Fix for optimizations when STACK_BIAS is true

[Jeffrey A Law]

  In message <199710021815.OAA03931@jenolan.rutgers.edu>you write:
  > 
  > This fixes the problems in constant offset optimizations in the
  > combiner when the target has a biased stack and frame pointer,
  > currently this is only sparc64.
  > 
  > Crap, I just noticed that I've been sending unified format diffs to
  > this list, I should start using context diffs instead shouldn't I?
  > 
  > Thu Oct  2 14:02:15 1997  David S. Miller  <davem@tanya.rutgers.edu>
  > 
  > 	* combine.c (force_to_mode, nonzero_bits): Correctly optimize
  > 	constant offset computations from objects with known alignment in
  > 	the presence of STACK_BIAS.
Thanks.  I've installed this patch.

jeff