LoDTensorArray>();
81
  } else if (var_type == proto::VarType::PLACE_LIST) {
Y
Yang Yu 已提交
82
    var->GetMutable<platform::PlaceList>();
83
  } else if (var_type == proto::VarType::READER) {
F
fengjiayi 已提交
84
    var->GetMutable<ReaderHolder>();
85 86
  } else if (var_type == proto::VarType::CHANNEL) {
    var->GetMutable<ChannelHolder>();
T
typhoonzero 已提交
87 88
  } else if (var_type == proto::VarType::RAW) {
    // GetMutable will be called in operator
Q
QI JUN 已提交
89 90
  } else {
    PADDLE_THROW(
Y
Yu Yang 已提交
91
        "Variable type %d is not in "
F
fengjiayi 已提交
92
        "[LOD_TENSOR, SELECTED_ROWS, FEED_MINIBATCH, FETCH_LIST, "
T
typhoonzero 已提交
93
        "LOD_RANK_TABLE, PLACE_LIST, READER, CHANNEL, RAW]",
Y
Yu Yang 已提交
94
        var_type);
Q
QI JUN 已提交
95 96 97
  }
}

L
Liu Yiqun 已提交
98 99 100
void Executor::CreateVariables(const ProgramDesc& pdesc, Scope* scope,
                               int block_id) {
  auto& global_block = pdesc.Block(block_id);
101 102 103 104 105 106 107 108 109 110 111 112 113 114

  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());
115
        InitializeVariable(ptr, var->GetType());
116 117 118 119
        VLOG(3) << "Create Variable " << var->Name()
                << " global, which pointer is " << ptr;
      } else {
        auto* ptr = scope->Var(var->Name());
120
        InitializeVariable(ptr, var->GetType());
121 122 123 124 125 126 127
        VLOG(3) << "Create Variable " << var->Name()
                << " locally, which pointer is " << ptr;
      }
    }
  } else {
    for (auto& var : global_block.AllVars()) {
      auto* ptr = scope->Var(var->Name());
128
      InitializeVariable(ptr, var->GetType());
129 130 131 132 133 134
      VLOG(3) << "Create variable " << var->Name() << ", which pointer is "
              << ptr;
    }
  }
}

Y
Yu Yang 已提交
135
void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id,
T
typhoonzero 已提交
136
                   bool create_local_scope, bool create_vars) {
X
Xin Pan 已提交
137
  platform::RecordBlock b(block_id);
138
  if (FLAGS_use_mkldnn) EnableMKLDNN(pdesc);
Q
Qiao Longfei 已提交
139 140
  auto ctx = Prepare(pdesc, block_id);
  RunPreparedContext(ctx.get(), scope, create_local_scope, create_vars);
Q
qijun 已提交
141 142
}

143 144 145 146 147 148 149
// 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(
150
    const BlockDesc& block,
L
Liu Yiqun 已提交
151
    const std::map<std::string, const LoDTensor*>& feed_targets,
152 153
    const std::string& feed_holder_name) {
  size_t feed_count = 0;
154
  for (auto* op : block.AllOps()) {
155 156
    if (op->Type() == kFeedOpType) {
      feed_count++;
L
Liu Yiqun 已提交
157
      // The input variable's name of feed_op should be feed_holder_name.
158 159 160 161 162 163 164 165 166 167 168 169 170 171 172
      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'");

173
    if (!feed_holder_name.empty()) {
L
Liu Yiqun 已提交
174
      // When feed operator are present, so should be feed_holder.
175 176 177 178 179 180 181
      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);
    }
182 183 184 185 186 187 188 189 190 191 192 193
  }

  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
194 195
    const BlockDesc& block,
    const std::map<std::string, LoDTensor*>& fetch_targets,
196 197
    const std::string& fetch_holder_name) {
  size_t fetch_count = 0;
198
  for (auto* op : block.AllOps()) {
199 200
    if (op->Type() == kFetchOpType) {
      fetch_count++;
L
Liu Yiqun 已提交
201
      // The output variable's name of fetch_op should be fetch_holder_name.
202 203 204 205 206 207 208 209 210 211 212 213 214 215 216
      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'");

217
    if (!fetch_holder_name.empty()) {
L
Liu Yiqun 已提交
218
      // When fetch operator are present, so should be fetch_holder.
219 220 221 222 223 224 225
      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);
    }
226 227 228 229 230 231
  }

  return fetch_count > 0;
}

void Executor::Run(const ProgramDesc& program, Scope* scope,
232 233
                   std::map<std::string, const LoDTensor*>* feed_targets,
                   std::map<std::string, LoDTensor*>* fetch_targets,
W
Wu Yi 已提交
234 235
                   bool create_local_scope, bool create_vars,
                   const std::string& feed_holder_name,
236
                   const std::string& fetch_holder_name) {
X
Xin Pan 已提交
237
  platform::RecordBlock b(kProgramId);
238
  if (FLAGS_use_mkldnn) EnableMKLDNN(program);
239
  bool has_feed_ops =
240
      has_feed_operators(program.Block(0), *feed_targets, feed_holder_name);
241
  bool has_fetch_ops =
242
      has_fetch_operators(program.Block(0), *fetch_targets, fetch_holder_name);
243 244

  ProgramDesc* copy_program = const_cast<ProgramDesc*>(&program);
S
sneaxiy 已提交
245
  std::unique_ptr<ProgramDesc> unique_ptr_of_copy_program;
246
  if (!has_feed_ops || !has_fetch_ops) {
S
sneaxiy 已提交
247 248
    unique_ptr_of_copy_program.reset(new ProgramDesc(program));
    copy_program = unique_ptr_of_copy_program.get();
249
  }
250 251
  auto* global_block = copy_program->MutableBlock(0);

