parallel_executor.cc 22.3 KB
Newer Older
Y
Yang Yang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.

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. */

#include "paddle/fluid/framework/parallel_executor.h"
Y
Yu Yang 已提交
16 17
#include "ThreadPool.h"
#include "executor.h"
Y
Yu Yang 已提交
18
#include "lod_tensor.h"
Y
Yu Yang 已提交
19
#include "lod_tensor_array.h"
Y
Yu Yang 已提交
20
#include "op_registry.h"
Y
Yu Yang 已提交
21
#include "paddle/fluid/operators/math/concat.h"
Y
Yang Yang 已提交
22 23

namespace paddle {
Y
Yu Yang 已提交
24 25
namespace framework {

Y
Yu Yang 已提交
26 27 28 29 30 31
#ifdef PADDLE_WITH_CUDA

// FIXME: CHECK the return value of x;
#define NCCL_INVOKE(x) x
#endif

Y
Yu Yang 已提交
32 33
struct OpHandle;

Y
Yu Yang 已提交
34 35 36 37 38
struct VarHandleBase {
  virtual ~VarHandleBase() {}
  virtual std::string DebugString() const = 0;

  OpHandle *generated_op_;
Y
Yu Yang 已提交
39
  std::unordered_set<OpHandle *> pending_ops_;
Y
Yu Yang 已提交
40 41 42 43 44 45 46 47 48
};

struct VarHandle : public VarHandleBase {
  std::string DebugString() const override {
    std::stringstream ss;
    ss << name_ << ":" << place_;
    return ss.str();
  }

Y
Yu Yang 已提交
49 50
  // version field currently is not used, however, just store the version to
  // debug easily.
Y
Yu Yang 已提交
51 52 53
  size_t version_;
  std::string name_;
  platform::Place place_;
Y
Yu Yang 已提交
54
};
Y
Yu Yang 已提交
55

Y
Yu Yang 已提交
56
struct DependencyVarHandle : public VarHandleBase {
Y
Yu Yang 已提交
57
  std::string DebugString() const override { return "Dependency Variable"; }
Y
Yu Yang 已提交
58 59 60
};

struct OpHandle {
Y
Yu Yang 已提交
61 62 63 64 65
  std::vector<VarHandleBase *> inputs_;
  std::vector<VarHandleBase *> outputs_;
  std::unordered_map<platform::Place, platform::DeviceContext *,
                     platform::PlaceHash>
      dev_ctx_;
Y
Yu Yang 已提交
66 67 68 69 70

  std::string DebugString() {
    std::stringstream ss;
    ss << "(";
    for (auto *var : inputs_) {
Y
Yu Yang 已提交
71
      ss << var->DebugString() << ", ";
Y
Yu Yang 已提交
72 73 74
    }
    ss << ") --> (";
    for (auto *var : outputs_) {
Y
Yu Yang 已提交
75
      ss << var->DebugString() << ", ";
Y
Yu Yang 已提交
76 77 78 79 80 81
    }
    ss << ")\n";
    return ss.str();
  }

  virtual ~OpHandle() {}
Y
Yu Yang 已提交
82

Y
Yu Yang 已提交
83
  virtual void Run() { PADDLE_THROW("Not implemented"); }
Y
Yu Yang 已提交
84
  virtual void Wait(platform::DeviceContext *waited_dev) {}
Y
Yu Yang 已提交
85 86 87 88
};

struct ComputationOpHandle : public OpHandle {
  std::unique_ptr<OperatorBase> op_;
Y
Yu Yang 已提交
89 90
  Scope *scope_;
  platform::Place place_;
Y
Yu Yang 已提交
91

Y
Yu Yang 已提交
92 93
  explicit ComputationOpHandle(const OpDesc &op_desc, Scope *scope,
                               platform::Place place)
Y
Yu Yang 已提交
94
      : op_(framework::OpRegistry::CreateOp(op_desc)),
Y
Yu Yang 已提交
95
        scope_(scope),
Y
Yu Yang 已提交
96 97 98 99 100 101 102
        place_(place) {}

  void Run() override {
    // Wait other op if necessary
    auto *cur_ctx = dev_ctx_[place_];
    for (auto *in : inputs_) {
      if (in->generated_op_ && in->generated_op_->dev_ctx_[place_] != cur_ctx) {
Y
Yu Yang 已提交
103
        in->generated_op_->Wait(cur_ctx);
Y
Yu Yang 已提交
104 105 106 107 108
      }
    }

    op_->Run(*scope_, place_);
  }
Y
Yu Yang 已提交
109 110 111 112

  void Wait(platform::DeviceContext *waited_dev) override {
    this->dev_ctx_.at(place_)->Wait();
  }
Y
Yu Yang 已提交
113 114
};

