// Copyright (c) 2018 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. #pragma once #include #include #include #include #include #include "paddle/fluid/framework/details/op_handle_base.h" #include "paddle/fluid/framework/ir/memory_optimize_pass/reference_count_pass_helper.h" namespace paddle { namespace framework { class Scope; namespace ir { class MemOptVarInfo; } // namespace ir namespace details { class EagerDeletionOpHandle : public OpHandleBase { public: EagerDeletionOpHandle(ir::Node *node, Scope *scope, const platform::Place &place, const std::unordered_set &vars, GarbageCollector *gc); ~EagerDeletionOpHandle(); std::string Name() const override; /** * Currently, EagerDeletionOpHandle has the highest priority. * This priority settings speed up gc 15% in Transformer * V100 8-GPU model. */ Priority GetPriority() const override { return kHighest; } protected: void RunImpl() override; void InitCUDA() override; std::vector GetLocalScopes() override { return {scope_}; } private: void ClearGarbages(std::deque> *garbages); void CallOnce(); Scope *scope_; platform::Place place_; std::vector var_infos_; // not own GarbageCollector *gc_; // not own std::vector vars_; #ifdef PADDLE_WITH_CUDA platform::CUDADeviceContext *dev_ctx_{nullptr}; cudaEvent_t event_{nullptr}; #endif }; } // namespace details } // namespace framework } // namespace paddle