#
…

2023-03-03

00:01xcpy0 has quit [Read error: Connection reset by peer]
00:01xcpy0 has joined #asahi-gpu
00:02i509vcbI was wondering if a description of the entire rendering pipeline in the hardware is somewhere. There is this article from powervr that describes the USC part of it: https://blog.imaginationtech.com/a-look-at-the-powervr-graphics-architecture-tile-based-rendering/
00:02i509vcb(I have noticed a lot of references to powervr in the driver talk of course)
00:53kesslerd_ has quit [Quit: Konversation terminated!]
01:09kesslerd has quit [Remote host closed the connection]
01:10aomizu has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
01:31alyssa has joined #asahi-gpu
01:31alyssaok at 7090 followers on mastodon and told transphobes to unfollow
01:31alyssalet's see where my follower count is at tomorrow :~)
02:05Tramtristwould be surprised if theres a material change to the negative on Mastodon...
02:36possiblemeatball has joined #asahi-gpu
02:41TramtristActually more likely to be unfollowed for not using content warning or.. whatever hidden texts
03:00hightower3 has joined #asahi-gpu
03:05aomizu has joined #asahi-gpu
03:07hightower2 has quit [Ping timeout: 480 seconds]
04:34possiblemeatball has quit [Quit: Quit]
04:54Hibyehello has joined #asahi-gpu
05:09Hibyehello has quit [Ping timeout: 480 seconds]
05:12Hibyehello has joined #asahi-gpu
05:19Hibyehello_ has joined #asahi-gpu
05:22Hibyehello has quit [Ping timeout: 480 seconds]
05:39alyssalina: dEQP results (particularly dEQP-GLES3) are regressed significantly since sync
05:39alyssaplease debug
05:39alyssaERROR - dEQP error: corrupted double-linked list
05:39alyssaTest case 'dEQP-GLES3.functional.occlusion_query.scissor'..
05:39alyssaglcts: ../src/gallium/drivers/asahi/agx_query.c:115: agx_get_query_result: Assertion `query->writer == NULL && "cleared when cleaning up batch"' failed.
05:39alyssalots of "fun" stuff here
05:46alyssaMESA_SHADER_CACHE_DISABLE=1 EGL_PLATFORM=surfaceless deqp-runner run --deqp ~/CTS/glcts --output output --jobs 8 --renderer-check G13 --caselist ~/CTS/gl_cts/data/mustpass/gles/aosp_mustpass/master/gles3-master.txt --skips ~/mesa/.gitlab-ci/all-skips.txt -- --deqp-surface-type=pbuffer --deqp-gl-config-name=rgba8888d24s8ms0 --deqp-surface-width=256 --deqp-surface-height=256 --deqp-log-images=disable
05:46alyssa--deqp-log-shader-sources=disable
05:46alyssaMESA_GLES_VERSION_OVERRIDE=3.1 ASAHI_MESA_DEBUG=deqp MESA_SHADER_CACHE_DISABLE=1 EGL_PLATFORM=surfaceless deqp-runner run --deqp ~/CTS/glcts --output output --jobs 8 --renderer-check G13 --caselist ~/CTS/gl_cts/data/mustpass/gles/aosp_mustpass/master/gles31-master.txt --skips ~/mesa/.gitlab-ci/all-skips.txt -- --deqp-surface-type=pbuffer --deqp-gl-config-name=rgba8888d24s8ms0 --deqp-surface-width=256
05:47alyssa--deqp-surface-height=256 --deqp-log-images=disable --deqp-log-shader-sources=disable
05:47alyssaThose csv files were the failures.csv for deqp-gles31 and deqp-gles3 just prior to merging explicit sync
05:47alyssaAny fails you see running deqp that aren't in that list are regressions and need to be fixed before we can merge.
05:48linaOof... gles2 was fine last I tried it, but I didn't try gles3... ^^;;
05:48alyssa(I realize we're not shipping GLES3 *yet*, but piles of bugs affecting GLES3 will also affect the GL2 that we are shipping.)
05:48alyssayeah, deqp-gles2 is mostly a toy smoke test
05:48alyssait's tiny, it doesn't cover much
05:48alyssaalso, piglit
05:49alyssapiglit i think has regressed due to my bugs, not (just) sync bugs
05:49alyssabut last time I tried, I couldn't get through a piglit run anymore without my system crashing and errr
05:49alyssathat sounds like maybe a tangential issue
05:50alyssaMESA_SHADER_CACHE_DISABLE=1 PIGLIT_PLATFORM=gbm piglit-runner run --output output --piglit-folder /home/alyssa/piglit --profile quick_gl
05:50alyssaOof, deqp-gles31 is hitting the OOM killer now. That's not great.
