16:35 <mntmn> i want to file an etnaviv/xwayland bug ~but~ the gitlab interface disappears (black tiles, then completely black content except for the cursor) when i'm in the text input field in chromium.

16:36 <mntmn> latest mesa and xserver/xwayland from git

16:37 <mntmn> the other issue is that glamor now fails to render specific icons/textures: http://dump.mntmn.com/screenshot-2022-05-21-18-12-57.png

16:37 <mntmn> with glamor disabled (via env variable): http://dump.mntmn.com/screenshot-2022-05-21-18-11-42.png

16:37 <mntmn> also, the alpha is messed up

16:42 <mntmn> gimp icons are still fine in mesa 21.3.8, and so is the text input field in chromium, so i guess it's a regression in mesa/etnaviv

16:58 <mntmn> mesa 22.0.3 from debian unstable also does not exhibit the bugs

17:03 <austriancoder> mntmn: 22.1.0 fine too? May you can do a git bisect?

17:04 <mntmn> yeah, doing a new build

17:42 <mntmn> 08bdee3f107647090df8d5f98f0994949e8c9a22 is still ok, which is already 22.2.0

17:52 <mntmn> started a real bisect.

20:03 <mntmn> > 7647023f3bb5785f15476b64c08b3ed01c46c536 is the first bad commit

20:03 <mntmn> > glsl: enable the use of the nir based varying linker

20:03 <mntmn>

20:03 <mntmn> cc austriancoder

20:07 <marex> mntmn: ETNA_MESA_DEBUG=tgsi fixes it ?

20:08 <mntmn> tgsi? that never rendered anything correctly

20:08 <austriancoder> mntmn: may I ask you for some debug output for good vs bad with NIR_DEBUG=print?

20:08 <marex> that was the default before ;-/

20:08 <mntmn> marex: yes, we were always shipping ETNA_MESA_DEBUG=nir because tgsi was always broken

20:08 <austriancoder> marex: tgis is gone

20:09 <mntmn> austriancoder: debug output from xwayland i presume?

20:09 <mntmn> austriancoder: or sway? or the application (gimp?)

20:12 <austriancoder> mntmn: the application

20:12 <mntmn> i suspect this only outputs something when mesa is built in debug mode?

20:12 <mntmn> (because i'm not getting any output)

20:15 <austriancoder> You can also provide a trace file (apitrace) of the application. I only need to see the difference the linker caused.

20:15 <austriancoder> mntmn could be...I always build with debug

20:16 <mntmn> ok, i think apitrace is easier for me

20:28 <mntmn> kinda strange, i have a trace, but when trying to replay it, i get "warning: no current context" for each line

20:49 <mntmn> austriancoder: http://dump.mntmn.com/chromium-bad.trace

20:50 <mntmn> austriancoder: i'm not able to trace gimp in a way so that i can also replay it, yet... because gimp is not doing the GL calls, but xwayland/glamor is. and tracing xwayland doesn't catch setting up a context (??)

21:02 <mntmn> austriancoder: if i play back chromium-bad.trace on "bad" mesa, it shows the bug. if i play it back in "good" mesa, it looks fine ^^

21:02 <mntmn> but i get a bunch of "major shader compiler issue"s on "good" mesa that i don't get on "bad" mesa

21:02 <mntmn> (when playing back)

21:03 <mntmn> 5381: message: major shader compiler issue 3: 0:1(12): warning: extension `GL_ARB_gpu_shader5' unsupported in vertex shader

21:03 <mntmn> 5381: message: major shader compiler issue 4: 0:2(12): warning: extension `GL_EXT_gpu_shader5' unsupported in vertex shader

21:03 <mntmn> 12485: message: major shader compiler issue 5: 0:1(12): warning: extension `GL_ARB_gpu_shader5' unsupported in fragment shader

21:03 <mntmn> 12485: message: major shader compiler issue 6: 0:2(12): warning: extension `GL_EXT_gpu_shader5' unsupported in fragment shader

21:03 <austriancoder> mntmn: I will trigger a CI run on current master and hope to see some fails there too

21:03 <mntmn> austriancoder: ok, cool

21:05 <mntmn> glamor is really good at surfacing etnaviv bugs in my experience

21:06 <mntmn> :D

21:08 <cphealy> austriancoder: can you share the details of the current CI infrastructure for etnaviv? Are there test runners somewhere that have GC7000 that run automatically on a regular basis?

21:09 <mntmn> i am a manual, irregular test runner

21:09 <cphealy> ;-)

21:09 <cphealy> Which Vivante GPUs do you test on?

21:10 <mntmn> GC7000!

21:10 <cphealy> What would be needed for this to be automatically run on a regular basis?

21:11 <austriancoder> cphealy: The CI has gc2000, gc3000 and gc7000 boards connected to it but atm only gc2000 is integrated into mesas gitlab. For gc7000 there is a merge request but need to fix a u-boot problem first.

21:12 <austriancoder> I hope to have it run automatically someday.. at the moment it is a manual triggered action in mesa merge requests

21:12 <cphealy> What are the HW requirements for someone to have a test runner setup integrated into mesa gitlab? I've got GC2000/GC3000/GC7000 HW and am interested in contributing.

21:12 <cphealy> ack

21:13 <marex> austriancoder: what's u-boot ?

21:13 <mntmn> austriancoder: n8n8

21:13 <marex> austriancoder: something I can help with ?

21:14 <austriancoder> cphealy: https://docs.mesa3d.org/ci/index.html#ci-farm-expectations

21:15 <austriancoder> marex: not sure yet.. I am slowly getting back to do anything etnaviv related .. but if I remember what the problem is/was I can ping you

21:16 <austriancoder> mntmn: https://gitlab.freedesktop.org/austriancoder/mesa/-/pipelines/592683 - I hope to see a non green CI run :)

21:26 <mntmn> looks like lots of errors?

21:33 <austriancoder> some errors are expected (and they get filtered away for good/bad reporting). But yeah.. it looks like there are some more reproducers. Makes fixing the issue a lot easier than working with traces.

21:38 <mntmn> cool

21:39 <mntmn> austriancoder: should i file a bug in mesa gitlab or not necessary?

21:41 <austriancoder> mntmn: it is always a good idea to do .. the bug can be referenced from the MR/git commit .

21:42 <mntmn> ok

21:53 <mntmn> https://gitlab.freedesktop.org/mesa/mesa/-/issues/6536