Y
Yu Yang 已提交
115 116 117 118 119 120 121 122 123 124 125 126 127
struct ScaleLossGradOpHandle : public OpHandle {
  float coeff_;
  Scope *scope_;
  platform::Place place_;

  explicit ScaleLossGradOpHandle(size_t num_dev, Scope *scope,
                                 platform::Place place)
      : coeff_(static_cast<float>(1.0 / num_dev)),
        scope_(scope),
        place_(place) {}

  void Run() override {
    std::string var_name = static_cast<VarHandle *>(this->outputs_[0])->name_;
Y
Yu Yang 已提交
128

Y
Yu Yang 已提交
129 130 131 132 133 134 135 136 137 138 139 140 141 142
    float *tmp = scope_->FindVar(var_name)
                     ->GetMutable<framework::LoDTensor>()
                     ->mutable_data<float>(make_ddim({1}), place_);

    if (platform::is_cpu_place(place_)) {
      *tmp = coeff_;
    } else {
      memory::Copy(
          boost::get<platform::CUDAPlace>(place_), tmp, platform::CPUPlace(),
          &coeff_, sizeof(float),
          static_cast<platform::CUDADeviceContext *>(this->dev_ctx_[place_])
              ->stream());
    }
  }
Y
Yu Yang 已提交
143 144 145 146

  void Wait(platform::DeviceContext *waited_dev) override {
    this->dev_ctx_.at(place_)->Wait();
  }
Y
Yu Yang 已提交
147 148
};

Y
Yu Yang 已提交
149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206
struct FetchedData {
 public:
  std::vector<framework::LoDTensor> tensors_;

  explicit FetchedData(size_t num_fetched) { tensors_.resize(num_fetched); }
};

struct FetchOpHandle : public OpHandle {
  std::shared_ptr<FetchedData> data_;
  size_t offset_;
  std::vector<Scope *> *local_scopes_;
  std::vector<LoDTensor> tensors_;

  ~FetchOpHandle() {
    for (auto *input_var : inputs_) {
      input_var->pending_ops_.erase(this);
    }
    for (auto &pair : dev_ctx_) {
      pair.second->Wait();
    }

    // Lazily merge tensors. Will faster code.
    MergeTensors();
  }

  void Run() override {
    tensors_.resize(inputs_.size());
    auto *var = static_cast<VarHandle *>(inputs_[0]);
    auto &var_name = var->name_;
    platform::CPUPlace cpu;
    auto &scopes = *local_scopes_;

    for (size_t i = 0; i < scopes.size(); ++i) {
      auto &scope = scopes[i];
      auto &t = scope->FindVar(var_name)->Get<framework::LoDTensor>();
      if (platform::is_gpu_place(var->place_)) {
        TensorCopy(t, cpu, *dev_ctx_[t.place()], &tensors_[i]);
      } else {
        tensors_[i].ShareDataWith(t);
        tensors_[i].set_lod(t.lod());
      }
    }
  }

  void Wait(platform::DeviceContext *waited_dev) override {
    PADDLE_THROW("Nobody should wait FetchOp. Unexpceted Error");
  }

 private:
  void MergeTensors() const {
    std::vector<const LoDTensor *> tensors_ptr;
    for (auto &t : tensors_) {
      tensors_ptr.emplace_back(&t);
    }
    data_->tensors_[offset_].MergeLoDTensor(tensors_ptr, platform::CPUPlace());
  }
};

Y
Yu Yang 已提交
207 208
class ParallelExecutorPrivate {
 public:
Y
Yu Yang 已提交
209 210 211
  explicit ParallelExecutorPrivate(size_t num_threads = 12)
      : pool_(num_threads) {}

Y
Stash  
Yu Yang 已提交
212 213
  std::vector<platform::Place> places_;

Y
Yu Yang 已提交
214
  std::vector<Scope *> local_scopes_;
Y
Yu Yang 已提交
215
  Scope *global_scope_;
Y
Yu Yang 已提交
216

