OPtimize signed+unsigned char arithmetic

include/opp/static_vector.h

@@ -12,8 +12,8 @@ template <class T, int N>
 class static_vector
 {
 protected:
-	char	_space[N * sizeof(T)];
 	enum { m = N }	_size;
+	char	_space[N * sizeof(T)];
 public:
 	typedef T 	element_type;
 

oscar64/ByteCodeGenerator.cpp

@@ -3461,11 +3461,14 @@ ByteCode ByteCodeBasicBlock::RelationalOperator(InterCodeProcedure* proc, const
 	else
 	{
 		if (ins->mSrc[1].mTemp < 0)
+		{
+			if (ins->mSrc[1].mType == IT_INT16 || ins->mSrc[1].mMemory == IM_ABSOLUTE)
 			{
 				ByteCodeInstruction	lins(BC_LOAD_REG_16);
 				lins.mRegister = BC_REG_TMP + proc->mTempOffset[ins->mSrc[0].mTemp];
 				lins.mRegisterFinal = ins->mSrc[0].mFinal;
 				mIns.Push(lins);
+
 				if (csigned)
 				{
 					ByteCodeInstruction	cins(BC_BINOP_CMPSI_16);
@@ -3479,7 +3482,36 @@ ByteCode ByteCodeBasicBlock::RelationalOperator(InterCodeProcedure* proc, const
 					mIns.Push(cins);
 				}
 			}
+			else
+			{
+				ByteCodeInstruction	bins(BC_LEA_ABS);
+				bins.mRegister = BC_REG_ACCU;
+				bins.mLinkerObject = ins->mSrc[1].mLinkerObject;
+				bins.mValue = int(ins->mSrc[1].mIntConst);
+				bins.mRelocate = true;
+				mIns.Push(bins);
+
+				if (csigned)
+				{
+					ByteCodeInstruction	cins(BC_BINOP_CMPSR_16);
+					cins.mRegister = BC_REG_TMP + proc->mTempOffset[ins->mSrc[0].mTemp];
+					cins.mRegisterFinal = ins->mSrc[0].mFinal;
+					mIns.Push(cins);
+				}
+				else
+				{
+					ByteCodeInstruction	cins(BC_BINOP_CMPUR_16);
+					cins.mRegister = BC_REG_TMP + proc->mTempOffset[ins->mSrc[0].mTemp];
+					cins.mRegisterFinal = ins->mSrc[0].mFinal;
+					mIns.Push(cins);
+				}
+
+				code = TransposeBranchCondition(code);
+			}
+		}
 		else if (ins->mSrc[0].mTemp < 0)
+		{
+			if (ins->mSrc[0].mType == IT_INT16 || ins->mSrc[0].mMemory == IM_ABSOLUTE)
 			{
 				ByteCodeInstruction	lins(BC_LOAD_REG_16);
 				lins.mRegister = BC_REG_TMP + proc->mTempOffset[ins->mSrc[1].mTemp];
@@ -3552,6 +3584,31 @@ ByteCode ByteCodeBasicBlock::RelationalOperator(InterCodeProcedure* proc, const
 				code = TransposeBranchCondition(code);
 			}
 			else
+			{
+				ByteCodeInstruction	bins(BC_LEA_ABS);
+				bins.mRegister = BC_REG_ACCU;
+				bins.mLinkerObject = ins->mSrc[0].mLinkerObject;
+				bins.mValue = int(ins->mSrc[0].mIntConst);
+				bins.mRelocate = true;
+				mIns.Push(bins);
+
+				if (csigned)
+				{
+					ByteCodeInstruction	cins(BC_BINOP_CMPSR_16);
+					cins.mRegister = BC_REG_TMP + proc->mTempOffset[ins->mSrc[1].mTemp];
+					cins.mRegisterFinal = ins->mSrc[1].mFinal;
+					mIns.Push(cins);
+				}
+				else
+				{
+					ByteCodeInstruction	cins(BC_BINOP_CMPUR_16);
+					cins.mRegister = BC_REG_TMP + proc->mTempOffset[ins->mSrc[1].mTemp];
+					cins.mRegisterFinal = ins->mSrc[1].mFinal;
+					mIns.Push(cins);
+				}
+			}
+		}
+		else
 		{
 			ByteCodeInstruction	lins(BC_LOAD_REG_16);
 			lins.mRegister = BC_REG_TMP + proc->mTempOffset[ins->mSrc[0].mTemp];

oscar64/InterCode.cpp

@@ -4616,7 +4616,7 @@ bool InterInstruction::PropagateConstTemps(const GrowingInstructionPtrArray& cte
 				InterType	t = mSrc[i].mType;
 				InterInstruction* ains = ctemps[mSrc[i].mTemp];
 
