[Bug rtl-optimization/70460] [4.9/5/6 Regression] Miscompilation of glibc on i686-linux starting with r234101

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70460

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Perhaps we can try to delegitimize the SET_SRC of def_insn or something similar
to handle at least some -fPIC case where we'd want to turn indirect jumps into
direct ones.
Alan, do you have particular testcases (in the testsuite or in mind) that you
wanted to handle with the new mini-pass?
Tried
int c;

void
foo (void)
{
  int x = 0;
  static int b[] = { &&lab1 - &&lab0, &&lab2 - &&lab0 };
  void *a = &&lab0 + b[x];
  goto *a;
lab1:
  c += 2;
lab2:
  c++;
lab0:
  ;
}

int
main ()
{
  foo ();
  if (c != 3)
    __builtin_abort ();
  foo ();
  if (c != 6)
    __builtin_abort ();
  return 0;
}
(very much surprised nothing at GIMPLE level optimizes this), but that didn't
used to be optimized at least with -fpic.