#
…

2023-07-05

00:20crabbedhaloablut has quit [Read error: Connection reset by peer]
00:21crabbedhaloablut has joined #asahi-gpu
00:27nsklaus has quit [Remote host closed the connection]
01:49chadmed has quit [Quit: Konversation terminated!]
01:50chadmed has joined #asahi-gpu
01:59balrog has quit [Ping timeout: 480 seconds]
02:01yuyichao has joined #asahi-gpu
02:12alyssalina: 42 files changed, 2624 insertions(+), 469 deletions(-)
02:13alyssawow, sure whittled down my patch stack a lot T_T
02:17mikelee has joined #asahi-gpu
02:19mikelee has quit [Remote host closed the connection]
02:42alyssaeh, we'll get through it :)
02:51mikelee has joined #asahi-gpu
02:53alyssaOpened some more "hopefully not at all scary" MRs to whittle things down
02:55alyssalina: https://gitlab.freedesktop.org/asahi/mesa/-/merge_requests/95 <-- this one you definitely want to look @
03:47lina has quit [Remote host closed the connection]
03:49lina has joined #asahi-gpu
03:50linaLooks like my previous kernel fix didn't fix the Firefox hangs, it just made them not spin a CPU core... ;;
03:59PyroPeter_ has joined #asahi-gpu
04:01cyrozap has quit [Quit: ZNC 1.8.2+deb2+b1 - https://znc.in]
04:01PyroPeter has quit [Ping timeout: 480 seconds]
04:01cyrozap has joined #asahi-gpu
04:06cyrozap has quit []
04:06cyrozap has joined #asahi-gpu
04:09mikelee has quit [Read error: Connection reset by peer]
04:09cyrozap has quit []
04:10cyrozap has joined #asahi-gpu
04:27mikelee has joined #asahi-gpu
05:13mikelee has quit [Remote host closed the connection]
05:14mikelee has joined #asahi-gpu
05:15mikelee has quit [Remote host closed the connection]
05:15mikelee has joined #asahi-gpu
05:15mikelee has quit [Remote host closed the connection]
05:16mikelee has joined #asahi-gpu
05:24crabbedhaloablut has quit []
05:26crabbedhaloablut has joined #asahi-gpu
05:37cylm has quit [Ping timeout: 480 seconds]
05:40mikelee has quit [Remote host closed the connection]
05:40mikelee has joined #asahi-gpu
05:41mikelee has quit [Remote host closed the connection]
05:41mikelee has joined #asahi-gpu
06:05hightower3 has quit []
06:13mikelee_ has joined #asahi-gpu
06:20mikelee has quit [Ping timeout: 480 seconds]
07:38bisko has joined #asahi-gpu
08:16chadmed has quit [Remote host closed the connection]
08:29nsklaus has joined #asahi-gpu
08:42chadmed has joined #asahi-gpu
10:03chadmed has quit [Remote host closed the connection]
10:03chadmed has joined #asahi-gpu
11:36linajannau: Do you remember what test was failing with the TVB thing?
11:37jannaulina: no
11:37jannauat least not out of my head
11:38linaDo you remember if it was dEQP-GLES3 or dEQP-GLES31 or something else?
11:38jannaulina: deqp-gles3/functional/shaders/builtin_functions/precision/matrixcompmult/highp_fragment/mat3
11:43jannaulina: with asahi.initial_tvb_size=64 and alyssa's es31 branch only atomic tests failed
11:45jannauyes, es31. the top 2 commits were unfinished tests for the heisenbug
11:49jannauyes, that looks like atomics tests
11:50jannauI suspect some atomics tests might succeed spuriously
11:50linaThese are the tests that failed without "atomic" in the name but I think they're related?
11:51linadEQP-GLES31.functional.compute.basic.image_barrier_multiple,Fail
11:51linadEQP-GLES31.functional.compute.basic.ssbo_cmd_barrier_multiple,Fail
11:51linadEQP-GLES31.functional.compute.indirect_dispatch.gen_in_compute.multiple_groups,Fail
11:51linadEQP-GLES31.functional.compute.indirect_dispatch.upload_buffer.multiple_groups,Fail
11:51linadEQP-GLES31.functional.image_load_store.early_fragment_tests.early_fragment_tests_stencil,Fail
11:51linadEQP-GLES31.functional.image_load_store.early_fragment_tests.no_early_fragment_tests_depth,Fail
11:52jannaulooks familiar, Alyssa looked at some/all of them and said they use atomics
12:09bisko has quit [Ping timeout: 480 seconds]
12:19linaTL;DR today I fixed a bunch of kernel stuff, both more sizing/param issues (I have testable test data for it now!) and also implemented the sync TVB growth event though I don't think we actually use/need it now, it was getting triggered by some bad settings I think. But you can test it with ASAHI_MESA_DEBUG=synctvb if you want (see open MR).
