优化调度器所花费的时间

benchmark/benchmark_async_mem.cpp

@@ -9,18 +9,28 @@
 
 const size_t N = 1000000;
 
+volatile size_t globalValue = 0;
+
 void resumable_main_benchmark_mem()
 {
 	using namespace std::chrono;
+	
+	resumef::state_t<void> st;
+	std::cout << sizeof(st) << " " << sizeof(resumef::promise_vt) << std::endl;
 
 	for (size_t i = 0; i < N; ++i)
 	{
-		GO
+		go[=]()->resumef::future_t<size_t>
 		{
-			for(size_t k = 0; k<100; ++k)
-				co_await resumef::sleep_for(10s);
+			for (size_t k = 0; k < 10; ++k)
+			{
+				globalValue += i * k;
+				co_yield k;
+			}
+			return 0;
 		};
 	}
 
 	resumef::this_scheduler()->run_until_notask();
+	_getch();
 }

librf/src/rf_task.cpp

@@ -6,6 +6,8 @@
 namespace resumef
 {
 	task_base::task_base()
+		: _next_node(nullptr)
+		, _prev_node(nullptr)
 	{
 #if RESUMEF_DEBUG_COUNTER
 		++g_resumef_task_count;

librf/src/rf_task.h

@@ -22,6 +22,8 @@ namespace resumef
 #if RESUMEF_ENABLE_MULT_SCHEDULER
 		virtual void bind(scheduler *) = 0;
 #endif
+		task_base * _next_node;
+		task_base * _prev_node;
 	};
 
 	//----------------------------------------------------------------------------------------------

librf/src/scheduler.cpp

@@ -82,7 +82,7 @@ namespace resumef
 	{
 		if (task)
 		{
-			scoped_lock<std::recursive_mutex> __guard(_mtx_ready);
+			scoped_lock<spinlock> __guard(_mtx_ready);
 #if RESUMEF_ENABLE_MULT_SCHEDULER
 			task->bind(this);
 #endif
@@ -93,22 +93,12 @@ namespace resumef
 	void scheduler::cancel_all_task_()
 	{
 		{
-			scoped_lock<std::recursive_mutex> __guard(_mtx_task);
-			for (auto task : this->_task)
-			{
-				task->cancel();
-				delete task;
-			}
-			this->_task.clear();
+			scoped_lock<lock_type> __guard(_mtx_task);
+			this->_task.clear(true);
 		}
 		{
-			scoped_lock<std::recursive_mutex> __guard(_mtx_ready);
-			for (auto task : this->_ready_task)
-			{
-				task->cancel();
-				delete task;
-			}
-			this->_ready_task.clear();
+			scoped_lock<spinlock> __guard(_mtx_ready);
+			this->_ready_task.clear(true);
 		}
 	}
 
@@ -116,7 +106,7 @@ namespace resumef
 	{
 		cancel_all_task_();
 
-		scoped_lock<std::recursive_mutex> __guard(_mtx_task);
+		scoped_lock<lock_type> __guard(_mtx_task);
 		this->_timer->clear();
 	}
 
@@ -127,13 +117,13 @@ namespace resumef
 			th_scheduler_ptr = this;
 #endif
 		{
-			scoped_lock<std::recursive_mutex> __guard(_mtx_task);
+			scoped_lock<lock_type> __guard(_mtx_task);
 
 			this->_timer->update();
 
 			using namespace std::chrono;
 
-			for (auto iter = this->_task.begin(); iter != this->_task.end(); )
+			for (auto task = this->_task.begin(); task != nullptr; )
 			{
 #if _DEBUG
 #define MAX_TIME_COST 10000us
@@ -142,24 +132,21 @@ namespace resumef
 #endif
 //				time_cost_evaluation<microseconds> eva(MAX_TIME_COST);
 
-				auto task = *iter;
 				if (task->is_suspend() || task->go_next(this))
 				{
 //					eva.add("coscheduler");
-					++iter;
+					task = task->_next_node;
 					continue;
 				}
 
-				iter = this->_task.erase(iter);
-				delete task;
+				task = this->_task.erase(task, false);
 			}
 
 			{
-				scoped_lock<std::recursive_mutex> __guard(_mtx_ready);
-				if (this->_ready_task.size() > 0)
+				scoped_lock<spinlock> __guard(_mtx_ready);
+				if (!this->_ready_task.empty())
 				{
-					this->_task.insert(this->_task.end(), this->_ready_task.begin(), this->_ready_task.end());
-					this->_ready_task.clear();
+					this->_task.merge_back(this->_ready_task);
 				}
 			}
 		}

librf/src/scheduler.h

@@ -6,6 +6,7 @@
 //#include <yvals.h>
 
 #include "rf_task.h"
+#include "task_list.h"
 #include "utils.h"
 #include "timer.h"
 