-				if (t != IT_POINTER || ains->mConst.mMemory == IM_ABSOLUTE)
+				if (t != IT_POINTER || ains->mConst.mMemory == IM_ABSOLUTE || ains->mConst.mMemory == IM_GLOBAL)
 				{
 					mSrc[i] = ains->mConst;
 					mSrc[i].mType = t;
@@ -24631,7 +24631,7 @@ void InterCodeProcedure::Close(void)
 {
 	GrowingTypeArray	tstack(IT_NONE);
 	
-	CheckFunc = !strcmp(mIdent->mString, "edit_line");
+	CheckFunc = !strcmp(mIdent->mString, "test_copy_ainit");
 	CheckCase = false;
 
 	mEntryBlock = mBlocks[0];

oscar64/NativeCodeGenerator.cpp

@@ -1994,6 +1994,10 @@ bool NativeCodeInstruction::SameLinkerObjectVariableRange(const NativeCodeInstru
 			return mAddress == ins.mAddress;
 		else if (mLinkerObject && mLinkerObject->mStripe > 1)
 			return mAddress / mLinkerObject->mStripe == ins.mAddress / mLinkerObject->mStripe;
+		else if (mMode == ASMIM_ABSOLUTE && mAddress < ins.mAddress)
+			return false;
+		else if (ins.mMode == ASMIM_ABSOLUTE && ins.mAddress < mAddress)
+			return false;
 		else
 			return true;
 	}
@@ -2889,21 +2893,45 @@ bool NativeCodeInstruction::BitFieldForwarding(NativeRegisterDataSet& data, AsmI
 		opmask = data.mRegs[mAddress].mMask;
 		opvalue = data.mRegs[mAddress].mValue;
 	}
+	else if (mMode == ASMIM_ABSOLUTE && mLinkerObject && (mLinkerObject->mFlags & LOBJF_CONST) && mLinkerObject->mReferences.Size() == 0 && mType != ASMIT_JSR)
+	{
+		opmask = 0xff;
+		opvalue = mLinkerObject->mData[mAddress];
+	}
 #if 1
-	else if ((mMode == ASMIM_ABSOLUTE || mMode == ASMIM_ABSOLUTE_X || mMode == ASMIM_ABSOLUTE_Y) && mLinkerObject && (mLinkerObject->mFlags & LOBJF_CONST) && mLinkerObject->mSize <= 256 && mLinkerObject->mReferences.Size() == 0 && mType != ASMIT_JSR)
+	else if ((mMode == ASMIM_ABSOLUTE_X || mMode == ASMIM_ABSOLUTE_Y) && mLinkerObject && (mLinkerObject->mFlags & LOBJF_CONST) && mLinkerObject->mSize <= 256 && mLinkerObject->mReferences.Size() == 0)
 	{
 		int mor = 0;
 		int mand = 0xff;
-		for (int i = 0; i < mLinkerObject->mSize; i++)
+		int ior = 0;
+		int iand = 0xff;
+
+		if (mMode == ASMIM_ABSOLUTE_X)
 		{
-			mor |= mLinkerObject->mData[i];
+			ior = data.mRegs[CPU_REG_X].mMask & data.mRegs[CPU_REG_X].mValue;
-			mand &= mLinkerObject->mData[i];
+			iand = (~data.mRegs[CPU_REG_X].mMask | data.mRegs[CPU_REG_X].mValue) & 0xff;
+		}
+		else if (mMode == ASMIM_ABSOLUTE_Y)
+		{
+			ior = data.mRegs[CPU_REG_Y].mMask & data.mRegs[CPU_REG_Y].mValue;
+			iand = (~data.mRegs[CPU_REG_Y].mMask | data.mRegs[CPU_REG_Y].mValue) & 0xff;
+		}
+
+		for (int i = 0; i < mLinkerObject->mSize - mAddress; i++)
+		{
+			if ((i & ~iand) == 0 && (i & ior) == ior)
+			{
+				mor |= mLinkerObject->mData[mAddress + i];
+				mand &= mLinkerObject->mData[mAddress + i];
+			}
 		}
 		opmask = (mand | ~mor) & 0xff;
 		opvalue = mand;
+
 #if 0
-		if (opmask)
+		if (CheckFunc)
-			printf("Check %s, %02x %02x\n", mLinkerObject->mIdent->mString, opmask, opvalue);
+			printf("Check %s + %d (%02x, %02x), %02x %02x\n", mLinkerObject->mIdent->mString, mAddress, ior, iand, opmask, opvalue);
+
 #endif
 	}
 #endif
@@ -11292,6 +11320,28 @@ NativeCodeBasicBlock* NativeCodeBasicBlock::BinaryOperator(InterCodeProcedure* p
 								mIns.Push(NativeCodeInstruction(ins, ASMIT_LDA, ASMIM_IMMEDIATE, 0));
 								mIns.Push(NativeCodeInstruction(ins, ASMIT_STA, ASMIM_ZERO_PAGE, treg + 1));
 							}
+#if 1
+							else if (ins->mOperator == IA_ADD && ins->mSrc[0].IsUByte() && ins->mSrc[1].IsSByte() && ins->mDst.mTemp != ins->mSrc[1].mTemp && ins->mSrc[1].mFinal)
+							{
+//								printf("ADD0 %s:%d, %d+%d->%d\n", mProc->mIdent->mString, mIndex, ins->mSrc[0].mTemp, ins->mSrc[1].mTemp, treg);
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_LDA, ASMIM_ZERO_PAGE, BC_REG_TMP + proc->mTempOffset[ins->mSrc[1].mTemp]));
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_EOR, ASMIM_IMMEDIATE, 0x80));
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_ROL, ASMIM_IMPLIED));
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_AND, ASMIM_IMMEDIATE, 0x01));
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_SBC, ASMIM_IMMEDIATE, 0x00));
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_STA, ASMIM_ZERO_PAGE, treg + 1));
+							}
+							else if (ins->mOperator == IA_ADD && ins->mSrc[1].IsUByte() && ins->mSrc[0].IsSByte() && ins->mDst.mTemp != ins->mSrc[0].mTemp && ins->mSrc[0].mFinal)
+							{
+//								printf("ADD1 %s:%d, %d+%d->%d\n", mProc->mIdent->mString, mIndex, ins->mSrc[0].mTemp, ins->mSrc[1].mTemp, treg);
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_LDA, ASMIM_ZERO_PAGE, BC_REG_TMP + proc->mTempOffset[ins->mSrc[0].mTemp]));
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_EOR, ASMIM_IMMEDIATE, 0x80));
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_ROL, ASMIM_IMPLIED));
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_AND, ASMIM_IMMEDIATE, 0x01));
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_SBC, ASMIM_IMMEDIATE, 0x00));
+								mIns.Push(NativeCodeInstruction(ins, ASMIT_STA, ASMIM_ZERO_PAGE, treg + 1));
+							}
+#endif
 							else
 #endif
 							{
@@ -11306,6 +11356,7 @@ NativeCodeBasicBlock* NativeCodeBasicBlock::BinaryOperator(InterCodeProcedure* p
 								mIns.Push(NativeCodeInstruction(ins, ASMIT_STA, ASMIM_ZERO_PAGE, treg + 1));
 							}
 						}
+
 					}
 				}
 			}
@@ -26580,6 +26631,40 @@ bool NativeCodeBasicBlock::JoinTailCodeSequences(NativeCodeProcedure* proc, bool
 #endif
 		CheckLive();
 
