fix glitches in dma and poor code in command unpacking which were causing some slowdown. clean up ioregview refresh triggering code and add a note about how to cut compile time of gpu.cpp in half.

desmume/src/GPU.cpp

@@ -42,6 +42,9 @@
 //#define FORCEINLINE
 //#define SSE2_NOINTRIN
 
+//compilation speed hack (cuts time exactly in half by cutting out permutations)
+//#define DISABLE_MOSAIC
+
 extern BOOL click;
 NDS_Screen MainScreen;
 NDS_Screen SubScreen;
@@ -2133,9 +2136,12 @@ static void GPU_ligne_layer(NDS_Screen * screen, u16 l)
 					//useful for debugging individual layers
 					//if(gpu->core == 1 || i16 != 2) continue;
 
+#ifndef DISABLE_MOSAIC
 					if(gpu->curr_mosaic_enabled)
 						gpu->modeRender<true>(i16);
-					else gpu->modeRender<false>(i16);
+					else 
+#endif
+						gpu->modeRender<false>(i16);
 				} //layer enabled
 			}
 		}

desmume/src/MMU.cpp

@@ -47,10 +47,6 @@
 #include "readwrite.h"
 #include "MMU_timing.h"
 
-#ifdef WIN32
-#include "windows/IORegView.h"
-#endif
-
 #ifdef DO_ASSERT_UNALIGNED
 #define ASSERT_UNALIGNED(x) assert(x)
 #else
@@ -767,6 +763,7 @@ static inline void MMU_VRAMmapControl(u8 block, u8 VRAMBankCnt)
 		MMU_VRAMmapRefreshBank(i);
 
 	//printf(vramConfiguration.describe().c_str());
+	//printf("vram remapped at vcount=%d\n",nds.VCount);
 
 	//if texInfo changed, trigger notifications
 	if(memcmp(&oldTexInfo,&MMU.texInfo,sizeof(MMU_struct::TextureInfo)))
@@ -1815,7 +1812,7 @@ u32 TGXSTAT::read32()
 	
 	ret |= gxfifo_irq; //user's irq flags
 
-	//printf("Returning gxstat read: %08X\n",ret);
+	//printf("vc=%03d Returning gxstat read: %08X\n",nds.VCount,ret);
 	return ret;
 }
 
@@ -1995,10 +1992,7 @@ if(_startmode==0 && wordcount==1) {
 	if(!doNotStart)
 		doSchedule();
 
-	//todo - make a driver stub for this so that we dont have to conditionalize it everywhere
+	driver->DEBUG_UpdateIORegView(BaseDriver::EDEBUG_IOREG_DMA);
-#ifdef WIN32
-	RefreshAllIORegViews();
-#endif
 }
 
 void DmaController::exec()
@@ -2050,7 +2044,7 @@ start:
 
 		if(triggered)
 		{
-			//if(procnum==0) printf("%08lld trig type %d dma#%d with words %d at src:%08X dst:%08X gxf:%d",nds_timer,startmode,chan,wordcount,saddr,daddr,gxFIFO.size);
+			//if(procnum==0) printf("vc=%03d %08lld trig type %d dma#%d w/words %d at src:%08X dst:%08X gxf:%d",nds.VCount,nds_timer,startmode,chan,wordcount,saddr,daddr,gxFIFO.size);
 			if(saddr ==0x023BCCEC && wordcount==118) {
 				int zzz=9;
 			}
@@ -2064,9 +2058,7 @@ start:
 		}
 	}
 
-#ifdef WIN32
+	driver->DEBUG_UpdateIORegView(BaseDriver::EDEBUG_IOREG_DMA);
-	RefreshAllIORegViews();
-#endif
 }
 
 void DmaController::doCopy()
@@ -2162,6 +2154,7 @@ void triggerDma(EDMAMode mode)
 void DmaController::tryTrigger(EDMAMode mode)
 {
 	if(startmode != mode) return;
+	if(!enable) return;
 