@@ -16,11 +17,14 @@ namespace resumef
 	struct scheduler : public std::enable_shared_from_this<scheduler>
 	{
 	private:
-		mutable std::recursive_mutex _mtx_ready;
-		std::deque<task_base *> _ready_task;
+		//typedef spinlock lock_type;
+		typedef std::recursive_mutex lock_type;
 
-		mutable std::recursive_mutex _mtx_task;
-		std::list<task_base *> _task;
+		mutable spinlock _mtx_ready;
+		task_list _ready_task;
+
+		mutable lock_type _mtx_task;
+		task_list _task;
 		timer_mgr_ptr	_timer;
 
 		RF_API void new_task(task_base * task);
@@ -47,11 +51,9 @@ namespace resumef
 
 		inline bool empty() const
 		{
-			scoped_lock<std::recursive_mutex, std::recursive_mutex> __guard(_mtx_ready, _mtx_task);
+			scoped_lock<spinlock, lock_type> __guard(_mtx_ready, _mtx_task);
 
-			return
-				(this->_task.size() + this->_ready_task.size()) == 0 &&
-				this->_timer->empty();
+			return this->_task.empty() && this->_ready_task.empty() && this->_timer->empty();
 		}
 
 		RF_API void break_all();

librf/src/spinlock.h

@@ -9,36 +9,54 @@ namespace resumef
 {
 	struct spinlock
 	{
-		volatile std::atomic_flag lck;
+		static const size_t MAX_ACTIVE_SPIN = 4000;
+		static const int FREE_VALUE = 0;
+		static const int LOCKED_VALUE = 1;
+
+		volatile std::atomic<int> lck;
 
 		spinlock()
 		{
-			lck.clear();
+			lck = FREE_VALUE;
 		}
 
 		void lock()
 		{
-			if (std::atomic_flag_test_and_set_explicit(&lck, std::memory_order_acquire))
+			using namespace std::chrono;
+
+			int val = FREE_VALUE;
+			if (!lck.compare_exchange_weak(val, LOCKED_VALUE, std::memory_order_acquire))
 			{
-				using namespace std::chrono;
+				size_t spinCount = 0;
 				auto dt = 1ms;
-				while (std::atomic_flag_test_and_set_explicit(&lck, std::memory_order_acquire))
+				do
 				{
-					std::this_thread::sleep_for(dt);
-					dt *= 2;
-				}
+					while (lck.load(std::memory_order_relaxed) != FREE_VALUE)
+					{
+						if (spinCount < MAX_ACTIVE_SPIN)
+							++spinCount;
+						else
+						{
+							std::this_thread::sleep_for(dt);
+							dt *= 2;
+						}
+					}
+
+					val = FREE_VALUE;
+				} while (!lck.compare_exchange_weak(val, LOCKED_VALUE, std::memory_order_acquire));
 			}
 		}
 
 		bool try_lock()
 		{
-			bool ret = !std::atomic_flag_test_and_set_explicit(&lck, std::memory_order_acquire);
+			int val = FREE_VALUE;
+			bool ret = lck.compare_exchange_weak(val, LOCKED_VALUE, std::memory_order_acquire);
 			return ret;
 		}
 
 		void unlock()
 		{
-			std::atomic_flag_clear_explicit(&lck, std::memory_order_release);
+			lck.store(FREE_VALUE, std::memory_order_release);
 		}
 	};
 }

librf/src/state.h

@@ -2,6 +2,7 @@
 #pragma once
 
 #include "def.h"
+#include "spinlock.h"
 #include "counted_ptr.h"
 #include <iostream>
 
@@ -13,7 +14,9 @@ namespace resumef
 	struct state_base
 	{
 	protected:
+		//typedef spinlock lock_type;
 		typedef std::recursive_mutex lock_type;
+
 		lock_type	_mtx;		//for value, _exception
 		RF_API void set_value_none_lock();
 #if RESUMEF_ENABLE_MULT_SCHEDULER

