Merge pull request #12764 from Sintendo/jitarm64-temp-regs

JitArm64: Skip temp regs where possible
This commit is contained in:
Admiral H. Curtiss 2024-05-21 22:06:21 +02:00 committed by GitHub
commit abc8aa2237
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 33 additions and 21 deletions

View File

@ -1392,13 +1392,17 @@ void JitArm64::subfic(UGeckoInstruction inst)
}
else
{
gpr.BindToRegister(d, d == a);
const bool allocate_reg = d == a;
gpr.BindToRegister(d, allocate_reg);
// d = imm - a
ARM64Reg WA = gpr.GetReg();
ARM64Reg RD = gpr.R(d);
ARM64Reg WA = allocate_reg ? gpr.GetReg() : RD;
MOVI2R(WA, imm);
CARRY_IF_NEEDED(SUB, SUBS, gpr.R(d), WA, gpr.R(a));
gpr.Unlock(WA);
CARRY_IF_NEEDED(SUB, SUBS, RD, WA, gpr.R(a));
if (allocate_reg)
gpr.Unlock(WA);
ComputeCarry();
}
@ -1431,10 +1435,9 @@ void JitArm64::addex(UGeckoInstruction inst)
}
case CarryFlag::InHostCarry:
{
ARM64Reg WA = gpr.GetReg();
MOVI2R(WA, i + j);
ADC(gpr.R(d), WA, ARM64Reg::WZR);
gpr.Unlock(WA);
ARM64Reg RD = gpr.R(d);
MOVI2R(RD, i + j);
ADC(RD, RD, ARM64Reg::WZR);
break;
}
case CarryFlag::ConstantTrue:
@ -1672,7 +1675,8 @@ void JitArm64::divwx(UGeckoInstruction inst)
{
const s32 divisor = s32(gpr.GetImm(b));
gpr.BindToRegister(d, d == a);
const bool allocate_reg = a == d;
gpr.BindToRegister(d, allocate_reg);
// Handle 0, 1, and -1 explicitly
if (divisor == 0)
@ -1709,7 +1713,6 @@ void JitArm64::divwx(UGeckoInstruction inst)
ARM64Reg RA = gpr.R(a);
ARM64Reg RD = gpr.R(d);
const bool allocate_reg = a == d;
ARM64Reg WA = allocate_reg ? gpr.GetReg() : RD;
TST(RA, RA);
@ -1729,13 +1732,13 @@ void JitArm64::divwx(UGeckoInstruction inst)
// Optimize signed 32-bit integer division by a constant
SignedMagic m = SignedDivisionConstants(divisor);
ARM64Reg WA = gpr.GetReg();
ARM64Reg WB = gpr.GetReg();
ARM64Reg RD = gpr.R(d);
ARM64Reg WA = gpr.GetReg();
ARM64Reg WB = allocate_reg ? gpr.GetReg() : RD;
ARM64Reg XD = EncodeRegTo64(RD);
ARM64Reg XA = EncodeRegTo64(WA);
ARM64Reg XB = EncodeRegTo64(WB);
ARM64Reg XD = EncodeRegTo64(RD);
SXTW(XA, gpr.R(a));
MOVI2R(XB, s64(m.multiplier));
@ -1768,7 +1771,9 @@ void JitArm64::divwx(UGeckoInstruction inst)
ADD(RD, WA, RD);
}
gpr.Unlock(WA, WB);
gpr.Unlock(WA);
if (allocate_reg)
gpr.Unlock(WB);
}
if (inst.Rc)
@ -2004,9 +2009,11 @@ void JitArm64::srawx(UGeckoInstruction inst)
}
else
{
gpr.BindToRegister(a, a == b || a == s);
const bool will_read = a == b || a == s;
gpr.BindToRegister(a, will_read);
ARM64Reg WA = gpr.GetReg();
const bool allocate_reg = will_read || js.op->wantsCA;
ARM64Reg WA = allocate_reg ? gpr.GetReg() : gpr.R(a);
LSL(EncodeRegTo64(WA), EncodeRegTo64(gpr.R(s)), 32);
ASRV(EncodeRegTo64(WA), EncodeRegTo64(WA), EncodeRegTo64(gpr.R(b)));
@ -2019,7 +2026,8 @@ void JitArm64::srawx(UGeckoInstruction inst)
ComputeCarry(WA);
}
gpr.Unlock(WA);
if (allocate_reg)
gpr.Unlock(WA);
}
if (inst.Rc)
@ -2098,15 +2106,19 @@ void JitArm64::rlwimix(UGeckoInstruction inst)
else
{
gpr.BindToRegister(a, true);
const bool allocate_reg = a == s;
ARM64Reg RA = gpr.R(a);
ARM64Reg WA = gpr.GetReg();
ARM64Reg WB = gpr.GetReg();
ARM64Reg WB = allocate_reg ? gpr.GetReg() : RA;
MOVI2R(WA, mask);
BIC(WB, gpr.R(a), WA);
BIC(WB, RA, WA);
AND(WA, WA, gpr.R(s), ArithOption(gpr.R(s), ShiftType::ROR, rot_dist));
ORR(gpr.R(a), WB, WA);
ORR(RA, WB, WA);
gpr.Unlock(WA, WB);
gpr.Unlock(WA);
if (allocate_reg)
gpr.Unlock(WB);
}
if (inst.Rc)