[Bug tree-optimization/62173] [5.0 regression] 64bit Arch can't ivopt while 32bit Arch can

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62173

--- Comment #20 from amker at gcc dot gnu.org ---
(In reply to Richard Biener from comment #18)
> It's probably not correct to simply transfer range info from *idx to
> iv->base.
> Instead SCEV analysis needs to track the range of CHREC_LEFT when it analyzes
> the SSA use-def chain.  That's of course a much bigger change :/
> 
> The patch may still help in some cases - I suppose the original testcase is
> reduced from sth else?

I see it's a tricky problem, and I have to admit that I don't understand it
very well yet.  The question is, is relax of POINTER_PLUS_EXPR constraint the
right way fixing this?  I do remember some other PRs (other than this one or
bug 52563) are caused by this constraint according to your comments.

Of course, take range information into consideration is always an improvement. 
Actually I have another possible example in iv elimination.  Curently GCC
simply rejects elimination of an iv use with a candidate if the cand is of
smaller type, but as long as we can prove value of iv use can be represented by
the smaller candidate, elimination is actually safe.  But seems to me, fixing
this issue with value information is like a side-effect?