[Bug tree-optimization/52445] [4.6/4.7 Regression] conditional store replacement causes segfault in generated code

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52445

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-03-01 
09:01:59 UTC ---
Not to mention that in this exact case, even if was always non-trapping, I
doubt it will ever be an optimization to "optimize"
  if (len_1(D) > 1)
    goto <bb 5>;
  else
    goto <bb 6>;

<bb 5>:
  MEM[(char *)buf_2(D) + 1B] = 0;

<bb 6>:
  return;
into:
  if (len_1(D) > 1)
    goto <bb 6>;
  else
    goto <bb 5>;

<bb 5>:
  cstore.2_7 = MEM[(char *)buf_2(D) + 1B];

<bb 6>:
  # cstore.2_9 = PHI <cstore.2_7(5), 0(4)>
  MEM[(char *)buf_2(D) + 1B] = cstore.2_9;
  return;

because the latter we then expand into:
        jbe     .L8
        movb    %al, 1(%rdi)
        ret
...
.L8:
        movzbl  1(%rdi), %eax
        movb    %al, 1(%rdi)
        ret

So if the conditional bb contains just the potentionally cselim optimized
store, perhaps we should punt.

Plus for C++11/C11 memory model we probably need to disable cselim optimization
altogether.