[Bug tree-optimization/46038] New: Vectorizer generates misaligned address for vld1 qn, [rn:alignment]

[belagod at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46038

           Summary: Vectorizer generates misaligned address for vld1 qn,
                    [rn:alignment]
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassig...@gcc.gnu.org
        ReportedBy: bela...@gcc.gnu.org
                CC: ju...@gcc.gnu.org


This test case:

extern void abort (void);

void foo (short *__restrict in, int n, int l,
       short *__restrict o) __attribute__((noinline));

void foo (short *__restrict in, int n, int l , 
      short *__restrict o)
{
  int A=0, lg, i;
   for (lg = 0; lg < n; lg++) 
    {
      A = 0;
      for (i = 0; i < l; i++)
         A += (int)in[i] +  (int)in[i+lg];
    }

   if (A != 256)
     abort ();
}

int main()
{
  short in[32];
  short o[4];
  int i;

  for (i=0;i<32;i++) in[i] = i;
  foo (in, 2, 16, o);
  return 0;
}

aborts.
It's compiled with -c -static -mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp
foo.c -marm -O3.

The first iteration's result 'A' is correct. The second iteration's result is
broken.

The bug seems to be during the second iteration of the outer loop in
foo(), the vld1.16 at 0x8514 seems to be getting a bogus base address
that isn't 8-byte aligned inspite of the assertion [rn:64]. This might be a
potential source of the bug.

The output of dmesg on a cortex-a8 is:

Alignment trap: not handling instruction edc30b00 at [<0000c64c>]
Unhandled fault: alignment exception (0x801) at 0xbe8818c9