[Bug middle-end/55555] [4.8 Regression] miscompilation at -O2 (number_of_iterations)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55555



--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> 2012-12-14 
14:11:59 UTC ---

The unrolling puts __builtin_unreachable ()s into the inner duplicated loops:



First iteration, good:



  <bb 16>:

  # lxp_30 = PHI <0(2)>

  t_32 = pol_x[lxp_30];

  _33 = (long int) lxp_30;

  _34 = _33 * 4;

  l_35 = _34 + -1;



  <bb 17>:

  # S_36 = PHI <1(16), S_45(18)>

  if (S_36 <= 4)

    goto <bb 18>;

  else

    goto <bb 19>;



  <bb 18>:

  _38 = S_36 + l_35;

  _39 = coef_x[_38];

  _40 = S_36 + -1;

  _41 = s[_40];

  _42 = _41 * t_32;

  _43 = _39 + _42;

  coef_x[_38] = _43;

  S_45 = S_36 + 1;

  goto <bb 17>;



  <bb 19>:

  lxp_46 = lxp_30 + 1;



  <bb 20>:



Peeled iteration, bad:



  <bb 3>:

  # lxp_1 = PHI <lxp_46(20)>

  t_9 = pol_x[lxp_1];

  _10 = (long int) lxp_1;

  _11 = _10 * 4;

  l_12 = _11 + -1;

  goto <bb 5>;



  <bb 4>:

  _13 = S_3 + l_12;

  __builtin_unreachable ();

  _14 = coef_x[_13];

  _15 = S_3 + -1;

  _16 = s[_15];

  _17 = _16 * t_9;

  _18 = _14 + _17;

  __builtin_unreachable ();

  coef_x[_13] = _18;

  S_20 = S_3 + 1;



  <bb 5>:

  # S_3 = PHI <1(3), S_20(4)>

  if (S_3 <= 4)

    goto <bb 4>;

  else

    goto <bb 6>;



  <bb 6>:

  lxp_21 = lxp_1 + 1;

  if (1 == 0)

    goto <bb 15>;

  else

    goto <bb 7>;



  <bb 15>:



  <bb 22>:

  __builtin_unreachable ();



  <bb 7>:

  __asm__ __volatile__("" :  : "r" &coef_x : "memory");

  goto <bb 13>;



Note that the outer loop body looks ok - it is the inner body that get's

mangled in a bogus way.



That's because the loop bounds derived from the inner loop accesses

are bogus:



remove_exits_and_undefined_stmts (loop=0x7ffff67ec770, npeeled=1)

    at /space/rguenther/src/svn/trunk/gcc/tree-ssa-loop-ivcanon.c:481

481       bool changed = false;

(gdb) n

483       for (elt = loop->bounds; elt; elt = elt->next)

(gdb) 

488           if (!elt->is_exit

(gdb) p elt 

$8 = (nb_iter_bound *) 0x7ffff6925e38

(gdb) p *elt

$9 = {stmt = 0x7ffff6905c80, bound = {low = 0, high = 0}, is_exit = false, 

  next = 0x7ffff6925dc0}

(gdb) p elt->stmt

$10 = (gimple) 0x7ffff6905c80

(gdb) call debug_gimple_stmt ($10)

# .MEM_19 = VDEF <.MEM_6>

coef_x[_13] = _18;



as far as I can see even with lxp == 1 we have at most an index of 4 + 1*4 - 1.



Statement _14 = coef_x[_13];

 is executed at most 0 (bounded by 0) + 1 times in loop 1.



_that's_ bogus.