fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[x64] Fix AVX2 optimization path for VECTOR_SHL_V128

Browse Source 
Ensure the optimized path matches the fallback behavior
This commit is contained in:
Herman S.
2025-10-25 17:47:18 +09:00
parent b3a131698c
commit 77e320f79a
1 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
src/xenia/cpu/backend/x64/x64_seq_vector.cc
View File

@@ -947,9 +947,9 @@ struct VECTOR_SHL_V128
e.vpshufb(e.xmm2, e.xmm2, e.GetXmmConstPtr(XMMIntsToBytes)); e.vpshufb(e.xmm2, e.xmm2, e.GetXmmConstPtr(XMMIntsToBytes));
e.vpshufb(e.xmm3, e.xmm3, e.GetXmmConstPtr(XMMIntsToBytes)); e.vpshufb(e.xmm3, e.xmm3, e.GetXmmConstPtr(XMMIntsToBytes));
e.vpunpckldq(e.xmm0, e.xmm0, e.xmm1); e.vpunpckldq(e.xmm0, e.xmm0, e.xmm2);
e.vpunpckldq(e.xmm2, e.xmm2, e.xmm3); e.vpunpckldq(e.xmm1, e.xmm1, e.xmm3);
e.vpunpcklqdq(i.dest, e.xmm0, e.xmm2); e.vpunpcklqdq(i.dest, e.xmm0, e.xmm1);
return; return;
} else { } else {
vec128_t constmask = i.src2.constant(); vec128_t constmask = i.src2.constant();
@@ -1001,9 +1001,9 @@ struct VECTOR_SHL_V128
e.vpshufb(e.xmm2, e.xmm2, e.GetXmmConstPtr(XMMIntsToBytes)); e.vpshufb(e.xmm2, e.xmm2, e.GetXmmConstPtr(XMMIntsToBytes));
e.vpshufb(e.xmm3, e.xmm3, e.GetXmmConstPtr(XMMIntsToBytes)); e.vpshufb(e.xmm3, e.xmm3, e.GetXmmConstPtr(XMMIntsToBytes));
e.vpunpckldq(e.xmm0, e.xmm0, e.xmm1); e.vpunpckldq(e.xmm0, e.xmm0, e.xmm2);
e.vpunpckldq(e.xmm2, e.xmm2, e.xmm3); e.vpunpckldq(e.xmm1, e.xmm1, e.xmm3);
e.vpunpcklqdq(i.dest, e.xmm0, e.xmm2); e.vpunpcklqdq(i.dest, e.xmm0, e.xmm1);
return; return;
} }