[Bug tree-optimization/23471] New: a*a (for signed ints with -fno-wrapv) is always postive

[pinskia at gcc dot gnu dot org]

Hmm, this is both a VRP bug and a folding bug.  Even though a is VARYING, we 
know the that a*a will 
always be postive (this with -fno-wrapv which is default for C, C++ and 
fortran).
void link_error(void);
void f(int a)
{
  int b = a;
  b*=a;
  if (b < 0)
   link_error();
}

---
There should be no references to link_error().

There is another way of fixing this via the fold:
void link_error(void);
void f(int a)
{
  if (a*a < 0)
   link_error();
}

But that will miss:
void link_error(void);
void f(int a)
{
  if (a*a*a*a*a*a < 0)
   link_error();
}

---
Yes that is most likely to overflow but that is the whole point of -fno-wrapv.

Also ICC does not do this optimization at all.

-- 
           Summary: a*a (for signed ints with -fno-wrapv) is always postive
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P2
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23471