failed bootstrap

[Ulrich Weigand]

Paolo Carlini wrote:

> Andrew Pinski wrote:
> 
> > i686-pc-linux-gnu: 
> 
> Bootstrap completes and regtesting is ok reverting this commit:
> 
> 2003-01-31  Ulrich Weigand  <uweigand@de.ibm.com>
> 
> 	* reload.c (find_reloads): Do not use the mode specified in the insn
> 	pattern as reload mode for address operands.  Do not generate optional
> 	reloads for operands where a mandatory reload was already pushed.
> 	Generate optional reloads only in the final pass though find_reloads.
> 	(have_replacement_p): New function.

I've analyzed the i686 failure now; what happens here is that an
optional reload is *merged* with an address reload.  This means 
that not generating optional reloads can *change* the set of
mandatory reloads, and therefore it is not a good idea after all
to generate optional reloads only in the last pass.

I have reverted the reload.c commit now (on mainline and branch);
I'll have to think of some other way to fix my problem.

Sorry for the breakage :-(

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  weigand@informatik.uni-erlangen.de