[Bug target/92462] [arm32] -ftree-pre makes a variable to be wrongly hoisted out

[wilco at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92462

--- Comment #23 from Wilco <wilco at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #22)
> Fixed on trunk.  Can arm people verify?  I checked the DSE dump only.  Bonus
> if you manage to create a testcase for the testsuite failing before, passing
> now.
> 
> The patch is simple enough to backport if it works.

I will have a look. But only checking the low bit is still way too dangerous.
Aliasing checks should be conservative and 100% accurate, not use random
heuristics which hope for the best. So if we want to keep the AND code then it
needs to look for a mask with all top bits set as the absolute minimum.