Method CompiledProgram

paddle/fluid/framework/compiled_program.cc:472–516

}

CompiledProgram::CompiledProgram(const std::vector<phi::Place> &places,
                                 const std::vector<std::string> &bcast_vars,
                                 const std::string &loss_var_name,
                                 Scope *scope,
                                 const std::vector<Scope *> &local_scopes,
                                 const BuildStrategy &build_strategy,
                                 ir::Graph *graph)
    : member_(new CompiledProgramPrivate(places, scope)) {
  PADDLE_ENFORCE_EQ(
      !places.empty(),
      true,
      common::errors::Unavailable("NPU is not supported in CompiledProgram."));
  InitP2P(places);
  InitReaderQueueDeviceCount(
      graph, *(member_->global_scope_), member_->places_.size());
  // Initialize necessary info of member_ with strategy.
  InitProgramPrivateMemberInfo(build_strategy, places.size());

  // Step 1. Create local scopes and Clone graph into multi device
  CreateLocalScopes(scope, local_scopes, /*create_new*/ true);
  std::vector<ir::Graph *> graphs = CloneGraphToMultiDevices(graph);
  PrepareNCCLCommunicator(scope);

  // broadcast parameters from the 0th device to others:
  auto need_broadcast = [&]() -> bool {
    if (member_->build_strategy_.num_trainers_ > 1) {  // NOLINT
      // 1. num_tariners would be grater than 1 for nccl distributed training.
      return true;
    } else if (member_->local_scopes_.size() != 1 && local_scopes.empty()) {
      // 2. Only one trainer process, but CompiledProgram hold multiple
      // devices.
      return true;
    }
    return false;
  };
  if (need_broadcast()) {
    BCastParamsToDevices(bcast_vars, member_->build_strategy_.trainer_id_);
  }

  // Step 2. Convert main_program to SSA form and dependency graph. Also, insert
  // ncclOp
  std::vector<ir::Graph *> async_graphs =
      CompileGraphWithBuildStrategy(graph, &graphs, loss_var_name);
  graph = member_->ApplyMemoryOptimizePass(graph);
}

void CompiledProgram::BCastParamsToDevices(const std::vector<std::string> &vars,
                                           int trainer_id) const {