[+] Non-mutually exclusive binary semaphore / event wait path

[+] ThreadingConfig::gPreferFutexEvent

Include/Aurora/Runtime.hpp

@@ -379,11 +379,12 @@ namespace Aurora
         AuUInt64 uAdaptiveSpinCUCnt8               : 4 { 2     };
         AuUInt64 uAdaptiveSpinCUCnt16              : 4 { 4     };
         AuUInt64 bPreferFutexRWLock                : 1 { true  };
-        AuUInt64 bWinXpThrough7BlazeOptimizerPower : 7 { 6     }; // dont worry about it. we dont care about old portables. lets try to make older win32 targets tweak the scheduling in our favor a bit.
+        AuUInt64 bWinXpThrough7BlazeOptimizerPower : 12 { 300  }; // dont worry about it. we dont care about old portables. lets try to make older win32 targets tweak the scheduling in our favor a bit.
         AuUInt64 bPreferLinuxPrimitivesFutexNoSpin : 1 { false };
         AuUInt64 bPreferUnixPrimitivesNoSpin       : 1 { false };
         AuUInt64 bAlwaysRWLockWriteBiasOnReadLock  : 1 { false };
         AuUInt64 bEnableRWLockWriteBiasOnReadLock  : 1 { true  };
+        AuUInt64 bPreferFutexEvent                 : 1 { true  };
     };
 
     struct DummyConfig

Source/Threading/Primitives/AuEvent.cpp

@@ -8,6 +8,7 @@
 #include <Source/RuntimeInternal.hpp>
 #include "AuEvent.hpp"
 #include "SMTYield.hpp"
+#include "../AuWakeInternal.hpp"
 
 namespace Aurora::Threading::Primitives
 {
@@ -44,27 +45,50 @@ namespace Aurora::Threading::Primitives
             uStartTime = Time::SteadyClockNS();
             uEndTime = uStartTime + uTimeout;
         }
-        
-        AU_LOCK_GUARD(this->mutex_);
 
-        while (!AtomicIsEventSetLogicNoSpinNoLock())
+        if (gPreferFutexEvent)
         {
-            AuUInt32 uTimeoutNS {};
+            while (!AtomicIsEventSetLogicNoSpinNoLock())
-
-            if (uTimeout)
             {
-                uStartTime = Time::SteadyClockNS();
+                EventBits bits;
-                if (uStartTime >= uEndTime)
+                bits.state = AuAtomicLoad(&this->state_);
+
+                if (bits.bTriggered)
                 {
-                    return false;
+                    if (!InternalLTSWaitOnAddressHighRes(&this->state_, &bits.state, sizeof(bits.state), uEndTime))
+                    {
+                        return false;
+                    }
+                }
+                else
+                {
+                    SMPPause();
+                }
+            }
+        }
+        else
+        {
+            AU_LOCK_GUARD(this->mutex_);
+
+            while (!AtomicIsEventSetLogicNoSpinNoLock())
+            {
+                AuUInt32 uTimeoutNS {};
+
+                if (uTimeout)
+                {
+                    uStartTime = Time::SteadyClockNS();
+                    if (uStartTime >= uEndTime)
+                    {
+                        return false;
+                    }
+
+                    uTimeoutNS = uEndTime - uStartTime;
                 }
 
-                uTimeoutNS = uEndTime - uStartTime;
+                if (!this->condition_.WaitForSignalNsEx(&this->mutex_, uTimeoutNS))
-            }
+                {
-
+                    continue;
-            if (!this->condition_.WaitForSignalNsEx(&this->mutex_, uTimeoutNS))
+                }
-            {
-                continue;
             }
         }
 
@@ -139,17 +163,7 @@ namespace Aurora::Threading::Primitives
             }
         }
 
