On Tue, Aug 17, 2004 at 11:38:09AM +0200, Andreas Schwab wrote: > Agreed, although probably much less in C++. But if they are references > more than once in a function then cse'ing the address is usually faster > anyway. I wasn't aware that m68k had so many free registers that it is unconcerned about spilling once the function grows to a non-trivial size. r~