[*] Optimize >4 thread wakeups on <= Win7, under Semaphore

Source/Threading/Primitives/AuSemaphore.NT.cpp

@@ -224,8 +224,12 @@ namespace Aurora::Threading::Primitives
         }
         else
         {
-            this->mutex.Lock(); // do not remove
+            // realistically, we cant use the sleep counter optimization trick under windows 7
-            AuAtomicAdd<AuUInt32>(&this->dwState_, count);
+            // we would have to expand our already oversized by 8, 24-byte x86_64 semaphore for a trivial perf boost
+            // we cant efficiently access the conditions state or atomic guarantees...
+
+            this->mutex.Lock(); // do not [re]move this lock fence
+            AuAtomicAdd<AuUInt32>(&this->dwState_, count); // this could be moved anywhere above the unlock, including above the lock.
             this->mutex.Unlock();
 
             if (count == 1)
@@ -234,6 +238,15 @@ namespace Aurora::Threading::Primitives
             }
             else
             {
+                if (count >= 3) // ...this is the only optimization we can hope to achieve
+                {
+                    // we can always save a few cycles by doing an atomic broadcast on a contended semaphore
+                    // waking up the wrong amount of threads probably doesn't matter at this point, on these target platforms
+                    this->var.Broadcast();
+                }
+                else // ...otherwise, do the handshake just a few times
+                {
+                    // doing the condvar handshake for the exact amount of threads you need, once the cond is contended, can pay off
                     for (AU_ITERATE_N(i, count))
                     {
                         (void)i;
@@ -242,6 +255,7 @@ namespace Aurora::Threading::Primitives
                 }
             }
         }
+    }
     
     void SemaphoreImpl::Unlock()
     {