+		if (mEntryBlocks.Size() == 1 && mIns.Size() >= 1 && mIns[0].mType == ASMIT_LDA && mIns[0].mMode == ASMIM_ZERO_PAGE && !(mIns[0].mLive & LIVE_CPU_REG_Z))
+		{
+			NativeCodeBasicBlock* eb = mEntryBlocks[0];
+			int index, addr;
+			if (eb->HasTailSTA(addr, index) && addr == mIns[0].mAddress)
+			{
+				for (int i = index; i < eb->mIns.Size(); i++)
+					eb->mIns[i].mLive |= LIVE_CPU_REG_A;
+				eb->mExitRequiredRegs += CPU_REG_A;
+				mEntryRequiredRegs += CPU_REG_A;
+				mIns.Remove(0);
+				changed = true;
+			}
+			else if (eb->HasTailSTY(addr, index) && addr == mIns[0].mAddress)
+			{
+				for (int i = index; i < eb->mIns.Size(); i++)
+					eb->mIns[i].mLive |= LIVE_CPU_REG_Y;
+				eb->mExitRequiredRegs += CPU_REG_Y;
+				mEntryRequiredRegs += CPU_REG_Y;
+				mIns[0].mType = ASMIT_TYA;
+				mIns[0].mMode = ASMIM_IMPLIED;
+				changed = true;
+			}
+			else if (eb->HasTailSTX(addr, index) && addr == mIns[0].mAddress)
+			{
+				for (int i = index; i < eb->mIns.Size(); i++)
+					eb->mIns[i].mLive |= LIVE_CPU_REG_X;
+				eb->mExitRequiredRegs += CPU_REG_X;
+				mEntryRequiredRegs += CPU_REG_X;
+				mIns[0].mType = ASMIT_TXA;
+				mIns[0].mMode = ASMIM_IMPLIED;
+				changed = true;
+			}
+		}
 #if 1
 		if (mIns.Size() >= 1 && mIns[0].mType == ASMIT_TAX && !(mIns[0].mLive & (LIVE_CPU_REG_A | LIVE_CPU_REG_Z)) && !mEntryRegA)
 		{
@@ -28918,6 +29003,22 @@ bool NativeCodeBasicBlock::FindImmediateStore(int at, int reg, const NativeCodeI
 	return false;
 }
 
