|Site sponsored by IGEL|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[directfb-dev] Re: [PATCH] Preliminary AGPONLY support
On Mon, Dec 16, 2002 at 02:39:17PM +0100, Denis Oliver Kropp wrote: > > So with this optimized routine write-combining is almost as fast as > > write-back. I just tried with "uncachable" too and sys->agp went to 130 > > MB/s and agp->sys reamined at 30 MB/s. So with write combining reads are > > uncacheable. No very nice at all. Some googling revealed some linux-kernel > > discussion about the very same thing. Apparently the AGP protocol requires > > this :( > > I can't imagine that it's essential for framebuffer memory. Maybe command > buffers are only required to. I don't know much about this stuff but according to my common sense the problems occur when the gfx card writes to the memory and the CPU has already cached the old data. Since command buffers are write only so I don't think the cache can have stale data there. Well unless there's some progress/status stuff in the AGP memory that the gfx card writes to. I seems to remember that the Mach64 DRI driver did something like this but I'm not sure. > What's the performance of mem2agpcpy() for sys-sys, sys-vid and vid-sys > compared to dfb_memcpy()? The same. They look pretty much identical. I have a Duron so it's using the mmxext version. -- Ville Syrjälä email@example.com http://www.sci.fi/~syrjala/ -- Info: To unsubscribe send a mail to firstname.lastname@example.org with "unsubscribe directfb-dev" as subject.