More MTGS improvements and cleanups. I actually changed some of how the MTGS ring buffer is managed this time, so it might introduce some bugs (lets hope not!).

Removed the SPU2timestretch auto-enable/disable code.  Also added checks for frameskipping support to the F4 keypress (just in case).

git-svn-id: http://pcsx2-playground.googlecode.com/svn/trunk@344 a6443dda-0b58-4228-96e9-037be469359c

pcsx2/FiFo.c

@@ -133,14 +133,6 @@ void WriteFIFO(u32 mem, u64 *value) {
 			data[1] = value[1];
 			GSgifTransferDummy(2, (u32*)data, 1);
 			GSRINGBUF_DONECOPY((u8*)data, 16);
-
-			if( !CHECK_DUALCORE ) {
-#if defined(_WIN32) && !defined(WIN32_PTHREADS)
-				SetEvent(g_hGsEvent);
-#else
-				pthread_cond_signal(&g_condGsEvent);
-#endif
-			}
 		}
 		else {
 			FreezeXMMRegs(1);

pcsx2/GS.cpp

@@ -253,6 +253,19 @@ static __forceinline void gsSetEventWait()
 	}
 }
 
+// mem and size are the ones from GSRingBufCopy
+void GSRINGBUF_DONECOPY(const u8 *mem, u32 size)
+{
+	u8* temp = (u8*)(mem) + (size); 
+
+	assert( temp <= GS_RINGBUFFEREND); 
+
+	if( temp == GS_RINGBUFFEREND ) temp = GS_RINGBUFFERBASE; 
+	InterlockedExchangePointer((void**)&g_pGSWritePos, temp);
+	if( !CHECK_DUALCORE ) GS_SETEVENT();
+}
+
+
 void gsShutdown()
 {
 	if( CHECK_MULTIGS ) {
@@ -310,80 +323,88 @@ u8* GSRingBufCopy(void* mem, u32 size, u32 type)
 	// is reading it.
 
 	u8* writepos = g_pGSWritePos;
-	u8* tempbuf;
 	assert( size < GS_RINGBUFFERSIZE );
 	assert( writepos < GS_RINGBUFFEREND );
 	assert( ((uptr)writepos & 15) == 0 );
 	assert( (size&15) == 0);
 
 	size += 16;
-	tempbuf = *(volatile PU8*)&g_pGSRingPos;
+	if( writepos + size > GS_RINGBUFFEREND )
-	if( writepos + size > GS_RINGBUFFEREND ) {
+	{
+		// If the incoming packet doesn't fit, then start over from
+		// the start of the ring buffer (it's a lot easier than trying
+		// to wrap the packet around the end of the buffer).
 
-		// skip to beginning
+		// We have to be careful not to leapfrog our readposition.  If it's 
-		while( writepos < tempbuf || tempbuf == GS_RINGBUFFERBASE ) {
+		// greater than the current write position then we need to stall
+		// until it loops around:
+
+		//const u8* readps = *(volatile PU8*)&g_pGSRingPos;
+		while( *(volatile PU8*)&g_pGSRingPos > writepos ) // && readpos == GS_RINGBUFFERBASE )
 			gsSetEventWait();
-			tempbuf = *(volatile PU8*)&g_pGSRingPos;
-		}
 
-		// notify GS
+		GSRingBufSimplePacket( GS_RINGTYPE_RESTART, 0, 0, 0 );
-		if( writepos != GS_RINGBUFFEREND ) {
+
-			InterlockedExchangePointer((void**)writepos, GS_RINGTYPE_RESTART);
+		writepos = GS_RINGBUFFERBASE;
-		}
+
+		// stall until the read position is past the end of our incoming block:
+		while( writepos+size >= *(volatile PU8*)&g_pGSRingPos )
+			gsSetEventWait();
 
 		InterlockedExchangePointer((void**)&g_pGSWritePos, GS_RINGBUFFERBASE);
-		writepos = GS_RINGBUFFERBASE;
 	}
-    else if( writepos + size == GS_RINGBUFFEREND ) {
+    else if( writepos + size == GS_RINGBUFFEREND )
-        while(tempbuf == GS_RINGBUFFERBASE && tempbuf != writepos) {
+	{
-			gsSetEventWait();
+		// Yay.  Perfect fit.  What are the odds?
-            tempbuf = *(volatile PU8*)&g_pGSRingPos;
+		// ... apparently very slim because the old code just performed the equivalent
-        }
+		// of a gsWaitGS (stalling the GS until the ring buffer emptied completely) and
-    }
+		// no one noticed enough to fix it. :)
 