Y
Yu Yang 已提交
217 218 219 220 221 222 223 224 225 226 227 228 229 230 231
#ifdef PADDLE_WITH_CUDA
  struct NCCLContext {
    std::unique_ptr<platform::CUDADeviceContext> ctx_;
    ncclComm_t comm;

    explicit NCCLContext(int dev_id) {
      ctx_.reset(new platform::CUDADeviceContext(platform::CUDAPlace(dev_id)));
    }

    cudaStream_t stream() const { return ctx_->stream(); }

    int device_id() const {
      return boost::get<platform::CUDAPlace>(ctx_->GetPlace()).device;
    }

Y
Update  
Yu Yang 已提交
232 233
    static void InitNCCLContext(std::unordered_map<int, NCCLContext> &contexts,
                                const std::vector<platform::Place> &places) {
Y
Yu Yang 已提交
234 235 236 237 238
      std::vector<ncclComm_t> comms;
      std::vector<int> devs;
      comms.resize(contexts.size());
      devs.reserve(contexts.size());

Y
Update  
Yu Yang 已提交
239 240
      for (auto &p : places) {
        devs.push_back(boost::get<platform::CUDAPlace>(p).device);
Y
Yu Yang 已提交
241 242 243 244 245 246
      }

      NCCL_INVOKE(platform::dynload::ncclCommInitAll(
          &comms[0], static_cast<int>(contexts.size()), &devs[0]));

      int i = 0;
Y
Update  
Yu Yang 已提交
247 248
      for (auto &dev_id : devs) {
        contexts.at(dev_id).comm = comms[i++];
Y
Yu Yang 已提交
249 250 251 252
      }
    }
  };

Y
Update  
Yu Yang 已提交
253
  std::unordered_map<int, NCCLContext> communication_streams_;
Y
Yu Yang 已提交
254 255 256 257 258 259 260 261

  NCCLContext &GetNCCLCtx(platform::Place p) {
    int dev_id = boost::get<platform::CUDAPlace>(p).device;
    return communication_streams_.at(dev_id);
  }

#endif

Y
Yu Yang 已提交
262 263 264 265 266 267 268 269 270 271 272 273 274
  platform::DeviceContext *CommunicationDevCtx(const platform::Place &place) {
    if (platform::is_cpu_place(place) || local_scopes_.size() == 1) {
      return const_cast<platform::DeviceContext *>(
          platform::DeviceContextPool::Instance().Get(place));
    } else {
#ifdef PADDLE_WITH_CUDA
      return GetNCCLCtx(place).ctx_.get();
#else
      PADDLE_THROW("Not compiled with CUDA")
#endif
    }
  }

Y
Yu Yang 已提交
275 276 277 278 279 280
  platform::Place main_place_;

  std::unordered_map<platform::Place,
                     std::unordered_map<std::string, std::map<int, VarHandle>>,
                     platform::PlaceHash>
      vars_;
Y
Yu Yang 已提交
281 282
  std::unordered_set<std::unique_ptr<VarHandleBase>> dep_vars_;

Y
Yu Yang 已提交
283
  std::vector<std::unique_ptr<OpHandle>> ops_;
Y
Yu Yang 已提交
284

Y
Yu Yang 已提交
285
  // Use a simpler thread pool, might be faster.
Y
Yu Yang 已提交
286
  ThreadPool pool_;
Y
Yu Yang 已提交
287 288

  std::unique_ptr<platform::EnforceNotMet> exception_;
Y
Yu Yang 已提交
289 290 291 292
};

// TODO(yy): Move this function somewhere
ncclDataType_t ToNCCLDataType(std::type_index type) {
Y
Stash  
Yu Yang 已提交
293 294 295 296 297 298 299 300 301
  if (type == typeid(float)) {  // NOLINT
    return ncclFloat;
  } else if (type == typeid(double)) {  // NOLINT
    return ncclDouble;
  } else if (type == typeid(int)) {  // NOLINT
    return ncclInt;
  } else {
    PADDLE_THROW("Not supported");
  }
Y
Yu Yang 已提交
302 303
}

Y
Yu Yang 已提交
304 305 306 307 308 309 310 311 312 313 314 315 316 317 318
struct NCCLAllReduceOpHandle : public OpHandle {
  ParallelExecutorPrivate *member_;

  explicit NCCLAllReduceOpHandle(ParallelExecutorPrivate *member)
      : member_(member) {}

  void Run() override {
    if (this->inputs_.size() == 1) {
      return;  // No need to all reduce when GPU count = 1;
    } else {
      auto &var_name = static_cast<VarHandle *>(this->inputs_[0])->name_;

      int dtype = -1;
      size_t numel = 0;

