rs6000: PREFERRED_RELOAD_CLASS

[Geoff Keating]

> cc: gcc-patches@gcc.gnu.org
> Date: Fri, 21 Jul 2000 15:12:18 -0400
> From: David Edelsohn <dje@watson.ibm.com>
> 
> >>>>> Geoff Keating writes:
> 
> Geoff> Anyway, we still don't want to be performing memory copies using FPRs.
> Geoff> The latency makes it faster to do the copy using GPRs.
> 
> 	This is not always the case and one may want the bandwidth over
> the latency.  I do not think that making this change so cavalierly is
> wise.

The change was extensively tested.

> 	Also, I believe that some parts of the PowerPC port make decisions
> based on moving 8 byte small structures in FPRs.  I am not sure how
> HARD_REGNO_OK versus PREFFERED_RELOAD_CLASS interact to allow or prevent
> that after your change.  I am very concerned about this change and the
> possibility that it will have unintended negative consequences.

The change has no effect in that case.  If a reload is required to be
in the FPRs, this change won't do anything.

I prefer to assume that the change has generally positive
consequences, considering the evidence.  I don't believe that forever
worrying about phantoms is conducive to progress.

-- 
- Geoffrey Keating <geoffk@cygnus.com>