#
…

2023-07-25

01:01hightower3 has joined #asahi-gpu
01:05mikelee has joined #asahi-gpu
01:07fenix144 has quit [Remote host closed the connection]
01:07mikelee has quit [Remote host closed the connection]
01:08hightower2 has quit [Ping timeout: 480 seconds]
01:09Z750 has quit [Quit: Ping timeout (120 seconds)]
01:10Z750 has joined #asahi-gpu
01:12mikelee has joined #asahi-gpu
01:31mikelee has quit [Remote host closed the connection]
01:33mikelee has joined #asahi-gpu
01:33jeisom has quit [Quit: Leaving]
01:36mikelee has quit [Remote host closed the connection]
01:39nepeat has quit [Read error: Connection reset by peer]
01:39ave36309 has quit [Read error: Connection reset by peer]
01:40ave36309 has joined #asahi-gpu
01:40nepeat has joined #asahi-gpu
01:48jeisom has joined #asahi-gpu
01:48ave36309 has quit [Read error: Connection reset by peer]
01:48ave36309 has joined #asahi-gpu
01:52ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
01:53ourdumbfuture has joined #asahi-gpu
02:02jeisom has quit [Ping timeout: 480 seconds]
02:21ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
02:26ourdumbfuture has joined #asahi-gpu
03:00ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
03:08cylm has quit [Ping timeout: 480 seconds]
03:14crabbedhaloablut has joined #asahi-gpu
03:53nst has quit [Ping timeout: 480 seconds]
03:58nst has joined #asahi-gpu
05:32Z750 has quit [Quit: Ping timeout (120 seconds)]
05:33Z750 has joined #asahi-gpu
05:57Cyrinux9474 has joined #asahi-gpu
05:59Cyrinux947 has quit [Ping timeout: 480 seconds]
09:04c10l563 has quit []
09:04c10l563 has joined #asahi-gpu
09:16bisko has joined #asahi-gpu
09:27darkapex has quit [Remote host closed the connection]
09:29darkapex has joined #asahi-gpu
09:34Z750 has quit [Quit: Ping timeout (120 seconds)]
09:35Z750 has joined #asahi-gpu
10:26nsklaus has joined #asahi-gpu
10:39c10l563 has quit []
10:40c10l563 has joined #asahi-gpu
11:32alyssa has joined #asahi-gpu
11:32alyssalina: when you get a chance, could you prioritize review of https://gitlab.freedesktop.org/asahi/mesa/-/merge_requests/111 ? it's not too big, but
11:33alyssa1. it's churny around UAPI bits so I expect it will conflict badly with whatever you work on next, better merge it sooner then ;)
11:33alyssa2. it's mucking with UAPI bits and I don't have a ton of confidence in my ability to get the logic right, so I do need an actual review from you and not a rubberstamp ;)
11:34alyssai'm also unsure if we want to go further than that
11:34alyssabe able to queue up arbitrary sequences of VDM and CDM work
11:34alyssaI could see this interfacing needing to be extended yet again for efficient mipmap generation (with the blit PDM queue?)
11:35linaalyssa: Sure ^^
11:35linaWhat I said with the barriers works right?
11:37alyssaif you hate the code and would rather rewrite yourself given the new requirements I'm fine with that outcome too ;)
11:37alyssacurrently the only new requirement is combined CDM+VDM submit for graphics batches
11:38alyssabut I can imagine a lot of optimizations where we'd want the more general thing
11:38alyssa(potentiallly even submitting multiple batches at once to reduce user/kernel roundtrips)
11:38jeisom has joined #asahi-gpu
11:39linaYeah, there was always room for that
11:39linaE.g. when doing mipmap gen we can totally do that
11:39linaBasically when flushing writers we'd batch up all the submits and send it all at once
11:39alyssaYeah, makes sense
11:39alyssaI don't need that for my secret project
11:40lina"secret" project ^^
11:40alyssabut if you're going to rework the code rather than merge what I have, might be worth architecting around those cases too? idk
11:45alyssait turns out there are a lot of yaks I need to shave :>
11:45alyssaannoying one will be rewriting a big chunk of resource management code
11:45alyssabut AGXV needs that too so.. not a loss :>
11:45alyssa(turns out I unwittingly hardcoded a GLism)
11:46linaOops ^^
11:46alyssahavent' decided how I want to untangle that yet
11:50linaalyssa: Unrelated, but have you seen the Google Sheets issue? I haven't had a chance to look at it yet, just curious whether you have any ideas. Basically as you scroll the visible parts of the canvas gets seriously blurred proportional to the distance to cell A1.
