https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127467
Bug ID: 127467
Summary: [17 Regression] avoid_avx512_perm not dropping back to
256 bit perm
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: haochen.jiang at intel dot com
Target Milestone: ---
Created attachment 65625
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=65625&action=edit
Derived testcase
With testcase, under -O3 -march=x86-64-v4 -mprefer-vector-width=512
-mtune-ctrl=avx512_avoid_perm
It somehow missed vectorization for part of the attached code. It should
generate the same (or similar) code as -O3 -march=x86-64-v4
-mprefer-vector-width=256 did.
Reproduced in godbolt:
https://godbolt.org/z/cExjY4v5d