 	//hmm dont trigger it if its already running! 
 	//but paused things need triggers to continue
@@ -2696,13 +2689,13 @@ void FASTCALL _MMU_ARM9_write16(u32 adr, u16 val)
 #endif
 					if(val & (1<<15))
 					{
-						//LOG("Main core on top\n");
+						//printf("Main core on top (vcount=%d)\n",nds.VCount);
 						MainScreen.offset = 0;
 						SubScreen.offset = 192;
 					}
 					else
 					{
-						//LOG("Main core on bottom\n");
+						//printf("Main core on bottom (vcount=%d)\n",nds.VCount);
 						MainScreen.offset = 192;
 						SubScreen.offset = 0;
 					}

desmume/src/MMU_timing.h

@@ -44,6 +44,7 @@
 
 	// makes non-sequential accesses slower than sequential ones.
 #define ACCOUNT_FOR_NON_SEQUENTIAL_ACCESS
+	//(SOMETIMES THIS IS A BIG SPEED HIT!)
 
 	// enables emulation of code fetch waits.
 #define ACCOUNT_FOR_CODE_FETCH_CYCLES

desmume/src/NDSSystem.cpp

@@ -1953,9 +1953,12 @@ static void execHardware_hstart_vcount()
 	u16 vmatch = T1ReadWord(MMU.ARM9_REG, 4);
 	if(nds.VCount==((vmatch>>8)|((vmatch<<1)&(1<<8))))
 	{
+		//arm9 vmatch
 		T1WriteWord(MMU.ARM9_REG, 4, T1ReadWord(MMU.ARM9_REG, 4) | 4);
-		if(T1ReadWord(MMU.ARM9_REG, 4) & 32)
+		if(T1ReadWord(MMU.ARM9_REG, 4) & 32) {
+			//printf("VMATCH FIRING! vc=%03d\n",nds.VCount);
 			NDS_makeARM9Int(2);
+		}
 	}
 	else
 		T1WriteWord(MMU.ARM9_REG, 4, T1ReadWord(MMU.ARM9_REG, 4) & 0xFFFB);
@@ -1963,6 +1966,7 @@ static void execHardware_hstart_vcount()
 	vmatch = T1ReadWord(MMU.ARM7_REG, 4);
 	if(nds.VCount==((vmatch>>8)|((vmatch<<1)&(1<<8))))
 	{
+		//arm7 vmatch
 		T1WriteWord(MMU.ARM7_REG, 4, T1ReadWord(MMU.ARM7_REG, 4) | 4);
 		if(T1ReadWord(MMU.ARM7_REG, 4) & 32)
 			NDS_makeARM7Int(2);
@@ -2153,7 +2157,7 @@ bool nds_loadstate(EMUFILE* is, int size)
 
 //#define LOG_ARM9
 //#define LOG_ARM7
-//static bool dolog = false;
+//static bool dolog = true;
 
 FORCEINLINE void arm9log()
 {
@@ -2311,8 +2315,8 @@ void NDS_exec(s32 nb)
 
 #ifndef NDEBUG
 			//what we find here is dependent on the timing constants above
-			if(nds_timer>next && (nds_timer-next)>22)
+			//if(nds_timer>next && (nds_timer-next)>22)
-				printf("curious. please report: over by %d\n",(int)(nds_timer-next));
+			//	printf("curious. please report: over by %d\n",(int)(nds_timer-next));
 #endif
 
