[Bug middle-end/30364] [4.1/4.2/4.3 Regression] Wrong variable ranges due to constant folding

[guillaume dot melquiond at ens-lyon dot fr]

------- Comment #1 from guillaume dot melquiond at ens-lyon dot fr  2007-01-04 10:15 -------
Is it really a middle-end issue?

It could also be seen as a front-end issue, as it does produce "a + b - 40",
doesn't it? If the front-end had given "(a - 20) + (b - 20)" to the middle-end,
then the correct ranges would have been computed. When adjusting the testcase
in such a way, GCC does generate correct code:

int f(int a, int b)
{
  if (a > 0x7FFFFFF0) return 0;
  if (b > 0x7FFFFFF0) return 0;

  int d = a - 20;
  int e = b - 20;
  int c = d + e;
  return c > 0x7FFFFFF0;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30364