Y
Update  
Yu Yang 已提交
319 320
      platform::dynload::ncclGroupStart();

Y
Yu Yang 已提交
321 322 323
      for (size_t i = 0; i < member_->local_scopes_.size(); ++i) {
        auto &p = member_->places_[i];
        auto *s = member_->local_scopes_[i];
Y
Yu Yang 已提交
324 325 326 327 328 329 330 331 332 333 334 335 336 337
        int dev_id = boost::get<platform::CUDAPlace>(p).device;

        auto &lod_tensor = s->FindVar(var_name)->Get<framework::LoDTensor>();
        void *buffer = const_cast<void *>(lod_tensor.data<void>());
        if (dtype == -1) {
          dtype = ToNCCLDataType(lod_tensor.type());
        }

        if (numel == 0) {
          numel = static_cast<size_t>(lod_tensor.numel());
        }

        auto &nccl_ctx = member_->communication_streams_.at(dev_id);

Y
Update  
Yu Yang 已提交
338 339 340
        platform::dynload::ncclAllReduce(
            buffer, buffer, numel, static_cast<ncclDataType_t>(dtype), ncclSum,
            nccl_ctx.comm, nccl_ctx.stream());
Y
Yu Yang 已提交
341 342
      }

Y
Update  
Yu Yang 已提交
343
      platform::dynload::ncclGroupEnd();
Y
Yu Yang 已提交
344 345
    }
  }
Y
Yu Yang 已提交
346 347

  void Wait(platform::DeviceContext *waited_dev) override {
Y
Debug  
Yu Yang 已提交
348
    VLOG(3) << "Wait NCCL AllReduce";
Y
Yu Yang 已提交
349 350
    this->dev_ctx_.at(waited_dev->GetPlace())->Wait();
  }
Y
Yu Yang 已提交
351 352
};

Y
Yu Yang 已提交
353 354 355 356 357 358
ParallelExecutor::ParallelExecutor(
    const std::vector<platform::Place> &places,
    const std::unordered_set<std::string> &params,
    const ProgramDesc &startup_program, const ProgramDesc &main_program,
    const std::string &loss_var_name, Scope *scope)
    : member_(new ParallelExecutorPrivate()) {
Y
Stash  
Yu Yang 已提交
359
  member_->places_ = places;
Y
Yu Yang 已提交
360
  member_->global_scope_ = scope;
Y
Yu Yang 已提交
361 362 363 364
  // Step 1. RunStartupProgram and Bcast the params to devs.
  Executor exe(places[0]);
  exe.Run(startup_program, scope, 0);
  // Create local scopes
Y
Yu Yang 已提交
365 366
  for (size_t i = 0; i < member_->places_.size(); ++i) {
    member_->local_scopes_.push_back(&scope->NewScope());
Y
Yu Yang 已提交
367 368 369 370
  }
  member_->main_place_ = places[0];

  // Bcast Parameters to all GPUs
Y
Yu Yang 已提交
371
  BuildNCCLCommunicator();
Y
Yu Yang 已提交
372 373 374
  if (platform::is_gpu_place(member_->main_place_) &&
      member_->local_scopes_.size() != 1) {  // Is CUDA
    BCastParamsToGPUs(startup_program);
Y
Yu Yang 已提交
375 376 377 378 379 380
  }
  // Startup Program has been run. All local scopes has correct parameters.

  // Step 2. Convert main_program to SSA form and dependency graph. Also, insert
  // ncclOp
  ConstructDependencyGraph(params, main_program, loss_var_name);
Y
Yu Yang 已提交
381 382

  // Step 3. Create vars in each scope;
Y
Yu Yang 已提交
383
  for (auto *scope : member_->local_scopes_) {
Y
Yu Yang 已提交
384 385 386 387 388 389 390 391
    for (auto *var : main_program.Block(0).AllVars()) {
      if (scope->FindVar(var->Name()) != nullptr) {
        continue;
      }