05:50alyssaYeah. deqp-gles31 is going down in under a minute (8GB mini)
05:51linaOh yeah, deqp-EGL hits the OOM killer on 8GB with >1 thread (or maybe 2 fit, barely), I don't know why...
05:51linaIt's not a hard leak but I don't know how much memory it's *supposed* to use...
05:51alyssanow, as in, this is a regression
05:52alyssaunsure if it's related to sync. it's kinda hard to bisect over hard uapi changes.
05:52alyssabut this was working before at any rate
05:54alyssaand we can't release if the driver is like this. what we're shipping now was rock solid. if we can't even run basic test suites anymore without crashing, it's almost guaranteed that something will regress for a user if we push it out like this.
05:55linaSo I'm going to move the SHAREABLE fix thing to before the batch tracking because that was actually regressed (well, it was always technically wrong, it just started mattering) by the UAPI change, not the actual explicit sync
05:55linathe flush_resource thing
05:56linaAnd I'll do something to make batches not roll around all the time. I think I can just do a poll on all submitted syncobjs twice after each submission (you can do several in one syscall, doing it twice means we potentially clean up 2 batches for each new batch, which should keep memory usage under control)
05:57linaI don't think we want to do anything fancy with IRQs delivered to userspace... the equivalent of that with the existing UAPI is to just have a thread poll on syncobjs and do the cleanup there, but that sounds like a "might be a good idea some day, probably not now" thing...
05:58alyssaand o look it's your stream o'clock which means it's my sleep o'clock
06:49aomizu has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
07:13systwi has quit [Ping timeout: 480 seconds]
07:14linaalyssa: So I think a lot of the regressions are actually from the git work/rebase for upstreaming the device stuff ^^;; (fd leaks...)
07:14linadEQP-EGL was all kinds of broken now...
07:14linaAnd at least some are due to the flush_resource stuff
07:25systwi has joined #asahi-gpu
07:43aomizu has joined #asahi-gpu
07:52Emantor_ is now known as Emantor
08:05linaOkay, the flush_resource thing was completely broken... Maybe that's what was breaking Sway... ^^;;
08:32Ziemas has quit [Ping timeout: 480 seconds]
08:38MajorBiscuit has joined #asahi-gpu
08:42i509vcb has quit [Quit: Connection closed for inactivity]
08:49aomizu has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
08:49merry has quit [Quit: ZNC - https://znc.in]
08:50mort_ has quit [Quit: Ping timeout (120 seconds)]
08:50mort_ has joined #asahi-gpu
08:50merry has joined #asahi-gpu
08:51aomizu has joined #asahi-gpu
08:58aomizu has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
09:07linaalyssa: I have a flake regression in dEQP-GLES2 and I don't know where it comes from... it's easy to reproduce running dEQP-GLES2.functional.fragment_ops.interaction.basic_shader.77 in a loop. The images have random bad pixels... any ideas? It still happens in sync mode, and also with the flush_resource stuff disabled...
09:08linaI don't recall ever seeing this back when I tested dEQP-GLES2 with the explicit sync stuff... could it be an upstream mesa regression recently? I tried just UAPI on top of mesa/mesa:main and it still happens...
09:40Ziemas has joined #asahi-gpu
09:52m42uko has quit []
09:52m42uko has joined #asahi-gpu
09:52nyilas has joined #asahi-gpu
09:59linaalyssa: I can't repro the pink square thing with stk but I can with nautilus. Fixed ^^
10:00linaIt turns out you can render to a BO and *then* export it and so we need to post-facto insert a DMA-BUF fence into it at export time...
10:31c10l has quit [Ping timeout: 480 seconds]
10:49aomizu has joined #asahi-gpu
10:54aomizu has quit []
11:53nyilas has quit [Remote host closed the connection]
11:56aomizu has joined #asahi-gpu
12:15linaalyssa: Ummm so I'm running into ASAN issues with constant buffers and blitting and I can't figure out why this isn't broken for everyone...
12:30linaOK so I fixed that one with a state tracker patch... but I still don't know why this isn't breaking other drivers...
12:30linaAnd now I'm running into others... and I'm getting the feeling none of this has anything to do with explicit sync and it's just all very broken and I don't get it...
