[Bug middle-end/71762] [4.9 Regression] ifcombine wrong codegen with uninitialized data

[ibuclaw at gdcproject dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71762

--- Comment #1 from Iain Buclaw <ibuclaw at gdcproject dot org> ---
Backporting r217638 to gcc-4.9 fixes the problem too.

Allowing propagation of __cond60 and __cond59 comparisons into comparisons of
len elides the wrong optimization.

---
__cond60_9 = len_8(D) > 1;
if (__cond60_9 != 0)
  goto <bb 6>;
else
  goto <bb 3>;

<bb 3>:
__cond59_13 = len_8(D) == 1;
if (__cond59_13 != 0)
  goto <bb 4>;
else
  goto <bb 5>;

/* ... */

<bb 9>:
if (__cond60_9 != 0)
  goto <bb 12>;
else
  goto <bb 10>;
---

Becomes:

---
if (len_8(D) > 1)
  goto <bb 6>;
else
  goto <bb 3>;

<bb 3>:
if (len_8(D) == 1)
  goto <bb 4>;
else
  goto <bb 5>;

/* ... */

<bb 9>:
if (len_8(D) <= 1)
  goto <bb 10>;
else
  goto <bb 12>;
---