ADDED option to disable multi-threaded pixel filtering

REDUCED maximum number of cores to use to 8

git-svn-id: https://svn.code.sf.net/p/vbam/code/trunk@473 a31d4220-a93d-0410-bf67-fe4944624d44

src/win32/Direct3D.cpp

@@ -173,7 +173,7 @@ Direct3DDisplay::Direct3DDisplay()
 	pfthread_data = NULL;
 	hThreads = NULL;
 	nThreads = theApp.maxCpuCores;
-	if( nThreads > 16 ) nThreads = 16;
+	if( nThreads > 8 ) nThreads = 8;
 }
 
 
@@ -444,75 +444,75 @@ void Direct3DDisplay::render()
 		return;
 	} else {
 		u32 pitch = theApp.sizeX * ( systemColorDepth >> 3 ) + 4;
+
 		if( theApp.filterFunction ) {
-			u8 *start = pix + pitch;
+			if( theApp.filterMT ) {
-			int src_height_per_thread = theApp.sizeY / nThreads;
+				u8 *start = pix + pitch;
-			int src_height_remaining = theApp.sizeY - ( ( theApp.sizeY / nThreads ) * nThreads );
+				int src_height_per_thread = theApp.sizeY / nThreads;
-			u32 src_bytes_per_thread = pitch * src_height_per_thread;
+				int src_height_remaining = theApp.sizeY - ( ( theApp.sizeY / nThreads ) * nThreads );
+				u32 src_bytes_per_thread = pitch * src_height_per_thread;
 
-			int dst_height_per_thread = src_height_per_thread * theApp.filterMagnification;
+				int dst_height_per_thread = src_height_per_thread * theApp.filterMagnification;
-			u32 dst_bytes_per_thread = lr.Pitch * dst_height_per_thread;
+				u32 dst_bytes_per_thread = lr.Pitch * dst_height_per_thread;
 
-			unsigned int i = nThreads - 1;
+				unsigned int i = nThreads - 1;
 
-			// Use Multi Threading
+				// Use Multi Threading
-			do {
+				do {
-				// create last thread first because it could have more work than the others (for eg. if nThreads = 3)
+					// create last thread first because it could have more work than the others (for eg. if nThreads = 3)
-				// (last thread has to process the remaining lines if (height / nThreads) is not an integer)
+					// (last thread has to process the remaining lines if (height / nThreads) is not an integer)
 
