[Bug tree-optimization/80275] [6/7 Regression] Poor (but valid) code generated by optimizer passing optimizer list to function

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80275

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2
             Status|UNCONFIRMED                 |ASSIGNED
      Known to work|                            |5.4.1
           Keywords|                            |missed-optimization
   Last reconfirmed|                            |2017-04-03
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
     Ever confirmed|0                           |1
            Summary|Poor (but valid) code       |[6/7 Regression] Poor (but
                   |generated by optimizer      |valid) code generated by
                   |passing optimizer list to   |optimizer passing optimizer
                   |function                    |list to function
   Target Milestone|---                         |6.4

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  In GCC 5 cunrolli is able to compute number of iterations of the
loop(s) and unrolls them so we can optimize away the stuff.

niter issue.  trunk gets

Analyzing # of iterations of loop 1
  exit condition [(const int *) &D.15512 + 4, + , 4](no_overflow) != &MEM[(void
*)&D.15512 + 12B]
  bounds on difference of bases: -18446744073709551615 ... 18446744073709551611
Applying pattern match.pd:148, generic-match.c:7290
  result:
    # of iterations ((unsigned long) &MEM[(void *)&D.15512 + 12B] - (unsigned
long) ((const int *) &D.15512 + 4)) /[ex] 4 & 4611686018427387903, bounded by
4611686018427387902
...
Loop 1 iterates at most -2 times.
Loop 1 likely iterates at most -2 times.

IL is quite similar.  So we fail to simplify &MEM[&D.15512 + 12B] - (unsigned
long) ((const int *) &D.15512 + 4)) for a start.  I would guess at the
associate
case in fold-const.