microVU: New improved flag speedhack.

- Now does Mac Flags as well pretty safely
- Speedup is anywhere from nothing to ~30fps (latter is DQ8 + MTVU)

pcsx2/x86/microVU_Analyze.inl

@@ -291,7 +291,7 @@ __ri void flagSet(mV, bool setMacFlag) {
 	for (int i = mVUcount, j = 0; i > 0; i--, j++) {
 		j += mVUstall;
 		incPC2(-2);
-		if (sFLAG.doFlag && (j >= 3)) { 
+		if (sFLAG.doFlag && (j >= 3) && sFLAG.skipWrite != 1) {
 			if (setMacFlag) { mFLAG.doFlag = 1; }
 			else { sFLAG.doNonSticky = 1; }
 			break; 
@@ -317,7 +317,7 @@ __ri void mVUanalyzeSflag(mV, int It) {
 
 __ri void mVUanalyzeFSSET(mV) {
 	mVUlow.isFSSET = 1;
-	mVUlow.readFlags = 1;
+	mVUlow.readFlags = 2;
 }
 
 //------------------------------------------------------------------
@@ -345,7 +345,7 @@ __ri void mVUanalyzeMflag(mV, int Is, int It) {
 
 __fi void mVUanalyzeCflag(mV, int It) {
 	mVUinfo.swapOps = 1;
-	mVUlow.readFlags = 1;
+	mVUlow.readFlags = 2;
 	if (mVUcount < 4) { 
 		if (!(mVUpBlock->pState.needExactMatch & 4)) // The only time this should happen is on the first program block
 			DevCon.WriteLn(Color_Green, "microVU%d: pState's cFlag Info was expected to be set [%04x]", getIndex, xPC);

pcsx2/x86/microVU_Flags.inl

@@ -31,7 +31,7 @@ __fi void mVUstatusFlagOp(mV) {
 	int i = mVUcount;
 	bool runLoop = true;
 
-	if (sFLAG.doFlag) {
+	if (sFLAG.doFlag && sFLAG.skipWrite != 1) {
 		sFLAG.doNonSticky = true;
 	}
 	else {
@@ -41,7 +41,7 @@ __fi void mVUstatusFlagOp(mV) {
 				runLoop = false;
 				break;
 			}
-			else if (sFLAG.doFlag) {
+			else if (sFLAG.doFlag && sFLAG.skipWrite != 1) {
 				sFLAG.doNonSticky = true;
 				break;
 			}
@@ -92,30 +92,128 @@ void sortFullFlag(int* fFlag, int* bFlag) {
 	}
 }
 
-#define sFlagCond (sFLAG.doFlag || mVUlow.isFSSET || mVUinfo.doDivFlag)
+#define sFlagCond ((sFLAG.doFlag && sFLAG.skipWrite != 1) || mVUlow.isFSSET || mVUinfo.doDivFlag )
 #define sHackCond (mVUsFlagHack && !sFLAG.doNonSticky)
 
 // Note: Flag handling is 'very' complex, it requires full knowledge of how microVU recs work, so don't touch!
 __fi void mVUsetFlags(mV, microFlagCycles& mFC) {
-
 	int endPC  = iPC;
-	u32 aCount = 1; // Amount of instructions needed to get valid mac flag instances for block linking
+	u32 aCount = 0; // Amount of instructions needed to get valid mac flag instances for block linking
+	bool writeProtect = false;
 
 	// Ensure last ~4+ instructions update mac/status flags (if next block's first 4 instructions will read them)
 	for(int i = mVUcount; i > 0; i--, aCount++) {
 		if (sFLAG.doFlag) { 
-			if (__Mac)
+
+			if (__Mac) {
 				mFLAG.doFlag = true;
+				writeProtect = true;
+			}
 
-			if (__Status)
+			if (__Status) {
 				sFLAG.doNonSticky = true;
+				writeProtect = true;
+			}
+			if (writeProtect == true)
 				
-			if (aCount >= 4)
+			if (aCount >= 3){
 				break;
 			}
+		}
 		incPC2(-2);
 	}
 