+int NativeCodeBasicBlock::FindImmediateGlobalStore(int at, const NativeCodeInstruction& ins)
+{
+	at--;
+	while (at >= 0 && !ins.MayBeChangedOnAddress(mIns[at]))
+		at--;
+	if (at > 0 && mIns[at].mMode == ASMIM_ABSOLUTE && mIns[at].mLinkerObject == ins.mLinkerObject && mIns[at].mAddress == ins.mAddress &&
+		mIns[at - 1].mType == ASMIT_LDA && mIns[at - 1].mMode == ASMIM_IMMEDIATE)
+	{
+		return mIns[at - 1].mAddress;
+	}
+	if (at < 0 && mEntryBlocks.Size() == 1)
+		return mEntryBlocks[0]->FindImmediateGlobalStore(mEntryBlocks[0]->mIns.Size(), ins);
+
+	return -1;
+}
+
 bool NativeCodeBasicBlock::CheckPatchFailUse(void)
 {
 	if (mPatchStart)
@@ -31167,6 +31268,165 @@ bool NativeCodeBasicBlock::PatchBoolBitPropagation(const NativeCodeBasicBlock* b
 	return changed;
 }
 
+bool NativeCodeBasicBlock::CollectRegBoolInstructionsForward(int reg, ExpandingArray<NativeCodeBasicBlock*>& cblocks, ExpandingArray<NativeCodeInstruction*>& lins)
+{
+	if (!mPatchStart)
+	{
+		mPatchStart = true;
+		if (!mEntryRequiredRegs[reg])
+			return true;
+
+		int i = 0;
+		while (i < mIns.Size() && !mIns[i].ReferencesZeroPage(reg))
+			i++;
+		if (i < mIns.Size())
+		{
+			if (mIns[i].UsesZeroPage(reg))
+			{
+				if (i == mIns.Size() - 1 &&
+					(mBranch == ASMIT_BEQ || mBranch == ASMIT_BNE) &&
+					(mIns[i].mType == ASMIT_LDA && !(mIns[i].mLive & LIVE_CPU_REG_A) ||
+					 mIns[i].mType == ASMIT_LDX && !(mIns[i].mLive & LIVE_CPU_REG_X) ||
+					 mIns[i].mType == ASMIT_LDY && !(mIns[i].mLive & LIVE_CPU_REG_Y)))
+				{
+					cblocks.Push(this);
+				}
+				else
+					return false;
+			}
+			else if (mIns[i].ChangesZeroPage(reg))
+				return true;
+		}
+
+		for (int i = 0; i < mEntryBlocks.Size(); i++)
+		{
+			if (!mEntryBlocks[i]->CollectRegBoolInstructionsBackward(reg, cblocks, lins))
+				return false;
+		}
+
+		if (mTrueJump && !mTrueJump->CollectRegBoolInstructionsForward(reg, cblocks, lins))
+			return false;
+		if (mFalseJump && !mFalseJump->CollectRegBoolInstructionsForward(reg, cblocks, lins))
+			return false;
+	}
+
+	return true;
+}
+
+bool NativeCodeBasicBlock::CollectRegBoolInstructionsBackward(int reg, ExpandingArray<NativeCodeBasicBlock*>& cblocks, ExpandingArray<NativeCodeInstruction*>& lins)
+{
+	if (!mPatchExit)
+	{
+		mPatchExit = true;
+
+		int sz = mIns.Size();
+		if (mFalseJump && (mBranch == ASMIT_BEQ || mBranch == ASMIT_BNE) && sz > 0 && mIns[sz - 1].mMode == ASMIM_ZERO_PAGE && !(mIns[sz - 1].mLive & LIVE_MEM) && mIns[sz - 1].mAddress == reg)
+		{
+			if (mIns[sz - 1].mType == ASMIT_LDA && !(mIns[sz - 1].mLive & LIVE_CPU_REG_A) ||
+				mIns[sz - 1].mType == ASMIT_LDX && !(mIns[sz - 1].mLive & LIVE_CPU_REG_X) ||
+				mIns[sz - 1].mType == ASMIT_LDY && !(mIns[sz - 1].mLive & LIVE_CPU_REG_Y))
+			{
+				cblocks.Push(this);
+				sz -= 1;
+			}
+		}
+
+		int i = sz - 1;
+		while (i >= 0 && !mIns[i].ReferencesZeroPage(reg))
+			i--;
+
+		bool found = false;
+		if (i >= 0)
+		{
+			if (i > 0 &&
+				mIns[i].mType == ASMIT_STA && mIns[i - 1].mType == ASMIT_LDA && mIns[i - 1].mMode == ASMIM_IMMEDIATE && !(mIns[i].mLive & LIVE_CPU_REG_A) ||
+				mIns[i].mType == ASMIT_STX && mIns[i - 1].mType == ASMIT_LDX && mIns[i - 1].mMode == ASMIM_IMMEDIATE && !(mIns[i].mLive & LIVE_CPU_REG_X) ||
