[*] TODO: (Finish me) Win7 ILoopQueue doesn't seem to like the in process high perf timer io primitive so much. It works fine under Linux and WaitForMultipleObject WinXP-WinRT.

Windows 7 Perf hack

Source/IO/Loop/LSTimerNoKernelScheduler.cpp

@@ -16,11 +16,12 @@ namespace Aurora::IO::Loop
     void LSTimerIPDispatcher::RunAsync()
     {
         AU_LOCK_GUARD(this->mutex);
-        if (this->pParent)
+
+        // TODO: not as optimized
+        // TODO: we now just added a AuSharedFromThis requirement for no good reason
+        if (auto pThat = AuTryLockMemoryType(this->pParent))
         {
-            AU_LOCK_GUARD(this->pParent->cs);
+            pThat->Set();
-            this->pParent->Set();
-            this->pParent = nullptr;
         }
     }
 
@@ -29,16 +30,12 @@ namespace Aurora::IO::Loop
         reschedStepNsOrZero_(AuMSToNS<AuUInt64>(reschedStepMsOrZero)),
         maxIterationsOrZero_(maxIterationsOrZero)
     {
-        this->targetTime_ = AuTime::CurrentClockNS();
+        this->targetTime_ = AuTime::SteadyClockNS();
     }
 
     LSTimerIP::~LSTimerIP()
     {
-        if (this->pDispatcher)
+      
-        {
-            AU_LOCK_GUARD(this->pDispatcher->mutex);
-            this->pDispatcher->pParent = nullptr;
-        }
     }
 
     void LSTimerIP::UpdateTimeWall(AuUInt64 absTimeMs)
@@ -82,35 +79,17 @@ namespace Aurora::IO::Loop
         AU_DEBUG_MEMCRUNCH;
         AU_LOCK_GUARD(this->cs);
         
-        AuSPtr<LSTimerIPDispatcher> pNext;
-
-        {
-            AU_LOCK_GUARD(this->dispatcherMutex);
-
         if (this->pDispatcher)
         {
             AU_LOCK_GUARD(this->pDispatcher->mutex);
-
+            AuResetMember(this->pDispatcher->pParent);
-                if (this->pDispatcher->pParent)
-                {
-                    // abort
-                    this->pDispatcher->pParent = nullptr;
-                    pNext = this->pDispatcher = AuMakeShared<LSTimerIPDispatcher>();
-                }
-                else
-                {
-                    pNext = this->pDispatcher;
-                }
-            }
-            else
-            {
-                pNext = this->pDispatcher = AuMakeShared<LSTimerIPDispatcher>();
-            }
         }
    
+        // TODO: optimization was removed. readd it.
+        auto pNext = this->pDispatcher = AuMakeShared<LSTimerIPDispatcher>();
         SysAssert(pNext, "impossible OOM during timer op");
 
-        pNext->pParent = this;
+        pNext->pParent = this->SharedFromThis();;
         AuAsync::Schedule(absTimeNs, nullptr, {}, pNext);
         this->targetTime_ = absTimeNs;
     }
@@ -119,11 +98,10 @@ namespace Aurora::IO::Loop
     {
         AU_LOCK_GUARD(this->cs);
 
-        AU_LOCK_GUARD(this->dispatcherMutex);
         if (this->pDispatcher)
         {
             AU_LOCK_GUARD(this->pDispatcher->mutex);
-            this->pDispatcher->pParent = nullptr;
+            AuResetMember(this->pDispatcher->pParent);
             this->pDispatcher = nullptr;
         }
     }
@@ -168,18 +146,9 @@ namespace Aurora::IO::Loop
 
     bool LSTimerIP::OnTrigger(AuUInt handle)
     {
-        AU_LOCK_GUARD(this->cs);
+        return IsSignaledNoSpinIfUserland();
-
-        if (IsSignaledNoSpinIfUserland())
-        {
         return true;
     }
-        else
-        {
-            LSEvent::Reset();
-            return false;
-        }
-    }
 
     bool LSTimerIP::TryTakeWaitNS(AuUInt64 uTimeoutAbs)
     {
@@ -194,13 +163,19 @@ namespace Aurora::IO::Loop
         {
             return true;
         }
+        else
+        {
+            uNextTime = AuAtomicLoad(&this->targetTime_);
+        }
 
-        auto uNow = AuTime::CurrentClockNS();
+        auto uNow = AuTime::SteadyClockNS();
         if (uNow >= uNextTime)
         {
             return this->IsSignaledNonblocking();
         }
 