 			//if we were waiting for an irq, don't wait too long:
@@ -2347,6 +2351,7 @@ void execHardware_interrupts()
 		if ( armcpu_irqException(&NDS_ARM9))
 #endif
 		{
+			//printf("ARM9 interrupt! flags: %08X ; mask: %08X ; result: %08X\n",MMU.reg_IF[0],MMU.reg_IE[0],MMU.reg_IF[0]&MMU.reg_IE[0]);
 			//nds.ARM9Cycle = nds.cycles;
 		}
 	}

desmume/src/driver.h

@@ -53,6 +53,13 @@ public:
 	virtual bool EMU_IsFastForwarding() { return false; }
 	virtual bool EMU_HasEmulationStarted() { return true; }
 	virtual bool EMU_IsAtFrameBoundary() { return true; }
+
+	enum eDebug_IOReg
+	{
+		EDEBUG_IOREG_DMA
+	};
+
+	virtual void DEBUG_UpdateIORegView(eDebug_IOReg category) { }
 };
 extern BaseDriver* driver;
 

desmume/src/gfx3d.cpp

@@ -96,39 +96,18 @@ public:
 	{
 		reset();
 	}
-	void test() {
-		printf("test 1: 0x00412321\n");
-		receive(0x00412321); receive(1); receive(2); receive(3); receive(0);
-		//21: 1
-		//23: 2
-		//23: 3
-		//41: 4 (a dud parameter)
-		printf("test 2: 0x00002321\n");
-		receive(0x00002321); receive(1); receive(2); receive(3);
-		//21: 1
-		//23: 2
-		//23: 3
-		printf("test 3: 0x11111111\n"); 
-		receive(0x11111111); 
-		receive(1);
-		//11: 1 (a dud parameter)
-		printf("test 4: 0x00000011, 0x29111111\n"); 
-		receive(0x00000011); receive(0x29111111); receive(1);
-		//11: 17
-		//11: 688984337
-		//11: 1
-		printf("test 5: 0x00004127\n");
-		receive(0x00004127); receive(1); receive(0);//final one should be a dud
-	}
 	void reset() {
 		countdown = 0;
-		commandsPending = std::queue<u8>();
+		commandCursor = 4;
-		countdowns = std::queue<u8>();
+		for(int i=0;i<4;i++) {
+			commandsPending[i].command = 0;
+			commandsPending[i].countdown = 0;
+		}
 	}
 	//todo - things in here other than the very first thing involving GFX3D_NOP_NOARG_HACK I am not too sure about.
 	void receive(u32 val) {
 		bool hack = false;
-		if(commandsPending.size()>0 && (commandsPending.front() == 0x15 || commandsPending.front() == GFX3D_NOP_NOARG_HACK || commandsPending.front() == 0x11 || commandsPending.front() == 0x41) && val != 0) {
+		if(size()>0 && (front().command == 0x15 || front().command == GFX3D_NOP_NOARG_HACK || front().command == 0x11 || front().command == 0x41) && val != 0) {
 			//apparently a zero is swallowed in this case but if another value is sent 
 			//processing will continue
 			//if(commandsPending.front() == GFX3D_NOP_NOARG_HACK)
@@ -136,7 +115,7 @@ public:
 			//else
 			{
 				//printf("gxf: sending hack %02X: (dummy=0)\n", commandsPending.front());
-				GFX_FIFOsend(commandsPending.front(),0);
+				GFX_FIFOsend(front().command,0);
 			}
 			hack = true;
 			goto hackTrigger;
@@ -147,21 +126,20 @@ public:
 			//if(commandsPending.front() == GFX3D_NOP_NOARG_HACK)
 			//{}
 			//else 
-				GFX_FIFOsend(commandsPending.front(),val);
+				GFX_FIFOsend(front().command,val);
 			hackTrigger:
 			countdown--;
 			while(countdown==0) {
-				commandsPending.pop();
+				dequeue();
-				countdowns.pop();
 			trigger:
 				//dont set hack to false if you jumped from below! it needs to be true for when you jump down from above.
 				//oh my what a mess.
-				if(countdowns.empty()) break;
+				if(size()==0) break;
-				countdown = countdowns.front();
+				countdown = front().countdown;
 				if(!countdown) {
-					if(commandsPending.front() != INVALID_COMMAND /*&& commandsPending.front() != GFX3D_NOP_NOARG_HACK*/) {
+					if(front().command != INVALID_COMMAND /*&& commandsPending.front() != GFX3D_NOP_NOARG_HACK*/) {
 						//printf("[%06d]gxf: sending %02X: (dummy=0)\n", currFrameCounter,commandsPending.front());
-						GFX_FIFOsend(commandsPending.front(),0);
+						GFX_FIFOsend(front().command,0);
 					}
 				}
 			}
@@ -182,11 +160,11 @@ decode:
 				u8 cmd = commands[i];
 				u8 type = commandTypes[i];
 				if(type == INVALID_COMMAND) {
-					commandsPending.push(INVALID_COMMAND);
+					commandsPending[i].command = INVALID_COMMAND;
 				} else {
 					if(type == UNDEFINED_COMMAND) 
-						commandsPending.push(GFX3D_NOP_NOARG_HACK); //enqueue a single undefined command we know how to handle
+						commandsPending[i].command = GFX3D_NOP_NOARG_HACK;  //enqueue a single undefined command we know how to handle
-					else commandsPending.push(cmd);
+					else commandsPending[i].command = cmd;
 				}
 				if(type == UNDEFINED_COMMAND || type == 0x00) {
 					//these are valid commands with no parameters. they might need special handling
@@ -199,35 +177,44 @@ decode:
 						}
 					}
 					if(safe) {
-						countdowns.push(0);
+						commandsPending[i].countdown = 0;
 					} else {
 						//we need to receive a dummy parameter in this case
-						countdowns.push(1);
+						commandsPending[i].countdown = 1;
 					}
 				} else if(type != INVALID_COMMAND) {
-					countdowns.push(type);
+					commandsPending[i].countdown = type;
-				} else countdowns.push(0);
+				} else commandsPending[i].countdown = 0;
 			}
 
