Bug 91986

Summary:	Artifacts in models rendering with GeForce 6150 (NV4x chipsets)
Product:	Mesa	Reporter:	Mauro Rossi <issor.oruam>
Component:	Drivers/DRI/nouveau	Assignee:	Nouveau Project <nouveau>
Status:	RESOLVED MOVED	QA Contact:	Nouveau Project <nouveau>
Severity:	major
Priority:	medium	CC:	masao-takahashi
Version:	git
Hardware:	x86-64 (AMD64)
OS:	Linux (All)
Whiteboard:
i915 platform:		i915 features:
Attachments:	X-Wing model artifacts visible on with mesa 11.1git X-Wing model artifacts visible on with mesa 10.6.5 Kernel log

Description Mauro Rossi 2015-09-12 15:47:30 UTC

Created attachment 118224 [details]
X-Wing model artifacts visible on with mesa 11.1git

X-Wing model and other models in FreeSpace2 with StarWars Original Trilogy mod shows rendering artifacts.

Filing a separate bug from https://bugs.freedesktop.org/show_bug.cgi?id=91551
as requested for NV4C chipset,
also because the artifacts are different.

Here I see big black poligons filling the screen, like vertexes of the model invading the whole screen. 

NOTE: using mesa 10.6 the artifacts are not covering the entire GUI,
but I see some kind of vertex/surfaces problem, but the shaders maps are fine but the vertexes/surfaces are completely distorted.

Here is the link of the old trace that is able to produce the artifacs described in the different mesa versions. (X-Wing is the model visible before the Tie pilot)

http://www.mediafire.com/download/949r35rra31rpte/Tie_pilot_Nvidia_GT210_fs2_open_3.7.2.trace.7z

Attached pictures of the artifact for 11.1git and 10.6

Attached dmesg that show some nouveau errors, but most probably unrelated to the bug.

Comment 1 Mauro Rossi 2015-09-12 15:50:39 UTC

Created attachment 118225 [details]
X-Wing model artifacts visible on with mesa 10.6.5

Comment 2 Mauro Rossi 2015-09-12 15:53:26 UTC

Created attachment 118226 [details]
Kernel log

Comment 3 Ilia Mirkin 2015-09-12 20:00:56 UTC

A small number of nv4x fixes did go into mesa 11, would be nice to know what affected the change between 10.6 and git.

May be interesting to know if this magically starts working with NV30_SWTNL=true in the environment (which will force vertex shaders to execute on the CPU).

Comment 4 Mauro Rossi 2015-09-12 22:56:46 UTC

Hi Ilia,

tested with NV30_SWTNL=true 

With both mesa 11.1git (oibaf ppa) and 10.6.5 (xorg-edgers ppa) the problems disappears

As a side note, with NV30_SWTNL=true I see minor side effects on all models, including the ones that were perfect (Tie Pilot, Tie Fighter), like missing translucency and normal maps have been disabled.

Just my curiosity, are these side effects to be expected when using  NV30_SWTNL=true ?

Mauro

Comment 5 Ilia Mirkin 2015-09-12 23:20:59 UTC

(In reply to Mauro Rossi from comment #4)
> Just my curiosity, are these side effects to be expected when using 
> NV30_SWTNL=true ?

Strictly speaking, swtnl should work identically to hwtnl. However until recently, the swtnl path was *totally* broken. I fixed it up, but I wouldn't exclude the possibility of some remaining fail that I didn't notice.

Comment 6 Mauro Rossi 2015-09-13 12:00:33 UTC

>Strictly speaking, swtnl should work identically to hwtnl. However until >recently, the swtnl path was *totally* broken. I fixed it up, but I wouldn't >exclude the possibility of some remaining fail that I didn't notice.

Hi, thanks Ilia, my question was to check the expected behavior in that case. 

Do you recommed to open a different bug?

I have collected a trace of Millenium Falcon model particularly showing the opaqueness/missing translucency with NV4x.

Mauro

Comment 7 GitLab Migration User 2019-09-18 20:41:04 UTC

-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/1083.

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.