[Bug 38800] glXSwapBuffersMscOML is slow on AMD Fusion but not on Intel 945 w/Atom

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=38800

--- Comment #16 from Michel Dänzer <michel at daenzer.net> 2011-07-05 10:37:21 PDT ---
(In reply to comment #15)
> Whatever resolution I choose, perf top shows me that 50% of my used CPU time is
> spent in r600_fence_finish from the r600g DRI driver.

That makes some sense, as r600_fence_finish() is currently more or less a
polling loop. This CPU usage is probably glFinish() waiting for the frame to
finish rendering.

(BTW, IIRC the intel drivers used to have a broken glFinish() implementation
which didn't always wait for all rendering to finish. I wonder if your tests
might have been affected by that)


> I also see about 5% of my used CPU time in radeon_fp_native_mode from the
> kernel driver; again, this is resolution independent.

Hmm, this might indicate it's constantly trying to get a native mode for the
fake LVDS. Have you tried force-disabling it, to see if that makes any
difference?


> I'm forced to conclude that it's not CPU-side slowdown - is there any way (like
> Intel's intel_gpu_top and intel_gpu_dump) to see roughly what the GPU is up to?

Not yet, unfortunately.

However, I think we first need to narrow down further where exactly the CPU is
blocking, and why.

-- 
Configure bugmail: https://bugs.freedesktop.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.