[Bug middle-end/30322] New: fold folds -i-1+i+1 to ~i+i+1

[rguenth at gcc dot gnu dot org]

int bar(int i)
{
  return -i-1+i+1;
}

optimizes to

;; Function bar (bar)

Analyzing Edge Insertions.
bar (i)
{
<bb 2>:
  return i + 1 + ~i;

}

(instead of return 0; which we catch at rtl level later)

Either i+1 needs to be folded to -~i, we need to re-associate earlier
or something needs to do tree-combining and recognize that ~i and -(i+1)
is the same.

This happens in tramp3d as loop exit condition and shows up as

if (d.22083 != (int) ((unsigned int) d + 1 + (unsigned int) D.902833 +
(unsigned int) ~d)) goto <L3>; else goto <L6>;

instead of just

if (d.22083 != D.902833) goto <L3>; else goto <L6>;


[I think most of these "interesting" foldings should be deferred]


-- 
           Summary: fold folds -i-1+i+1 to ~i+i+1
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: missed-optimization, TREE
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rguenth at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30322