Bug 29703

Summary: demos crashing on rv710 r600g
Product: Mesa Reporter: Michal Suchanek <hramrach>
Component: Drivers/DRI/R600Assignee: Default DRI bug account <dri-devel>
Status: RESOLVED FIXED QA Contact:
Severity: normal    
Priority: medium    
Version: git   
Hardware: x86-64 (AMD64)   
OS: Linux (All)   
Whiteboard:
i915 platform: i915 features:

Description Michal Suchanek 2010-08-20 07:06:44 UTC
some curently broken demos:

locks up GPU:

shadowtex
teapot
dlist-*
deriv
bounce
arbocclude

segfaults the program:

arbfplight
arbfslight
bitmap

as a bonus some interesting effects:

failed clear - shows garbage or some pictures displayed in other applications:
copypix
clear*

array draws a mesh on r600c but draws nothing on r600g (paints the background over the mesh?)

firecube flashes

dissolv works very fast on r600g but the patterns are odd and it is impossibly slow on r600c (much slower than software)
Comment 1 Michal Suchanek 2010-08-24 02:04:21 UTC
after latest changes

- shadowtex works (the shadows are pixelated somewhat and the m option has seams - not sure if that is error in rendering or the demo is too simplistic to get such details right). Pressing v gives wrong colors - vertex program is a fail.

- teapot looks generally OK except it is partially transparent (where there are holes in the top the light goes through although there is bottom still) it gives 13fps which is quite slow

- dlist-* looks mostly OK except it does not redraw after window resize

- arbfplight locks up GPU
- arbfslight aborts
- bitmap locks up GPU
- copypix segfaults (regression, previously just failed)
- clear-fbo*, clear-scissor now sometimes lock up GPU (regression, previously just failed)
- dissolve now sometimes locks up GPU (regression, previously just failed)

- fbo_firecube and array still fail
Comment 2 Michal Suchanek 2010-08-30 02:13:32 UTC
Now there are no more GPU crashes with today's mesa on linux 2.6.36-rc2 + r600 fence patch and the above demos.

copypix segfaults in mesa

dissolve - wrong pattern

fbo_firecube - flashes

array renders significanly differently from software

dlist- does not crash but some may misrender

clear* some may misrender

arbf?light works within broken lighting expectations

bitmap text not shown

shadowtex works with the issues as listed above

teapot works
Comment 3 Michal Suchanek 2010-09-02 01:26:17 UTC
I added dmesg logging to piglit so I identifed the piglit tests that lock up the GPU:

fbo-drawbuffers-*

Also holding ^C after starting a piglit run does not terminate the runner, only the individual tests and leads to complete system crash. The X server crashes and the kernel keeps faulting continuously.
Comment 4 Michal Suchanek 2010-09-02 05:03:35 UTC
fbo_firecube segfaults or locks up GPU within minutes but usually not immediately
Comment 5 Michal Suchanek 2010-09-20 04:50:34 UTC
currently none of these named demos crashes the card.

Thanks
Comment 6 Török Edwin 2010-09-20 07:44:09 UTC
On rv730 with r600g the 'gloss' demo locks up the GPU too
Comment 7 Michal Suchanek 2010-10-06 06:05:28 UTC
The draw-vertices piglit test locks up the GPU or aborts.
Comment 8 Michal Suchanek 2010-11-03 04:04:04 UTC
draw-vertices now works reliably with r600c from 7.9 and r600g from git.
Comment 9 Jerome Glisse 2010-11-04 12:20:11 UTC
Fixed

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.