#
…

2023-06-15

00:09ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
00:12ourdumbfuture has joined #asahi-gpu
00:34mlp has quit [Read error: Connection reset by peer]
00:35mlp has joined #asahi-gpu
02:06odak_ has joined #asahi-gpu
02:21odak_ has quit [Quit: odak_]
02:21odak_ has joined #asahi-gpu
02:30linaalyssa: What I call clusters is what Apple calls "mGPU"s. M1 has just one. We need to test whatever you end up with on t600x and check whether the core ID register is globally unique or we need to get the mGPU ID somewhere else and add it in... hopefully it's unique.
03:18ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
03:24PyroPeter has joined #asahi-gpu
03:26pyropeter3 has quit [Ping timeout: 480 seconds]
03:31odak_ has quit [Quit: odak_]
03:32odak_ has joined #asahi-gpu
03:54maria has quit [Remote host closed the connection]
03:55maria has joined #asahi-gpu
04:00possiblemeatball has joined #asahi-gpu
04:36linaalyssa: BTW, the firefox register spilling thing was some crazy shadertoy, not webrender, so don't worry about that ^^
04:46possiblemeatball has quit [Quit: Quit]
05:11systwi has quit [Ping timeout: 480 seconds]
05:16cylm has joined #asahi-gpu
05:32hightower2 has quit [Ping timeout: 480 seconds]
06:07systwi has joined #asahi-gpu
06:17nsklaus has joined #asahi-gpu
06:52odak_ has quit [Quit: odak_]
07:11nimprod3l has joined #asahi-gpu
07:29cylm has quit [Ping timeout: 480 seconds]
07:59cylm has joined #asahi-gpu
08:07mkurz has quit [Quit: Konversation terminated!]
08:07mkurz has joined #asahi-gpu
08:28nimprod3l has quit [Quit: Leaving]
08:40hightower2 has joined #asahi-gpu
09:21cylm has quit [Ping timeout: 480 seconds]
09:29djorz has quit [Ping timeout: 480 seconds]
10:20zocker has quit [Quit: Lost terminal]
10:21alyssalina: right, ok. Interestingly Apple gets the "# of concurrent tiles per core" parameter from a uniform, and doesn't ever explicitly use "# of cores" in the shader
10:21alyssameaning, the shaders they produce are portable across GPUs regardless of configuration
10:21alyssa(just needing the driver to use the right constants for the implementation)
10:22alyssaOhh.. I guess the hierarchy in Apple terms is like
10:22alyssa"core -> cluster -> mGPU"?
10:22alyssawhereas you call those
10:22alyssa"frags -> core -> cluster"?
10:26alyssaas for the shadertoy, yes if you run shadertoys you get to pick up the pieces
10:27hightower3 has joined #asahi-gpu
10:27alyssa has quit [Quit: leaving]
10:33hightower2 has quit [Ping timeout: 480 seconds]
10:40ourdumbfuture has joined #asahi-gpu
10:43cylm has joined #asahi-gpu
10:58ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
11:02ourdumbfuture has joined #asahi-gpu
11:26hightower3 has quit [Ping timeout: 480 seconds]
11:34linaalyssa: Frags is Apple terminology and I don't actually know what it means
11:34linaWhat is a cluster according to apple? I thought I made that one up
11:35linaAs far as I'm concerned it's just core -> mGPU/cluster
11:37ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
11:41linaalyssa: For compute context switching, they have a table as part of their data structure that maps core numbers, and my understanding is that is used for GPUs with disabled cores to avoid allocating memory that will never be used, since I'm guessing the core ID has gaps on those. So I'm surprised they don't use it for eMRT?
11:42linaAlso on M2 Pro/Max machines, all GPUs have "missing" cores due to the way they designed the chips... (clusters have uneven numbers of cores and they implemented it by just having phantom cores logically which are always marked disabled)...
11:45ourdumbfuture has joined #asahi-gpu
12:03odak_ has joined #asahi-gpu
12:13chadmed has quit [Remote host closed the connection]
12:14chadmed has joined #asahi-gpu
13:01mlp has quit [Read error: Connection reset by peer]
13:02mlp has joined #asahi-gpu
13:02alyssa has joined #asahi-gpu
13:03alyssaI'm glad I took the easy way out because I'm getting close to something workable
13:03linaNice! ^^
13:03alyssaHopefully I can finish that off today, it's only 9am lol
13:03alyssa(I've been up for a few hours I don't know my sleep schedule these days)
13:03ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
13:03alyssaAfter this, the only fail left will be due to nir_opt_preamble speculating loads
13:04alyssalina: what's the story with soft fault?
13:04alyssa[Although probably I should figure out how to fix that properly since Adreno doesn't have soft fault and it also uses the same broken pass..]
