--- Comment #16 from Thomas Hellström <thomas@...> 2008-02-29 10:57:05 PST ---
(In reply to comment #15)
> (In reply to comment #13)
> > Created an attachment (id=14639)
--> (http://bugs.freedesktop.org/attachment.cgi?id=14639) [details] [details]
> > Fix-check patch
> > OK, so I've attached a patch that should revert back and make it possible to
> > add the fixes one by one.
> > Could you
> > 1) Apply the patch, and verify that performance is as expected.
> this one gets about 100.6 fps
> > 2) Define FIX1 at the top of i915_dma.c, recheck.
> this one gets about 99.7 fps
> > 3) Define FIX2 at the top of i915_dma.c, recheck.
> this one gets about 92.6 fps
OK, so the problem is FIX2.
What this patch is doing is to save all buffer object validation replies until
we're completely sure we don't hit an EAGAIN. If we hit an EAGAIN, the IOCTL
should be rerun with the original arguments and they are overwritten by the
EAGAINS are not common during normal operation, but should be fairly common
with X server command submissions, due to silken mouse operation.
I can't really see how this could affect performance on i965 unless the app is
already heavily CPU bound and uses a lot of small textures.
It's possible that this can be restructured to restore the original arguments
only if we hit an EAGAIN. I don't have access to an i965 until in a week or so
Configure bugmail: http://bugs.freedesktop.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.