      InitializeVariable(scope->Var(var->Name()), var->GetType());
    }
  }
Y
Yu Yang 已提交
392 393 394 395 396
}

void ParallelExecutor::ConstructDependencyGraph(
    const std::unordered_set<std::string> &params,
    const ProgramDesc &main_program, const std::string &loss_var_name) const {
Y
Yu Yang 已提交
397
  std::unordered_set<std::string> grads;
Y
Yu Yang 已提交
398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413
  for (auto &each_param : params) {
    grads.insert(each_param + "@GRAD");
  }

  bool is_forwarding = true;
  for (auto *op : main_program.Block(0).AllOps()) {
    bool change_forward = false;

    if (!is_forwarding) {
      // FIXME(yy): Do not hard code like this
      if (op->OutputArgumentNames().size() == 1 &&
          op->OutputArgumentNames()[0] == loss_var_name + "@GRAD") {
        continue;  // Drop fill 1. for backward coeff;
      }
    }

Y
Yu Yang 已提交
414 415 416 417 418
    for (size_t i = 0; i < member_->places_.size(); ++i) {
      auto &p = member_->places_[i];
      auto *s = member_->local_scopes_[i];

      member_->ops_.emplace_back(new ComputationOpHandle(*op, s, p));
Y
Yu Yang 已提交
419
      auto *op_handle = member_->ops_.back().get();
Y
Yu Yang 已提交
420 421
      op_handle->dev_ctx_[p] = const_cast<platform::DeviceContext *>(
          platform::DeviceContextPool::Instance().Get(p));
Y
Yu Yang 已提交
422 423 424 425

      auto var_names = op->InputArgumentNames();

      for (auto &each_var_name : var_names) {
Y
Yu Yang 已提交
426
        VarHandle *var = GetVarHandle(each_var_name, p);
Y
Yu Yang 已提交
427
        op_handle->inputs_.emplace_back(var);
Y
Yu Yang 已提交
428
        var->pending_ops_.emplace(op_handle);
Y
Yu Yang 已提交
429 430 431 432
      }
      var_names = op->OutputArgumentNames();

      for (auto &each_var_name : var_names) {
Y
Yu Yang 已提交
433
        GenerateVar(op_handle, each_var_name, p);
Y
Yu Yang 已提交
434 435 436 437 438
      }

      if (is_forwarding) {
        if (var_names.size() == 1 && var_names[0] == loss_var_name) {
          // Insert ScaleCost OpHandle
Y
Yu Yang 已提交
439
          member_->ops_.emplace_back(new ScaleLossGradOpHandle(
Y
Yu Yang 已提交
440
              this->member_->local_scopes_.size(), s, p));
Y
Yu Yang 已提交
441
          op_handle = member_->ops_.back().get();
Y
Yu Yang 已提交
442

Y
Yu Yang 已提交
443
          op_handle->dev_ctx_[p] = member_->CommunicationDevCtx(p);
Y
Yu Yang 已提交
444

Y
Yu Yang 已提交
445 446 447 448 449 450
          // FIXME: Currently ScaleLossGradOp only use device_count as scale
          // factor. So it does not depend on any other operators.
          // VarHandle *loss = GetVarHandle(loss_var_name, place);
          // loss->pending_ops_.emplace_back(op_handle);
          // op_handle->inputs_.emplace_back(loss);

Y
Yu Yang 已提交
451
          GenerateVar(op_handle, loss_var_name + "@GRAD", p);
Y
Yu Yang 已提交
452 453 454 455 456 457 458 459 460 461 462 463 464 465
          change_forward = true;
        }
      }
    }

    if (change_forward) {
      is_forwarding = false;
    }

    if (!is_forwarding) {
      auto var_names = op->OutputArgumentNames();
      for (auto &og : var_names) {
        if (grads.count(og) != 0) {  // is param grad
          // Insert NCCL AllReduce Op
Y
Yu Yang 已提交
466
          member_->ops_.emplace_back(new NCCLAllReduceOpHandle(member_));
Y
Yu Yang 已提交
467 468
          auto *op_handle = member_->ops_.back().get();

Y
Yu Yang 已提交
469 470 471
          for (size_t i = 0; i < member_->places_.size(); ++i) {
            auto &p = member_->places_[i];
            auto &vars = member_->vars_[p][og];
Y
Yu Yang 已提交
472 473 474 475 476 477

            if (vars.empty()) {  // This device has no data. continue.
              continue;
            }
            auto *prev_grad = &vars[vars.size() - 1];
            op_handle->inputs_.emplace_back(prev_grad);
Y
Yu Yang 已提交
478
            prev_grad->pending_ops_.emplace(op_handle);
Y
Yu Yang 已提交
479
            auto &var = vars[vars.size()];
Y
Yu Yang 已提交
480
            var.place_ = p;
Y
Yu Yang 已提交
481 482 483 484
            var.generated_op_ = op_handle;
            var.name_ = og;
            var.version_ = vars.size() - 1;
            op_handle->outputs_.emplace_back(&var);
Y
Yu Yang 已提交
485

