https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92889
Bug ID: 92889
Summary: GCC-8 considers the _mm_gf2p8affine_epi64_epi8
intrinsic to be symmetric
Product: gcc
Version: 8.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: drucker.nir at gmail dot com
Target Milestone: ---
Created attachment 47461
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=47461&action=edit
Source file for reproducing the bug
GCC version: gcc-8 (Ubuntu 8.3.0-6ubuntu1~18.04.1) 8.3.0
Platform: Dell XPS 13 7390 2-in-1 with the 10th Intel(C) Core^TM Generation
(Micro architecture Codename ``Ice Lake''[ICL]) Intel(c) Core^TM i7-1065G7 CPU
1.30GHz.
The source file gfni_bug.c is a stand-alone file for reproducing the bug.
The file prints two vectors that are not supposed to be equal.
To compile use:
Correct compilation: gcc-8 -O3 -march=native gfni_bug.c -DGOOD_TEST=1
Bad compilation: gcc-8 -O3 -march=native gfni_bug.c
The problem is with the new _mm_gf2p8affine_epi64_epi8 (x86_64 GF-NI)
intrinsic.
When calling
OUT = _mm_gf2p8affine_epi64_epi8(A,B,0)
gcc-8 translates it to assembler in two different ways:
1) vgf2p8affineqb $0x0, A, B, OUT //Good
2) vgf2p8affineqb $0x0, B, A, OUT //Bad
Where only one of these can be correct because the operation is not symmetric.
Thanks,
Nir Drucker
