[Bug tree-optimization/57359] store motion causes wrong code for union access at -O3

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57359

--- Comment #28 from Richard Biener <rguenth at gcc dot gnu.org> ---
OK, so more "advanced" testcases are a bit difficult because the
ref_always_accessed_p logic is too simple and it's required for
store-motion of accesses in conditional paths.  Basically if
we have if (test) { *p = 1; } else { *p = 0; } LIM doesn't figure
*p is accessed unconditionally (that's a missed optimization).

Anyway, here's a testcase circumventing that by using static storage
and showing that conditionally executed SM (with and without
-fallow-store-data-races) fail.  It's supposed to be a testcase for a "merge"
point of two
different sequenced store sequences we need to check.

typedef int A;
typedef float B;

float x[256];

void __attribute__((noinline,noclone))
foo(long unk, long ack)
{
  B *q = x;
  for (long i = 0; i < unk; ++i)
    {
      if (ack & i)
        {
          *((A*)q + 4) = 1;
          q[i] = 42;
        }
      else
        {
          q[i] = 42;
          *((A*)q + 4) = 1;
        }
    }
}

int main(void)
{
  foo(5, 6);
  if (x[4] != 42) __builtin_abort();
  return 0;
}