mirrors/dynarmic
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

emit_x64: pmaxuw and pminuw require SSE 4.1

Browse source 
This commit is intended to close citra-emu/citra#3137.

pmaxuw and pminuw were used to perform unsigned comparisons; we emulate
these using a signed comparison by offsetting the inputs by 0x8000 for
CPUs that do not support SSE 4.1.
This commit is contained in:
MerryMage 2017-11-19 23:14:56 +00:00
parent c1495ca5da
commit 08f638d447
1 changed files with 30 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

39
src/backend_x64/emit_x64.cpp
View file

@ -1554,11 +1554,22 @@ void EmitX64::EmitPackedAddU16(RegAlloc& reg_alloc, IR::Block& block, IR::Inst*
Xbyak::Reg32 reg_ge = reg_alloc.ScratchGpr().cvt32(); Xbyak::Reg32 reg_ge = reg_alloc.ScratchGpr().cvt32();
Xbyak::Xmm tmp = reg_alloc.ScratchXmm(); Xbyak::Xmm tmp = reg_alloc.ScratchXmm();
code->movdqa(tmp, xmm_a); if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
code->pminuw(tmp, xmm_b); code->movdqa(tmp, xmm_a);
code->pcmpeqw(tmp, xmm_b); code->pminuw(tmp, xmm_b);
code->movd(reg_ge, tmp); code->pcmpeqw(tmp, xmm_b);
code->not_(reg_ge); code->movd(reg_ge, tmp);
code->not_(reg_ge);
} else {
// !(b <= a+b) == b > a+b
Xbyak::Xmm tmp_b = reg_alloc.ScratchXmm();
code->movdqa(tmp, xmm_a);
code->movdqa(tmp_b, xmm_b);
code->paddw(tmp, code->MConst(0x80008000));
code->paddw(tmp_b, code->MConst(0x80008000));
code->pcmpgtw(tmp_b, tmp); // *Signed* comparison!
code->movd(reg_ge, tmp_b);
}
ExtractMostSignificantBitFromPackedBytes(code, reg_alloc, reg_ge); ExtractMostSignificantBitFromPackedBytes(code, reg_alloc, reg_ge);
reg_alloc.DefineValue(ge_inst, reg_ge); reg_alloc.DefineValue(ge_inst, reg_ge);
@ -1672,10 +1683,20 @@ void EmitX64::EmitPackedSubU16(RegAlloc& reg_alloc, IR::Block& block, IR::Inst*
reg_ge = reg_alloc.ScratchGpr().cvt32(); reg_ge = reg_alloc.ScratchGpr().cvt32();
Xbyak::Xmm xmm_ge = reg_alloc.ScratchXmm(); Xbyak::Xmm xmm_ge = reg_alloc.ScratchXmm();
code->movdqa(xmm_ge, xmm_a); if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
code->pmaxuw(xmm_ge, xmm_b); code->movdqa(xmm_ge, xmm_a);
code->pcmpeqw(xmm_ge, xmm_a); code->pmaxuw(xmm_ge, xmm_b); // Requires SSE 4.1
code->movd(reg_ge, xmm_ge); code->pcmpeqw(xmm_ge, xmm_a);
code->movd(reg_ge, xmm_ge);
} else {
// (a >= b) == !(b > a)
code->paddw(xmm_a, code->MConst(0x80008000));
code->paddw(xmm_b, code->MConst(0x80008000));
code->movdqa(xmm_ge, xmm_b);
code->pcmpgtw(xmm_ge, xmm_a); // *Signed* comparison!
code->movd(reg_ge, xmm_ge);
code->not_(reg_ge);
}
} }
code->psubw(xmm_a, xmm_b); code->psubw(xmm_a, xmm_b);