This is the mail archive of the
mailing list for the GCC project.
Re: powerpc & unaligned block moves with fp registers
- To: David Edelsohn <dje at watson dot ibm dot com>
- Subject: Re: powerpc & unaligned block moves with fp registers
- From: Mark Mitchell <mark at codesourcery dot com>
- Date: Fri, 09 Nov 2001 11:27:07 -0800
- cc: Richard Kenner <kenner at vlsi1 dot ultra dot nyu dot edu>, "gcc at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>
--On Friday, November 09, 2001 09:52:29 AM -0500 David Edelsohn
>>>>>> Mark Mitchell writes:
> Mark> I would prefer to see GCC just ask the back end "How do you want to
> Mark> perform a memory-to-memory copy of thus and such many bits, knowing
> Mark> thus and such about the current alignment of the memory?"
> And I consider BLKmode move (movstrsi, etc.) the way to do that.
> It already provides an alignment and number of bytes. On PowerPC, BLKmove
> chooses the most efficient method, including using DImode.
Yes, I think BLKmode is a fine way to implement the interface.
> One could call BLKmode move always, but my model was using
> SLOW_UNALIGNED_ACCESS to determine when the backend considred things
> "complicated" or "non-obvious" that the common code should punt to BLKmode
This would certainly work, but I think it would be nicer to just always
use BLKmode and have the back ends call back to a common routine if they
want command handling. I would like the middle end/back end interface
to be small, and then provide a library of common routines that back
ends could use.
That's just my bias on the design; there's nothing terribly magic
Mark Mitchell email@example.com
CodeSourcery, LLC http://www.codesourcery.com