frederik
/
oscar64
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Replace JMP with branch when possible

This commit is contained in:
drmortalwombat 2022-03-02 21:18:06 +01:00
parent 2a815a9709
commit 9ec2b4fdef
5 changed files with 368 additions and 79 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
autotest/arraytest.c
View File

@ -79,6 +79,7 @@ int main(void)
} }
assert(sum(a, 100) == 450); assert(sum(a, 100) == 450);
copy(b, a, 100); copy(b, a, 100);
assert(sum(b, 100) == 450); assert(sum(b, 100) == 450);
reverse(c, a, 100); reverse(c, a, 100);

74
oscar64/InterCode.cpp
View File

@ -5463,6 +5463,20 @@ void InterCodeBasicBlock::PerformValueForwarding(const GrowingInstructionPtrArra
{ {
InterInstruction* ins = mInstructions[i]; InterInstruction* ins = mInstructions[i];
// Normalize kommutative operands if one is constant
#if 1
if (ins->mCode == IC_BINARY_OPERATOR &&
(ins->mOperator == IA_MUL || ins->mOperator == IA_ADD || ins->mOperator == IA_AND || ins->mOperator == IA_OR || ins->mOperator == IA_XOR) &&
ltvalue[ins->mSrc[1].mTemp] && ltvalue[ins->mSrc[0].mTemp] &&
ltvalue[ins->mSrc[1].mTemp]->mCode == IC_CONSTANT && ltvalue[ins->mSrc[0].mTemp]->mCode != IC_CONSTANT)
{
InterOperand op = ins->mSrc[0];
ins->mSrc[0] = ins->mSrc[1];
ins->mSrc[1] = op;
}
#endif
#if 1 #if 1
if (ins->mCode == IC_BINARY_OPERATOR && ins->mOperator == IA_MUL && ins->mDst.mType == IT_INT16 && spareTemps + 1 < tvalid.Size()) if (ins->mCode == IC_BINARY_OPERATOR && ins->mOperator == IA_MUL && ins->mDst.mType == IT_INT16 && spareTemps + 1 < tvalid.Size())
{ {
@ -5528,6 +5542,66 @@ void InterCodeBasicBlock::PerformValueForwarding(const GrowingInstructionPtrArra
ins->mSrc[0].mTemp = cai->mDst.mTemp; ins->mSrc[0].mTemp = cai->mDst.mTemp;
} }
} }
else if (mi0 && mi1 && mi0->mCode == IC_CONSTANT && mi1->mCode == IC_BINARY_OPERATOR && mi1->mOperator == IA_ADD)
{
InterInstruction* ai0 = ltvalue[mi1->mSrc[0].mTemp], * ai1 = ltvalue[mi1->mSrc[1].mTemp];
if (ai0 && ai0->mCode == IC_CONSTANT)
{
InterInstruction* nai = new InterInstruction();
nai->mCode = IC_BINARY_OPERATOR;
nai->mOperator = IA_MUL;
nai->mSrc[0].mTemp = mi1->mSrc[1].mTemp;
nai->mSrc[0].mType = IT_INT16;
nai->mSrc[1].mTemp = ins->mSrc[0].mTemp;
nai->mSrc[1].mType = IT_INT16;
nai->mDst.mTemp = spareTemps++;
nai->mDst.mType = IT_INT16;
mInstructions.Insert(i, nai);
ltvalue[nai->mDst.mTemp] = nullptr;
InterInstruction* cai = new InterInstruction();
cai->mCode = IC_CONSTANT;
cai->mDst.mTemp = spareTemps++;
cai->mDst.mType = IT_INT16;
cai->mConst.mIntConst = ai0->mConst.mIntConst * mi0->mConst.mIntConst;
mInstructions.Insert(i, cai);
ltvalue[cai->mDst.mTemp] = nullptr;
ins->mOperator = IA_ADD;
ins->mSrc[0].mTemp = nai->mDst.mTemp;
ins->mSrc[1].mTemp = cai->mDst.mTemp;
}
else if (ai1 && ai1->mCode == IC_CONSTANT)
{
InterInstruction* nai = new InterInstruction();
nai->mCode = IC_BINARY_OPERATOR;
nai->mOperator = IA_MUL;
nai->mSrc[0].mTemp = mi1->mSrc[0].mTemp;
nai->mSrc[0].mType = IT_INT16;
nai->mSrc[1].mTemp = ins->mSrc[0].mTemp;
nai->mSrc[1].mType = IT_INT16;
nai->mDst.mTemp = spareTemps++;
nai->mDst.mType = IT_INT16;
mInstructions.Insert(i, nai);
ltvalue[nai->mDst.mTemp] = nullptr;
InterInstruction* cai = new InterInstruction();
cai->mCode = IC_CONSTANT;
cai->mDst.mTemp = spareTemps++;
cai->mDst.mType = IT_INT16;
cai->mConst.mIntConst = ai1->mConst.mIntConst * mi0->mConst.mIntConst;
mInstructions.Insert(i, cai);
ltvalue[cai->mDst.mTemp] = nullptr;
ins->mOperator = IA_ADD;
ins->mSrc[0].mTemp = nai->mDst.mTemp;