+	if (mVUsFlagHack){
+		bool flagReadFound = false;
+		bool updateCalcFound = false;
+		int flagReadptr = 0;
+		incPC2((aCount * 2));
+		int instanceCycles = 0;
+
+		// Go through checking flags if they need to be written back or not, ignoring the last 4ish of course
+		for (int i = mVUcount; i > 0; i--) {
+			if (sFLAG.doFlag) {
+				if (writeProtect && (mVUcount - i) <= aCount) { //Don't actually clear any of the last ops in the block, just check for reads
+					//DevCon.Warning("aCnt %d, mVUcnt %d mVUcnt-i %d", aCount, mVUcount, (mVUcount - i));
+					sFLAG.skipWrite = 2;
+				}
+
+				if (flagReadFound) {
+					if (instanceCycles >= 4) {
+						if (mVUinfo.uOp.VF_write.reg != 0) { //Keep going because this is a mac/status calculation op only
+							if (updateCalcFound == false)
+								sFLAG.skipWrite = 2;
+							//DevCon.Warning("Flag read mac found at %d and %d cycles, rolling back to %d", i, instanceCycles, i + (flagReadptr - 1) * 2);
+							flagReadFound = false;
+							updateCalcFound = false;
+							instanceCycles = 0;
+							//Roll back in case there was other flag reads
+
+							incPC2((flagReadptr - 1) * 2);
+							i += flagReadptr - 1;
+							flagReadptr = 0;
+						}
+						else {
+							//DevCon.Warning("Calculation found!");
+							updateCalcFound = true;
+							sFLAG.skipWrite = 2;
+						}
+					}
+					//else DevCon.Warning("waiting On %d cycles, flagptr %d pos %d", instanceCycles, flagReadptr, i);
+					//Else do nothing, we will come back to this one					
+				}
+				else {
+					if (sFLAG.skipWrite != 2) { //Kill only non-write protected flag updates
+
+						sFLAG.skipWrite = 1;
+						//DevCon.Warning("ignoring at %d", i);
+					}
+					//else DevCon.Warning("Write protected at %d", i);
+				}
+			}
+
+
+
+			if (flagReadFound) {
+				if (i == 1) {
+					//Nothing found by the end of the block
+					flagReadFound = false;
+
+					//Roll back in case there was other flag reads
+					if (flagReadptr > 0) {
+						incPC2((flagReadptr)* 2);
+						i += flagReadptr - 1;
+						flagReadptr = 0;
+					}
+				}
+				else { //Relevant flag update instructions not yet found, keep checking the stalls to get the right one
+					if (flagReadptr > 0){
+						//DevCon.Warning("adding cycles Swap ops %d", mVUinfo.swapOps);
+						instanceCycles += mVUstall;
+						if (mVUstall == 0)
+							instanceCycles += 1;
+
+					}
+					flagReadptr++;
+				}
+			}
+			if (mVUlow.readFlags == 1 && !flagReadFound) {
+				flagReadFound = true;
+				flagReadptr = 0;
+				instanceCycles = 0;
+				//DevCon.Warning("Found Flag on %d Stall = %d swapops = %d", i, mVUstall, mVUinfo.swapOps);
+
+				if (mVUinfo.swapOps)
+					instanceCycles += mVUstall;
+				
+				instanceCycles += 1;
+				//DevCon.Warning("%d cycles on flag read", instanceCycles);
+				flagReadptr++;
+			}
+			incPC2(-2);
+		}
+	}
+	
 	// Status/Mac Flags Setup Code
 	int xS = 0, xM = 0, xC = 0;
 	u32 ff0=0, ff1=0, ffOn=0, fInfo=0;

pcsx2/x86/microVU_IR.h

@@ -133,7 +133,7 @@ struct microLowerOp {
 	bool backupVI;	// Backup VI reg to memory if modified before branch (branch uses old VI value unless opcode is ILW or ILWR)
 	bool memReadIs;	// Read Is (VI reg) from memory (used by branches)
 	bool memReadIt;	// Read If (VI reg) from memory (used by branches)
-	bool readFlags; // Current Instruction reads Status, Mac, or Clip flags
+	u8   readFlags; // Current Instruction reads Status, Mac, or Clip flags
 };
 
 struct microFlagInst {
@@ -142,6 +142,7 @@ struct microFlagInst {
 	u8	 write;		  // Points to the instance that should be written to (s-stage write)
 	u8	 lastWrite;	  // Points to the instance that was last written to (most up-to-date flag)
 	u8	 read;		  // Points to the instance that should be read by a lower instruction (t-stage read)
+	u8	 skipWrite;	  // Makes sure the flags dont written (MAC Flag hack)
 };
 
 struct microFlagCycles {

pcsx2/x86/microVU_Upper.inl

@@ -31,7 +31,7 @@ static void mVUupdateFlags(mV, const xmm& reg, const xmm& regT1in = xEmptyReg, c
 	static const u16 flipMask[16] = {0, 8, 4, 12, 2, 10, 6, 14, 1, 9, 5, 13, 3, 11, 7, 15};
 
 	//SysPrintf("Status = %d; Mac = %d\n", sFLAG.doFlag, mFLAG.doFlag);
-	if (!sFLAG.doFlag && !mFLAG.doFlag) { return; }
+	if ((!sFLAG.doFlag && !mFLAG.doFlag) || sFLAG.skipWrite == 1) { return; }
 
 	const xmm& regT1 = regT1b ? mVU.regAlloc->allocReg() : regT1in;
 
@@ -125,7 +125,7 @@ static void setupPass1(microVU& mVU, int opCase, bool isACC, bool noFlagUpdate)
 	opCase3 { mVUanalyzeFMAC1(mVU, ((isACC) ? 0 : _Fd_), _Fs_, 0); }
 	opCase4 { mVUanalyzeFMAC1(mVU, ((isACC) ? 0 : _Fd_), _Fs_, 0); }
 
-	if (noFlagUpdate) {
+	if (noFlagUpdate) { //Max/Min Ops
 		sFLAG.doFlag = false;
 	}
 }