xenia-project/xenia
1
0
Fork 0
mirror of https://github.com/xenia-project/xenia.git synced 2026-03-15 17:55:01 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

[x64] Add AVX512 optimizations for PERMUTE_V128

Browse source
This commit is contained in:
Max-Tepafray 2023-05-25 17:27:22 -05:00 committed by GitHub
parent 8ba102932d
commit e10756d150
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 0 additions and 28 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

28
src/xenia/cpu/backend/x64/x64_seq_vector.cc
View file

@ -679,19 +679,6 @@ struct VECTOR_SUB
// src1/src2.
e.vpsubd(e.xmm1, src1, src2);
if (e.IsFeatureEnabled(kX64EmitAVX512Ortho)) {
// If the result is less or equal to the first operand then
// we did not underflow
Opmask not_underflow = e.k1;
// _mm_cmple_epu32_mask
e.vpcmpud(not_underflow, e.xmm1, src1, 0x2);
// Copy over values that did not underflow, write zero
// everywhere else
e.vmovdqa32(dest | not_underflow | e.T_z, e.xmm1);
return;
}
// If result is greater than either of the inputs, we've
// underflowed (only need to check one input)
// if (res > src1) then underflowed
@ -703,21 +690,6 @@ struct VECTOR_SUB
} else {
e.vpsubd(e.xmm1, src1, src2);
if (e.IsFeatureEnabled(kX64EmitAVX512Ortho |
kX64EmitAVX512DQ)) {
e.vmovdqa32(e.xmm3, src1);
e.vpternlogd(e.xmm3, e.xmm1, src2, 0b00011000);
const Opmask saturate = e.k1;
e.vpmovd2m(saturate, e.xmm3);
e.vpsrad(e.xmm2, e.xmm1, 31);
e.vpxord(e.xmm2, e.xmm2, e.GetXmmConstPtr(XMMSignMaskI32));
e.vpblendmd(dest | saturate, e.xmm1, e.xmm2);
return;
}
// We can only overflow if the signs of the operands are
// opposite. If signs are opposite and result sign isn't the
// same as src1's sign, we've overflowed. if ((s32b)((src1 ^