+				mIns[i].mType == ASMIT_STY && mIns[i - 1].mType == ASMIT_LDY && mIns[i - 1].mMode == ASMIM_IMMEDIATE && !(mIns[i].mLive & LIVE_CPU_REG_Y))
+			{
+				lins.Push(&(mIns[i - 1]));
+				found = true;
+			}
+			else
+				return false;
+		}
+
+		if (!found)
+		{
+			if (!mEntryBlocks.Size())
+				return false;
+
+			for (int i = 0; i < mEntryBlocks.Size(); i++)
+			{
+				if (!mEntryBlocks[i]->CollectRegBoolInstructionsBackward(reg, cblocks, lins))
+					return false;
+			}
+		}
+
+		if (mTrueJump && !mTrueJump->CollectRegBoolInstructionsForward(reg, cblocks, lins))
+			return false;
+		if (mFalseJump && !mFalseJump->CollectRegBoolInstructionsForward(reg, cblocks, lins))
+			return false;
+	}
+
+	return true;
+}
+
+
+
+bool NativeCodeBasicBlock::PatchBitBoolConstOrigin(void)
+{
+	bool changed = false;
+
+	if (!mVisited)
+	{
+		mVisited = true;
+
+		int sz = mIns.Size();
+		if (mFalseJump && (mBranch == ASMIT_BEQ || mBranch == ASMIT_BNE) && sz > 0 && mIns[sz - 1].mMode == ASMIM_ZERO_PAGE && !(mIns[sz - 1].mLive & LIVE_MEM))
+		{
+			if (mIns[sz - 1].mType == ASMIT_LDA && !(mIns[sz - 1].mLive & LIVE_CPU_REG_A) ||
+				mIns[sz - 1].mType == ASMIT_LDX && !(mIns[sz - 1].mLive & LIVE_CPU_REG_X) ||
+				mIns[sz - 1].mType == ASMIT_LDY && !(mIns[sz - 1].mLive & LIVE_CPU_REG_Y))
+			{
+				ExpandingArray<NativeCodeBasicBlock*>	lblocks;
+				ExpandingArray<NativeCodeInstruction*>	lins;
+
+				mProc->ResetPatched();
+
+				if (CollectRegBoolInstructionsBackward(mIns[sz - 1].mAddress, lblocks, lins))
+				{
+					for (int i = 0; i < lins.Size(); i++)
+					{
+						if (lins[i]->mAddress != 0)
+							lins[i]->mAddress = 0xff;
+					}
+					for (int i = 0; i < lblocks.Size(); i++)
+					{
+						if (!lblocks[i]->mPatched)
+						{
+							lblocks[i]->mPatched = true;
+							int sz = lblocks[i]->mIns.Size();
+							lblocks[i]->mIns[sz - 1].mType = ASMIT_BIT;
+							if (lblocks[i]->mBranch == ASMIT_BNE)
+								lblocks[i]->mBranch = ASMIT_BMI;
+							else
+								lblocks[i]->mBranch = ASMIT_BPL;
+						}
+					}
+					changed = true;
+				}
+			}
+		}
+
+		if (mTrueJump && mTrueJump->PatchBitBoolConstOrigin())
+			changed = true;
+		if (mFalseJump && mFalseJump->PatchBitBoolConstOrigin())
+			changed = true;
+	}
+
+	return changed;
+}
+
 bool NativeCodeBasicBlock::CheckGlobalAddressSumYPointer(const NativeCodeBasicBlock * block, int reg, int index, int at, int yval)
 {
 	if (!mPatched)
@@ -34887,7 +35147,21 @@ bool NativeCodeBasicBlock::PartialBackwardValuePropagation(void)
 				i++;
 			}
 		}
-
+#if 1
+		for (int i = 0; i < mIns.Size(); i++)
+		{
+			if ((mIns[i].mType == ASMIT_LDA || mIns[i].mType == ASMIT_LDX || mIns[i].mType == ASMIT_LDY) && mIns[i].mMode == ASMIM_ABSOLUTE)
+			{
+				int val = FindImmediateGlobalStore(i, mIns[i]);
+				if (val >= 0)
+				{
+					mIns[i].mMode = ASMIM_IMMEDIATE;
+					mIns[i].mAddress = val;
+					changed = true;
+				}
+			}
+		}
+#endif
 		if (mTrueJump && mTrueJump->PartialBackwardValuePropagation())
 			changed = true;
 		if (mFalseJump && mFalseJump->PartialBackwardValuePropagation())
@@ -39990,6 +40264,31 @@ bool NativeCodeBasicBlock::OptimizeSimpleLoopInvariant(NativeCodeProcedure* proc
 		}
 	}
 