13:05linaIt's a kernel parameter, asahi.fault_control=0xb
13:05alyssaso just need a bit of uapi to tell userspace whether it's on or not?
13:06ourdumbfuture has joined #asahi-gpu
13:06linaYeah ^^
13:06linaIt should just be a compatible feature flag
13:09linaIf you want to test, just flip the kernel arg and plumb it into a new feature you fake on or something, I can put it into the kernel next time I sit down to hack on it ^^
13:09linaI think different bits in fault_control might enable different things, but I don't know what's what yet...
13:09linaI think there's probably one global enable and then specific feature bits or something...
13:10linaI'm not sure if we should just pass that to userspace as-is as an integer without understanding it though, it is parsed by the firmware as far as I know, not straight a hardware register.
13:10alyssaInteresting
13:10alyssaHow'd you find that 0xb value?
13:10alyssa("macOS uses it")
13:10linaThat's the macO-yes
13:11linaAnd I guess 0 was disable/fault
13:11alyssaOk, how'd you fgure out that's fault control and that setting not-0xb is safe?
13:11linaGuesswork...
13:11alyssasensible
13:11alyssa13:10 < lina> I'm not sure if we should just pass that to userspace as-is as an integer without understanding it though, it is parsed by the firmware as far as I know, not straight a hardware register.
13:11linaThere are a few "interesting" clusters of flags in the giant initdata structures and that's one of them
13:11linaWell away from all the power management gunk
13:11alyssaIMO, do *not* pass through the value unparsed if it's not 100% hardware defined
13:12alyssasince that's a ticking uapi given the fw randomly changes on uprevs
13:12linaYeah, so then just a feature flag, only problem is once we do understand it we might want to split that into separate feature flags... but that's okay I guess
13:12linaI guess if == 0xb then set the flag, otherwise don't set it, for now
13:12linaSince we know that's the safe macOS behavior
13:12alyssaYeah, ++
13:13alyssaif we add more bits later... ugly UAPI > breaking userspace
13:13linaYou could try adding this to the kernel yourself, maybe try a bit of kernel rust~ ✨
13:14alyssaMight be fun
13:14alyssaThis isn't blocking CTS though
13:15alyssasince we need to support the !softfault case even if we end up shipping softfault
13:15alyssa(if only because it makes Mesa debugging massively easier)
13:16linaTBH I want to enable softfault by default, it's going to make things a lot more stable for people I bet, since hard faults can kill unrelated jobs and also trigger other existing bugs...
13:16alyssaWhat I mean is, even if literally every other user in existence has soft fault enabled, I want hard faults on my dev machines for my own use, and I don't want enabling faults to regress CTS lol
13:17linaI suspect that setting only affects stuff fetched from shader cores (textures and device loads), so I think we'll still get hard faults for other brokenness like bad command streams or TVB problems
13:17linaAnd yeah, fair ^^
13:18alyssasoft fault was the hummus in my macOS mesa debug PITA
13:18linawwwwwwwww
13:19linaHey I know you enjoyed my mesa fault decoding code ^^
13:20alyssaIt's growing on me like bacteria
13:25alyssaso... might have a branch passing CTS on Monday
13:26alyssanot sure if you'll be able to kick off CTS runs next week though if your machines are in random boxes ;)
13:26alyssaahead-of-schedule is a good thing, hey
13:26linaThe machines are not going in boxes ^^
13:27linaBut I'm more worried about whether I'll be in a state to kick off CTS runs and fix the remaining M2 Pro/Max things...
13:28linaSee signal...
13:32mort_5 is now known as mort_
13:34_jannau__are the complicated requirements for CTS runs? I could cover M1 Max, M1 Ultra and M2
13:35alyssa_jannau__: ooh, that might be helpful indeed instead of me doing 1 run and Lina doing 17000 :-p
13:35alyssaand no, not particularly complicated
13:36alyssaI mean the CTS is kinda annoying but. "git clone this repo, cmake build it, build mesa, run ./cts-runner, wait a zillion hours, zip up all the files produced and send me them"
13:37alyssathere's a little bit more work to do the actual submission but that's on my end
13:38odak_ has quit [Quit: odak_]
13:38odak_ has joined #asahi-gpu
13:42alyssaThe most annoying part is that the CTS is slow
13:42alyssaGLES3.1 on Mali took ~12h if I remember
13:42alyssayour M1 Max will not have that problem :-p
13:43alyssa(CTS is mostly CPU-bound. and the "real" CTS runs are basically single-threaded.)
13:44alyssa(For development and CI, anholt's deqp-runner shards the CTS across all available CPUs, which scales almost linearly. But that's not valid for official CTS submissions.)