-			countdown = countdowns.front();
+			commandCursor = 0;
+			countdown = front().countdown;
 			if(countdown==0) 
 				goto trigger;
 		}
 	}
-	std::queue<u8> commandsPending;
+
-	std::queue<u8> countdowns;
+	struct CommandItem {
+		u8 command, countdown;
+	} commandsPending[4];
+
+	u32 commandCursor;
 	u8 countdown;
 
+private:
+	void dequeue() { commandCursor++; }
+	CommandItem& front() { return commandsPending[commandCursor]; }
+	u32 size() { return 4-commandCursor; }
+public:
+
 	void savestate(EMUFILE *f)
 	{
+		//TODO - next time we invalidate savestates, simplify this format.
 		write32le(0,f); //version
-		std::queue<u8> temp = commandsPending;
+		write32le(size(),f);
-		write32le(temp.size(),f);
+		for(u32 i=commandCursor;i<4;i++) write8le(commandsPending[i].command,f);
-		while(!temp.empty()) { write8le(temp.front(),f); temp.pop(); }
+		write32le(0,f);
-		temp = countdowns;
+		for(u32 i=commandCursor;i<4;i++) write8le(commandsPending[i].countdown,f);
-		write32le(temp.size(),f);
-		while(!temp.empty()) { write8le(temp.front(),f); temp.pop(); }
-		
 		write8le(countdown,f);
 	}
 	
@@ -237,14 +224,14 @@ decode:
 		if(read32le(&version,f) != 1) return false;
 		if(version != 0) return false;
 
-		assert(commandsPending.size()==0);
+		u32 tempsize;
-		assert(countdowns.size()==0);
+		read32le(&tempsize,f);
-
+		commandCursor = 4-tempsize;
-		u32 temp;
+		for(u32 i=0;i<commandCursor;i++) commandsPending[i].command = 0;
-		read32le(&temp,f);
+		for(u32 i=commandCursor;i<4;i++) read8le(&commandsPending[i].command,f);
-		for(u32 i=0;i<temp;i++) { u8 temp8; read8le(&temp8,f); commandsPending.push(temp8); }
+		read32le(&tempsize,f);
-		read32le(&temp,f);
+		for(u32 i=0;i<commandCursor;i++) commandsPending[i].countdown = 0;
-		for(u32 i=0;i<temp;i++) { u8 temp8; read8le(&temp8,f); countdowns.push(temp8); }
+		for(u32 i=commandCursor;i<4;i++) read8le(&commandsPending[i].countdown,f);
 
 		read8le(&countdown,f);
 
