[Bug tree-optimization/93586] [10 Regression] wrong code at -O1 on x86_64-linux-gnu

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93586

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
GIMPLE testcase for the DSE issue not relying on uninitialized reads.  We
DSE the = {} zeroing.

int __GIMPLE(ssa) foo(int j)
{
  int c[1][10][1];
  int _1;

__BB(2):
  c[0][1][0] = 1;
  c[0][1] = _Literal (int[1]) {};
  _1 = c[0][j_2(D)][0];
  return _1;
}

int main()
{
  if (foo (1) != 0)
    __builtin_abort ();
  return 0;
}