13:57ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
14:09hightower2 has joined #asahi-gpu
14:52alyssaok.. my blending is not working with multiple draws together
14:52alyssadraw #1 writes to an image, draw #2 reads it
14:53alyssai'm putting wait_pix's in but that doesn't seem to be strong enough
14:53alyssaI guess I'll look at how rasterizer order groups work in metal
14:54alyssa(This functionality is definitely supposed to work)
14:58odak_ has quit [Quit: odak_]
14:58odak_ has joined #asahi-gpu
15:01nimprod3l has joined #asahi-gpu
15:16ourdumbfuture has joined #asahi-gpu
15:23odak_ has quit [Quit: odak_]
15:42i509vcbHmm agx_border says customBorderColorWithoutFormat can't be supported but zink requires that to be supported. Shader hack later I guess could cover that?
15:57mkurz has quit [Ping timeout: 480 seconds]
16:00nimprod3l has quit [Quit: Leaving]
16:23alyssai509vcb: zink doesn't really require customBorderColorWithoutFormat, it already doesn't use it on turnip for perf reasons
16:23alyssaeasy to patch Zink to do what we need
16:23alyssaThe bigger issue is DXVK/VKD3D
16:26mkurz has joined #asahi-gpu
16:41rhysmdnz has quit [Quit: Bridge terminating on SIGTERM]
16:41Guest2821 has quit [Quit: Bridge terminating on SIGTERM]
16:45Jamie has joined #asahi-gpu
16:45rhysmdnz has joined #asahi-gpu
16:45Jamie is now known as Guest3184
16:48possiblemeatball has joined #asahi-gpu
17:09ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
17:24mkurz has quit [Remote host closed the connection]
17:26alyssaMORE spicy barriers? ah bah oui!
17:27alyssa2, 2, 10
17:27alyssa3, 2, 10
17:27alyssa2, 1, 10
17:27alyssathe 0, 0, 4 is notably different here
17:28alyssaalso interesting that it's inserting these after the write, implicitly
17:28alyssaalso, sample_mask 255, 1 at the top of the program
17:28alyssaalso never did figure out what wait_pix 512, 3 was about
17:29alyssathe barriers are only inserted when raster order groups are used
17:29alyssaflush the write to other invocations, I guess
17:30ourdumbfuture has joined #asahi-gpu
17:33alyssa0, 0, 4 is the barrier I had discovered by bruteforce
17:33alyssaand called flush_memory_to_texture
17:34alyssabut actually the problem is HSR, i think
17:35alyssayeah ok lol
17:35alyssayep ok blend works now
17:39alyssanext up, fix partial renders
17:39alyssawhich should be trivial
17:47alyssastill seeing some HSR related issues..
17:47possiblemeatball has quit [Quit: Quit]
17:47alyssabut fewer at least
17:47alyssagood enough to play supertuxkart with all render targets spilled to memory
17:48alyssa(aka "my Apple M1 is an immediate mode renderer now!")
18:00cylm has quit [Read error: Connection reset by peer]
18:15alyssaOoh tasty new bit in the PBE descriptor for sRGB
18:15alyssalooks like bit 125
18:16aafeke_ has joined #asahi-gpu
18:32alyssalina: I am now observing that MSRTT makes eMRT significantly more annoying
18:32alyssaso yet another reason to defer that until we have a use case
18:32alyssathe only thing I know that uses it is WebGL in Chromium
18:33zzywysm has joined #asahi-gpu
18:33alyssafor iPhone SoCs, it'd be a BIG deal for perf there
18:33alyssafor the desktop class stuff, possibly we can eat the it
18:41alyssaOh uffffffffff
18:42alyssafragment shader side effects + sample shading = 🤯
18:56ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
19:01aafeke_ has quit [Quit: aafeke_]
19:01aafeke_ has joined #asahi-gpu
19:09ourdumbfuture has joined #asahi-gpu
19:23mkurz has joined #asahi-gpu
19:57possiblemeatball has joined #asahi-gpu
20:16aafeke_ has quit [Quit: aafeke_]
20:17aafeke_ has joined #asahi-gpu
20:27aafeke_ has quit [Quit: aafeke_]
20:27aafeke_ has joined #asahi-gpu
20:31alyssamany hacks involved but passed the KHR-GLES tests with compression + preambles disabled
20:33hightower2 has quit [Remote host closed the connection]
20:48alyssa has quit [Quit: leaving]
21:36aafeke_ has quit [Ping timeout: 480 seconds]
22:19possiblemeatball has quit [Quit: Quit]
22:44skoobasteeve has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
22:45skoobasteeve has joined #asahi-gpu
22:47skoobasteeve has quit []
22:48skoobasteeve has joined #asahi-gpu
23:16nsklaus has quit [Ping timeout: 480 seconds]
…