[+] while_bc

Include/Aurora/Threading/Waitables/FutexCondWaitable.hpp

@@ -122,7 +122,11 @@ namespace Aurora::Threading::Waitables
             AuUInt32 uWaitCount {};
             AuUInt32 uWaiters {};
 
+        #if defined(AURORA_RUNTIME_FUTEX_AGGRESSIVE_COND_WAKE)
             while ((uWaiters = AuAtomicLoad(&this->uAtomicSleeping)))
+        #else
+            if ((uWaiters = AuAtomicLoad(&this->uAtomicSleeping)))
+        #endif
             {
                 AuAtomicAdd(&this->uAtomicState, uWaiters);
                 WakeNOnAddress((const void *)&this->uAtomicState, uWaiters);

Source/Threading/Primitives/AuConditionEx.cpp

@@ -8,6 +8,7 @@
 #include <Source/RuntimeInternal.hpp>
 #include "AuConditionEx.hpp"
 #include "AuSemaphore.Generic.hpp"
+#include "SMTYield.hpp" // for: while_bc
 
 namespace Aurora::Threading::Primitives
 {
@@ -144,7 +145,7 @@ namespace Aurora::Threading::Primitives
         AuUInt32 uWaitCount {};
         AuUInt32 uWaiters {};
 
-        while ((uWaiters = AuAtomicLoad(&this->uWaiters_)))
+        while_bc ((uWaiters = AuAtomicLoad(&this->uWaiters_)))
         {
             this->s_.Unlock(uWaiters);
             uWaitCount = uWaiters;

Source/Threading/Primitives/AuConditionVariable.Generic.cpp

@@ -90,7 +90,7 @@ namespace Aurora::Threading::Primitives
         AuUInt32 uWaitCount {};
         AuUInt32 uWaiters {};
 
-        while ((uWaiters = AuAtomicLoad(&this->uSleeping_)))
+        while_bc (uWaiters = AuAtomicLoad(&this->uSleeping_))
         {
             AuAtomicAdd(&this->uState_, uWaiters);
             InternalLTSWakeCount(&this->uState_, uWaiters);

Source/Threading/Primitives/AuConditionVariable.Linux.cpp

@@ -180,7 +180,7 @@ namespace Aurora::Threading::Primitives
         AuUInt32 uWaitCount {};
         AuUInt32 uWaiters {};
 
-        while ((uWaiters = AuAtomicLoad(&this->uSleeping_)))
+        while_bc ((uWaiters = AuAtomicLoad(&this->uSleeping_)))
         {
             AuAtomicAdd(&this->uState_, uWaiters);
             futex_wake(&this->uState_, uWaiters);

Source/Threading/Primitives/SMTYield.hpp

@@ -22,6 +22,31 @@ extern "C"
 
 #define SPIN_FOUR 1
 
+#define while_bc___(exp, ex) \
+    AuUInt32 __wsc ## ex {}; \
+    while ((exp) && ((__wsc ## ex++) < 2))
+#define while_bc__(exp, ex) while_bc___(exp, ex)
+#define while_bc_(exp, ex) while_bc__(exp, AU_WHAT(ex))
+
+#if defined(AURORA_RUNTIME_ALWAYS_SPIN_ON_BROADCAST)
+    #define while_bc(exp) while (exp)
+#elif defined(AURORA_RUNTIME_ALWAYS_CHECK_ONCE_ON_BROADCAST)
+    #define while_bc(exp) if (exp)
+#elif defined(__COUNTER__)
+    #define while_bc(exp) while_bc_(exp, __COUNTER__)
+#else
+    #define while_bc(exp) while_bc_(exp, __LINE__)
+#endif
+
+// Replace condition variable broadcasts: `if/while (waiters) { signal() }` loops with while_bc.
+// while_bc will attempt to rebroadcast if another sleeper turns up. On unicore systems, 
+// depending on the scheduler, spinning in this fashion may result in a deadlock in a tight
+// enough wait loop. In other systems, having a `while (waiters) { signal(); }` may help
+// improve performance when other threads are tying up the system scheduler on the wake side.
+// That way some threads can be late in, and we dont have to worry so much about there being a 
+// wake-up spin lock under real world use cases. To strike a balance between the two conditions,  
+// we add a little bit of extra branching overhead to ensure we don't spin more than 2-3 times.
+
 namespace Aurora::Threading::Primitives
 {
     namespace ThrdCfg