Thread data registry (#40912)

* add align for WorkQueue

* add spinlock

* merge develop

* merge

* Add EventsWaiter

* Revert "Add EventsWaiter"

This reverts commit e206173aa9be7401b83a53581627bfaf557c8fb2.

* Update ThreadDataRegistry
Co-authored-by: Nliutiexing <liutiexing@google.com>

paddle/fluid/framework/new_executor/workqueue/events_waiter.cc

@@ -142,20 +142,24 @@ std::string EventsWaiter::WaitEvent() {
   }
 
   trigger_event_.store(kEmptyEventId, std::memory_order_relaxed);
-  waiting_.store(false, std::memory_order_relaxed);
   std::string evt_name =
       triggered == kEmptyEventId ? "NoEventNotifier" : GetEventName(triggered);
   VLOG(10) << "Consume event id:" << triggered << ", name:" << evt_name;
   // lazy deletion
   {
+    triggered = trigger_event_;
     std::lock_guard<paddle::memory::SpinLock> guard(events_lock_);
     if (deleted_events_.size() > 0) {
       for (auto evt : deleted_events_) {
+        if (evt == triggered) {
+          continue;
+        }
         events_.erase(evt);
       }
       deleted_events_.clear();
     }
   }
+  waiting_.store(false, std::memory_order_relaxed);
   return evt_name;
 }
 

paddle/fluid/framework/new_executor/workqueue/thread_data_registry.h

@@ -15,6 +15,7 @@
 #pragma once
 
 #include <functional>
+#include <memory>
 #include <mutex>
 #include <thread>
 #include <type_traits>
@@ -23,13 +24,8 @@
 namespace paddle {
 namespace framework {
 
-static uint64_t main_tid =
-    std::hash<std::thread::id>()(std::this_thread::get_id());
-
 template <typename T>
 class ThreadDataRegistry {
-  class ThreadDataHolder;
-
  public:
   // Singleton
   static ThreadDataRegistry& GetInstance() {
@@ -52,73 +48,92 @@ class ThreadDataRegistry {
   template <typename Alias = T, typename = std::enable_if_t<
                                     std::is_copy_constructible<Alias>::value>>
   std::unordered_map<uint64_t, T> GetAllThreadDataByValue() {
-    std::unordered_map<uint64_t, T> data_copy;
-    std::lock_guard<std::mutex> lock(lock_);
-    data_copy.reserve(tid_map_.size());
-    for (auto& kv : tid_map_) {
-      data_copy.emplace(kv.first, kv.second->GetData());
-    }
-    return data_copy;
+    return impl_->GetAllThreadDataByValue();
   }
 
   // Returns current snapshot of all threads. Make sure there is no thread
   // create/destory when using it.
   std::unordered_map<uint64_t, std::reference_wrapper<T>>
   GetAllThreadDataByRef() {
-    std::unordered_map<uint64_t, std::reference_wrapper<T>> data_ref;
-    std::lock_guard<std::mutex> lock(lock_);
-    data_ref.reserve(tid_map_.size());
-    for (auto& kv : tid_map_) {
-      data_ref.emplace(kv.first, std::ref(kv.second->GetData()));
-    }
-    return data_ref;
+    return impl_->GetAllThreadDataByRef();
   }
 
-  void RegisterData(uint64_t tid, ThreadDataHolder* tls_obj) {
-    std::lock_guard<std::mutex> lock(lock_);
-    tid_map_[tid] = tls_obj;
-  }
+ private:
+  // types
+  class ThreadDataHolder;
+  class ThreadDataRegistryImpl {
+   public:
+    void RegisterData(uint64_t tid, ThreadDataHolder* tls_obj) {
+      std::lock_guard<std::mutex> lock(lock_);
+      tid_map_[tid] = tls_obj;
+    }
 
-  void UnregisterData(uint64_t tid) {
-    if (tid == main_tid) {
-      return;
+    void UnregisterData(uint64_t tid) {
+      std::lock_guard<std::mutex> lock(lock_);
+      tid_map_.erase(tid);
     }
-    std::lock_guard<std::mutex> lock(lock_);
-    tid_map_.erase(tid);
-  }
 
- private:
+    template <typename Alias = T, typename = std::enable_if_t<
+                                      std::is_copy_constructible<Alias>::value>>
+    std::unordered_map<uint64_t, T> GetAllThreadDataByValue() {
+      std::unordered_map<uint64_t, T> data_copy;
+      std::lock_guard<std::mutex> lock(lock_);
+      data_copy.reserve(tid_map_.size());
+      for (auto& kv : tid_map_) {
+        data_copy.emplace(kv.first, kv.second->GetData());
+      }
+      return data_copy;
+    }
+
+    std::unordered_map<uint64_t, std::reference_wrapper<T>>
+    GetAllThreadDataByRef() {
+      std::unordered_map<uint64_t, std::reference_wrapper<T>> data_ref;
+      std::lock_guard<std::mutex> lock(lock_);
+      data_ref.reserve(tid_map_.size());
+      for (auto& kv : tid_map_) {
+        data_ref.emplace(kv.first, std::ref(kv.second->GetData()));
+      }
+      return data_ref;
+    }
+
+   private:
+    std::mutex lock_;
+    std::unordered_map<uint64_t, ThreadDataHolder*> tid_map_;  // not owned
+  };
+
   class ThreadDataHolder {
    public:
-    ThreadDataHolder() {
+    explicit ThreadDataHolder(
+        std::shared_ptr<ThreadDataRegistryImpl> registry) {
+      registry_ = std::move(registry);
       tid_ = std::hash<std::thread::id>()(std::this_thread::get_id());
-      ThreadDataRegistry::GetInstance().RegisterData(tid_, this);
+      registry_->RegisterData(tid_, this);
     }
 
-    ~ThreadDataHolder() {
-      ThreadDataRegistry::GetInstance().UnregisterData(tid_);
-    }
+    ~ThreadDataHolder() { registry_->UnregisterData(tid_); }
 
     T& GetData() { return data_; }
 
    private:
+    std::shared_ptr<ThreadDataRegistryImpl> registry_;
     uint64_t tid_;
     T data_;
   };
 
-  ThreadDataRegistry() = default;
+  // methods
+  ThreadDataRegistry() { impl_ = std::make_shared<ThreadDataRegistryImpl>(); }
 
   ThreadDataRegistry(const ThreadDataRegistry&) = delete;
 
   ThreadDataRegistry& operator=(const ThreadDataRegistry&) = delete;
 
   T& CurrentThreadData() {
-    static thread_local ThreadDataHolder thread_data;
+    static thread_local ThreadDataHolder thread_data(impl_);
     return thread_data.GetData();
   }
 
-  std::mutex lock_;
-  std::unordered_map<uint64_t, ThreadDataHolder*> tid_map_;  // not owned
+  // data
+  std::shared_ptr<ThreadDataRegistryImpl> impl_;
 };
 
 }  // namespace framework