Re: Speedup invalid_mode_change_p

[Jan Hubicka <jh at suse dot cz>]

> On Thursday, February 27, 2003, at 04:09 AM, Jan Hubicka wrote:
> >this patch saves 5-10% of nonoptimizing combine.c compilation time.
> 
> If applicable to 3.3 and reasonably safe, would you mind regtesting and 
> submitting there?  invalid_unit_mode_change_p accounts for 0.6% of the 
> time in Finder_FE multiunit compiles.
> 
> [ we are really interested in seeing 3.3 compile fast with -O0 ]

The patch has been tested 3.3 and in fact I wanted to mark it as
3.3/mainline patch, so I will install it into both branches.  On i386
reg-class is twice as slow as gcc-3.0 one, so it definitly accounts as
an regression.

Honza
> 
> Thanks.