ins->mSrc[1].mTemp = cai->mDst.mTemp;
}
}
} }
#endif #endif
#if 1 #if 1

293
oscar64/NativeCodeGenerator.cpp
View File

@ -666,6 +666,44 @@ void NativeCodeInstruction::ReplaceYRegWithXReg(void)
mLive |= LIVE_CPU_REG_Y; mLive |= LIVE_CPU_REG_Y;
} }
void NativeCodeInstruction::ReplaceXRegWithYReg(void)
{
switch (mType)
{
case ASMIT_LDX:
mType = ASMIT_LDY;
break;
case ASMIT_STX:
mType = ASMIT_STY;
break;
case ASMIT_CPX:
mType = ASMIT_CPY;
break;
case ASMIT_TXA:
mType = ASMIT_TYA;
break;
case ASMIT_TAX:
mType = ASMIT_TAY;
break;
case ASMIT_INX:
mType = ASMIT_INY;
break;
case ASMIT_DEX:
mType = ASMIT_DEY;
break;
}
if (mMode == ASMIM_ABSOLUTE_X)
{
assert(HasAsmInstructionMode(mType, ASMIM_ABSOLUTE_Y));
mMode = ASMIM_ABSOLUTE_Y;
}
if (mLive & LIVE_CPU_REG_Y)
mLive |= LIVE_CPU_REG_X;
}
bool NativeCodeInstruction::ChangesYReg(void) const bool NativeCodeInstruction::ChangesYReg(void) const
{ {
return mType == ASMIT_TAY || mType == ASMIT_LDY || mType == ASMIT_INY || mType == ASMIT_DEY || mType == ASMIT_JSR; return mType == ASMIT_TAY || mType == ASMIT_LDY || mType == ASMIT_INY || mType == ASMIT_DEY || mType == ASMIT_JSR;
@ -2914,34 +2952,75 @@ static AsmInsType TransposeBranchCondition(AsmInsType code)
} }
int NativeCodeBasicBlock::PutJump(NativeCodeProcedure* proc, NativeCodeBasicBlock* target) int NativeCodeBasicBlock::PutJump(NativeCodeProcedure* proc, NativeCodeBasicBlock* target, int offset)
{ {
if (target->mIns.Size() == 1 && target->mIns[0].mType == ASMIT_RTS) if (target->mIns.Size() == 1 && target->mIns[0].mType == ASMIT_RTS)
{ {
PutByte(0x60); PutByte(0x60);
return 1; return 1;
} }
else #if 1
else if (offset >= -126 && offset <= 129)
{ {
PutByte(0x4c); if (mNDataSet.mRegs[CPU_REG_C].mMode == NRDM_IMMEDIATE)
{
if (mNDataSet.mRegs[CPU_REG_C].mValue)
PutByte(AsmInsOpcodes[ASMIT_BCS][ASMIM_RELATIVE]);
else
PutByte(AsmInsOpcodes[ASMIT_BCC][ASMIM_RELATIVE]);
LinkerReference rl; PutByte(offset - 2);
rl.mObject = nullptr; return 2;
rl.mOffset = mCode.Size(); }
rl.mFlags = LREF_LOWBYTE | LREF_HIGHBYTE; else if (mNDataSet.mRegs[CPU_REG_Z].mMode == NRDM_IMMEDIATE)
rl.mRefObject = nullptr; {
rl.mRefOffset = target->mOffset; if (mNDataSet.mRegs[CPU_REG_Z].mValue)
mRelocations.Push(rl); PutByte(AsmInsOpcodes[ASMIT_BNE][ASMIM_RELATIVE]);
else
PutByte(AsmInsOpcodes[ASMIT_BEQ][ASMIM_RELATIVE]);
PutWord(0); PutByte(offset - 2);
return 3; return 2;
}
} }
#endif
PutByte(0x4c);
LinkerReference rl;
rl.mObject = nullptr;
rl.mOffset = mCode.Size();
rl.mFlags = LREF_LOWBYTE | LREF_HIGHBYTE;
rl.mRefObject = nullptr;
rl.mRefOffset = target->mOffset;
mRelocations.Push(rl);
PutWord(0);
return 3;
} }
int NativeCodeBasicBlock::JumpByteSize(NativeCodeBasicBlock* target) int NativeCodeBasicBlock::BranchByteSize(int from, int to)
{
if (to - from >= -126 && to - from <= 129)
return 2;
else
return 5;
}
int NativeCodeBasicBlock::JumpByteSize(NativeCodeBasicBlock* target, int offset)
{ {
if (target->mIns.Size() == 1 && target->mIns[0].mType == ASMIT_RTS) if (target->mIns.Size() == 1 && target->mIns[0].mType == ASMIT_RTS)
return 1; return 1;
#if 1
else if (offset >= -126 && offset <= 129)
{
if (mNDataSet.mRegs[CPU_REG_C].mMode == NRDM_IMMEDIATE)
return 2;
else if (mNDataSet.mRegs[CPU_REG_Z].mMode == NRDM_IMMEDIATE)
return 2;
else
return 3;
}
#endif
else else
return 3; return 3;
} }
@ -9112,6 +9191,74 @@ bool NativeCodeBasicBlock::ReduceLocalYPressure(void)
return changed; return changed;
} }
bool NativeCodeBasicBlock::ReplaceXRegWithYReg(int start, int end)
{
