EGCS bad on loop invariant optimization ?

[Jeffrey A Law]

  In message < 199804192208.SAA19223@jfc. >you write:
  > 
  > Try this patch.
  > 
  > Sun Apr 19 18:07:14 1998  John Carr  <jfc@mit.edu>
  > 
  > 	* alias.c (true_dependence): If neither mode is QI or BLK and the
  > 	modes are different, X and MEM do not alias.
Looks correct to me.

One day it would be nice to distinguish items which have the same
size, but are different types (ie pointers vs int/log, but we need
help from the front end, or a first-class Pmode for that).

jeff