12:48aomizu has quit [Max SendQ exceeded]
12:51aomizu has joined #asahi-gpu
12:53linaSooo --deqp-surface-type=fbo is very, very broken... ;;
13:06aomizu has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
13:07kesslerd has joined #asahi-gpu
13:31tertu has quit [Ping timeout: 480 seconds]
13:48aomizu has joined #asahi-gpu
13:53possiblemeatball has joined #asahi-gpu
13:56hightower3 has quit [Ping timeout: 480 seconds]
14:41hightower2 has joined #asahi-gpu
14:59c10l has joined #asahi-gpu
15:01aomizu has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
15:29LinuxM1 has joined #asahi-gpu
15:30alyssalina: dEQP-GLES2.functional.fragment_ops.interaction.basic_shader.77 could easily be an upstream regression, yeah
15:30alyssamaybe the colour masking opts are still not quite right
15:31alyssaI use --deqp-surface-type=pbuffer, same as CI, fwiw
15:32LinuxM1 has quit []
15:34linaalyssa: I'm stuck on max draw buffers right now, TIB offsets >= 64 do not work. I'm going to macOS to see if I'm missing something here, even kernel side...
15:34kesslerd has quit [Remote host closed the connection]
15:35alyssalina: That's an existing fail, not a regression
15:35alyssai mean you're welcome to debug it but it's not a blocker
15:36linaIt wasn't on your existing GLES3 fail list... but I see it is in the GLES31 one...
15:36alyssaoh. yeah. lmao
15:36alyssaThose tests used to be in dEQP-GLES31 but then Khronos realized "Wait there's literally nothing in here that depends on GLES31" so they moved it to GLES3
15:36alyssaand I haven't uprevved deqp in a long time
15:36linaOh...........
15:37possiblemeatball has quit [Quit: Quit]
15:37alyssastill grateful for help debugging them because I got as far as "yeah tib offsets >= 64 broken ???? why ????"
15:37linaWell either way at least for GLES3 I'm pretty sure I don't have any sync-related regressions... it's that, some depth/stencil stuff, and some pack/unpack stuff. All of which sound completely unrelated to sync.
15:37linaI think I spent the past 2-3 hours getting there...
15:37alyssaAwesome :)
15:38possiblemeatball has joined #asahi-gpu
15:38linaI do feel like sometimes I spend hours figuring out stuff you've already figured out ;;
15:40alyssaif I had figured it out, I would've fixed it ;)
15:40linaI mean I spent that long getting to where you were ;;
15:40alyssai'm sorry to hear that
15:41linaI'm not sure how long it's been but it's been a while of digging through render target code to get to offsets > 64 (which I got to just now) ^^;;
15:42alyssathe way I got there was by looking at the results.xml for the failing test
15:42alyssaand seeing that all render targets were correct except the very last
15:42alyssaand then looking at the store for the very last and seeing it was offset 64 and thinking "yeah that's probably it then"
15:42alyssaand that's when I shrugged it off and moved onto other things
15:43linaYeah... I saw only the last one was wrong, but it took me a while to even notice the 64 offset...
15:43alyssai'm not sure what you want me to say/do
15:43linaMaybe we should have a notes doc on what's broken/why to avoid duplicate work ^^;;
15:44alyssasounds like it would get out of sync very quickly
15:54alyssalina: lldb should work
15:54alyssaoh. that's incredibly cursed lol
16:00jhan has joined #asahi-gpu
16:05Mary has quit [Quit: The Lounge - https://thelounge.chat]
16:09Mary has joined #asahi-gpu
16:35bluetail9I like that you both motivate eachother.
16:44Mary has quit [Quit: The Lounge - https://thelounge.chat]
16:50Mary has joined #asahi-gpu
16:56jhan has quit [Remote host closed the connection]
17:08possiblemeatball has quit [Quit: Quit]
17:09possiblemeatball has joined #asahi-gpu
17:13jhan has joined #asahi-gpu
17:17MajorBiscuit has quit [Quit: WeeChat 3.6]
17:22jhan has quit [Ping timeout: 480 seconds]
17:46possiblemeatball has quit [Quit: Quit]
17:55linaalyssa: Pushed to my agx/next, I think that should be clean. Needs a kernel uprev due to a minor UAPI change.
17:55linaTop commit on that is some guesswork for the things we were still not setting properly in the cmdbuf, no idea how it interacts with multisampling.
