On Mer, 2004-12-01 at 12:19, Thomas HellstrÃm wrote: > Actually, I've been running with a 512Kb userspace / kernel buffer, and > I've not seen any noticable drop in performance, but I'm not sure that the > submissions actually will be that large with the programs I've tested.
Might be worth finding the size needed - remembering a 512K buffer can be verified and copied in 32K chunks anyway. > If the perfomance drop I experience with running the verifier on AGP > memory is due to the fact that none of it is previously cached or that the > processor cannot cache WC memory lines while reading? If it is due to the > latter then we should be OK with large cacheable buffers. Otherwise we > will run into trouble when the buffer does not fit in the cache. AGP space is uncached. We could pull the pages out of AGP, write to them and put them back but we don't really support that right now. As a result all I/O to AGP pages is strictly synchronous. Pulling stuff from cachable memory and verifying it will prefetch lines from memory on the later processors (and you can also hint this with the prefetch(address) call in the kernel (prefetch about 320 bytes ahead is normally right). I'd expect the user pages are already in L2 cache anyway from when Mesa wrote the bits. ------------------------------------------------------- SF email is sponsored by - The IT Product Guide Read honest & candid reviews on hundreds of IT Products from real users. Discover which products truly live up to the hype. Start reading now. http://productguide.itmanagersjournal.com/ -- _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
