PR rtl-optimization/28071

[Roger Sayle]

On Thu, 27 Jul 2006, Jan Hubicka wrote:
> 2006-07-27  Jan Hubicka  <jh@suse.cz>
>
> 	PR rtl-optimization/28071
> 	* cselib.c (cselib_process_insn): Don't remove useless values too
> 	often for very large hashtables.

This is OK for mainline.

It might be interesting to also instrument this code to check whether
its reasonable to increase MAX_USELESS_VALUES.  It is bizarre that we
have quadratic behaviour keeping the hashtable clean, when the overhead
of probing a dirty table is at worst linear, and best case has no/little
performance overhead best if our hashing is working.

I like your approach of testing the fraction of useless values.
I just wonder if the values "32" and "25%" can be improved upon.

Good catch.

Roger
--