ShuriZma
/
dolphin
mirror of https://github.com/dolphin-emu/dolphin.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

JitArm64: Optimize rlwinmx. 

The new code adds fast paths for most usages which fits in one
instruction with one cycle latency.
This commit is contained in:
Markus Wick 2017-08-11 23:32:36 +02:00
parent 0a8e1bd864
commit 5ee7f86199
1 changed files with 22 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
Source/Core/Core/PowerPC/JitArm64/JitArm64_Integer.cpp
View File

@ -516,11 +516,28 @@ void JitArm64::rlwinmx(UGeckoInstruction inst)
gpr.BindToRegister(a, a == s);
ARM64Reg WA = gpr.GetReg();
ArithOption Shift(gpr.R(s), ST_ROR, 32 - inst.SH);
MOVI2R(WA, mask);
AND(gpr.R(a), WA, gpr.R(s), Shift);
gpr.Unlock(WA);
if (!inst.SH)
{
// Immediate mask
ANDI2R(gpr.R(a), gpr.R(s), mask);
}
else if (inst.ME == 31 && 31 < inst.SH + inst.MB)
{
// Bit select of the upper part
UBFX(gpr.R(a), gpr.R(s), 32 - inst.SH, 32 - inst.MB);
}
else if (inst.ME == 31 - inst.SH && 32 > inst.SH + inst.MB)
{
// Bit select of the lower part
UBFIZ(gpr.R(a), gpr.R(s), inst.SH, 32 - inst.SH - inst.MB);
}
else
{
ARM64Reg WA = gpr.GetReg();
MOVI2R(WA, mask);
AND(gpr.R(a), WA, gpr.R(s), ArithOption(gpr.R(s), ST_ROR, 32 - inst.SH));
gpr.Unlock(WA);
}
if (inst.Rc)
ComputeRC(gpr.R(a), 0);