252
  if (!has_feed_ops) {
253 254
    // create feed_holder variable
    auto* feed_holder = global_block->Var(feed_holder_name);
255
    feed_holder->SetType(proto::VarType::FEED_MINIBATCH);
256 257 258
    feed_holder->SetPersistable(true);

    int i = 0;
259
    for (auto& feed_target : (*feed_targets)) {
260 261 262 263 264 265 266 267 268 269 270 271 272 273 274
      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++;
    }
  }

275
  if (!has_fetch_ops) {
276 277
    // create fetch_holder variable
    auto* fetch_holder = global_block->Var(fetch_holder_name);
278
    fetch_holder->SetType(proto::VarType::FETCH_LIST);
279 280 281
    fetch_holder->SetPersistable(true);

    int i = 0;
282
    for (auto& fetch_target : (*fetch_targets)) {
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297
      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++;
    }
  }

298
  auto ctx = Prepare(*copy_program, 0);
W
Wu Yi 已提交
299 300 301
  RunPreparedContext(ctx.get(), scope, feed_targets, fetch_targets,
                     create_local_scope, create_vars, feed_holder_name,
                     fetch_holder_name);
302 303
}

Q
Qiao Longfei 已提交
304 305
std::unique_ptr<ExecutorPrepareContext> Executor::Prepare(
    const ProgramDesc& program, int block_id) {
Q
Qiyang Min 已提交
306 307
  std::unique_ptr<ExecutorPrepareContext> ctx(
      new ExecutorPrepareContext(program, block_id));
Y
Yu Yang 已提交
308 309 310 311 312
  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
313
  return ctx;
Y
Yu Yang 已提交
314 315
}

T
refine  
typhoonzero 已提交
316
std::vector<std::shared_ptr<ExecutorPrepareContext>> Executor::Prepare(
T
typhoonzero 已提交
317 318 319 320 321 322 323 324 325 326 327 328 329 330
    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
331
void Executor::RunPreparedContext(ExecutorPrepareContext* ctx, Scope* scope,
Q
qiaolongfei 已提交
332 333
                                  bool create_local_scope, bool create_vars,
                                  bool keep_kids) {
Y
Yu Yang 已提交
334 335 336 337
  Scope* local_scope = scope;
  if (create_vars) {
    if (create_local_scope) {
      local_scope = &scope->NewScope();
338 339
    }
    CreateVariables(ctx->prog_, local_scope, ctx->block_id_);
L
Liu Yiqun 已提交
340
  }
Y
Yu Yang 已提交
341 342

  for (auto& op : ctx->ops_) {
Q
qiaolongfei 已提交
343
    VLOG(4) << place_ << " " << op->DebugStringEx(local_scope);
344
    op->Run(*local_scope, place_);
Q
qiaolongfei 已提交
345 346 347 348
    // 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
349

Y
Yu Yang 已提交
350 351 352 353 354
    if (FLAGS_benchmark) {
      VLOG(2) << "Memory used after operator " + op->Type() + " running: "
              << memory::memory_usage(place_);
    }
  }
355
  platform::DeviceContextPool::Instance().Get(place_)->Wait();
Q
qiaolongfei 已提交
356
  if (local_scope != scope) {
Y
Yu Yang 已提交
357
    scope->DeleteScope(local_scope);
358
  } else {
Q
qiaolongfei 已提交
359 360 361 362 363
    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
364 365
      // 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
366 367
      scope->DropKids();
    }
Y
Yu Yang 已提交
368
  }
Q
qiaolongfei 已提交
369

Y
Yu Yang 已提交
370 371 372 373 374 375 376 377
  if (FLAGS_benchmark) {
    VLOG(2) << "-------------------------------------------------------";
    VLOG(2) << "Memory used after deleting local scope: "
            << memory::memory_usage(place_);
    VLOG(2) << "-------------------------------------------------------";
  }
}

378 379
void Executor::RunPreparedContext(
    ExecutorPrepareContext* ctx, Scope* scope,
380
    std::map<std::string, const LoDTensor*>* feed_targets,
W
Wu Yi 已提交
381 382 383
    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) {
384 385
  auto& global_block = ctx->prog_.Block(ctx->block_id_);

386
  PADDLE_ENFORCE(
387
      has_feed_operators(global_block, *feed_targets, feed_holder_name),
388 389
      "Program in ExecutorPrepareContext should has feed_ops.");
  PADDLE_ENFORCE(
390
      has_fetch_operators(global_block, *fetch_targets, fetch_holder_name),
391 392
      "Program in the prepared context should has fetch_ops.");

393 394 395 396 397
  // 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"));
398 399
      SetFeedVariable(scope, *(*feed_targets)[feed_target_name],
                      feed_holder_name, idx);
400 401 402
    }
  }

W
Wu Yi 已提交
403
  RunPreparedContext(ctx, scope, create_local_scope, create_vars);
404 405 406 407 408 409

  // 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"));
410
      *(*fetch_targets)[fetch_target_name] =
411 412 413 414 415
          GetFetchVariable(*scope, fetch_holder_name, idx);
    }
  }
}

416 417 418 419 420 421 422 423 424 425 426
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);
      }
    }
  }
427 428 429
#else
  LOG(WARNING)
      << "'MKLDNN' is not supported, Please re-compile with WITH_MKLDNN option";
430 431 432
#endif
}

Q
qijun 已提交
433 434
}  // namespace framework
}  // namespace paddle