for (int i = start; i < end; i++)
{
NativeCodeInstruction& ins(mIns[i]);
ins.ReplaceXRegWithYReg();
}
return true;
}
bool NativeCodeBasicBlock::ReduceLocalXPressure(void)
{
bool changed = false;
if (!mVisited)
{
mVisited = true;
int start = 0;
while (start < mIns.Size())
{
const NativeCodeInstruction& ins(mIns[start]);
if ((ins.mType == ASMIT_LDX || ins.mType == ASMIT_TAX) && ins.mMode != ASMIM_INDIRECT_Y && ins.mMode != ASMIM_ABSOLUTE_Y && !(ins.mLive & LIVE_CPU_REG_Y))
{
int end = start + 1;
while (end < mIns.Size())
{
const NativeCodeInstruction& eins(mIns[end]);
if (eins.mType == ASMIT_LDX || eins.mType == ASMIT_TAX)
{
ReplaceXRegWithYReg(start, end);
changed = true;
break;
}
else if (eins.ChangesYReg() || (eins.mMode == ASMIM_ABSOLUTE_X && !HasAsmInstructionMode(eins.mType, ASMIM_ABSOLUTE_Y)))
{
break;
}
else if (!(eins.mLive & LIVE_CPU_REG_X))
{
end++;
ReplaceXRegWithYReg(start, end);
changed = true;
break;
}
end++;
}
start = end;
}
else
start++;
}
if (mTrueJump && mTrueJump->ReduceLocalXPressure())
changed = true;
if (mFalseJump && mFalseJump->ReduceLocalXPressure())
changed = true;
}
return changed;
}
bool NativeCodeBasicBlock::LocalRegisterXYMap(void) bool NativeCodeBasicBlock::LocalRegisterXYMap(void)
{ {
bool changed = false; bool changed = false;
@ -9919,7 +10066,7 @@ bool NativeCodeBasicBlock::PatchSingleUseGlobalLoad(int reg, int at, const Nativ
return changed; return changed;
} }
bool NativeCodeBasicBlock::CheckForwardSumYPointer(int reg, int base, int at, int yval) bool NativeCodeBasicBlock::CheckForwardSumYPointer(int reg, int base, int index, int at, int yval)
{ {
if (!mPatched) if (!mPatched)
{ {
@ -9938,7 +10085,7 @@ bool NativeCodeBasicBlock::CheckForwardSumYPointer(int reg, int base, int at, in
{ {
NativeCodeInstruction& ins(mIns[at]); NativeCodeInstruction& ins(mIns[at]);
if (ins.mMode == ASMIM_ZERO_PAGE && (ins.mAddress == reg || ins.mAddress == reg + 1)) if (ins.mMode == ASMIM_ZERO_PAGE && (ins.mAddress == reg || ins.mAddress == reg + 1 || ins.mAddress == index))
return false; return false;
else if (ins.mMode == ASMIM_INDIRECT_Y && ins.mAddress == reg) else if (ins.mMode == ASMIM_INDIRECT_Y && ins.mAddress == reg)
{ {
@ -9960,7 +10107,7 @@ bool NativeCodeBasicBlock::CheckForwardSumYPointer(int reg, int base, int at, in
return false; return false;
yval = -1; yval = -1;
} }
else if (ins.ChangesZeroPage(base) || ins.ChangesZeroPage(base + 1)) else if (ins.ChangesZeroPage(base) || ins.ChangesZeroPage(base + 1) || ins.ChangesZeroPage(index))
return false; return false;
else if (ins.ChangesYReg()) else if (ins.ChangesYReg())
yval = -1; yval = -1;
@ -9968,16 +10115,16 @@ bool NativeCodeBasicBlock::CheckForwardSumYPointer(int reg, int base, int at, in
at++; at++;
} }
if (mTrueJump && !mTrueJump->CheckForwardSumYPointer(reg, base, 0, yval)) if (mTrueJump && !mTrueJump->CheckForwardSumYPointer(reg, base, index, 0, yval))
return false; return false;
if (mFalseJump && !mFalseJump->CheckForwardSumYPointer(reg, base, 0, yval)) if (mFalseJump && !mFalseJump->CheckForwardSumYPointer(reg, base, index, 0, yval))
return false; return false;
} }
return true; return true;
} }
bool NativeCodeBasicBlock::PatchForwardSumYPointer(int reg, int base, int at, int yval) bool NativeCodeBasicBlock::PatchForwardSumYPointer(int reg, int base, int index, int at, int yval)
{ {
bool changed = false; bool changed = false;
@ -10003,7 +10150,7 @@ bool NativeCodeBasicBlock::PatchForwardSumYPointer(int reg, int base, int at, in
if (ins.mLive & LIVE_CPU_REG_Y) if (ins.mLive & LIVE_CPU_REG_Y)