+	if (sz >= 2 &&
+		mIns[0].mType == ASMIT_LDX && mIns[0].mMode == ASMIM_ABSOLUTE &&
+		mIns[1].mType == ASMIT_INC && mIns[0].SameEffectiveAddress(mIns[1]))
+	{
+		int i = 2;
+		while (i < mIns.Size() && !mIns[i].ChangesXReg() && !mIns[0].MayBeChangedOnAddress(mIns[i]))
+			i++;
+		if (i == mIns.Size())
+		{
+			if (!prevBlock)
+				return OptimizeSimpleLoopInvariant(proc, full);
+			for (int i = 0; i < mIns.Size(); i++)
+				mIns[i].mLive |= LIVE_CPU_REG;
+			mIns[1].mType = ASMIT_INX; mIns[1].mMode = ASMIM_IMPLIED;
+			prevBlock->mIns.Push(mIns[0]);
+			exitBlock->mIns.Insert(0, NativeCodeInstruction(mIns[0].mIns, ASMIT_STX, mIns[0]));
+			mIns.Remove(0);
+			prevBlock->mExitRequiredRegs += CPU_REG_X;
+			mEntryRequiredRegs += CPU_REG_X;
+			mExitRequiredRegs += CPU_REG_X;
+			exitBlock->mEntryRequiredRegs += CPU_REG_X;
+			changed = true;
+		}
+	}
+
 	if (sz >= 2 && mIns[0].mType == ASMIT_LDX && mIns[0].mMode == ASMIM_ZERO_PAGE)
 	{
 		int	i = mIns.Size() - 1;
@@ -45206,6 +45505,43 @@ void NativeCodeBasicBlock::BlockSizeReduction(NativeCodeProcedure* proc, int xen
 		}
 #endif
 
+		// Check for small condition block
+		if (mFalseJump)
+		{
+			NativeCodeBasicBlock* cblock = nullptr, * eblock = nullptr;
+
+			if (mTrueJump == mFalseJump->mTrueJump && !mFalseJump->mFalseJump && mTrueJump->mNumEntries == 2 && mFalseJump->mNumEntries == 1)
+			{
+				cblock = mFalseJump;
+				eblock = mTrueJump;
+			}
+			else if (mFalseJump == mTrueJump->mTrueJump && !mTrueJump->mFalseJump && mFalseJump->mNumEntries == 2 && mTrueJump->mNumEntries == 1)
+			{
+				cblock = mTrueJump;
+				eblock = mFalseJump;
+			}
+
+			if (eblock && cblock->mIns.Size() == 1 && cblock->mIns[0].mType == ASMIT_INC)
+			{
+				int i = 0;
+				while (i < eblock->mIns.Size() && !eblock->mIns[i].ReferencesCarry())
+					i++;
+				if (i < eblock->mIns.Size() && eblock->mIns[i].mType == ASMIT_CLC)
+				{
+					if (mBranch == ASMIT_BCC && eblock == mTrueJump || mBranch == ASMIT_BCS && eblock == mFalseJump)
+					{
+						cblock->mIns.Push(NativeCodeInstruction(eblock->mIns[i].mIns, ASMIT_CLC));
+						eblock->mIns.Remove(i);
+						mExitRequiredRegs += CPU_REG_C;
+						cblock->mExitRequiredRegs += CPU_REG_C;
+						eblock->mEntryRequiredRegs += CPU_REG_C;
+						for (int j = 0; j < i; j++)
+							eblock->mIns[j].mLive |= LIVE_CPU_REG_C;
+					}
+				}
+			}
+		}
+
 		bool	carrySet = false, carryClear = false;
 		if (mEntryBlocks.Size() == 1 && center >= 0)
 		{
@@ -45252,7 +45588,7 @@ void NativeCodeBasicBlock::BlockSizeReduction(NativeCodeProcedure* proc, int xen
 