17:55lina(Or whether it does anything()
17:56alyssaLooks like I can squash it in with the "add UAPI"? :P
17:56alyssas/:P/:)/
17:57linaIf you think it's correct, sure ^^
17:57linaAlso I just force pushed
17:57alyssaI don't think it's incorrect! :p
18:00linaalyssa: Also I had to add an argument to 1228f92097d due to a mesa:main change, but it's just copypasta, please check that it's correct ^^;;
18:00alyssalooks good
18:01alyssaand yeah I held back on submitting the MR for that because I knew that the func sig was going to have to change
18:01linaDidn't get around to reviewing MRs, sorry... (nor to addressing all the feedback on the uapi-prep one, though at least I pushed updates to it with the bugfixes)
18:01alyssasometimes that's how it goes :)
18:06alyssalina: Latest agx/next fixes the magenta rectangles on STK, thanks :)
18:07linaDidn't repro on stk at all for me but it did on nautilus ^^
18:08alyssaI hope you agree in retrospect this was an important bug to fix that didn't really have anything to do with stk or nautilus :)
18:09linaYes, also I hate implicit sync ^^;;;
18:09alyssainteresting
18:09alyssaI would have thought after all that you would hate explicit sync instead :p
18:09linaExplicit sync was easy...
18:09linaMaking it work with the implicit sync world on the other side on the other hand...
18:10alyssaah, well, yes
18:11jhan has joined #asahi-gpu
18:12alyssadeqp results look good again
18:12alyssasome regressions for me to fix but back to a world of sanity at least
18:12linaAlso I think iris has the magenta rectangles bug, I don't see anything in that MR to address the case that fixed it ^^
18:13alyssalina: but isn't that our thing?
18:13alyssaDoing things better than the industry? :~)
18:14alyssalina: btw on the subject of known fails
18:14alyssaI *think* I figured out what the deqp-gles3 cube maps fails are about, *maybe*
18:14alyssaseamless cube maps are supposed to ignore the wrap mode in GL and VK
18:14alyssabut I think for some reason AGX (and Metal?) honour the repeat wrap mode even for seamless cube maps
18:15alyssawhich is all kinds of wrong
18:15alyssaif this is indeed the issue -- I haven't confirmed it, it's basically just from staring at the fails/pass list -- then for GL it shouldn't be too bad to fix
18:15alyssaadd a hack in mesa/st to fix up the wrap mode and it
18:16alyssa'll be ok
18:16alyssafor VK it'll be a bit more annoying because we have no way of knowing whether a given sampler is for a cube map or not
18:17alyssathe only solution I can come up with for Vulkan is duplicating samplers to have a regular version and a cube map version, and toggle between the two in the shader depending on whether loading from a cube map or not
18:17alyssawhich is very much not great but not catastrophic
18:18alyssaHowever.. I don't see any references to this issue in Metal docs nor MoltenVK nor ANGLE
18:18alyssaand I don't see the Apple Metal blob fixing this up
18:18alyssaso it's hard to say what's actually happening here
18:19alyssait's entirely possible that there's, like, a bit in some control register to toggle this behaviour
18:19jhan has quit [Ping timeout: 480 seconds]
18:20alyssaRGX has a TPU control register, maybe AGX does too, idk
18:25kesslerd has joined #asahi-gpu
18:27alyssareally hoping AGX has a TPU control register that can toggle this behaviour tbh
18:27alyssaI can implement the workarounds if needed but I'd much rather not
18:30alyssaRGX has this bit
18:30alyssaTAG_CEMEDGE_DONTFILTER
18:30alyssaPixel data master. Disable filtering over edges/corners for CEM. When set to 1, HW will be seemfull, ie, always stay in the current map, always honour
18:30alyssathe addressmode. When set to 0, HW will be seemless, ie, ignore addressmode, filter between faces at the edges/corners
18:31i509vcb has joined #asahi-gpu
18:31alyssaThat's not what we're looking for but it gives a sense that AGX might have an analogous reg
18:43alyssalina: rsrc->bo can be corrupted in agx_batch_cleanup
18:43alyssahappens running dEQP-GLES31.functional.compute.*
18:43alyssa==53478== Invalid read of size 8
18:43alyssa==53478== at 0x5FBD6F0: ??? (in /home/alyssa/mesa/build/src/gallium/targets/dri/asahi_dri.so)
18:43alyssa==53478== Address 0x7b45180 is 128 bytes inside a block of size 480 free'd
18:44alyssa==53478== at 0x4887B40: free (in /usr/libexec/valgrind/vgpreload_memcheck-arm64-linux.so)
18:44alyssa==53478== by 0x5DDD3DF: ??? (in /home/alyssa/mesa/build/src/gallium/targets/dri/asahi_dri.so)
18:44alyssa==53478== Block was alloc'd at
18:44alyssa==53478== at 0x4889F94: calloc (in /usr/libexec/valgrind/vgpreload_memcheck-arm64-linux.so)
18:44alyssa==53478== by 0x5FC1687: ??? (in /home/alyssa/mesa/build/src/gallium/targets/dri/asahi_dri.so)
18:44alyssause-after-free, yeet0
18:45alyssaoh because we're reference counting BOs, but not resources
18:46alyssaso this is busted
18:50alyssahow did this work before
18:51linaI had enough crazy memory safety issues today... (see MR, you have no idea how long I spent tracking down that state tracker/blitter interaction one and deciding how to fix it...)
