Impaczus/rpcs3
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

simd_builder: fixups

Browse Source 
Fix resetting vmask in reduce() step.
Fix AVX-512 loads in vec_load_unaligned().
Fix bzhi reg size in build_look().
This commit is contained in:
Nekotekina
2022-09-08 17:06:09 +03:00
committed by Ivan
parent 5d91caebe9
commit a9437d69ab
2 changed files with 21 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Utilities/JIT.h
+12 -3
View File
@@ -323,11 +323,20 @@ namespace asmjit
// Build single last iteration (masked)
this->test(reg_cnt, reg_cnt);
this->jz(exit);
this->bzhi(reg_cnt, x86::Mem(consts[~u128()], 0), reg_cnt);
this->kmovq(x86::k7, reg_cnt);
if (esize == 1 && vsize == 64)
{
this->bzhi(reg_cnt.r64(), x86::Mem(consts[~u128()], 0), reg_cnt.r64());
this->kmovq(x86::k7, reg_cnt.r64());
}
else
{
this->bzhi(reg_cnt.r32(), x86::Mem(consts[~u128()], 0), reg_cnt.r32());
this->kmovd(x86::k7, reg_cnt.r32());
}
vmask = 7;
build();
vmask = -1;
// Rollout reduction step
this->bind(exit);