 		i = 0;
 		j = 0;
-		int	accuVal = 0, accuMask = 0;
+		int	accuVal = 0, accuMask = 0, xregVal = 0, xregMask = 0, yregVal = 0, yregMask = 0;
 		bool	accuFlags = false;
 		while (i < mIns.Size())
 		{
@@ -45398,12 +45734,74 @@ void NativeCodeBasicBlock::BlockSizeReduction(NativeCodeProcedure* proc, int xen
 					accuFlags = true;
 				}
 				break;
+			case ASMIT_LDX:
+				if (mIns[i].mMode == ASMIM_IMMEDIATE)
+				{
+					xregVal = mIns[i].mAddress;
+					xregMask = 0xff;
+				}
+				else
+					xregMask = 0;
+				accuFlags = false;
+				break;
+			case ASMIT_LDY:
+				if (mIns[i].mMode == ASMIM_IMMEDIATE)
+				{
+					yregVal = mIns[i].mAddress;
+					yregMask = 0xff;
+				}
+				else
+					yregMask = 0;
+				accuFlags = false;
+				break;
+
 			case ASMIT_LDA:
 				if (mIns[i].mMode == ASMIM_IMMEDIATE)
 				{
 					accuVal = mIns[i].mAddress;
 					accuMask = 0xff;
 				}
+				else if (mIns[i].mMode == ASMIM_ABSOLUTE && mIns[i].mLinkerObject && (mIns[i].mLinkerObject->mFlags & LOBJF_CONST) && mIns[i].mLinkerObject->mReferences.Size() == 0)
+				{
+					accuVal = mIns[i].mLinkerObject->mData[mIns[i].mAddress];
+					accuMask = 0xff;
+				}
+#if 1
+				else if ((mIns[i].mMode == ASMIM_ABSOLUTE_X || mIns[i].mMode == ASMIM_ABSOLUTE_Y) && mIns[i].mLinkerObject && (mIns[i].mLinkerObject->mFlags & LOBJF_CONST) && mIns[i].mLinkerObject->mSize <= 256 && mIns[i].mLinkerObject->mReferences.Size() == 0)
+				{
+					int mor = 0;
+					int mand = 0xff;
+					int ior = 0;
+					int iand = 0xff;
+
+					if (mIns[i].mMode == ASMIM_ABSOLUTE_X)
+					{
+						ior = xregMask & xregVal;
+						iand = (~xregMask | xregVal) & 0xff;
+					}
+					else if (mIns[i].mMode == ASMIM_ABSOLUTE_Y)
+					{
+						ior = yregMask & yregVal;
+						iand = (~yregMask | yregVal) & 0xff;
+					}
+
+					for (int j = 0; j < mIns[i].mLinkerObject->mSize - mIns[i].mAddress; j++)
+					{
+						if ((j & ~iand) == 0 && (j & ior) == ior)
+						{
+							mor |= mIns[i].mLinkerObject->mData[mIns[i].mAddress + j];
+							mand &= mIns[i].mLinkerObject->mData[mIns[i].mAddress + j];
+						}
+					}
+
+					accuMask = (mand | ~mor) & 0xff;
+					accuVal = mand;
+#if 0
+					if (CheckFunc)
+						printf("Check %s + %d (%02x, %02x), %02x %02x\n", mLinkerObject->mIdent->mString, mAddress, ior, iand, opmask, opvalue);
+#endif
+				}
+#endif
 				else
 				{
 					accuVal = 0;
@@ -45546,7 +45944,23 @@ void NativeCodeBasicBlock::BlockSizeReduction(NativeCodeProcedure* proc, int xen
 				accuFlags = mIns[i].mMode == ASMIM_IMPLIED;
 				break;
 			case ASMIT_TAX:
+				xregMask = accuMask;
+				xregVal = accuVal;
+				accuFlags = true;
+				break;
 			case ASMIT_TAY:
+				yregMask = accuMask;
+				yregVal = accuVal;
+				accuFlags = true;
+				break;
+			case ASMIT_TXA:
+				accuMask = xregMask;
+				accuVal = xregVal;
+				accuFlags = true;
+				break;
+			case ASMIT_TYA:
+				accuMask = yregMask;
+				accuVal = yregVal;
 				accuFlags = true;
 				break;
 			default:
@@ -45555,6 +45969,10 @@ void NativeCodeBasicBlock::BlockSizeReduction(NativeCodeProcedure* proc, int xen
 					carryClear = false;
 					carrySet = false;
 				}
+				if (mIns[i].ChangesXReg())
+					xregMask = 0;
+				if (mIns[i].ChangesYReg())
+					yregMask = 0;
 