+        // TODO: this added mess isn't even hit.
+
         // Win32 note: there is quite a bit of hackery behind SleepNs
         // Win 7 and lower may spin for a bit
         // Various CPUs may enter a user-space montior condition
@@ -208,14 +183,42 @@ namespace Aurora::IO::Loop
         // Alderlake et al can tpause
         if (uTimeoutAbs)
         {
-            AuThreading::SleepNs(AuMin(uNextTime, uTimeoutAbs) - uNow);
+            if (uTimeoutAbs < uNextTime)
+            {
+                AuThreading::SleepNs(uTimeoutAbs - uNow);
+                return false;
+            }
+
+            AuThreading::SleepNs(uNextTime - uNow);
+
+            while (uNextTime < AuTime::SteadyClockNS())
+            {
+                if (this->IsSignaledNonblocking())
+                {
+                    return true;
+                }
+            }
+
             return this->IsSignaledNonblocking();
         }
         else
         {
             AuThreading::SleepNs(uNextTime - uNow);
-            return this->IsSignaledNonblocking();
+            if (this->IsSignaledNonblocking())
+            {
+                return true;
             }
+
+            while (AuAtomicLoad(&this->targetTime_) < AuTime::SteadyClockNS())
+            {
+                if (this->IsSignaledNonblocking())
+                {
+                    return true;
+                }
+            }
+        }
+
+        return false;
     }
 
     ELoopSource LSTimerIP::GetType()
@@ -225,7 +228,7 @@ namespace Aurora::IO::Loop
 
     bool LSTimerIP::TryInit()
     {
-        return LSEvent::TryInit(false, false, true);
+        return LSEvent::TryInit(false, true, true);
     }
 
     bool LSTimerIP::IsSignaled()

Source/IO/Loop/LSTimerNoKernelScheduler.hpp

@@ -7,6 +7,7 @@
 ***/
 #pragma once
 
+#include <Source/Async/IAsyncRunnable.hpp>
 #include "LSHandle.hpp"
 #include "LSEvent.hpp"
 
@@ -16,13 +17,13 @@ namespace Aurora::IO::Loop
 
     struct LSTimerIPDispatcher : AuAsync::IAsyncRunnable
     {
-        AuMutex mutex;
+        AuCriticalSection mutex;
-        LSTimerIP *pParent {};
+        AuWPtr<LSTimerIP> pParent {};
 
         void RunAsync() override;
     };
 
-    struct LSTimerIP : LSEvent, ITimer
+    struct LSTimerIP : LSEvent, ITimer, AuEnableSharedFromThis<LSTimerIP>
     {
         LSTimerIP(AuUInt32 reschedStepMsOrZero, AuUInt32 maxIterationsOrZero, bool bSingleshot);
         ~LSTimerIP();
@@ -53,15 +54,14 @@ namespace Aurora::IO::Loop
         void UpdateTimeInternal(AuUInt64 absTimeMs);
         void UpdateTimeInternalSteady(AuUInt64 absTimeMs);
 
-    private:
+    //private:
         friend struct LSTimerIPDispatcher;
         AuUInt32 maxIterationsOrZero_ {};
         AuUInt64 reschedStepNsOrZero_ {}, targetTime_ {};
-        AuUInt32 count_ {};
         bool bSingleshot {};
+        AuUInt32 count_ {};
 
         AuCriticalSection cs;
-        AuMutex dispatcherMutex;
         AuSPtr<LSTimerIPDispatcher> pDispatcher;
         bool IsSignaledNonblocking();
     };

Source/IO/Loop/Loop.cpp

@@ -9,6 +9,7 @@
 #include "Loop.hpp"
 #include "ILoopSourceEx.hpp"
 #include "LSIOHandle.hpp"
+#include "LSTimerNoKernelScheduler.hpp"
 #include <Source/Threading/Primitives/SMTYield.hpp>
 
 namespace Aurora::IO::Loop
@@ -70,6 +71,20 @@ namespace Aurora::IO::Loop
             return;
         }
 
+        if (auto pSemaphore = AuDynamicCast<Loop::LSTimerIP>(pLoopSource))
+        {
+            // TOOD:Hack:
+            AU_LOCK_GUARD(pSemaphore->cs);
+            if (pSemaphore->maxIterationsOrZero_)
+            {
+                AuAtomicSub(&pSemaphore->count_, 1u);
+            }
+
+            pSemaphore->UpdateTimeSteadyNs(pSemaphore->targetTime_ - pSemaphore->reschedStepNsOrZero_);
+            pSemaphore->Set();
+            return;
+        }
+
         if (auto pMutex = AuDynamicCast<Loop::ILSMutex>(pLoopSource))
         {
             pMutex->Unlock();