18:52linaYou get to figure this one out ^^
18:52alyssahaha alright :)
18:52alyssaget some sleep, sis :)
18:54alyssaoh ugh ok this is very broken
18:54alyssain panfrost
18:55linaOh yeah the fence stuff is definitely broken in panfrost (at least the missing null checks in fence_reference) FYI
18:56linaStory of explicit sync: cargo cult code from other drivers then realize they're all broken ^^ (first panfrost, then iris/xe)
18:57opticron has quit [Ping timeout: 480 seconds]
18:57alyssas/explicit//
19:03alyssaI'm just going to remove the hash table
19:03alyssaAvoid a can of worms that way
19:03alyssaProbably faster too
19:07linaalyssa: Make sure to keep the writer_syncobj clear in that path somehow, that's important (without it you could end up importing the wrong fence on export, we don't want to import if there is no current writer)
19:07linaI'm not sure how bad that would be but I can see it being anywhere from a perf issue to a deadlock issue.
19:31alyssayeah, I'm copying your existing logic
19:31alyssajust changing the data structures to something with a lot less pointers
19:32linaGot it ^^
19:32linaAnd I should get some sleep
19:32alyssato avoid reference count hell
19:32linanini~!
19:32alyssaglcts: ../src/gallium/drivers/asahi/agx_state.h:399: agx_writer_add: Assertion `(*value) == 0 && "there should be no existing writer"' failed.
19:32alyssai love assert
19:34alyssaoh ugh this looks like an actual issue
19:50alyssa"GPU is crashed cannot submit"
19:56alyssado I win a prize
19:56alyssa(pushed mesa branch "splat", run deqp-gles31 with deqp-runner and it crashed in a few minutes)
19:59alyssa(It's a broken Mesa branch but that shouldn't be able to provoke.. that)
20:05alyssaoddly deqp-gles3 results look great
20:08Misthios has quit [Quit: Misthios]
20:09Misthios has joined #asahi-gpu
20:10Misthios has quit []
20:11Misthios has joined #asahi-gpu
20:17alyssafixed the packing issue
20:17alyssatoo much late night code for me :p
20:32nyilas has joined #asahi-gpu
20:32jhan has joined #asahi-gpu
20:40jhan has quit [Ping timeout: 480 seconds]
20:50tertu has joined #asahi-gpu
20:55opticron has joined #asahi-gpu
21:06louisadamian has joined #asahi-gpu
21:33Tramtristyou have gained +5 follows from yesterday in case you havent checked :p
21:36louisadamian has quit [Remote host closed the connection]
21:37louisadamian has joined #asahi-gpu
21:43alyssalina: trying to run piglit is crashing the gpu within minutes
21:43alyssakernel bugs? firmware bugs? both?
21:44alyssadon't think this is related to explicit sync, I think the mesa side regressed a few months back and exposed the kernel bug
21:44alyssabut it's kinda hard to find/fix the mesa regression when the gpu crashes
21:44louisadamian has quit [Remote host closed the connection]
21:45alyssa(and without being able to run piglit I can't tell how we're doing regressionwise since the last release. piglit results were crap then too but not "the gpu crashes in minutes" bad)
21:46alyssaat any rate I have a wip branch that gets deqp{2,3,31} where they should be
21:46alyssaso piglit is the next (and hopefully snafu before we can release
21:47louisadamian has joined #asahi-gpu
21:48alyssathere's a LOT of changes this time around.. I don't want regressions :)
21:48alyssaand i keep reading about people saying the gpu support is flawless, I want to be sure it stays that way ^^
22:00jhan has joined #asahi-gpu
22:11louisadamian has quit [Remote host closed the connection]
22:16jhan has quit [Ping timeout: 480 seconds]
23:20stsmwg has joined #asahi-gpu
23:32jhan has joined #asahi-gpu
23:35stsmwg has quit [Quit: Lost terminal]
23:40jhan has quit [Ping timeout: 480 seconds]
23:47possiblemeatball has joined #asahi-gpu
23:51jhan has joined #asahi-gpu
23:55aomizu has joined #asahi-gpu
…