11:51linaIt looks like it's sampling a texture linearly with texcoords that are noticeably off or something like that
11:51lina"fp16" you say, but "no16" doesn't fix it!
11:53linaIt could also be that it's intended to be point sampling and it's the linear sampling that shouldn't be there...
11:53alyssaI do remember hitting that once
11:53alyssa11:51 lina | "fp16" you say, but "no16" doesn't fix it!
11:53alyssano16 doesn't fully disable fp16
11:54alyssano16 tells the GLSL compiler not to generate fp16 instructions
11:54alyssathe backend AGX compiler may still choose to do itself regardless of flag
11:54alyssayou'll wan tto comment out the nir_lower_mediump_io call in agx_compile.c for that
11:56linaalyssa: No change
11:57alyssaIt does reek of fp16 coord bugs so idk
11:59linawait wait
11:59* alyssawait-waits
12:01alyssameanwhile my submit refactor thing seems to break both firefox and chrome, brrrr
12:01linaThat DID fix it *together* with fp16
12:01lina*Âno16
12:02linaOhh and one more thing!
12:02linaNo wait, that was just me being confused
12:03linaDo you think this is a driver bug? If not I think we need to extend no16 to also drop that lowering, and enable it for browsers because there is waaaaay too much broken WebGL out there...
12:03alyssaHard to say
12:03alyssaAt this point I would be fine with the latter.
12:03alyssaI don't think I've seen a single workload where no16 has moved the fps meter.
12:03ourdumbfuture has joined #asahi-gpu
12:04alyssamight help for, like, terrible shadertoys.
12:04alyssabut.. hm
12:04alyssalina: What browser is this on?
12:04linafirefox
12:04alyssaand what API are you rendering with?
12:04alyssaGL or GLES i mean
12:04mkurz has quit [Ping timeout: 480 seconds]
12:04linaI know we try to skip fp16 for texcoords but I'm guessing there's some ALU in there which breaks the trivial pass
12:04alyssabig GL doesn't even have mediump
12:04linaGLES I assume? Whatever firefox does...
12:05alyssa12:04 lina | I know we try to skip fp16 for texcoords but I'm guessing there's some ALU in there which breaks the trivial pass
12:05alyssayeah.. that was working around a sway bug
12:11linaOK yeah, I see texture sampling that uses min() or max() of some input and something else
12:11linaSo that would break the heuristic
12:11alyssaI can extend the heuristic more but also Google Sheets is presumably broken but also
12:11alyssaI'm curious how this works on any arm chromebooks..
12:13linaFound the bad shader (probably), it's exactly as expected (sampling after a clamp()) and everything is mediump
12:14alyssaso deffo a google sheets bug. wheeee.
12:15alyssaand you're sure this is GLES?
12:15alyssaactually it doesn't matter, webgl is gles
12:15_jannau__does chrome os uses angle on arm chromebooks?
12:16alyssa_jannau__: I don't think they used to. IDK what they do now
12:19_jannau__the paragraph starting with "Portions of the ANGLE shader compiler are used as a shader validator and translator by WebGL implementations ..." on https://github.com/google/angle sounds like angle could be used and hides this issue
12:19Cyrinux9474 has quit []
12:21alyssaI'd believe it
12:25alyssalina: I'm having troubles reproducing
12:25Cyrinux9474 has joined #asahi-gpu
12:29linaalyssa: Maybe needs HiDPI or page zoom? (guess)
12:31alyssawfm on my mini, maybe hidpi then idk
12:31* alyssaknee deep into debugging the new subimt
12:33alyssaworking in debug builds but not release, ah bah oui
12:33alyssaDRM_IOCTL_ASAHI_SUBMIT render failed: Invalid argument (800x600 tile 32x32 layers 1 samples 1)
12:33alyssachill asahi
12:33linaHappens for me with 100% display scale too (on KDE)...
12:33linaalyssa: Remember there is a submit count limit (64 IIRC per ioctl)
12:34alyssaI'm just submitting 1
12:34linaBarriers?
12:35alyssa--well I got that wrong too but still doesnt work :p
12:36* alyssapushes nonetheless
12:40alyssalina: what browser versions?
12:40alyssaalso can you repro in chromium? what about chromium with the `--use-gl=egl` flag?