 				if (mIns[i].ChangesAccu())
 				{
@@ -54280,7 +54698,27 @@ bool NativeCodeBasicBlock::PeepHoleOptimizerExits(int pass)
 			changed = true;
 		}
 	}
-
+#if 1
+	else if (sz >= 4 &&
+		mIns[sz - 4].mType == ASMIT_EOR && mIns[sz - 4].mMode == ASMIM_IMMEDIATE && mIns[sz - 4].mAddress == 0x80 &&
+		mIns[sz - 3].mType == ASMIT_ROL && mIns[sz - 3].mMode == ASMIM_IMPLIED &&
+		mIns[sz - 2].mType == ASMIT_AND && mIns[sz - 2].mMode == ASMIM_IMMEDIATE && mIns[sz - 2].mAddress == 0x01 &&
+		mIns[sz - 1].mType == ASMIT_SBC && mIns[sz - 1].mMode == ASMIM_IMMEDIATE && mIns[sz - 1].mAddress == 0x00 && !(mIns[sz - 1].mLive & LIVE_CPU_REG_A) && !mExitRequiredRegs[CPU_REG_Z] && !mExitRequiredRegs[CPU_REG_C])
+	{
+		if (mBranch == ASMIT_BEQ)
+		{
+			mIns.Remove(sz - 4);
+			mBranch = ASMIT_BNE;
+			changed = true;
+		}
+		else if (mBranch == ASMIT_BNE)
+		{
+			mIns.Remove(sz - 4);
+			mBranch = ASMIT_BEQ;
+			changed = true;
+		}
+	}
+#endif
 	else if (sz >= 2 &&
 		mIns[sz - 2].mType == ASMIT_LDA && mIns[sz - 2].mMode == ASMIM_IMMEDIATE && mIns[sz - 2].mAddress == 0 &&
 		mIns[sz - 1].mType == ASMIT_SBC && mIns[sz - 1].mMode == ASMIM_IMMEDIATE && mIns[sz - 1].mAddress == 0 && !(mIns[sz - 1].mLive & (LIVE_CPU_REG_A | LIVE_CPU_REG_C)) && !mExitRequiredRegs[CPU_REG_Z])
@@ -55930,7 +56368,7 @@ void NativeCodeProcedure::Compile(InterCodeProcedure* proc)
 		
 	mInterProc->mLinkerObject->mNativeProc = this;
 
-	CheckFunc = !strcmp(mIdent->mString, "main");
+	CheckFunc = !strcmp(mIdent->mString, "deco");
 
 	int	nblocks = proc->mBlocks.Size();
 	tblocks = new NativeCodeBasicBlock * [nblocks];
@@ -57831,6 +58269,13 @@ void NativeCodeProcedure::Optimize(void)
 				changed = true;
 		}
 
+		if (step == 23)
+		{
+			ResetVisited();
+			if (mEntryBlock->PatchBitBoolConstOrigin())
+				changed = true;
+		}
+
 #if _DEBUG
 		ResetVisited();
 		mEntryBlock->CheckAsmCode();
@@ -57856,7 +58301,7 @@ void NativeCodeProcedure::Optimize(void)
 		}
 
 #if 1
-		if (!changed && step < 23)
+		if (!changed && step < 24)
 		{
 			ResetIndexFlipped();
 

oscar64/NativeCodeGenerator.h

@@ -465,8 +465,6 @@ public:
 	bool MergeBasicBlocks(void);
 	bool RemoveJumpToBranch(void);
 
-	void MarkLoopHead(void);
-
 	struct DominatorStacks
 	{
 		ExpandingArray< NativeCodeBasicBlock* >	d1, d2;
@@ -575,6 +573,7 @@ public:
 	bool CombineSameYtoX(int xpos, int ypos, int end);
 
 	bool FindImmediateStore(int at, int reg, const NativeCodeInstruction*& ains);
+	int FindImmediateGlobalStore(int at, const NativeCodeInstruction& ins);
 
 	bool JoinXYCrossBlock(void);
 	bool CanCombineSameXtoYCrossBlock(int from);
@@ -812,6 +811,12 @@ public:
 	bool CheckBoolBitPropagation(const NativeCodeBasicBlock* block, int at, int reg);
 	bool PatchBoolBitPropagation(const NativeCodeBasicBlock* block, int at, int reg);
 
+	bool CollectRegBoolInstructionsForward(int reg, ExpandingArray<NativeCodeBasicBlock*>& cblocks, ExpandingArray<NativeCodeInstruction*>& lins);
+	bool CollectRegBoolInstructionsBackward(int reg, ExpandingArray<NativeCodeBasicBlock*>& cblocks, ExpandingArray<NativeCodeInstruction*>& lins);
+
+	bool CollectCheckRegOriginBlocks(int at, int reg, ExpandingArray<NativeCodeBasicBlock*>& lblocks, ExpandingArray<NativeCodeInstruction*>& lins);
+	bool PatchBitBoolConstOrigin(void);
+
 	// reg : base register pair to replace
 	// index: index register
 	// at : start position in block