alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://freenode.irclog.whitequark.org/panfrost - <daniels> avoiding X is a huge feature
yann has quit [Ping timeout: 244 seconds]
stikonas has quit [Remote host closed the connection]
tomboy64 has quit [Ping timeout: 240 seconds]
tomboy64 has joined #panfrost
vstehle has quit [Ping timeout: 264 seconds]
rhyskidd has quit [Quit: rhyskidd]
<chrisf> this devicetree is so bogus :(
icecream95 has joined #panfrost
davidlt has joined #panfrost
vstehle has joined #panfrost
cwabbott has quit [Quit: cwabbott]
cwabbott has joined #panfrost
buzzmarshall has quit [Remote host closed the connection]
<tomeu> chrisf: mmind00 will know everything about it :)
swiftgeek has quit [Ping timeout: 256 seconds]
swiftgeek has joined #panfrost
guillaume_g has joined #panfrost
davidlt has quit [Ping timeout: 256 seconds]
yann has joined #panfrost
tomboy64 has quit [Remote host closed the connection]
tomboy64 has joined #panfrost
rhyskidd has joined #panfrost
stikonas has joined #panfrost
raster has joined #panfrost
<icecream95> alyssa: I'm seeing some odd (and broken) code for shader-stencil-export now: https://gitlab.freedesktop.org/snippets/1088/raw
nlhowell has quit [Ping timeout: 256 seconds]
_whitelogger has joined #panfrost
nlhowell has joined #panfrost
nlhowell has quit [Ping timeout: 240 seconds]
<icecream95> alyssa: The dual-source write now happens in the second writeout bundle, so only some fragments get it. Any ideas on how to fix this?
icecream95 has quit [Ping timeout: 264 seconds]
nlhowell has joined #panfrost
guillaume_g has quit [Quit: Konversation terminated!]
guillaume_g has joined #panfrost
nlhowell has quit [Ping timeout: 256 seconds]
nlhowell has joined #panfrost
<alyssa> icecream95: ugh, that patch was specifically supposed to *prevent* code sequences like that from being generating ;-;
<alyssa> *generated
<alyssa> hm, how did it work before?
<alyssa> okay, figured out the stencil one
<alyssa> okay, figured out the stencil one
cwabbott_ has joined #panfrost
cwabbott has quit [Ping timeout: 272 seconds]
cwabbott_ is now known as cwabbott
adjtm has joined #panfrost
<mmind00> chrisf tomeu: any problems? :-)
nlhowell1 has joined #panfrost
nlhowell has quit [Read error: Connection reset by peer]
buzzmarshall has joined #panfrost
Depau_ has joined #panfrost
nerdboy has quit [Ping timeout: 240 seconds]
Depau has quit [Ping timeout: 240 seconds]
davidlt has joined #panfrost
<alyssa> tomeu: I wonder if we could switch to full GLES3 runs on T860 and drop the GLES2/T860 job
<alyssa> deqp-gles3 contains effectively a copy of deqp-gles2 (and the very few b/w compat differences would be caught by the T720/T760/T820 GLES2 jobs)
<alyssa> would be more pressure on the rk3399's but maybe acceptable now that only:changes is in
<alyssa> daniels: ^^
<alyssa> GLES3.0 I mean
adjtm has quit [Quit: Leaving]
<daniels> I think the only real answer is 'try it for a week and see if it falls apart horribly'
<alyssa> fair
<alyssa> Doing a single gles3 job (on a single runner, not sharded) is 11min
nlhowell1 has quit [Ping timeout: 264 seconds]
<alyssa> By contrast, the gles2 job on t720 is 8.5min
<alyssa> I think I'm happy to eat an extra 2.5min on panfrost CI to add thousands of tests into CI
<alyssa> I'm hopeful we can flip on GLES3.0 by the end of the month
<alyssa> FP16 too
<alyssa> (For T760/T820/T860, I mean)
<alyssa> At this point, that means finishing up multisampling and landing it, getting the CI landed, getting depth/stencil reload sorted, and fixing more framebuffer tests
guillaume_g has quit [Quit: Konversation terminated!]
