cancle threadpool before deconstruction interpretorcore (#37034)

* cancle thread when exit, test=develop

* gc to unique_ptr, test=develop

* refine, test=develop

* fix namespace, test=develop

paddle/fluid/framework/new_executor/interpretercore.cc

@@ -39,9 +39,11 @@ InterpreterCore::InterpreterCore(const platform::Place& place, BlockDesc* block,
     : place_(place),
       block_(block),
       global_scope_(global_scope),
-      stream_analyzer_(place),
-      async_work_queue_(kHostNumThreads, &main_thread_blocker_) {
+      stream_analyzer_(place) {
   is_build_ = false;
+  async_work_queue_.reset(
+      new interpreter::AsyncWorkQueue(kHostNumThreads, &main_thread_blocker_));
+  gc_.reset(new InterpreterCoreGarbageCollector());
 
   feed_names_ = feed_names;
 
@@ -55,6 +57,13 @@ InterpreterCore::InterpreterCore(const platform::Place& place, BlockDesc* block,
   // convert to run graph
 }
 
+InterpreterCore::~InterpreterCore() {
+  // cancle gc's thread
+  gc_.reset(nullptr);
+
+  async_work_queue_.reset(nullptr);
+}
+
 paddle::framework::FetchList InterpreterCore::Run(
     const std::vector<framework::LoDTensor>& feed_tensors) {
   auto FeedInput = [&] {
@@ -349,16 +358,16 @@ void InterpreterCore::RunInstruction(const Instruction& instr_node) {
 
 void InterpreterCore::ExecuteInstructionList(
     const std::vector<Instruction>& vec_instr) {
-  async_work_queue_.PrepareAtomicDeps(dependecy_count_);
-  async_work_queue_.PrepareAtomicVarRef(vec_meta_info_);
+  async_work_queue_->PrepareAtomicDeps(dependecy_count_);
+  async_work_queue_->PrepareAtomicVarRef(vec_meta_info_);
   op_run_number_ = 0;
 
   exception_holder_.Clear();
 
   for (size_t i = 0; i < dependecy_count_.size(); ++i) {
     if (dependecy_count_[i] == 0) {
-      async_work_queue_.AddTask(vec_instr.at(i).KernelType(),
-                                [&, i] { RunInstructionAsync(i); });
+      async_work_queue_->AddTask(vec_instr.at(i).KernelType(),
+                                 [&, i] { RunInstructionAsync(i); });
     }
   }
 
@@ -380,7 +389,7 @@ void InterpreterCore::ExecuteInstructionList(
 void InterpreterCore::RunNextInstructions(
     const Instruction& instr, std::queue<size_t>* reserved_next_ops) {
   auto& next_instr = instr.NextInstructions();
-  auto& atomic_deps = async_work_queue_.AtomicDeps();
+  auto& atomic_deps = async_work_queue_->AtomicDeps();
   auto IsReady = [&](size_t next_id) {
     return atomic_deps[next_id]->fetch_sub(1, std::memory_order_relaxed) == 1;
   };
@@ -389,7 +398,7 @@ void InterpreterCore::RunNextInstructions(
     // move all sync_ops into other threads
     for (auto next_id : next_instr.SyncRunIds()) {
       if (IsReady(next_id)) {
-        async_work_queue_.AddTask(
+        async_work_queue_->AddTask(
             vec_instruction_[next_id].KernelType(),
             [&, next_id] { RunInstructionAsync(next_id); });
       }
@@ -409,7 +418,7 @@ void InterpreterCore::RunNextInstructions(
     // move async_ops into async_thread
     for (auto next_id : next_instr.EventRunIds()) {
       if (IsReady(next_id)) {
-        async_work_queue_.AddTask(
+        async_work_queue_->AddTask(
             vec_instruction_[next_id].KernelType(),
             [&, next_id] { RunInstructionAsync(next_id); });
       }
@@ -425,7 +434,7 @@ void InterpreterCore::RunNextInstructions(
           continue;
         }
         // move rest ops into other threads
-        async_work_queue_.AddTask(
+        async_work_queue_->AddTask(
             vec_instruction_[next_id].KernelType(),
             [&, next_id] { RunInstructionAsync(next_id); });
       }
@@ -483,7 +492,7 @@ void InterpreterCore::RunInstructionAsync(size_t instr_id) {
 void InterpreterCore::CheckGC(const Instruction& instr) {
   size_t instr_id = instr.Id();
   auto& var_scope = *global_scope_;
-  auto& atomic_var_ref = async_work_queue_.AtomicVarRef();
+  auto& atomic_var_ref = async_work_queue_->AtomicVarRef();
 
   for (auto var_id : instr.GCCheckVars()) {
     bool is_ready =
@@ -493,8 +502,8 @@ void InterpreterCore::CheckGC(const Instruction& instr) {
       continue;
     }
     if (is_ready) {
-      gc_.Add(var_scope.Var(var_id), gc_event_.at(instr_id),
-              &instr.DeviceContext());
+      gc_->Add(var_scope.Var(var_id), gc_event_.at(instr_id),
+               &instr.DeviceContext());
     }
   }
 }

paddle/fluid/framework/new_executor/interpretercore.h

@@ -44,6 +44,8 @@ class InterpreterCore {
                   VariableScope* global_scope,
                   const std::vector<std::string>& feed_names);
 
+  ~InterpreterCore();
+
   paddle::framework::FetchList Run(
       const std::vector<framework::LoDTensor>& feed_tensors);
 
@@ -94,11 +96,11 @@ class InterpreterCore {
   StreamAnalyzer stream_analyzer_;
   EventManager event_manager_;
   EventsWaiter main_thread_blocker_;
-  interpreter::AsyncWorkQueue async_work_queue_;
+  std::unique_ptr<interpreter::AsyncWorkQueue> async_work_queue_;
   details::ExceptionHolder exception_holder_;
   std::shared_ptr<EventsWaiter::EventNotifier> exception_notifier_{nullptr};
 
-  InterpreterCoreGarbageCollector gc_;
+  std::unique_ptr<InterpreterCoreGarbageCollector> gc_;
   std::vector<paddle::platform::DeviceEvent> gc_event_;
   std::atomic<size_t> op_run_number_{0};
 };

paddle/fluid/framework/new_executor/interpretercore_garbage_collector.cc

@@ -28,6 +28,10 @@ InterpreterCoreGarbageCollector::InterpreterCoreGarbageCollector() {
   queue_ = CreateSingleThreadedWorkQueue(options);
 }
 
+InterpreterCoreGarbageCollector::~InterpreterCoreGarbageCollector() {
+  queue_.reset(nullptr);
+}
+
 void InterpreterCoreGarbageCollector::Add(
     std::shared_ptr<memory::Allocation> garbage,
     paddle::platform::DeviceEvent& event, const platform::DeviceContext* ctx) {

paddle/fluid/framework/new_executor/interpretercore_garbage_collector.h

@@ -35,6 +35,8 @@ class InterpreterCoreGarbageCollector {
  public:
   InterpreterCoreGarbageCollector();
 
+  ~InterpreterCoreGarbageCollector();
+
   void Add(std::shared_ptr<memory::Allocation> garbage,  // NOLINT
            paddle::platform::DeviceEvent& event,         // NOLINT
            const platform::DeviceContext* ctx);