[prev in list] [next in list] [prev in thread] [next in thread] 

List:       mesa3d-dev
Subject:    Re: [Mesa3d-dev] [Mesa3d-cvs] CVS Update: Mesa (branch: trunk)
From:       Keith Whitwell <keith () tungstengraphics ! com>
Date:       2005-05-23 14:01:17
Message-ID: 4291E22D.3050802 () tungstengraphics ! com
[Download RAW message or body]

Keith Whitwell wrote:
> Alex Deucher wrote:
> 
>> Couldn't this sse memcopy stuff be done as a generic sse_memcopy so
>> other drivers could plug it in as well (_mesa_sse_memcopy() or some
>> such) or is it specific to VIAs?
> 
> 
> It certainly would - I'll probably hoist it out of the via driver fairly 
> shortly.  The only issue at the moment is the use of "movntps" ie 
> non-temporal moves.  These work well in the via driver as the texture 
> destination is unusual for being in video ram.  However, other drivers 
> do plenty of memcpys to uncached memory as well, just not the 
> ctx->Driver.TextureMemCpy() operation.
> 
> Having seen how much it helps vertex emit even with a naive application, 
> I'd say it's going to be a useful piece of code, but maybe not every 
> driver will end up using it to hook out TextureMemCpy.  We can of course 
> add in a version which uses regular moves for those applications.

Grr.  Naive is right - most of the speedup due to me being tricked by 
cpuspeed...  As coded the sse path doesn't get touched for reasonable 
sized vertices.

Keith


-------------------------------------------------------
This SF.Net email is sponsored by Oracle Space Sweepstakes
Want to be the first software developer in space?
Enter now for the Oracle Space Sweepstakes!
http://ads.osdn.com/?ad_id=7412&alloc_id=16344&op=click
_______________________________________________
Mesa3d-dev mailing list
Mesa3d-dev@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mesa3d-dev
[prev in list] [next in list] [prev in thread] [next in thread]