https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92308
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |missed-optimization
Target| |arm
CC| |rguenth at gcc dot gnu.org
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Note on GIMPLE canonicalization is quite important so that would attempt to
undo any such "CSE of constants". That's also in the way of appplying auto-inc
style addressing here.
There's the SLSR (GIMPLE straight-line strength reduction) pass that is
probably
the best place to perform this optimization. But there's both CSE,
forwprop and copyprop (doing CCP) that would undo this.
So I'm not convinced GIMPLE is the correct place to fix this.