-				// configure thread
+					// configure thread
-				pfthread_data[i].filterFunction = theApp.filterFunction;
+					pfthread_data[i].filterFunction = theApp.filterFunction;
-				pfthread_data[i].sourcePointer = start + ( i * src_bytes_per_thread );
+					pfthread_data[i].sourcePointer = start + ( i * src_bytes_per_thread );
-				pfthread_data[i].sourcePitch = pitch;
+					pfthread_data[i].sourcePitch = pitch;
-				pfthread_data[i].deltaPointer = (u8*)theApp.delta; // TODO: check if thread-safe
+					pfthread_data[i].deltaPointer = (u8*)theApp.delta; // TODO: check if thread-safe
-				pfthread_data[i].destPointer = ( (u8*)lr.pBits ) + ( i * dst_bytes_per_thread );
+					pfthread_data[i].destPointer = ( (u8*)lr.pBits ) + ( i * dst_bytes_per_thread );
-				pfthread_data[i].destPitch = lr.Pitch;
+					pfthread_data[i].destPitch = lr.Pitch;
-				pfthread_data[i].width = theApp.sizeX;
+					pfthread_data[i].width = theApp.sizeX;
-				
+					
-				if( i == ( nThreads - 1 ) ) {
+					if( i == ( nThreads - 1 ) ) {
-					// last thread
+						// last thread
-					pfthread_data[i].height = src_height_per_thread + src_height_remaining;
+						pfthread_data[i].height = src_height_per_thread + src_height_remaining;
-				} else {
+					} else {
-					// other thread
+						// other thread
-					pfthread_data[i].height = src_height_per_thread;
+						pfthread_data[i].height = src_height_per_thread;
+					}
+
+					// create thread
+					hThreads[i] = CreateThread(
+						NULL,
+						0,
+						pfthread_func,
+						&pfthread_data[i],
+						0,
+						NULL );
+					assert( hThreads[i] != NULL );
+				} while ( i-- );
+
+				// Wait until every thread has finished.
+				WaitForMultipleObjects(
+					nThreads,
+					hThreads,
+					TRUE,
+					INFINITE );
+
+				// Close all thread handles.
+				for( i = 0 ; i < nThreads ; i++ ) {
+					CloseHandle( hThreads[i] );
 				}
-
+			} else {
-				// create thread
+				// multi-threading disabled
-				hThreads[i] = CreateThread(
+				theApp.filterFunction(
-					NULL,
+					pix + pitch,
-					0,
+					pitch,
-					pfthread_func,
+					(u8*)theApp.delta,
-					&pfthread_data[i],
+					(u8*)lr.pBits,
-					0,
+					lr.Pitch,
-					NULL );
+					theApp.sizeX,
-				assert( hThreads[i] != NULL );
+					theApp.sizeY
-			} while ( i-- );
+					);
-
-			// Wait until every thread has finished.
-			WaitForMultipleObjects(
-				nThreads,
-				hThreads,
-				TRUE,
-				INFINITE );
-
-			// Close all thread handles.
-			for( i = 0 ; i < nThreads ; i++ ) {
-				CloseHandle( hThreads[i] );
 			}
-
-
-			/* without threads
-			// pixel filter enabled
-			theApp.filterFunction(
-				pix + pitch,
-				pitch,
-				(u8*)theApp.delta,
-				(u8*)lr.pBits,
-				lr.Pitch,
-				theApp.sizeX,
-				theApp.sizeY
-				);
-			*/
 		} else {
 			// pixel filter disabled
 			switch( systemColorDepth )

src/win32/MainWnd.cpp

@@ -437,6 +437,8 @@ BEGIN_MESSAGE_MAP(MainWnd, CWnd)
   ON_WM_NCRBUTTONDOWN()
   ON_COMMAND(ID_OUTPUTAPI_XAUDIO2, &MainWnd::OnOutputapiXaudio2)
   ON_UPDATE_COMMAND_UI(ID_OUTPUTAPI_XAUDIO2, &MainWnd::OnUpdateOutputapiXaudio2)
+  ON_COMMAND(ID_PIXELFILTER_MULTI, &MainWnd::OnPixelfilterMultiThreading)
+  ON_UPDATE_COMMAND_UI(ID_PIXELFILTER_MULTI, &MainWnd::OnUpdatePixelfilterMultiThreading)
   END_MESSAGE_MAP()
 
 

src/win32/MainWnd.h

@@ -425,6 +425,8 @@ public:
 	afx_msg void OnNcRButtonDown(UINT nHitTest, CPoint point);
 	afx_msg void OnOutputapiXaudio2();
 	afx_msg void OnUpdateOutputapiXaudio2(CCmdUI *pCmdUI);
+	afx_msg void OnPixelfilterMultiThreading();
+	afx_msg void OnUpdatePixelfilterMultiThreading(CCmdUI *pCmdUI);
 };
 
 /////////////////////////////////////////////////////////////////////////////

src/win32/MainWndOptions.cpp

@@ -1854,3 +1854,13 @@ void MainWnd::OnEmulatorBiosfiles()
 		theApp.biosFileNameGB = dlg.m_pathGB;
 	}
 }
+
+void MainWnd::OnPixelfilterMultiThreading()
+{
+	theApp.filterMT = !theApp.filterMT;
+}
+
+void MainWnd::OnUpdatePixelfilterMultiThreading(CCmdUI *pCmdUI)
+{
+	pCmdUI->SetCheck( theApp.filterMT ? 1 : 0 );
+}

