Re: Patch 2/9: Split up and reorganize some functions

[Jeff Law <law at redhat dot com>]

On 06/18/10 08:05, Bernd Schmidt wrote:
>   I've also
> fixed up some confusion about when to clear allocno_saved_at_call.
Yea, I had to look at that a few times to convince myself that 
everything was OK.

> Note that there is a small change in behaviour: in 
> inc/dec_register_pressure,
> we no longer recompute nregs as we iterate through the classes.  I 
> think it's
> more correct this way, and I haven't seen any difference in code 
> generation.
There may be oddball architectures where this matters, but I can't think 
of one offhand.  One could argue that for such an architecture that the 
regs should be in different classes.  If we wanted to be absolutely sure 
we could test for this situation once per compilation unit and ICE -- 
that might save a developer on one of these targets considerable 
debugging time.

I think I'd give port maintainers a quick chance to chime in as to 
whether or not their port has registers in a class where the number of 
hard regs to represent a given mode varies within the class.   Otherwise 
it looks fine.

If you wanted to get the bulk of these changes in and old off on the 
behaviour change, then that'd be fine by me.

jeff