<daniels> or shard into 2?
<daniels> we have 4 Kevins online, so that's two jobs every 10 minutes which is plenty
<alyssa> That's an option too, of course
nlhowell1 has joined #panfrost
davidlt has quit [Ping timeout: 246 seconds]
adjtm has joined #panfrost
nerdboy has joined #panfrost
adjtm has quit [Remote host closed the connection]
adjtm has joined #panfrost
cwabbott has quit [Quit: cwabbott]
cwabbott has joined #panfrost
<chrisf> sigh, building deqp *on* the arm device has some weird issues
<chrisf> i suppose i should put my deqp hat on and fix that
<alyssa> bbrezillon: I'm rebasing your depth wallpapering stuff
<alyssa> A lot of the hacks there disappear now that we have better control over the hardware, I think
<alyssa> Down to a single test regressing from the rebase..
<alyssa> dEQP-GLES2.functional.depth_stencil_clear.depth_stencil_masked
adjtm has quit [Ping timeout: 264 seconds]
<tomeu> alyssa: I would just shard
<tomeu> want me to add to my todo?
<alyssa> tomeu: \shrugs/
<tomeu> chrisf: may be interesting to see how we build and run deqp in mesa-ci
<tomeu> alyssa: should be quite trivial
<alyssa> yeah
<alyssa> alyssa/mesa:full-plus-wallpaper is my current experiment branch
<tomeu> most of the work is coming up with a subset of tests that are stable enough
<alyssa> that has multisampling, bbrezillon's Z/S patches, and full gles3 CI in there
<alyssa> fixes SuperTuxKart's deferred shading renderer so can do some stunning lighting, but far too slow rn to be playable
<alyssa> Manhattan is still pretty broken
adjtm has joined #panfrost
<alyssa> Oh, I also still need MRT wallpapering
<alyssa> though that's kinda hard to get right with u_blitter i think
jgmdev has quit [Quit: Connection closed for inactivity]
<alyssa> (You really need to reload all RTs at once)
<chrisf> tomeu: sure. if there are hacks involved im keen to get them upstream as well
<tomeu> chrisf: there's a couple of patches that get applied, but those were upstreamed and it could well be that there's a release with them
<tomeu> upgrading deqp in CI may require quite some coordination between contributors to a bunch of drivers though
davidlt has joined #panfrost
<chrisf> tomeu: im not familiar with how the mesa-ci plumbing is laid out -- where can i find these patches?
<chrisf> tomeu: the first quirk i noticed was that it misdetects DE_CPU as DE_CPU_X86_64
<tomeu> chrisf: oh, the patches seem to have been dropped now
<tomeu> but there's 3 seds in .gitlab-ci/build-deqp-gl.sh
<chrisf> tomeu: was it the surfaceless stuff?
<tomeu> yep
<tomeu> marcheu's iirc
<chrisf> right -- that should have all gone in some time ago. i know i dragged my feet on it :S
<tomeu> so it was you! :p
<chrisf> yeah, we had a fun time when that support was getting built-- the original team all left :)
<chrisf> so not a lot of time for anything beyond "get android out the door and avoid having the wheels fall off anything"
davidlt has quit [Ping timeout: 246 seconds]
davidlt has joined #panfrost
<bbrezillon> alyssa: cool
raster has quit [Quit: Gettin' stinky!]
davidlt has quit [Ping timeout: 264 seconds]
nlhowell1 has quit [Ping timeout: 246 seconds]
* chrisf frowns at memory consumption when compiling spirv-tools
TheKit has quit [Ping timeout: 256 seconds]
TheKit has joined #panfrost
icecream95 has joined #panfrost
<chrisf> tomeu: do we expect deqp to start properly on g31?
<chrisf> (expected, or have i goofed: https://hastebin.com/nuvajuhudo.rb)
<chrisf> oh, i see -- gles3 isnt enabled by default, and surfaceless platform in deqp doesnt initialize properly without it
<chrisf> at least on this branch; it might be ok on the opengles-cts branches
<chrisf> i should really tidy that up :(