Y
Yu Yang 已提交
486
            op_handle->dev_ctx_[p] = member_->CommunicationDevCtx(p);
Y
Yu Yang 已提交
487 488 489 490 491
          }
        }
      }
    }
  }
Y
Yu Yang 已提交
492

Y
Yu Yang 已提交
493 494 495
  /*
    Dependency graph has been constructed. However, there are still data
    harzaeds need to be handled.
Y
Yu Yang 已提交
496
   */
Y
Yu Yang 已提交
497 498
  PolishGraphToSupportDataHarzaeds();
}
Y
Yu Yang 已提交
499

Y
Yu Yang 已提交
500 501 502 503 504 505 506 507
/**
 * We only handle write after read(WAR), since it should not have a write
 * after write in program. If there are write after write operators, we need
 * prune them.
 *
 * https://en.wikipedia.org/wiki/Hazard_(computer_architecture)#Write_after_read_(WAR)
 */
void ParallelExecutor::PolishGraphToSupportDataHarzaeds() const {
Y
Yu Yang 已提交
508 509 510 511 512 513 514 515 516 517 518
  for (auto &place_pair : member_->vars_) {
    for (auto &name_pair : place_pair.second) {
      if (name_pair.second.size() <= 1) {
        return;
      }
      auto it_new = name_pair.second.rbegin();
      auto it_old = name_pair.second.rbegin();
      ++it_old;
      for (; it_old != name_pair.second.rend(); it_new = it_old, ++it_old) {
        auto *write_op = it_new->second.generated_op_;
        auto &read_ops = it_old->second.pending_ops_;
Y
Yu Yang 已提交
519 520 521 522 523 524
        auto *ex_write_op = it_old->second.generated_op_;

        if (ex_write_op == nullptr) {  // Nobody write this var.
          continue;
        }

Y
Yu Yang 已提交
525 526
        for (auto *read_op : read_ops) {
          // Manually add a dependency var from read_op to write_op;
Y
Yu Yang 已提交
527 528 529 530
          if (read_op == write_op) {
            // Read Write is the same op.
            continue;
          }
Y
Yu Yang 已提交
531 532

          auto *dep_var = new DependencyVarHandle();
Y
Yu Yang 已提交
533

Y
Yu Yang 已提交
534 535 536
          dep_var->generated_op_ = read_op;
          read_op->outputs_.emplace_back(dep_var);

Y
Yu Yang 已提交
537
          dep_var->pending_ops_.emplace(write_op);
Y
Yu Yang 已提交
538 539 540 541 542 543
          write_op->inputs_.emplace_back(dep_var);
          member_->dep_vars_.emplace(dep_var);
        }
      }
    }
  }
Y
Yu Yang 已提交
544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578
}

void ParallelExecutor::GenerateVar(OpHandle *op_handle,
                                   const std::string &each_var_name,
                                   const platform::Place &place) const {
  auto &vars = member_->vars_[place][each_var_name];
  size_t version = vars.size();
  auto &var = vars[version];
  var.version_ = version;
  var.generated_op_ = op_handle;
  var.name_ = each_var_name;
  var.place_ = place;
  op_handle->outputs_.emplace_back(&var);
}

VarHandle *ParallelExecutor::GetVarHandle(const std::string &each_var_name,
                                          const platform::Place &place) const {
  auto &var_holders = member_->vars_[place];
  auto &var_holder = var_holders[each_var_name];
  VarHandle *var = nullptr;
  if (var_holder.empty()) {
    auto &init_var = var_holder[0];
    init_var.place_ = place;
    init_var.name_ = each_var_name;
    init_var.generated_op_ = nullptr;
    init_var.version_ = 0;
    var = &init_var;
  } else {
    var = &var_holder.rbegin()->second;
  }
  return var;
}

void ParallelExecutor::BCastParamsToGPUs(
    const ProgramDesc &startup_program) const {
Y
Yu Yang 已提交
579
#ifdef PADDLE_WITH_CUDA
Y
Yu Yang 已提交
580
  auto *main_scope = member_->local_scopes_[0];
Y
Yu Yang 已提交
581

