Bug in reload_cse_move2add()

Richard Henderson rth@redhat.com
Mon Jan 15 22:56:00 GMT 2001


On Mon, Jan 15, 2001 at 05:32:29PM -0200, Alexandre Oliva wrote:
> Here's a patch that implements this change.  It fixes the bug that
> showed up in ARM bootstraps (Richard, could you please give it a spin
> on your arm-netbsd, just to make sure?)...

Did you distill a test case from the ARM bootstrap failure?
If not (or maybe even if), here is a test that fails on i686
with the code as-is.

Your current patch does fix this, but we can't install that
if it's still killing ARM.  We do need to come to resolution
on this soon, though.



r~



extern void abort(void);
static void foo(int x) { }

const int data[8] = { -1, -2, -3, -4, -5, -6, -7, -8 };

int
main (void)
{
  int s, i;

  for (s = 0; s < 3; ++s)
    foo (s);

  for (s = 1; s < 3; ++s)
    i = data[s];

  if (i != -3)
    abort ();
}


More information about the Gcc-patches mailing list