[Help][RS780E][mipsel]Screen corruption sometimes

[Michel Dänzer]

On Don, 2011-08-04 at 16:59 +0800, Chen Jie wrote: 
> 
> 2011/8/4 Michel Dänzer <michel at daenzer.net>:
> > On Don, 2011-08-04 at 14:46 +0800, Chen Jie wrote:
> >>
> >> We're running RS780E on a MIPS64 compatible platform, and encounter
> >> screen corruption sometimes:
> >>  * Do 'x11perf -compwinwin500' -- Window decoration and the background
> >> corrupt (http://dev.lemote.com/files/upload/software/temp/x11perf-corrupt.png)
> >>  * Hardware Accelerated video playback(video-pipe branch of mesa) --
> >> both playback and the desktop:
> >>   - http://dev.lemote.com/files/upload/software/temp/newmobcal1920-playback.png
> >>   - http://dev.lemote.com/files/upload/software/temp/newmobcal1920-desktop-corrupt.png
> >>   - Related thread:
> >> http://lists.freedesktop.org/archives/mesa-dev/2011-August/010064.html
> >
> > The video decoding corruption may be due to the pipe-video code not
> > working properly on all hardware yet.
> >
> > The rest looks like more or less random BO corruption.
> The x11perf test causes the corruption each time.

You can try booting with radeon.test=1 and see in dmesg if the transfer
tests triggered by that succeed.


> >> Any idea? (BTW, don't know if related, but the bandwidth of separated
> >> video ram is really low, about 14MB/s for write op)
> >
> > How do you measure it?
> mmap the VRAM and write to it. Attachment is the test code.

I wonder if memset might use byte-sized writes, but anyway direct CPU
access to VRAM is known to be slow with discrete GPUs, especially
without write combining. The userspace drivers mostly avoid it in favour
of GPU transfers. You can try x11perf -shmput500 -shmget500 or booting
with radeon.benchmark=1 to get some numbers for those.


-- 
Earthling Michel Dänzer           |                   http://www.amd.com
Libre software enthusiast         |          Debian, X and DRI developer