Finished mVU macro, and set it on by default.

Basically this means whenever COP2 recs are used, its calling mVU instructions instead of sVU instructions.

Note: Theres a very-minor incompatibility problem when using interpreters/sVU for VU0 with mVU Macro. But I'll fix it later (and it probably doesn't effect much).


git-svn-id: http://pcsx2.googlecode.com/svn/trunk@1710 96395faa-99c1-11dd-bbfe-3dabce05a288

common/include/Pcsx2Config.h

@@ -61,7 +61,7 @@ extern SessionOverrideFlags g_Session;
 #define CHECK_MULTIGS	(Config.Options&PCSX2_GSMULTITHREAD)
 #define CHECK_MICROVU0	(Config.Options&PCSX2_MICROVU0)
 #define CHECK_MICROVU1	(Config.Options&PCSX2_MICROVU1)
-//#define CHECK_MACROVU0	// ifndef = Use sVU for VU macro, ifdef = Use mVU for VU macro
+#define CHECK_MACROVU0	// ifndef = Use sVU for VU macro, ifdef = Use mVU for VU macro
 #define CHECK_EEREC (!g_Session.ForceDisableEErec && Config.Options&PCSX2_EEREC)
 #define CHECK_VU0REC (!g_Session.ForceDisableVU0rec && Config.Options&PCSX2_VU0REC)
 #define CHECK_VU1REC (!g_Session.ForceDisableVU1rec && (Config.Options&PCSX2_VU1REC))

pcsx2/x86/microVU_Alloc.inl

@@ -77,13 +77,37 @@ microVUt(void) mVUallocSFLAGc(int reg, int regT, int fInstance) {
 	OR32RtoR(reg, regT);
 }
 
+// Denormalizes Status Flag
+microVUt(void) mVUallocSFLAGd(uptr memAddr, bool setAllflags) {
+	MOV32MtoR(gprF0, memAddr);
+	MOV32RtoR(gprF1, gprF0);
+	SHR32ItoR(gprF1, 3);
+	AND32ItoR(gprF1, 0x18);
+
+	MOV32RtoR(gprF2, gprF0);
+	SHL32ItoR(gprF2, 11);
+	AND32ItoR(gprF2, 0x1800);
+	OR32RtoR (gprF1, gprF2);
+
+	SHL32ItoR(gprF0, 14);
+	AND32ItoR(gprF0, 0x3cf0000);
+	OR32RtoR (gprF1, gprF0);
+
+	if (setAllflags) {
+		MOV32RtoR(gprF0, gprF1);
+		MOV32RtoR(gprF2, gprF1);
+		MOV32RtoR(gprF3, gprF1);
+	}
+}
+
 microVUt(void) mVUallocMFLAGa(mV, int reg, int fInstance) {
 	MOVZX32M16toR(reg, (uptr)&mVU->macFlag[fInstance]);
 }
 
 microVUt(void) mVUallocMFLAGb(mV, int reg, int fInstance) {
 	//AND32ItoR(reg, 0xffff);
-	MOV32RtoM((uptr)&mVU->macFlag[fInstance], reg);
+	if (fInstance < 4) MOV32RtoM((uptr)&mVU->macFlag[fInstance], reg);			// microVU
+	else			   MOV32RtoM((uptr)&mVU->regs->VI[REG_MAC_FLAG].UL, reg);	// macroVU
 }
 
 microVUt(void) mVUallocCFLAGa(mV, int reg, int fInstance) {

pcsx2/x86/microVU_Branch.inl

@@ -59,8 +59,13 @@ microVUt(void) mVUendProgram(mV, microFlagCycles* mFC, int isEbit) {
 	}
 
 	// Save Flag Instances
+#ifdef CHECK_MACROVU0
+	getFlagReg(fStatus, fStatus);
+	MOV32RtoM((uptr)&mVU->regs->VI[REG_STATUS_FLAG].UL,	fStatus);
+#else
 	mVUallocSFLAGc(gprT1, gprT2, fStatus);
 	MOV32RtoM((uptr)&mVU->regs->VI[REG_STATUS_FLAG].UL,	gprT1);
+#endif
 	mVUallocMFLAGa(mVU, gprT1, fMac);
 	mVUallocCFLAGa(mVU, gprT2, fClip);
 	MOV32RtoM((uptr)&mVU->regs->VI[REG_MAC_FLAG].UL,	gprT1);

pcsx2/x86/microVU_Execute.inl

@@ -40,24 +40,15 @@ void mVUdispatcherA(mV) {
 	SSE_LDMXCSR((uptr)&g_sseVUMXCSR);
 
