Experimental, alternative busy wait implementation

Disabled with a `constexpr` flag because on a desktop with discrete GPU
this slowed down by about 20%. But the CPU load is about zero. Need to
test on iGPUs, thermal shenanigans might make a difference there.

1 files changed, 66 insertions, 0 deletions

diff --git a/Whisper/Utils/DelayExecution.cpp b/Whisper/Utils/DelayExecution.cpp
new file mode 100644
index 0000000..d238e53
--- /dev/null
+++ b/Whisper/Utils/DelayExecution.cpp
@@ -0,0 +1,66 @@
+#include "stdafx.h"
+#include "DelayExecution.h"
+
+namespace
+{
+	constexpr bool useHighRezTimer = false;
+
+	constexpr int64_t sleepMicroseconds = 200;
+
+	inline HRESULT sleepImpl( HANDLE timer )
+	{
+		constexpr int64_t sleepTicks = sleepMicroseconds * 10;
+
+		LARGE_INTEGER li;
+		// Negative values indicate relative time
+		li.QuadPart = -sleepTicks;
+		if( !SetWaitableTimerEx( timer, &li, 0, nullptr, nullptr, nullptr, 0 ) )
+			return getLastHr();
+		const DWORD res = WaitForSingleObject( timer, 50 );
+		if( res == WAIT_OBJECT_0 )
+			return S_OK;
+		if( res == WAIT_FAILED )
+			return getLastHr();
+		return E_FAIL;
+	}
+}
+
+void DelayExecution::sleepOnTheTimer( const DelayExecution& delay )
+{
+	HRESULT hr = sleepImpl( delay.timer );
+	if( SUCCEEDED( hr ) )
+		return;
+	logWarningHr( hr, u8"DelayExecution.sleepOnTheTimer" );
+}
+
+void DelayExecution::spinWait( const DelayExecution& )
+{
+	for( size_t i = 0; i < 1024; i++ )
+		_mm_pause();
+}
+
+void DelayExecution::sleep( const DelayExecution& )
+{
+	Sleep( 0 );
+}
+
+DelayExecution::DelayExecution()
+{
+	if constexpr( useHighRezTimer )
+	{
+		constexpr DWORD flags = CREATE_WAITABLE_TIMER_HIGH_RESOLUTION;
+		HANDLE h = CreateWaitableTimerEx( nullptr, nullptr, flags, TIMER_ALL_ACCESS );
+		if( nullptr != h )
+		{
+			timer.Attach( h );
+			pfn = &sleepOnTheTimer;
+			return;
+		}
+
+		const HRESULT hr = getLastHr();
+		logWarningHr( hr, u8"CreateWaitableTimerEx" );
+	}
+
+	pfn = &spinWait;
+	// pfn = &sleep;
+}
\ No newline at end of file