[Bug tree-optimization/70527] Missed fold for "(long int) x * 12 - (long int)(x + 1) * 12"

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70527

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2016-04-04
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Note the issue is (non-)canonicalization of (long)(x + 1).  For optimal
re-assoc you'd want to canonicalize this to (long)x + 1L (with x + 1
having undefined behavior on overflow) and then to (long)x * 12 + 12.

But generally you do _not_ want to do this because you lose value-range info
and carry out the op in a larger mode (the former means the reverse transform
is not valid).

Generic folding has very limited capabilities here and GIMPLE reassoc currently
refuses to associate signed ops at all.