librf/src/task_list.h

@@ -0,0 +1,168 @@
+#pragma once
+
+namespace resumef
+{
+	struct task_list
+	{
+		using value_type = task_base;
+		using size_type = std::size_t;
+		using difference_type = std::ptrdiff_t;
+		using pointer = value_type *;
+		using const_pointer = const value_type *;
+		using reference = value_type & ;
+		using const_reference = const value_type&;
+
+		//using iterator = typename _Mybase::iterator;
+		//using const_iterator = typename _Mybase::const_iterator;
+		//using reverse_iterator = std::reverse_iterator<iterator>;
+		//using const_reverse_iterator = std::reverse_iterator<const_iterator>;
+		
+	private:
+		pointer			_M_header;
+		pointer			_M_last;
+	public:
+		task_list()
+			: _M_header(nullptr)
+			, _M_last(nullptr)
+		{
+		}
+
+		~task_list()
+		{
+			clear(true);
+		}
+
+		task_list(const task_list& _Right) = delete;
+		task_list& operator=(const task_list& _Right) = delete;
+
+		task_list(task_list&& _Right)
+		{
+			_M_header = _Right._M_header;
+			_Right._M_header = nullptr;
+			_M_last = _Right._M_last;
+			_Right._M_last = nullptr;
+		}
+
+		task_list& operator=(task_list&& _Right)
+		{	// assign by moving _Right
+			if (this != std::addressof(_Right))
+			{	// different, assign it
+				clear(true);
+				_Move_assign(_Right);
+			}
+			return (*this);
+		}
+
+		void clear(bool cancel_)
+		{
+			pointer header = _M_header;
+			_M_header = _M_last = nullptr;
+
+			for (; header; )
+			{
+				pointer temp = header;
+				header = header->_next_node;
+
+				if (cancel_)
+					temp->cancel();
+				delete temp;
+			}
+		}
+
+		void push_back(pointer node)
+		{
+			assert(node != nullptr);
+			assert(node->_next_node == nullptr);
+			assert(node->_prev_node == nullptr);
+			_Push_back(node);
+		}
+
+		pointer erase(pointer node, bool cancel_)
+		{
+			assert(node != nullptr);
+
+			pointer const next = node->_next_node;
+			pointer const prev = node->_prev_node;
+			if (next)
+				next->_prev_node = prev;
+			if (prev)
+				prev->_next_node = next;
+
+			if (_M_header == node)
+				_M_header = next;
+			if (_M_last == node)
+				_M_last = prev;
+
+			if (cancel_)
+				node->cancel();
+			delete node;
+
+			return next;
+		}
+
+		void merge_back(task_list & _Right)
+		{
+			if (this == std::addressof(_Right) || _Right._M_header == nullptr)
+				return;
+
+			if (_M_header == nullptr)
+			{
+				_Move_assign(_Right);
+			}
+			else
+			{
+				assert(_M_last != nullptr);
+
+				_M_last->_next_node = _Right._M_header;
+				_Right._M_header->_prev_node = _M_last;
+				_M_last = _Right._M_last;
+
+				_Right._M_header = _Right._M_last = nullptr;
+			}
+		}
+
+		bool empty() const
+		{
+			return _M_header == nullptr;
+		}
+
+		pointer begin() const
+		{
+			return _M_header;
+		}
+
+		pointer end() const
+		{
+			return nullptr;
+		}
+
+	private:
+		void _Push_back(pointer node)
+		{
+			if (_M_header == nullptr)
+			{
+				assert(_M_last == nullptr);
+				_M_header = _M_last = node;
+			}
+			else
+			{
+				assert(_M_last != nullptr);
+				_M_last->_next_node = node;
+				node->_prev_node = _M_last;
+
+				_M_last = node;
+			}
+		}
+
+		void _Move_assign(task_list& _Right)
+		{
+			assert(this != std::addressof(_Right));
+				
+			_M_header = _Right._M_header;
+			_Right._M_header = nullptr;
+			_M_last = _Right._M_last;
+			_Right._M_last = nullptr;
+		}
+	};
+
+}

tutorial/test_async_resumable.cpp

@@ -7,8 +7,8 @@
 #include "librf.h"
 
 
-//static const int N = 100000000;
-static const int N = 10;
+static const int N = 10000000;
+//static const int N = 10;
 
 template <typename T>
 void dump(std::string name, int n, T start, T end)
@@ -47,6 +47,10 @@ void resumable_switch(int coro)
 void resumable_main_resumable()
 {
 	resumable_switch(1);
+	resumable_switch(10);
+	resumable_switch(100);
 	resumable_switch(1000);
-	resumable_switch(1000000);
+	resumable_switch(10000);
+
+	//resumable_switch(10000000);
 }

vs_proj/librf.cpp

@@ -22,10 +22,9 @@ extern void resumable_main_benchmark_mem();
 
 int main(int argc, const char * argv[])
 {
+	//resumable_main_resumable();
 	resumable_main_when_all();
-	//resumable_main_multi_thread();
-	return 0;
-
+	resumable_main_multi_thread();
 	resumable_main_yield_return();
 	resumable_main_timer();
 	resumable_main_suspend_always();

vs_proj/librf.vcxproj

@@ -203,6 +203,7 @@
     <ClInclude Include="..\librf\src\event.h" />
     <ClInclude Include="..\librf\src\future.h" />
     <ClInclude Include="..\librf\src\generator.h" />
+    <ClInclude Include="..\librf\src\task_list.h" />
     <ClInclude Include="..\librf\src\mutex.h" />
     <ClInclude Include="..\librf\src\rf_task.h" />
     <ClInclude Include="..\librf\src\scheduler.h" />

vs_proj/librf.vcxproj.filters

@@ -159,5 +159,8 @@
     <ClInclude Include="..\librf\src\when.h">
       <Filter>librf\src</Filter>
     </ClInclude>
+    <ClInclude Include="..\librf\src\task_list.h">
+      <Filter>librf\src</Filter>
+    </ClInclude>
   </ItemGroup>
 </Project>