-	while( writepos < tempbuf && (writepos+size >= tempbuf || (writepos+size == GS_RINGBUFFEREND && tempbuf == GS_RINGBUFFERBASE)) ) {
+		while( writepos < *(volatile PU8*)&g_pGSRingPos )
-		gsSetEventWait();
+			gsSetEventWait();
-		tempbuf = *(volatile PU8*)&g_pGSRingPos;
+    }
+	else
+	{
+		// two conditionals in the following while() loop, so precache
+		// the readpos for more efficient behavior:
+		const u8* readpos = *(volatile PU8*)&g_pGSRingPos;
+
+		// generic gs wait/stall.
+		// Waits until the readpos is outside the scope of the write area.
+		while( writepos < readpos && writepos+size >= readpos ) // || (writepos+size == GS_RINGBUFFEREND && readpos == GS_RINGBUFFERBASE)) ) {
+		{
+			gsSetEventWait();
+			readpos = *(volatile PU8*)&g_pGSRingPos;
+		}
 	}
 
 	// just copy
-	*(u32*)writepos = type|(((size-16)>>4)<<16);
+	*(u32*)writepos = type | (((size-16)>>4)<<16);
 	return writepos+16;
 }
 
 void GSRingBufSimplePacket(int type, int data0, int data1, int data2)
 {
 	u8* writepos = g_pGSWritePos;
-	u8* tempbuf;
+	u8* future_writepos = writepos+16;
 
-	assert( writepos + 16 <= GS_RINGBUFFEREND );
+	assert( future_writepos <= GS_RINGBUFFEREND );
 
-	tempbuf = *(volatile PU8*)&g_pGSRingPos;
+    if( future_writepos == GS_RINGBUFFEREND )
-	while(writepos < tempbuf && writepos+16 >= tempbuf ) {
+        future_writepos = GS_RINGBUFFERBASE;
+
+	while( future_writepos == *(volatile PU8*)&g_pGSRingPos )
 		gsSetEventWait();
-		tempbuf = *(volatile PU8*)&g_pGSRingPos;
-	}
 
 	*(u32*)writepos = type;
 	*(u32*)(writepos+4) = data0;
 	*(u32*)(writepos+8) = data1;
 	*(u32*)(writepos+12) = data2;
 
-	writepos += 16;
+	InterlockedExchangePointer((void**)&g_pGSWritePos, future_writepos);
-    if( writepos == GS_RINGBUFFEREND ) {
 
-        while(tempbuf == GS_RINGBUFFERBASE && tempbuf != g_pGSWritePos) {
+	if( !CHECK_DUALCORE )
-			gsSetEventWait();
-            tempbuf = *(volatile PU8*)&g_pGSRingPos;
-        }
-
-        writepos = GS_RINGBUFFERBASE;
-    }
-	InterlockedExchangePointer((void**)&g_pGSWritePos, writepos);
-
-	if( !CHECK_DUALCORE ) {
 		GS_SETEVENT();
-	}
 }
 
 void gsReset()
@@ -939,8 +960,6 @@ static __forceinline void WRITERING_DMA(u32 *pMem, u32 qwc) {
 			GSRINGBUF_DONECOPY(pgsmem, sizetoread); 
 			GSgifTransferDummy(2, pMem, qwc); 
 		} 
-		
-		if( !CHECK_DUALCORE ) GS_SETEVENT(); 
 	} 
 	else { 
 		GSGIFTRANSFER3(pMem, qwc); 
@@ -1500,7 +1519,8 @@ void* GSThreadProc(void* lpParam)
 					break;
 
 				case GS_RINGTYPE_FRAMESKIP:
-					GSsetFrameSkip(*(int*)(g_pGSRingPos+4));
+					if( GSsetFrameSkip != NULL )
+						GSsetFrameSkip(*(int*)(g_pGSRingPos+4));
 					break;
 				case GS_RINGTYPE_MEMWRITE8:
 					g_MTGSMem[*(int*)(g_pGSRingPos+4)] = *(u8*)(g_pGSRingPos+8);

pcsx2/GS.h

@@ -97,15 +97,6 @@ extern u8* g_pGSWritePos;
 
 //#endif
 
-// mem and size are the ones from GSRingBufCopy
-static __forceinline void GSRINGBUF_DONECOPY(const u8 *mem, u32 size) { 
-	u8* temp = (u8*)(mem) + (size); 
-	assert( temp <= GS_RINGBUFFEREND); 
-//	MTGS_RECWRITE(mem, size); 
-	if( temp == GS_RINGBUFFEREND ) temp = GS_RINGBUFFERBASE; 
-	InterlockedExchangePointer((void**)&g_pGSWritePos, temp);	
-}
-
 #if defined(_WIN32) && !defined(WIN32_PTHREADS)
 
 #define GS_SETEVENT() SetEvent(g_hGsEvent)
@@ -125,6 +116,8 @@ void gsInit();
 void gsShutdown();
 void gsReset();
 
+// mem and size are the ones from GSRingBufCopy
+extern void GSRINGBUF_DONECOPY(const u8 *mem, u32 size);
 extern void gsWaitGS();
 