@@ -363,7 +350,6 @@ static u32 clInd = 0;
 
 static u32 clInd2 = 0;
 BOOL isSwapBuffers = FALSE;
-bool isVBlank = false;
 
 static u32 BTind = 0;
 static u32 PTind = 0;
@@ -553,7 +539,6 @@ void gfx3d_reset()
 	
 	clInd2 = 0;
 	isSwapBuffers = FALSE;
-	isVBlank = false;
 
 	GFX_PIPEclear();
 	GFX_FIFOclear();
@@ -1713,6 +1698,8 @@ void gfx3d_execute3D()
 
 	if(GFX_PIPErecv(&cmd, &param))
 	{
+		//if (isSwapBuffers) printf("Executing while swapbuffers is pending: %d:%08X\n",cmd,param);
+
 		//since we did anything at all, incur a pipeline motion cost.
 		//also, we can't let gxfifo sequencer stall until the fifo is empty.
 		//see...
@@ -1746,6 +1733,7 @@ void gfx3d_execute3D()
 
 void gfx3d_glFlush(u32 v)
 {
+	//printf("-------------FLUSH------------- (vcount=%d\n",nds.VCount);
 	gfx3d.sortmode = BIT0(v);
 	gfx3d.wbuffer = BIT1(v);
 #if 0
@@ -1871,7 +1859,6 @@ static void gfx3d_doFlush()
 
 void gfx3d_VBlankSignal()
 {
-	isVBlank = true;
 	if (isSwapBuffers)
 	{
 #ifndef FLUSHMODE_HACK
@@ -1884,8 +1871,6 @@ void gfx3d_VBlankSignal()
 
 void gfx3d_VBlankEndSignal(bool skipFrame)
 {
-	isVBlank = false;
-
 	if (!drawPending) return;
 	drawPending = FALSE;
 

desmume/src/windows/IORegView.cpp

@@ -282,6 +282,7 @@ static const int kYMargin = 1;
 
 typedef std::vector<CIORegView*> TIORegViewList;
 static TIORegViewList liveIORegViews;
+bool anyLiveIORegViews = false;
 
 void RefreshAllIORegViews()
 {
@@ -303,6 +304,7 @@ CIORegView::CIORegView()
 	, yoff(0)
 {
 	liveIORegViews.push_back(this);
+	anyLiveIORegViews = true;
 }
 
 CIORegView::~CIORegView()
@@ -311,6 +313,7 @@ CIORegView::~CIORegView()
 	UnregWndClass("DeSmuME_IORegView");
 	//TODO - is this thread safe? which thread do these calls come from
 	liveIORegViews.erase(std::find(liveIORegViews.begin(),liveIORegViews.end(),this));
+	if(liveIORegViews.size()==0) anyLiveIORegViews = false;
 }
 
 /*--------------------------------------------------------------------------*/

desmume/src/windows/main.cpp

@@ -1819,6 +1819,13 @@ class WinDriver : public BaseDriver
 
 		return ESTEP_DONE;
 	}
+
+	virtual void DEBUG_UpdateIORegView(eDebug_IOReg category)
+	{
+		extern bool anyLiveIORegViews;
+		if(anyLiveIORegViews)
+			RefreshAllIORegViews();
+	}
 };
 
 std::string GetPrivateProfileStdString(LPCSTR lpAppName,LPCSTR lpKeyName,LPCSTR lpDefault)