12:43tirr has quit [Quit: ZNC 1.8.2 - https://znc.in]
12:44tirr has joined #asahi-gpu
12:45linaalyssa: Doesn't happen on Chromium with or without that flag
12:45alyssaINTERESTING
12:45linaFirefox 115.0.2
12:45alyssafirefox 102 here
12:45alyssalet me try on fedora
12:46linaOooooold...
12:46* alyssaapt upgrades
12:47_jannau__could be firefox esr
12:48alyssalina: very subtle but I think I can repro on fedora
12:48alyssaok definitely hit it on fedora
12:48alyssafirefox 115
12:48alyssaso... firefox regression maybe?
12:49alyssai'm fine no16 driconf cone of shaming firefox
12:49Z750 has quit [Quit: Ping timeout (120 seconds)]
12:49Z750 has joined #asahi-gpu
12:53linaWe still need to make that skip that mediump io thing
12:55linaThe shaders I dumped on Chromium don't have precision qualifiers at all...
12:56linaI think Firefox might be using ANGLE?
12:57alyssay'know i've always hated webgl
12:59_jannau__are both firefox and chromium reporting webgl 2 as supported?
12:59linaAh, not on Chromium...
12:59chadmed has quit [Quit: Konversation terminated!]
12:59chadmed has joined #asahi-gpu
13:02linaAh, but it does with --use-gl=egl
13:03linaWait but it says WebGL2: Software?!
13:03linaI think... it might be trying to use a Vulkan backend???
13:04alyssaok this is too cursed for me, i'm tapping out
13:05alyssaDRM_IOCTL_ASAHI_SUBMIT render failed: Invalid argument (1903x519 tile 32x32 layers 1 samples 1)
13:05alyssawhat do you want from me
13:05alyssaand why only in release builds of mesa
13:05linaUninitialized data?
13:05alyssaplausible
13:05linaAre you clearing the cmdbuf before filling it in?
13:06lina(All relevant structs)
13:06alyssalooks like it
13:09linaalyssa: OK, --ozone-platform=wayland is what gets me WebGL2 using the GLES backend properly
13:10tirr has quit [Quit: ZNC 1.8.2 - https://znc.in]
13:10tirr has joined #asahi-gpu
13:13linaI think I found the same shader and the precisions are right here...
13:29alyssavalgrind is clean?wat?
13:31linaalyssa: I think sheets might not be using WebGL at all??? Maybe this is an internal firefox bug....
13:32alyssaI'd buy that
13:32alyssathough it's worth noting I also only saw this in sheets
13:32linaYeah but it could be due to the Canvas blits it does or something
13:36linaI found it...
13:37linaThose are the broken shaders, prior to the silly optimization path. It uses mediump on the FS...
13:37alyssaNice work!
13:38linaI'll file a bug
13:38linaLet me test with Mesa shader replacement that it works...
13:39alyssahmm our heuristic should force highp for that shader
13:39alyssathat doesn't look like the same shader
13:40alyssaunless the clamping is added later
13:41linaYes but it's not that one, it's a similar one
13:41linaLet me see where the clamp comes from
13:41lina*that* one
13:44alyssalina: at this point I'm leaning towards no16'ing firefox given that mobile gpus are clearly not a priority
13:44alyssaand webgl perf in firefox sucks anyway afaik
13:44linaYeah...
13:44cylm has joined #asahi-gpu
13:44linaAs long as we fix it to actually work ^^
13:44alyssawhich part
13:44alyssaboth of those are firefox problems and decidedly not us problems
13:45linano16 not working
13:45alyssaoh, sure
13:45alyssaI'll write the patch for that if you can look at my submit thing?
13:51Dementor has quit [Remote host closed the connection]
13:52alyssauntested. probably works
13:53Dementor has joined #asahi-gpu
13:59cwabbottalyssa: lina: I bet you someone at some point had to get that code working on utgard (mali 400)
14:00cwabbottthat only supports mediump on fragment shaders, except that there's a special hack to pass varying coordinates directly to the texture unit in highp for exactly this case
14:00linaBut I'm guessing that doesn't work when there's ALU in between exactly like this case?
14:00linaIf so this should be broken there too...
14:01cwabbottoh yeah, was looking at the wrong shader
14:01linaWe have a hack in the driver that detects the easy case with no ALU and does exactly that, upgrades to fp32, because this is a common issue
14:01linaBut it's the second shader with the clamp() that breaks that
14:02cwabbottyeah, makes sense
14:02cwabbottyou can blame utgard for that issue
14:05fenix144 has joined #asahi-gpu
14:06fenix144 has quit [Remote host closed the connection]
14:08alyssacwabbott: there are a lot of things I'd like to blame Utgard for.