 // used for resetting GIF fifo

pcsx2/Misc.c

@@ -863,7 +863,8 @@ void ProcessFKeys(int fkey, int shift)
 			break;	
 
 		case 4:
-
+		{
+			const char* limitMsg;
 #ifdef PCSX2_NORECBUILD
             SysPrintf("frame skipping only valid for recompiler build\n");
 #else
@@ -882,33 +883,37 @@ void ProcessFKeys(int fkey, int shift)
 					if( CHECK_MULTIGS ) GSRingBufSimplePacket(GS_RINGTYPE_FRAMESKIP, 0, 0, 0);
 					else GSsetFrameSkip(0);
 					Cpu->ExecuteVU1Block = recExecuteVU1Block;
-					if(SPU2setTimeStretcher != NULL)
+					limitMsg = "None/Normal";
-						SPU2setTimeStretcher(1);
-					SysPrintf("Normal - Frame Limit Mode Changed\n");
 					break;
 				case PCSX2_FRAMELIMIT_LIMIT:
 					if( CHECK_MULTIGS ) GSRingBufSimplePacket(GS_RINGTYPE_FRAMESKIP, 0, 0, 0);
 					else GSsetFrameSkip(0);
 					Cpu->ExecuteVU1Block = recExecuteVU1Block;
 					//Quality option, turn off timestretching on the SPU2 plugin
-					if(SPU2setTimeStretcher != NULL)
+					limitMsg = "Limit";
-						SPU2setTimeStretcher(0);
-					SysPrintf("Limit - Frame Limit Mode Changed\n");
 					break;
 				case PCSX2_FRAMELIMIT_SKIP:
 					Cpu->ExecuteVU1Block = recExecuteVU1Block;
-					if(SPU2setTimeStretcher != NULL)
-						SPU2setTimeStretcher(1);
-					SysPrintf("Frame Skip - Frame Limit Mode Changed\n");
-					break;
 				case PCSX2_FRAMELIMIT_VUSKIP:
-					if(SPU2setTimeStretcher != NULL)
+					if( GSsetFrameSkip == NULL )
-						SPU2setTimeStretcher(1);
+					{
-					SysPrintf("VU Skip - Frame Limit Mode Changed\n");
+						Config.Options &= ~PCSX2_FRAMELIMIT_MASK;
+						SysPrintf("Notice: GS Plugin does not support frameskipping.\n");
+						limitMsg = "None/Normal";
+					}
+					else
+					{
+						limitMsg = (CHECK_FRAMELIMIT == PCSX2_FRAMELIMIT_SKIP) ? "Skip" : "VUSkip";
+					}
+
 					break;
 			}
+
+			SysPrintf("Frame Limit Mode Changed: %s\n", limitMsg );
+
 			// [Air]: Do we really want to save runtime changes to frameskipping?
-            //SaveConfig();
+			//SaveConfig();
+		}
 #endif
 			break;
 		// note: VK_F5-VK_F7 are reserved for GS

pcsx2/VifDma.c

@@ -1668,11 +1668,10 @@ static int Vif1TransDirectHL(u32 *data){
 			if( gsmem != NULL ) {
 				FreezeMMXRegs(1);
 				memcpy_fast(gsmem, (u32*)splittransfer[0], 16);
+				FreezeMMXRegs(0);
 				GSRINGBUF_DONECOPY(gsmem, 16);
 				GSgifTransferDummy(1, (u32*)splittransfer[0], 1);
 			}
-				FreezeMMXRegs(0);
-			if( !CHECK_DUALCORE  ) GS_SETEVENT();
 		}
 		else {
 			FreezeXMMRegs(1);
@@ -1718,8 +1717,6 @@ static int Vif1TransDirectHL(u32 *data){
 			GSRINGBUF_DONECOPY(gsmem, ret<<2);
 			GSgifTransferDummy(1, data, ret>>2);
 		}
-		
-		if( !CHECK_DUALCORE  ) GS_SETEVENT();
 	}
 	else {
 		
@@ -2331,9 +2328,6 @@ void dmaVIF1()
 
 				GSRINGBUF_DONECOPY(pMem, 0);
 				
-				if( !CHECK_DUALCORE  )
-					GS_SETEVENT();
-
                 g_MTGSVifStart = cpuRegs.cycle;
                 g_MTGSVifCount = 4000000; // a little less than 1/60th of a second
 				

pcsx2/x86/iVUmicro.c

@@ -5633,14 +5633,6 @@ void VU1XGKICK_MTGSTransfer(u32 *pMem, u32 addr)
 		FreezeMMXRegs(0);
 	    
 	    GSRINGBUF_DONECOPY(pmem, size);
-
-        if( !CHECK_DUALCORE ) {
-#if defined(_WIN32) && !defined(WIN32_PTHREADS)
-            SetEvent(g_hGsEvent);
-#else
-            pthread_cond_signal(&g_condGsEvent);
-#endif
-        }
 	}
 }