mIns.Insert(at + 1, NativeCodeInstruction(ASMIT_LDY, ASMIM_IMMEDIATE, yval)); mIns.Insert(at + 1, NativeCodeInstruction(ASMIT_LDY, ASMIM_IMMEDIATE, yval));
mIns.Insert(at, NativeCodeInstruction(ASMIT_LDY, ASMIM_ZERO_PAGE, reg)); mIns.Insert(at, NativeCodeInstruction(ASMIT_LDY, ASMIM_ZERO_PAGE, index));
at++; at++;
for (int i = 0; i < yval; i++) for (int i = 0; i < yval; i++)
{ {
@ -10028,9 +10175,9 @@ bool NativeCodeBasicBlock::PatchForwardSumYPointer(int reg, int base, int at, in
at++; at++;
} }
if (mTrueJump && mTrueJump->PatchForwardSumYPointer(reg, base, 0, yval)) if (mTrueJump && mTrueJump->PatchForwardSumYPointer(reg, base, index, 0, yval))
changed = true; changed = true;
if (mFalseJump && mFalseJump->PatchForwardSumYPointer(reg, base, 0, yval)) if (mFalseJump && mFalseJump->PatchForwardSumYPointer(reg, base, index, 0, yval))
changed = true; changed = true;
} }
@ -14747,6 +14894,7 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 0].mType == ASMIT_TXA && mIns[i + 0].mType == ASMIT_TXA &&
mIns[i + 1].mType == ASMIT_STA && (mIns[i + 1].mMode == ASMIM_ZERO_PAGE || mIns[i + 1].mMode == ASMIM_ABSOLUTE)) mIns[i + 1].mType == ASMIT_STA && (mIns[i + 1].mMode == ASMIM_ZERO_PAGE || mIns[i + 1].mMode == ASMIM_ABSOLUTE))
{ {
mIns[i + 0].mLive |= LIVE_CPU_REG_X;
mIns[i + 1].mType = ASMIT_STX; mIns[i + 1].mType = ASMIT_STX;
progress = true; progress = true;
} }
@ -14755,6 +14903,7 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 0].mType == ASMIT_TYA && mIns[i + 0].mType == ASMIT_TYA &&
mIns[i + 1].mType == ASMIT_STA && (mIns[i + 1].mMode == ASMIM_ZERO_PAGE || mIns[i + 1].mMode == ASMIM_ABSOLUTE)) mIns[i + 1].mType == ASMIT_STA && (mIns[i + 1].mMode == ASMIM_ZERO_PAGE || mIns[i + 1].mMode == ASMIM_ABSOLUTE))
{ {
mIns[i + 0].mLive |= LIVE_CPU_REG_Y;
mIns[i + 1].mType = ASMIT_STY; mIns[i + 1].mType = ASMIT_STY;
progress = true; progress = true;
} }
@ -14762,6 +14911,7 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 0].mType == ASMIT_TAX && mIns[i + 0].mType == ASMIT_TAX &&
mIns[i + 1].mType == ASMIT_STX && (mIns[i + 1].mMode == ASMIM_ZERO_PAGE || mIns[i + 1].mMode == ASMIM_ABSOLUTE)) mIns[i + 1].mType == ASMIT_STX && (mIns[i + 1].mMode == ASMIM_ZERO_PAGE || mIns[i + 1].mMode == ASMIM_ABSOLUTE))
{ {
mIns[i + 0].mLive |= LIVE_CPU_REG_A;
mIns[i + 1].mType = ASMIT_STA; mIns[i + 1].mType = ASMIT_STA;
progress = true; progress = true;
} }
@ -14769,6 +14919,7 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 0].mType == ASMIT_TAY && mIns[i + 0].mType == ASMIT_TAY &&
mIns[i + 1].mType == ASMIT_STY && (mIns[i + 1].mMode == ASMIM_ZERO_PAGE || mIns[i + 1].mMode == ASMIM_ABSOLUTE)) mIns[i + 1].mType == ASMIT_STY && (mIns[i + 1].mMode == ASMIM_ZERO_PAGE || mIns[i + 1].mMode == ASMIM_ABSOLUTE))
{ {
mIns[i + 0].mLive |= LIVE_CPU_REG_A;
mIns[i + 1].mType = ASMIT_STA; mIns[i + 1].mType = ASMIT_STA;
progress = true; progress = true;
} }
@ -14871,7 +15022,7 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 1].mType == ASMIT_STA && mIns[i + 1].mMode == ASMIM_ZERO_PAGE && !mIns[i + 1].SameEffectiveAddress(mIns[i + 0])) mIns[i + 1].mType == ASMIT_STA && mIns[i + 1].mMode == ASMIM_ZERO_PAGE && !mIns[i + 1].SameEffectiveAddress(mIns[i + 0]))
{ {
NativeCodeInstruction ins = mIns[i + 0]; NativeCodeInstruction ins = mIns[i + 0];
mIns[i + 0] = mIns[i + 1]; mIns[i + 0] = mIns[i + 1]; mIns[i + 0].mLive |= ins.mLive;
mIns[i + 1] = ins; mIns[i + 1] = ins;
progress = true; progress = true;
} }
