https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123801

--- Comment #4 from Siddhesh Poyarekar <siddhesh at gcc dot gnu.org> ---
> But it is not the vectorizer that is causing the issue really.

eep, sorry, fag end of the day, bleary eyes :/

> Which is also correct due to &probe_144795_l.0_1->D.4608.D.4603.__tmp10 and 
> &probe_144795_l.0_1->D.4608.D.4607.__tmp16 being the same value except guess 
> what waccess needs the __tmp16  variant.

That looks problematic though, not just because waccess sees the wrong thing,
but also because it semantically changes the bounds.  I reckon __bos doesn't
stumble on it because we have the early_objsz hack to work around a completely
different (but similar) problem of accesses being flattened to the *larger*
containing struct.

Reply via email to