mirror/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-01-30 21:32:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

SPU LLVM: Fix edgecase in icelake codegen

Browse Source
This commit is contained in:
Malcolm Jestadt 2020-12-29 03:15:28 -05:00 committed by Ivan
parent 35322b5d14
commit c952e99f3e
1 changed files with 1 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
rpcs3/Emu/Cell/SPURecompiler.cpp
View File

@ -7299,15 +7299,6 @@ public:
const auto as = byteswap(a); const auto as = byteswap(a);
const auto bs = byteswap(b); const auto bs = byteswap(b);
if (m_use_avx512_icl && (op.ra != op.rb || m_interp_magn))
{
const auto m = gf2p8affineqb(build<u8[16]>(0x02, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x02, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04), c, 0x7f);
const auto mm = select(noncast<s8[16]>(m) >= 0, splat<u8[16]>(0), m);
const auto ab = vperm2b(as, bs, c);
set_vr(op.rt4, select(noncast<s8[16]>(c) >= 0, ab, mm));
return;
}
const auto x = avg(noncast<u8[16]>(sext<s8[16]>((c & 0xc0) == 0xc0)), noncast<u8[16]>(sext<s8[16]>((c & 0xe0) == 0xc0))); const auto x = avg(noncast<u8[16]>(sext<s8[16]>((c & 0xc0) == 0xc0)), noncast<u8[16]>(sext<s8[16]>((c & 0xe0) == 0xc0)));
const auto ax = pshufb(as, c); const auto ax = pshufb(as, c);
const auto bx = pshufb(bs, c); const auto bx = pshufb(bs, c);
@ -7350,7 +7341,7 @@ public:
if (m_use_avx512_icl && (op.ra != op.rb || m_interp_magn)) if (m_use_avx512_icl && (op.ra != op.rb || m_interp_magn))
{ {
const auto m = gf2p8affineqb(build<u8[16]>(0x02, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x02, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04), c, 0x7f); const auto m = gf2p8affineqb(build<u8[16]>(0x02, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x02, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04, 0x04), c, 0x7f);
const auto mm = select(noncast<s8[16]>(m) >= 0, splat<u8[16]>(0), m); const auto mm = select(noncast<s8[16]>(c << 1) >= 0, splat<u8[16]>(0), m);
const auto cr = eval(~c); const auto cr = eval(~c);
const auto ab = vperm2b(b, a, cr); const auto ab = vperm2b(b, a, cr);
set_vr(op.rt4, select(noncast<s8[16]>(cr) >= 0, mm, ab)); set_vr(op.rt4, select(noncast<s8[16]>(cr) >= 0, mm, ab));