On Wed, 2008-12-31 at 18:40 +0000, Peter Clifton wrote: > Hi guys, > > I've been testing the beta release of the 2.6 series Intel driver > against a 2.6.28 kernel, with UXA and DRI2 enabled. > > (I had 'Option "DRI2" "off"' in the Xorg.conf, but apparently that was > ignored - perhaps I put it in the wrong place, or this is not the right > option to control DRI2). > > I've been noticing really poor performance whilst rendering lots of > small lines, something which x11perf -line10 seems to bear out: > > > x11perf - X11 performance program, version 1.2 > The X.Org Foundation server version 10599003 on :0.0 > from pcjc2lap > Wed Dec 31 18:36:35 2008 > > Sync time adjustment is 0.0708 msecs. > > 600000 reps @ 0.0100 msec (100000.0/sec): 10-pixel line > 600000 reps @ 0.0100 msec (100000.0/sec): 10-pixel line > 600000 reps @ 0.0100 msec (100000.0/sec): 10-pixel line > 600000 reps @ 0.0100 msec (100000.0/sec): 10-pixel line > 600000 reps @ 0.0100 msec ( 99900.0/sec): 10-pixel line > 3000000 trep @ 0.0100 msec (100000.0/sec): 10-pixel line > > > When profiling, it appears the hot-spot is in the kernel, with ~90% of > total system time spent waiting in there. > > I upgraded to a git version of sysprof today, and this is able to trace > further inside the kernel. Hopefully this debugging information will be > of use (I can send the whole capture file to anyone who wants it). > > This was traced whilst running x11perf -line10. The timings above were > recorded without sysprof tracing, in case the sysprof overhead skewed > the timings.
Thanks for the sysprof data. The fix for software fallback performance just got merged to the stable branch, so that should fix this problem in the next rc. -- Eric Anholt [email protected] [email protected]
signature.asc
Description: This is a digitally signed message part
_______________________________________________ xorg mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/xorg