Y
Yu Yang 已提交
582 583 584 585 586 587 588 589
  for (auto *var_desc : startup_program.Block(0).AllVars()) {
    if (var_desc->GetType() == proto::VarType::LOD_TENSOR) {
      auto &main_tensor =
          main_scope->FindVar(var_desc->Name())->Get<LoDTensor>();
      ncclDataType_t data_type = ToNCCLDataType(main_tensor.type());
      auto &dims = main_tensor.dims();
      size_t numel = main_tensor.numel();

Y
Stash  
Yu Yang 已提交
590
      platform::dynload::ncclGroupStart();
Y
Yu Yang 已提交
591

Y
Update  
Yu Yang 已提交
592 593 594 595 596 597
      for (size_t i = 0; i < member_->places_.size(); ++i) {
        auto place = member_->places_[i];
        void *buffer;
        if (i == 0) {
          buffer = const_cast<void *>(main_tensor.data<void>());
        } else {
Y
Yu Yang 已提交
598
          auto local_scope = member_->local_scopes_[i];
Y
Update  
Yu Yang 已提交
599 600 601 602 603
          auto *t = local_scope->Var(var_desc->Name())->GetMutable<LoDTensor>();
          t->Resize(dims);
          buffer = t->mutable_data(place, main_tensor.type());
        }

Y
Stash  
Yu Yang 已提交
604
        auto &nccl_ctx = member_->GetNCCLCtx(place);
Y
Update  
Yu Yang 已提交
605
        platform::dynload::ncclBcast(buffer, numel, data_type, 0, nccl_ctx.comm,
Y
Stash  
Yu Yang 已提交
606
                                     nccl_ctx.stream());
Y
Yu Yang 已提交
607
      }
Y
Stash  
Yu Yang 已提交
608 609 610
      platform::dynload::ncclGroupEnd();
    }
  }
Y
Yu Yang 已提交
611 612 613 614
#else
  PADDLE_THROW("Not compiled with CUDA");
#endif
}
Y
Yu Yang 已提交
615

Y
Yu Yang 已提交
616 617
void ParallelExecutor::BuildNCCLCommunicator() const {
#ifdef PADDLE_WITH_CUDA
Y
Yu Yang 已提交
618
  for (auto &place : member_->places_) {
Y
Yu Yang 已提交
619
    int dev_id = boost::get<platform::CUDAPlace>(place).device;
Y
Yu Yang 已提交
620

Y
Yu Yang 已提交
621 622
    member_->communication_streams_.emplace(
        dev_id, ParallelExecutorPrivate::NCCLContext(dev_id));
Y
Yu Yang 已提交
623
  }
Y
Yu Yang 已提交
624 625

  ParallelExecutorPrivate::NCCLContext::InitNCCLContext(
Y
Update  
Yu Yang 已提交
626
      member_->communication_streams_, member_->places_);
Y
Yu Yang 已提交
627
#endif
Y
Yu Yang 已提交
628 629
}

Y
Yu Yang 已提交
630 631 632
void ParallelExecutor::Run(const std::vector<std::string> &fetch_tensors,
                           const std::string &fetched_var_name) {
  auto fetched_data = std::make_shared<FetchedData>(fetch_tensors.size());
Y
Yu Yang 已提交
633
  // Version --> VarHandle
Y
Yu Yang 已提交
634
  member_->exception_.reset();
Y
Yu Yang 已提交
635
  std::unordered_map<VarHandleBase *, bool> pending_vars;
Y
Yu Yang 已提交
636 637 638 639 640
  std::unordered_map<OpHandle *, size_t> pending_ops;

  for (auto &place_pair : member_->vars_) {
    for (auto &name_pair : place_pair.second) {
      for (auto &version_pair : name_pair.second) {
Y
Yu Yang 已提交
641 642
        pending_vars[&version_pair.second] =
            version_pair.second.generated_op_ == nullptr;
Y
Yu Yang 已提交
643 644 645 646
      }
    }
  }

Y
Yu Yang 已提交
647 648 649 650
  for (auto &var : member_->dep_vars_) {
    pending_vars[var.get()] = var->generated_op_ == nullptr;
  }

Y
Yu Yang 已提交
651 652
  std::vector<OpHandle *> to_run;

Y
Yu Yang 已提交
653
  for (auto &op : member_->ops_) {
Y
Yu Yang 已提交
654 655 656 657 658 659 660
    if (op->inputs_.empty()) {  // Special case, Op has no input.
      to_run.emplace_back(op.get());
    } else {
      pending_ops.insert({op.get(), op->inputs_.size()});
    }
  }