 	// Load Regs
+#ifdef CHECK_MACROVU0
 	MOV32MtoR(gprF0, (uptr)&mVU->regs->VI[REG_STATUS_FLAG].UL);
 	MOV32RtoR(gprF1, gprF0);
-	SHR32ItoR(gprF1, 3);
-	AND32ItoR(gprF1, 0x18);
-
 	MOV32RtoR(gprF2, gprF0);
-	SHL32ItoR(gprF2, 11);
+	MOV32RtoR(gprF3, gprF0);
-	AND32ItoR(gprF2, 0x1800);
+#else
-	OR32RtoR (gprF1, gprF2);
+	mVUallocSFLAGd((uptr)&mVU->regs->VI[REG_STATUS_FLAG].UL, 1);
-
+#endif
-	SHL32ItoR(gprF0, 14);
+	
-	AND32ItoR(gprF0, 0x3cf0000);
-	OR32RtoR (gprF1, gprF0);
-
-	MOV32RtoR(gprF0, gprF1);
-	MOV32RtoR(gprF2, gprF1);
-	MOV32RtoR(gprF3, gprF1);
-
 	SSE_MOVAPS_M128_to_XMM(xmmT1, (uptr)&mVU->regs->VI[REG_MAC_FLAG].UL);
 	SSE_SHUFPS_XMM_to_XMM (xmmT1, xmmT1, 0);
 	SSE_MOVAPS_XMM_to_M128((uptr)mVU->macFlag, xmmT1);

pcsx2/x86/microVU_Macro.inl

@@ -41,19 +41,31 @@ void setupMacroOp(int mode, const char* opName) {
 	memset(&microVU0.prog.IRinfo.info[0], 0, sizeof(microVU0.prog.IRinfo.info[0]));
 	iFlushCall(FLUSH_EVERYTHING);
 	microVU0.regAlloc->reset();
-	if (mode & 1) { // Q-Reg will be Read
+	if (mode & 0x01) { // Q-Reg will be Read
 		SSE_MOVSS_M32_to_XMM(xmmPQ, (uptr)&microVU0.regs->VI[REG_Q].UL);
 	}
-	if (mode & 8) { // Clip Instruction
+	if (mode & 0x08) { // Clip Instruction
 		microVU0.prog.IRinfo.info[0].cFlag.write	 = 0xff;
 		microVU0.prog.IRinfo.info[0].cFlag.lastWrite = 0xff;
 	}
+	if (mode & 0x10) { // Update Status/Mac Flags
+		microVU0.prog.IRinfo.info[0].sFlag.doFlag		= 1;
+		microVU0.prog.IRinfo.info[0].sFlag.doNonSticky	= 1;
+		microVU0.prog.IRinfo.info[0].sFlag.write		= 0;
+		microVU0.prog.IRinfo.info[0].sFlag.lastWrite	= 0;
+		microVU0.prog.IRinfo.info[0].mFlag.doFlag		= 1;
+		microVU0.prog.IRinfo.info[0].mFlag.write		= 0xff;
+		MOV32MtoR(gprF0, (uptr)&microVU0.regs->VI[REG_STATUS_FLAG].UL);
+	}
 }
 
 void endMacroOp(int mode) {
-	if (mode & 2) { // Q-Reg was Written To
+	if (mode & 0x02) { // Q-Reg was Written To
 		SSE_MOVSS_XMM_to_M32((uptr)&microVU0.regs->VI[REG_Q].UL, xmmPQ);
 	}
+	if (mode & 0x10) { // Status/Mac Flags were Updated
+		MOV32RtoM((uptr)&microVU0.regs->VI[REG_STATUS_FLAG].UL, gprF0);
+	}
 	microVU0.regAlloc->flushAll();
 }
 
@@ -261,10 +273,16 @@ static void recCFC2() {
 	if (!_Rt_) return;
 	iFlushCall(FLUSH_EVERYTHING);
 
-	MOV32MtoR(EAX, (uptr)&microVU0.regs->VI[_Rd_].UL);
+	if (_Rd_ == REG_STATUS_FLAG) { // Normalize Status Flag
+		MOV32MtoR(gprF0, (uptr)&microVU0.regs->VI[REG_STATUS_FLAG].UL);
+		mVUallocSFLAGc(EAX, gprF0, 0);
+	}
+	else MOV32MtoR(EAX, (uptr)&microVU0.regs->VI[_Rd_].UL);
+
+	// FixMe: Should R-Reg have upper 9 bits 0?
 	MOV32RtoM((uptr)&cpuRegs.GPR.r[_Rt_].UL[0], EAX);
 
-	if( _Rd_ >= 16 ) {
+	if (_Rd_ >= 16) {
 		CDQ(); // Sign Extend
 		MOV32RtoM ((uptr)&cpuRegs.GPR.r[_Rt_].UL[1], EDX);
 	}
@@ -289,7 +307,14 @@ static void recCTC2() {
 			OR32ItoR (EAX, 0x3f800000);
 			MOV32RtoM((uptr)&microVU0.regs->VI[REG_R].UL, EAX);
 			break;
-		case REG_CMSAR1: // REG_CMSAR1
+		case REG_STATUS_FLAG:
+			if (_Rt_) { // Denormalizes flag into gprF1
+				mVUallocSFLAGd((uptr)&cpuRegs.GPR.r[_Rt_].UL[0], 0);
+				MOV32RtoM((uptr)&microVU0.regs->VI[_Rd_].UL, gprF1);
+			}
+			else MOV32ItoM((uptr)&microVU0.regs->VI[_Rd_].UL, 0);
+			break;
+		case REG_CMSAR1:
 			if (_Rt_) {
 				MOV32MtoR(EAX, (uptr)&cpuRegs.GPR.r[_Rt_].UL[0]);
 				PUSH32R(EAX);