[Bug tree-optimization/52396] New: Gcc failed to hoist loop invariant expression out of loop

[carrot at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52396

             Bug #: 52396
           Summary: Gcc failed to hoist loop invariant expression out of
                    loop
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassig...@gcc.gnu.org
        ReportedBy: car...@google.com
            Target: arm-linux-gnueabi


Compile the following source code with options -march=armv7-a -mthumb -Os -fpic

extern int array[];

int t0l(int len)
{
  int t = 0;
  int i;
  for (i=0; i<len; i++)
    t = t + array[i];
  return t;
}

ARM gcc 4.7 generates:


t0l:
    ldr    r1, .L4
    movs    r3, #0
    push    {r4, lr}
.LPIC8:
    add    r1, pc
    mov    r2, r3
    b    .L2
.L3:
    ldr    r4, .L4+4                // A
    ldr    r4, [r1, r4]             // B
    ldr    r4, [r4, r3, lsl #2]
    adds    r3, r3, #1
    adds    r2, r2, r4
.L2:
    cmp    r3, r0
    blt    .L3
    mov    r0, r2
    pop    {r4, pc}
.L5:
    .align    2
.L4:
    .word    _GLOBAL_OFFSET_TABLE_-(.LPIC8+4)
    .word    array(GOT)

Instructions AB are loop invariant, could be taken out of the loop without
increase code size.

If I change -Os to -O2, ivopts can successfully do this optimization.