@ -14982,7 +15133,7 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 0].mFlags = iins->mFlags; mIns[i + 0].mFlags = iins->mFlags;
if (!(flags & LIVE_CPU_REG_Y) && (flags & LIVE_CPU_REG_X)) if (!(flags & LIVE_CPU_REG_Y) && (flags & LIVE_CPU_REG_X))
mIns[i + 0].mType == ASMIT_LDX; mIns[i + 0].mType = ASMIT_LDX;
} }
else if (iins->mAddress == sreg) else if (iins->mAddress == sreg)
{ {
@ -15009,7 +15160,7 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 0].mAddress = iins->mAddress; mIns[i + 0].mAddress = iins->mAddress;
if (!(flags & LIVE_CPU_REG_Y) && (flags & LIVE_CPU_REG_X)) if (!(flags & LIVE_CPU_REG_Y) && (flags & LIVE_CPU_REG_X))
mIns[i + 0].mType == ASMIT_LDX; mIns[i + 0].mType = ASMIT_LDX;
} }
progress = true; progress = true;
@ -15768,6 +15919,27 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 1].mType = ASMIT_STA; mIns[i + 1].mLive |= mIns[i + 2].mLive; mIns[i + 1].mType = ASMIT_STA; mIns[i + 1].mLive |= mIns[i + 2].mLive;
progress = true; progress = true;
} }
else if (
mIns[i + 0].mType == ASMIT_LDX &&
mIns[i + 1].mType == ASMIT_STX &&
mIns[i + 2].mType == ASMIT_TXA && !(mIns[i + 2].mLive & LIVE_CPU_REG_X))
{
mIns[i + 0].mType = ASMIT_LDA; mIns[i + 0].mLive |= LIVE_CPU_REG_A;
mIns[i + 1].mType = ASMIT_STA; mIns[i + 1].mLive |= LIVE_CPU_REG_A;
mIns[i + 2].mType = ASMIT_NOP; mIns[i + 2].mMode = ASMIM_IMPLIED;
progress = true;
}
else if (
mIns[i + 0].mType == ASMIT_LDY &&
mIns[i + 1].mType == ASMIT_STY &&
mIns[i + 2].mType == ASMIT_TYA && !(mIns[i + 2].mLive & LIVE_CPU_REG_Y))
{
mIns[i + 0].mType = ASMIT_LDA; mIns[i + 0].mLive |= LIVE_CPU_REG_A;
mIns[i + 1].mType = ASMIT_STA; mIns[i + 1].mLive |= LIVE_CPU_REG_A;
mIns[i + 2].mType = ASMIT_NOP; mIns[i + 2].mMode = ASMIM_IMPLIED;
progress = true;
}
#if 1 #if 1
if ( if (
@ -15929,6 +16101,8 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 3].mType == ASMIT_STX && mIns[i + 3].mMode == ASMIM_ZERO_PAGE && mIns[i + 3].mType == ASMIT_STX && mIns[i + 3].mMode == ASMIM_ZERO_PAGE &&
!mIns[i + 1].ChangesZeroPage(mIns[i + 0].mAddress) && !mIns[i + 1].UsesZeroPage(mIns[i + 3].mAddress)) !mIns[i + 1].ChangesZeroPage(mIns[i + 0].mAddress) && !mIns[i + 1].UsesZeroPage(mIns[i + 3].mAddress))
{ {
mIns[i + 0].mLive |= LIVE_CPU_REG_A;
mIns.Insert(i + 1, NativeCodeInstruction(ASMIT_STA, ASMIM_ZERO_PAGE, mIns[i + 3].mAddress)); mIns.Insert(i + 1, NativeCodeInstruction(ASMIT_STA, ASMIM_ZERO_PAGE, mIns[i + 3].mAddress));
mIns[i + 4].mType = ASMIT_NOP; mIns[i + 4].mMode = ASMIM_IMPLIED; mIns[i + 4].mType = ASMIT_NOP; mIns[i + 4].mMode = ASMIM_IMPLIED;
@ -16671,7 +16845,7 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 5].mType = ASMIT_NOP; mIns[i + 5].mMode = ASMIM_IMPLIED; mIns[i + 5].mType = ASMIT_NOP; mIns[i + 5].mMode = ASMIM_IMPLIED;
progress = true; progress = true;
} }
#if 1
if (mIns[i + 0].mType == ASMIT_CLC && if (mIns[i + 0].mType == ASMIT_CLC &&
mIns[i + 1].mType == ASMIT_LDA && mIns[i + 1].mMode == ASMIM_ZERO_PAGE && mIns[i + 1].mType == ASMIT_LDA && mIns[i + 1].mMode == ASMIM_ZERO_PAGE &&
mIns[i + 2].mType == ASMIT_ADC && mIns[i + 2].mMode == ASMIM_ZERO_PAGE && mIns[i + 2].mType == ASMIT_ADC && mIns[i + 2].mMode == ASMIM_ZERO_PAGE &&
@ -16682,7 +16856,7 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
!(mIns[i + 6].mLive & LIVE_CPU_REG_A)) !(mIns[i + 6].mLive & LIVE_CPU_REG_A))
{ {
proc->ResetPatched(); proc->ResetPatched();
if (CheckForwardSumYPointer(mIns[i + 3].mAddress, mIns[i + 1].mAddress, i + 7, -1)) if (CheckForwardSumYPointer(mIns[i + 3].mAddress, mIns[i + 1].mAddress, mIns[i + 3].mAddress, i + 7, -1))
