adjust generation of component addresses

[Richard Henderson]

On Tue, Jul 09, 2002 at 01:27:52PM -0400, David Edelsohn wrote:
> 	Any thoughts about how to address this performance problem?

Someone needs to identify the specific section of code that 
gets pesimized by this patch.

> 	Did the simplify_gen_binary change apply simplify_plus_minus too
> broadly?  It currently is called with the "force" flag, which also was
> added by the patch, so that it will succeed even if the result is not
> "simplier".

Yes, simplify_plus_minus will always succeed, but that's because
we always want to perform the addition.  The old code used 
plus_constant to always perform the addition.  The difference
between plus_constant and simplify_plus_minus+force is that the
later will sort the addends into a canonical order.  I'm having
trouble imagining how this could ever be a problem...


r~