https://gcc.gnu.org/bugzilla/show_bug.cgi?id=73457
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Severity|normal |enhancement
--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
GCC 8+ can handle this at -O3 and above because we can vectorize the loop
(which GCC 7 could not do, I have not looked into why) so DOM is able to
optimize it after unrolling.
GCC 10+ can handle this at -O3 -fno-tree-loop-vectorize (with SLP still
happening) because of r10-1420.
I do think we should be able to optimize this for -O2 though.