Y
Yu Yang 已提交
661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692
  std::unordered_map<std::string, std::vector<VarHandleBase *>> fetched_vars;

  for (auto &fetch_var_name : fetch_tensors) {
    for (auto &pair : member_->vars_) {
      auto it = pair.second.find(fetch_var_name);
      if (it != pair.second.end()) {
        fetched_vars[fetch_var_name].push_back(&it->second.rbegin()->second);
      }
    }
  }

  std::vector<FetchOpHandle> fetch_ops;

  for (size_t i = 0; i < fetch_tensors.size(); ++i) {
    auto &var_name = fetch_tensors[i];
    auto &vars = fetched_vars[var_name];
    fetch_ops.emplace_back();
    FetchOpHandle *op = &fetch_ops.back();
    op->data_ = fetched_data;
    op->offset_ = i;
    op->local_scopes_ = &member_->local_scopes_;
    for (auto &p : member_->places_) {
      op->dev_ctx_[p] = this->member_->GetNCCLCtx(p).ctx_.get();
    }

    for (auto *var : vars) {
      var->pending_ops_.emplace(op);
      op->inputs_.emplace_back(var);
    }
    pending_ops.insert({op, op->inputs_.size()});
  }

Y
Yu Yang 已提交
693 694
  for (auto *op : to_run) {
    RunOp(pending_vars, op);
Y
Yu Yang 已提交
695 696
  }

Y
Yu Yang 已提交
697
  while (!pending_ops.empty()) {
Y
Yu Yang 已提交
698
    VarHandleBase *ready_var = nullptr;
Y
Yu Yang 已提交
699 700 701
    for (auto &pair : pending_vars) {
      if (pair.second) {
        ready_var = pair.first;
Y
Yu Yang 已提交
702 703
      }
    }
Y
Yu Yang 已提交
704 705

    if (ready_var == nullptr) {
Y
Yu Yang 已提交
706 707 708 709 710 711 712
      // FIXME use conditional var instead of busy wait.

      if (member_->exception_) {
        throw * member_->exception_;
      }

      std::this_thread::yield();
Y
Yu Yang 已提交
713
      continue;
Y
Yu Yang 已提交
714 715
    }

Y
Yu Yang 已提交
716 717
    pending_vars.erase(ready_var);

Y
Yu Yang 已提交
718
    to_run.clear();
Y
Yu Yang 已提交
719 720 721 722 723 724

    for (auto *op : ready_var->pending_ops_) {
      auto &deps = pending_ops[op];
      --deps;
      if (deps == 0) {
        to_run.emplace_back(op);
Y
Yu Yang 已提交
725 726 727 728 729
      }
    }

    for (auto *op : to_run) {
      pending_ops.erase(op);
Y
Yu Yang 已提交
730
      RunOp(pending_vars, op);
Y
Yu Yang 已提交
731 732
    }
  }
Y
Yu Yang 已提交
733 734 735
  fetch_ops.clear();
  *member_->global_scope_->Var(fetched_var_name)->GetMutable<LoDTensorArray>() =
      fetched_data->tensors_;
Y
Yu Yang 已提交
736
}
Y
Yu Yang 已提交
737 738 739 740 741 742 743 744 745 746 747

void ParallelExecutor::RunOp(
    std::unordered_map<VarHandleBase *, bool> &pending_vars,
    OpHandle *op) const {
  std::vector<bool *> ready_buffer;
  for (auto *var : op->outputs_) {
    ready_buffer.emplace_back(&pending_vars[var]);
  }

  auto op_run = [ready_buffer, op, this] {
    try {
Y
Yu Yang 已提交
748
      VLOG(10) << op->DebugString();
Y
Yu Yang 已提交
749 750 751 752 753 754 755 756 757 758 759 760 761
      op->Run();
      for (auto *ready : ready_buffer) {
        *ready = true;
      }
    } catch (platform::EnforceNotMet ex) {
      member_->exception_.reset(new platform::EnforceNotMet(ex));
    } catch (...) {
      LOG(FATAL) << "Unknown exception catched";
    }
  };

  member_->pool_.enqueue(op_run);
}
Y
Yu Yang 已提交
762
}  // namespace framework
Y
Yang Yang 已提交
763
}  // namespace paddle