-        this->mutex_.Lock();
+        this->DoSignal(bits);
-        this->mutex_.Unlock();
-
-        if (bits.bAtomicRelease)
-        {
-            this->condition_.Signal();
-        }
-        else
-        {
-            this->condition_.Broadcast();
-        }
     }
 
     bool EventImpl::TrySet()
@@ -176,19 +190,38 @@ namespace Aurora::Threading::Primitives
             }
         }
 
-        this->mutex_.Lock();
+        this->DoSignal(bits);
-        this->mutex_.Unlock();
 
-        if (bits.bAtomicRelease)
+        return true;
+    }
+
+    void EventImpl::DoSignal(const EventBits &bits)
+    {
+        if (gPreferFutexEvent)
         {
-            this->condition_.Signal();
+            if (bits.bAtomicRelease)
+            {
+                InternalLTSWakeOne(&this->state_);
+            }
+            else
+            {
+                InternalLTSWakeAll(&this->state_);
+            }
         }
         else
         {
-            this->condition_.Broadcast();
+            this->mutex_.Lock();
-        }
+            this->mutex_.Unlock();
 
-        return true;
+            if (bits.bAtomicRelease)
+            {
+                this->condition_.Signal();
+            }
+            else
+            {
+                this->condition_.Broadcast();
+            }
+        }
     }
 
     bool EventImpl::HasOSHandle(AuMach &mach)

Source/Threading/Primitives/AuEvent.hpp

@@ -64,6 +64,8 @@ namespace Aurora::Threading::Primitives
             AuUInt8 state_;
         };
 
+
+        void DoSignal(const EventBits &bits);
     #endif
     };
 }

Source/Threading/Primitives/SMTYield.cpp

@@ -78,8 +78,12 @@ namespace Aurora::Threading::Primitives
     {
         auto uCores = AuHwInfo::GetCPUInfo().uThreads;
 
+        bool bPermitWOAInternal = IsWaitOnRecommended();
         gUseFutexRWLock = ThrdCfg::gPreferFutexRWLock &&
-            IsWaitOnRecommended();
+            bPermitWOAInternal;
+
+        gPreferFutexEvent = ThrdCfg::gPreferFutexEvent &&
+            bPermitWOAInternal;
 
         gSpinAdaptiveThreadCount = uCores;
 
@@ -176,5 +180,6 @@ namespace Aurora::Threading::Primitives
         ThrdCfg::gPreferUnixPrimitivesNoSpin        = gRuntimeConfig.threadingConfig.bPreferUnixPrimitivesNoSpin;
         ThrdCfg::gAlwaysRWLockWriteBiasOnReadLock   = gRuntimeConfig.threadingConfig.bAlwaysRWLockWriteBiasOnReadLock;
         ThrdCfg::gEnableRWLockWriteBiasOnReadLock   = gRuntimeConfig.threadingConfig.bEnableRWLockWriteBiasOnReadLock;
+        ThrdCfg::gPreferFutexEvent                  = gRuntimeConfig.threadingConfig.bPreferFutexEvent;
     }
 }

Source/Threading/Primitives/SMTYield.hpp

@@ -45,6 +45,7 @@ namespace Aurora::Threading::Primitives
         inline AuUInt32     gAdaptiveSpinCUCnt8 {};
         inline AuUInt32     gAdaptiveSpinCUCnt16 {};
         inline bool         gPreferFutexRWLock {};
+        inline bool         gPreferFutexEvent {};
         inline bool         gWinXpThrough7BlazeOptimizerPower {};
         inline bool         gPreferLinuxPrimitivesFutexNoSpin {};
         inline bool         gPreferUnixPrimitivesNoSpin {};
@@ -57,6 +58,7 @@ namespace Aurora::Threading::Primitives
     inline AuUInt32 gSpinAdaptiveThreadCount {};
 
     inline AuUInt32 gUseFutexRWLock {};
+    inline AuUInt32 gPreferFutexEvent {};
 
     void InitAdaptiveThreshold();
     void InitAdaptiveThresholdFirstTime();