[Bug tree-optimization/55079] [4.8 regression] false positive -Warray-bounds (also seen at -O3 bootstrap)

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55079

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[4.8 regression] false      |[4.8 regression] false
                   |positive -Warray-bounds     |positive -Warray-bounds
                   |                            |(also seen at -O3
                   |                            |bootstrap)

--- Comment #8 from Jan Hubicka <hubicka at gcc dot gnu.org> 2012-11-14 19:31:16 UTC ---
Hi,
at -O3 bootstrap we have false positive on simplify-rtx.c
../../gcc/simplify-rtx.c: In function ‘rtx_def* simplify_plus_minus(rtx_code,
machine_mode, rtx, rtx)’:
../../gcc/simplify-rtx.c:4285:63: warning: array subscript is below array
bounds [-Warray-bounds]
           while (j-- && simplify_plus_minus_op_data_cmp (ops[j].op, save.op));


the siplified testcase is as follows:

int a[8];

void
test(unsigned int n)
{
  unsigned int i;
  unsigned int j;
  if (n<8)
  for (j=0;j<n;j++)
   {
     i = j;
     do
        a[i+1]=a[i];
     while (i--);
   }
}

here we unroll the inner do loop 7 times based on the array bounds derived from
[i] access.  
Afterwards VRP proves that j is always at most 7 and thus the loop walks out of
bounds. Here we may be able to determine that the loop accesses both i and i+1
and we could actually unroll only 6 times, but we handle each of array access
independently.