14:09cwabbottI wonder how it works on freedreno, given that it does all the same mediump stuff
14:10alyssaPlausible that it's broken there
14:10alyssaThis seemingly is a recent Firefox regression and so far only repros on Google Sheets
14:10alyssagiven most freedreno is on chrome, shrug
14:35amarioguy has joined #asahi-gpu
14:36ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
14:38ourdumbfuture has joined #asahi-gpu
14:44mkurz has joined #asahi-gpu
14:53bisko has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
15:19mkurz has quit [Ping timeout: 480 seconds]
15:24ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
15:28jonmasters__ has quit []
15:29jonmasters has joined #asahi-gpu
16:19ourdumbfuture has joined #asahi-gpu
16:58mkurz has joined #asahi-gpu
17:28darkapex has quit [Remote host closed the connection]
17:43rhysmdnz has quit [Quit: Bridge terminating on SIGTERM]
17:43Guest6702 has quit [Quit: Bridge terminating on SIGTERM]
17:47rhysmdnz has joined #asahi-gpu
17:47Jamie has joined #asahi-gpu
17:47Jamie is now known as Guest6958
18:19compassion1785 has quit [Read error: Connection reset by peer]
18:20compassion1785 has joined #asahi-gpu
18:29compassion1785 has quit [Quit: lounge quit]
18:30jeisom has quit [Quit: Leaving]
18:31jeisom has joined #asahi-gpu
18:38compassion1785 has joined #asahi-gpu
19:05ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
19:12ourdumbfuture has joined #asahi-gpu
19:23ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
19:40fenix144 has joined #asahi-gpu
20:17fenix144 has quit [Remote host closed the connection]
20:19fenix144 has joined #asahi-gpu
20:22fenix144 has quit [Remote host closed the connection]
20:24fenix144 has joined #asahi-gpu
20:26alyssalina: ...and the Oscar goes to clang's __builtin_dump_struct
20:26alyssafor making obvious the UAF I had due to a dangling pointer to a stack varibale
20:26alyssaella-0: thanks for showing me that trick :>
20:27ourdumbfuture has joined #asahi-gpu
20:27alyssaman I have rust envy right now
20:27alyssawould've caught this one without breaking a sweat
20:27alyssaobvious violation of lifetime rules
20:29fenix144 has quit [Remote host closed the connection]
20:29fenix144 has joined #asahi-gpu
20:30alyssafixed the stupid.
20:30alyssaMR is ready for review.
20:34mikelee has joined #asahi-gpu
20:49mikelee has quit [Ping timeout: 480 seconds]
21:11fenix144 has quit [Remote host closed the connection]
21:13fenix144 has joined #asahi-gpu
21:23ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
21:23ourdumbfuture has joined #asahi-gpu
21:43fenix144 has quit [Remote host closed the connection]
21:46fenix144 has joined #asahi-gpu
21:49MaryI tried to run vk-cts with agxv/main to kind of look what I could work on and got a complete freeze when I was away;;
21:53Mary"asahi: asahi::slotalloc::SlotAllocator<asahi::event::EventInner>: out of slots, blocking" 10 times, well I guess that's why ><
22:13fenix144 has quit [Remote host closed the connection]
22:18fenix144 has joined #asahi-gpu
22:48Mary(testing on 6.3.8-450.asahi.fc38.aarch64)
23:10fenix144 has quit [Remote host closed the connection]
23:11fenix144 has joined #asahi-gpu
23:12nsklaus has quit [Ping timeout: 480 seconds]
23:31fenix144 has quit [Remote host closed the connection]
23:37fenix144 has joined #asahi-gpu
23:37mikelee has joined #asahi-gpu
23:39fenix144 has quit [Remote host closed the connection]
23:46cylm_ has joined #asahi-gpu
23:46ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
23:48ourdumbfuture has joined #asahi-gpu
23:48cylm has quit [Ping timeout: 480 seconds]
23:50fenix144 has joined #asahi-gpu
23:51mikelee has quit [Ping timeout: 480 seconds]
23:53alyssaMary: Your website looks.. familiar ^_^
23:59Maryalyssa: I kind of wanted to go with something simple... and it ended up quite similar to yours and the other mary ^^'
23:59alyssaI meant the css ;D
23:59Mary(and Cantarell is a pretty font ngl)
23:59alyssaoh yeah i love Cantarell
23:59alyssawas my favourite font for a while
23:59alyssaflirted with Lato for a little while but I think I'm back to Cantarell
…