{ {
mIns[i + 0].mType = ASMIT_NOP; mIns[i + 0].mMode = ASMIM_IMPLIED; mIns[i + 0].mType = ASMIT_NOP; mIns[i + 0].mMode = ASMIM_IMPLIED;
mIns[i + 1].mType = ASMIT_NOP; mIns[i + 1].mMode = ASMIM_IMPLIED; mIns[i + 1].mType = ASMIT_NOP; mIns[i + 1].mMode = ASMIM_IMPLIED;
@ -16693,11 +16867,38 @@ bool NativeCodeBasicBlock::PeepHoleOptimizer(NativeCodeProcedure* proc, int pass
mIns[i + 6].mType = ASMIT_NOP; mIns[i + 6].mMode = ASMIM_IMPLIED; mIns[i + 6].mType = ASMIT_NOP; mIns[i + 6].mMode = ASMIM_IMPLIED;
proc->ResetPatched(); proc->ResetPatched();
if (PatchForwardSumYPointer(mIns[i + 3].mAddress, mIns[i + 1].mAddress, i + 7, -1)) if (PatchForwardSumYPointer(mIns[i + 3].mAddress, mIns[i + 1].mAddress, mIns[i + 3].mAddress, i + 7, -1))
progress = true; progress = true;
} }
} }
#endif
#if 1
if (
mIns[i + 0].mType == ASMIT_STA && mIns[i + 0].mMode == ASMIM_ZERO_PAGE &&
mIns[i + 1].mType == ASMIT_CLC &&
mIns[i + 2].mType == ASMIT_ADC && mIns[i + 2].mMode == ASMIM_ZERO_PAGE &&
mIns[i + 3].mType == ASMIT_STA && mIns[i + 3].mMode == ASMIM_ZERO_PAGE && mIns[i + 3].mAddress != mIns[i + 0].mAddress &&
mIns[i + 4].mType == ASMIT_LDA && mIns[i + 4].mMode == ASMIM_ZERO_PAGE && mIns[i + 4].mAddress == mIns[i + 2].mAddress + 1 &&
mIns[i + 5].mType == ASMIT_ADC && mIns[i + 5].mMode == ASMIM_IMMEDIATE && mIns[i + 5].mAddress == 0 &&
mIns[i + 6].mType == ASMIT_STA && mIns[i + 6].mMode == ASMIM_ZERO_PAGE && mIns[i + 6].mAddress == mIns[i + 3].mAddress + 1 &&
!(mIns[i + 6].mLive & LIVE_CPU_REG_A))
{
proc->ResetPatched();
if (CheckForwardSumYPointer(mIns[i + 3].mAddress, mIns[i + 2].mAddress, mIns[i + 0].mAddress, i + 7, -1))
{
mIns[i + 1].mType = ASMIT_NOP; mIns[i + 1].mMode = ASMIM_IMPLIED;
mIns[i + 2].mType = ASMIT_NOP; mIns[i + 2].mMode = ASMIM_IMPLIED;
mIns[i + 3].mType = ASMIT_NOP; mIns[i + 3].mMode = ASMIM_IMPLIED;
mIns[i + 4].mType = ASMIT_NOP; mIns[i + 4].mMode = ASMIM_IMPLIED;
mIns[i + 5].mType = ASMIT_NOP; mIns[i + 5].mMode = ASMIM_IMPLIED;
mIns[i + 6].mType = ASMIT_NOP; mIns[i + 6].mMode = ASMIM_IMPLIED;
proc->ResetPatched();
if (PatchForwardSumYPointer(mIns[i + 3].mAddress, mIns[i + 2].mAddress, mIns[i + 0].mAddress, i + 7, -1))
progress = true;
}
}
#endif
} }
#if 1 #if 1
@ -16920,14 +17121,6 @@ void NativeCodeBasicBlock::Close(NativeCodeBasicBlock* trueJump, NativeCodeBasic
} }
static int BranchByteSize(int from, int to)
{
if (to - from >= -126 && to - from <= 129)
return 2;
else
return 5;
}
NativeCodeBasicBlock* NativeCodeBasicBlock::BypassEmptyBlocks(void) NativeCodeBasicBlock* NativeCodeBasicBlock::BypassEmptyBlocks(void)
{ {
if (mBypassed) if (mBypassed)
@ -17065,13 +17258,13 @@ bool NativeCodeBasicBlock::CalculateOffset(int& total)
else else
{ {
total += BranchByteSize(total, mTrueJump->mOffset); total += BranchByteSize(total, mTrueJump->mOffset);
total += JumpByteSize(mFalseJump); total += JumpByteSize(mFalseJump, mFalseJump->mOffset - total);
} }
} }
else if (mTrueJump) else if (mTrueJump)
{ {
if (mTrueJump->mPlace != mPlace + 1) if (mTrueJump->mPlace != mPlace + 1)
total += JumpByteSize(mTrueJump); total += JumpByteSize(mTrueJump, mTrueJump->mOffset - total);
} }
if (mOffset + mSize != total) if (mOffset + mSize != total)
@ -17100,13 +17293,13 @@ void NativeCodeBasicBlock::CopyCode(NativeCodeProcedure * proc, uint8* target)
else else
{ {
end += PutBranch(proc, mBranch, mTrueJump->mOffset - end); end += PutBranch(proc, mBranch, mTrueJump->mOffset - end);
end += PutJump(proc, mFalseJump); end += PutJump(proc, mFalseJump, mFalseJump->mOffset - end);