src/win32/VBA.cpp

@@ -215,6 +215,8 @@ VBA::VBA()
   if( S_OK != CoInitializeEx( NULL, COINIT_APARTMENTTHREADED ) ) {
 	  systemMessage( IDS_COM_FAILURE, NULL );
   }
+
+  // ! keep in mind that many of the following values will be really initialized in loadSettings()
   mode320Available = false;
   mode640Available = false;
   mode800Available = false;
@@ -229,6 +231,7 @@ VBA::VBA()
   filterType = FILTER_NONE;
   filterWidth = 0;
   filterHeight = 0;
+  filterMT = false;
   fsAdapter = 0;
   fsWidth = 0;
   fsHeight = 0;
@@ -1543,6 +1546,8 @@ void VBA::loadSettings()
   if(filterType < 0 || filterType > 17)
     filterType = 0;
 
+  filterMT = ( 1 == regQueryDwordValue("filterEnableMultiThreading", 0) );
+
   disableMMX = regQueryDwordValue("disableMMX", false) ? true: false;
 
   disableStatusMessage = regQueryDwordValue("disableStatus", 0) ? true : false;
@@ -1681,7 +1686,7 @@ void VBA::loadSettings()
 
   updateThrottle( (unsigned short)regQueryDwordValue( "throttle", 0 ) );
 
-   linktimeout = regQueryDwordValue("LinkTimeout", 1000);
+  linktimeout = regQueryDwordValue("LinkTimeout", 1000);
 
   linklog = regQueryDwordValue("Linklog", false) ? true : false;
   if(linklog)
@@ -1704,6 +1709,11 @@ void VBA::loadSettings()
 
   oalBufferCount = regQueryDwordValue( "oalBufferCount", 5 );
 #endif
+
+  if( ( maxCpuCores == 1 ) && filterMT ) {
+	  // multi-threading use useless for just one core
+	  filterMT = false;
+  }
 }
 
 void VBA::updateFrameSkip()
@@ -2511,6 +2521,8 @@ void VBA::saveSettings()
 
   regSetDwordValue("filter", filterType);
 
+  regSetDwordValue("filterEnableMultiThreading", filterMT ? 1 : 0);
+
   regSetDwordValue("LCDFilter", filterLCD);
 
   regSetDwordValue("disableMMX", disableMMX);

src/win32/VBA.h

@@ -91,6 +91,7 @@ class VBA : public CWinApp
   int filterHeight;
   int filterMagnification;
   int filterLCD;
+  bool filterMT; // enable multi-threading for pixel filters
   int fsWidth;
   int fsHeight;
   int fsColorDepth;

src/win32/VBA.rc

@@ -1661,6 +1661,7 @@ BEGIN
                 MENUITEM "&Smart",                      ID_OPTIONS_FILTER_INTERFRAMEBLENDING_SMART
             END
             MENUITEM SEPARATOR
+            MENUITEM "Multi-Threading (D3D)",       ID_PIXELFILTER_MULTI
             MENUITEM "Disable &MMX",                ID_OPTIONS_FILTER_DISABLEMMX
         END
         POPUP "&Audio"

src/win32/resource.h

@@ -850,13 +850,14 @@
 #define ID_FILE_OPENGBC                 40357
 #define ID_FILE_OPEN_GBC                40358
 #define ID_OUTPUTAPI_XAUDIO2            40359
+#define ID_PIXELFILTER_MULTI            40360
 
 // Next default values for new objects
 // 
 #ifdef APSTUDIO_INVOKED
 #ifndef APSTUDIO_READONLY_SYMBOLS
 #define _APS_NEXT_RESOURCE_VALUE        163
-#define _APS_NEXT_COMMAND_VALUE         40360
+#define _APS_NEXT_COMMAND_VALUE         40361
 #define _APS_NEXT_CONTROL_VALUE         1284
 #define _APS_NEXT_SYMED_VALUE           103
 #endif