[Bug target/82447] [IVOPTS] Consider removing cmp instruction while iterating on an array of known bound

["amker at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82447

--- Comment #6 from amker at gcc dot gnu.org ---
(In reply to Jakub Jelinek from comment #5)
> (In reply to amker from comment #4)
> > (In reply to Jakub Jelinek from comment #2)
> > > Indeed.  But when the body is:
> > >         sum += a[size - 1 - i];
> > > while base 128U with step -1U is considered, it isn't found as the cheapest
> > > for some reason either.
> > 
> > On 64-bit targets, candidate [128, -1] is added in unsigned int type, while
> > address IV has pointer type.  IVOPTs simply skips the candidate because it
> > has smaller precision than the IV use.
> 
> Yeah, but can't we also add [128UL, -1UL] candidate if we see there are IVs
> of that bitsize?  Conversion from that to the unsigned int IV is cheap.

Yes we can.  For constant niters loop, currently it's ivcanon adding the
counting down IV, I mentioned during cauldron that we may want to get rid of
iv_canonicalization by adding all necessary IV candidates in IVOTPs.