} }
} }
else if (mTrueJump) else if (mTrueJump)
{ {
if (mTrueJump->mPlace != mPlace + 1) if (mTrueJump->mPlace != mPlace + 1)
end += PutJump(proc, mTrueJump); end += PutJump(proc, mTrueJump, mTrueJump->mOffset - end);
} }
assert(end == next); assert(end == next);
@ -17951,11 +18144,12 @@ void NativeCodeProcedure::Optimize(void)
} }
#endif #endif
#if 1 #if 1
if (step == 4) if (step == 4 || step == 5)
{ {
#if 1 #if 1
#if 1 #if 1
int xregs[256], yregs[256]; int xregs[256], yregs[256];
for (int i = 0; i < 256; i++) for (int i = 0; i < 256; i++)
xregs[i] = yregs[i] = 0; xregs[i] = yregs[i] = 0;
@ -17983,6 +18177,7 @@ void NativeCodeProcedure::Optimize(void)
ResetVisited(); ResetVisited();
mEntryBlock->GlobalRegisterXYCheck(xregs, yregs); mEntryBlock->GlobalRegisterXYCheck(xregs, yregs);
if (xregs[0] >= 0) if (xregs[0] >= 0)
{ {
int j = 1; int j = 1;
@ -18019,9 +18214,17 @@ void NativeCodeProcedure::Optimize(void)
#endif #endif
if (!changed) if (!changed)
{ {
ResetVisited(); if (step == 4)
if (mEntryBlock->LocalRegisterXYMap()) {
changed = true; ResetVisited();
mEntryBlock->ReduceLocalXPressure();
}
else
{
ResetVisited();
if (mEntryBlock->LocalRegisterXYMap())
changed = true;
}
} }
#endif #endif
} }
@ -18036,7 +18239,7 @@ void NativeCodeProcedure::Optimize(void)
changed = true; changed = true;
#endif #endif
#endif #endif
if (!changed && step < 5) if (!changed && step < 6)
{ {
step++; step++;
changed = true; changed = true;

12
oscar64/NativeCodeGenerator.h
View File

@ -105,6 +105,7 @@ public:
bool IsShift(void) const; bool IsShift(void) const;
void ReplaceYRegWithXReg(void); void ReplaceYRegWithXReg(void);
void ReplaceXRegWithYReg(void);
}; };
class NativeCodeBasicBlock class NativeCodeBasicBlock
@ -131,8 +132,9 @@ public:
NativeRegisterDataSet mDataSet, mNDataSet; NativeRegisterDataSet mDataSet, mNDataSet;
int PutBranch(NativeCodeProcedure* proc, AsmInsType code, int offset); int PutBranch(NativeCodeProcedure* proc, AsmInsType code, int offset);
int PutJump(NativeCodeProcedure* proc, NativeCodeBasicBlock* target); int PutJump(NativeCodeProcedure* proc, NativeCodeBasicBlock* target, int offset);
int JumpByteSize(NativeCodeBasicBlock * target); int JumpByteSize(NativeCodeBasicBlock * target, int offset);
int BranchByteSize(int from, int to);
NativeCodeBasicBlock* BypassEmptyBlocks(void); NativeCodeBasicBlock* BypassEmptyBlocks(void);
@ -237,6 +239,7 @@ public:
bool ReverseLoadCommutativeOpUp(int aload, int aop); bool ReverseLoadCommutativeOpUp(int aload, int aop);
bool ReplaceZeroPageUp(int at); bool ReplaceZeroPageUp(int at);
bool ReplaceYRegWithXReg(int start, int end); bool ReplaceYRegWithXReg(int start, int end);
bool ReplaceXRegWithYReg(int start, int end);
bool FindImmediateStore(int at, int reg, const NativeCodeInstruction*& ains); bool FindImmediateStore(int at, int reg, const NativeCodeInstruction*& ains);
@ -271,6 +274,7 @@ public:
void GlobalRegisterYMap(int reg); void GlobalRegisterYMap(int reg);
bool LocalRegisterXYMap(void); bool LocalRegisterXYMap(void);
bool ReduceLocalYPressure(void); bool ReduceLocalYPressure(void);
bool ReduceLocalXPressure(void);
bool CheckGlobalAddressSumYPointer(int reg, int at, int yval); bool CheckGlobalAddressSumYPointer(int reg, int at, int yval);
bool PatchGlobalAddressSumYPointer(int reg, int at, int yval, LinkerObject * lobj, int address); bool PatchGlobalAddressSumYPointer(int reg, int at, int yval, LinkerObject * lobj, int address);
@ -278,8 +282,8 @@ public:
bool CheckSingleUseGlobalLoad(int reg, int at, const NativeCodeInstruction& ains); bool CheckSingleUseGlobalLoad(int reg, int at, const NativeCodeInstruction& ains);
bool PatchSingleUseGlobalLoad(int reg, int at, const NativeCodeInstruction& ains); bool PatchSingleUseGlobalLoad(int reg, int at, const NativeCodeInstruction& ains);
bool CheckForwardSumYPointer(int reg, int base, int at, int yval); bool CheckForwardSumYPointer(int reg, int base, int index, int at, int yval);
bool PatchForwardSumYPointer(int reg, int base, int at, int yval); bool PatchForwardSumYPointer(int reg, int base, int index, int at, int yval);
}; };
class NativeCodeProcedure class NativeCodeProcedure

65
samples/games/hscrollshmup.c
View File

@ -171,12 +171,23 @@ void tiles_draw1(char * dp, char * tm)
struct Shot struct Shot
{ {
byte ty, x, ry, n; byte ty, x, ry, n;
sbyte dx; sbyte dx;
Shot * next;
} shots[18]; } shots[18];
Shot * firstShot; Shot * freeShot;
Shot * lastShot;
void shot_init(void)
{
shots[0].next = shots;
shots[0].ty = 6;
freeShot = shots + 1;
for(char i=1; i<17; i++)
shots[i].next = shots + i + 1;
shots[17].next = nullptr;
}
inline void shot_draw(char * dp, char i, char xp, char yp) inline void shot_draw(char * dp, char i, char xp, char yp)
{ {
@ -202,16 +213,14 @@ void shot_add(int dx, int sy)
char ey = (py >> 3) & 3; char ey = (py >> 3) & 3;
char ry = py & 7; char ry = py & 7;
Shot * s = lastShot - 1; Shot * s = freeShot;
while (s->ty > gy) freeShot = s->next;
{
s[1] = s[0];
s--;
}
s++;
lastShot++; Shot * p = shots;
lastShot->ty = 6; while (p->next->ty < gy)
p = p->next;
s->next = p->next;
p->next = s;
s->ty = gy; s->ty = gy;
s->ry = ry; s->ry = ry;
@ -241,7 +250,7 @@ void tiles_draw(unsigned x)
char yl = 0; char yl = 0;
char ci = 0; char ci = 0;
Shot * ss = firstShot, * ts = firstShot; Shot * ps = shots;
for(int iy=0; iy<5; iy++) for(int iy=0; iy<5; iy++)
{ {
@ -303,26 +312,27 @@ void tiles_draw(unsigned x)
dp[k] = 0xf8; dp[k] = 0xf8;
} }
Shot *ss = ps->next;
while (ss->ty == iy) while (ss->ty == iy)
{ {
ss->x += ss->dx; ss->x += ss->dx;
ss->n--; ss->n--;
shot_draw(dp, ci++, ss->x, ss->ry); shot_draw(dp, ci++, ss->x, ss->ry);
if (ss->n) if (ss->n)
ps = ss;
else
{ {
if (ss != ts) ps->next = ss->next;
*ts = *ss; ss->next = freeShot;
ts++; freeShot = ss;
} }
ss++; ss = ps->next;
} }
yl += 4; yl += 4;
} }
lastShot = ts;
lastShot->ty = 6;
Font[248 * 8 + 2] = ~(1 << xs); Font[248 * 8 + 2] = ~(1 << xs);
vic.ctrl2 = VIC_CTRL2_MCM + xs; vic.ctrl2 = VIC_CTRL2_MCM + xs;
@ -336,8 +346,8 @@ struct Enemy
} enemies[5]; } enemies[5];
int spx = 40; int spx = 40;
int vpx = 16; sbyte vpx = 16;
int ax = 0; sbyte ax = 0;
char spy = 100; char spy = 100;
char fdelay = 0; char fdelay = 0;
char edelay = 5; char edelay = 5;
@ -363,7 +373,7 @@ void enemies_move(void)
} }
} }
void enemies_spwan(void) void enemies_spawn(void)
{ {
char u = rand(); char u = rand();
@ -420,10 +430,7 @@ int main(void)
for(int i=0; i<24; i++) for(int i=0; i<24; i++)
stars[i] = rand() % 40 + 40 * (i & 3); stars[i] = rand() % 40 + 40 * (i & 3);
shots[0].ty = 0; shot_init();
firstShot = shots + 1;
lastShot = firstShot;
lastShot->ty = 6;
spr_set(0, true, 160, 100, 64, VCOL_BLUE, true, false, false); spr_set(0, true, 160, 100, 64, VCOL_BLUE, true, false, false);
spr_set(7, true, 160, 100, 64 + 16, VCOL_MED_GREY, true, false, false); spr_set(7, true, 160, 100, 64 + 16, VCOL_MED_GREY, true, false, false);
@ -500,7 +507,7 @@ int main(void)
{ {
edelay--; edelay--;
if (edelay < 5) if (edelay < 5)
enemies_spwan(); enemies_spawn();
} }
else else
{ {