12:19linaI got through GLES3 in that mode but GLES31 deadlocked or something, I'll look into it later with lockdep (might be a regression with synctvb, if so we don't care). Without any overrides things work ^^
12:38alyssaany progress on our bugs? or is this strategy of "well, it definitely fixed SOMETHING!"
12:40possiblemeatball has quit [Quit: Quit]
13:01chadmed has quit [Ping timeout: 480 seconds]
13:03chadmed has joined #asahi-gpu
13:39linaalyssa: Well the one jannau ran into with the TVB size at least... and I'm pretty sure I fixed other stuff too, yes ^^
13:53alyssait's my understanding the atomics with clustering isssue and the heisenbug are the hot items
13:53alyssa(blocking CTS)
13:54maria has quit [Ping timeout: 480 seconds]
13:55maria has joined #asahi-gpu
14:13WindowPa- has joined #asahi-gpu
14:13WindowPain has quit [Read error: Connection reset by peer]
15:11_jannau__I think only atomics are blocking CTS on t600x, the heisenbug didn't reproduce there
15:16cylm has joined #asahi-gpu
15:21Dementor has quit [Remote host closed the connection]
15:23Dementor has joined #asahi-gpu
15:36nela0 has joined #asahi-gpu
15:38nela has quit [Ping timeout: 480 seconds]
15:38nela0 is now known as nela
15:51alyssa_jannau__: let's do process of elimination
15:52alyssa_jannau__: can you git clone github.com/dougallj/applegpu on macOS on t600x, run https://paste.debian.net/1285009 and send the output?
15:54alyssa(Or anyone else with an M1 Pro/Max/Ultra)
16:04dottedmagm1max, if it matters
16:08balrog has joined #asahi-gpu
16:11alyssadottedmag: thanks
16:12alyssathat indeed sets 1 bit that isn't set on M1
16:16alyssaI wrote a patch and it wedged my M1
16:16alyssathat sounds like we're on the right track
16:16alyssa reason: AfFault,
16:17alyssaooh and green screen when shutting down
16:17alyssadid i wedge dcp too? powerful mesa patch
16:21alyssaGive that a try on t600x
16:21alyssawith the atomics tests
16:21alyssaif that changes anything, let me know drm_asahi_params_global::num_clusters_total so I can do the proper runtime detection and not wedge t8103
16:27jannaualyssa: all atomic tests in es31 pass now on t6002
16:27alyssaAlright.
16:27alyssaI mean, that's good but
16:28alyssaI had a hunch already that bit 45 controls cache of some kind
16:28alyssathis.. adds to that suspicion >:)
16:33* alyssadigs up her notes on cache bits
16:34jannaunum_clusters_total should be 2 / 4 / 8 for m1 pro / max / ultra
16:36jannauthe lower core count versions have always cores in all clusters disabled
16:37jannaushall I start a full CTS run?
16:38alyssajannau: Let me get you a non-stupid branch to use
16:42Guest4807 has quit [Quit: Bridge terminating on SIGTERM]
16:42rhysmdnz has quit [Quit: Bridge terminating on SIGTERM]
16:43alyssakicking off CTS on M1 to break sure I didn't break anything in the meantime
16:43alyssajannau: alyssa/mesa:agx/es31-v2
16:45Jamie has joined #asahi-gpu
16:45rhysmdnz has joined #asahi-gpu
16:45Jamie is now known as Guest5071
16:45alyssaThat doesn't have your wide_color fix on the mesa side, and the 10-bit thing still hasn't landed upstream anyway, so unfortunately still an x11_egl run
16:45alyssabut... should pass, probably, maybe? O:)
16:46alyssa(Strictly you could do a Wayland run with your patch and point to it and say "look it's your bug" but it would complicate things.)
16:47alyssaanyway, M1 CTS is running
16:47jannauwe could do wayland and disable 10-bit formats
16:47alyssaeasier just to do the x11_egl run for now
16:47alyssaanyway, M1 CTS is running
16:47jannaum1 ultra x11_egl CTS is running as well
16:47alyssawho will win? =D
16:47alyssanominally the ultra but the CTS is completely single threaded CPU bound
16:48jannauthe failing cts run was slower on the ultra than a succesful one on m1
16:48jannauiirc 41 minutes vs. 28 min
16:49alyssadeqp-runner does a bunch of debug stuff for failing tests, idk what the real CTS runner is doing
16:53balrog has quit [Quit: Bye]
16:56balrog has joined #asahi-gpu
17:37yuka has quit [Remote host closed the connection]
17:40alyssaM1 seemed to finish
17:40yuka has joined #asahi-gpu
17:40jannaufinished here as well after 52 minutes, still failed
17:41jannauKHR-GLES31.core.draw_indirect.advanced-twoPass-transformFeedback-arrays
17:41alyssawhat this time?
17:41jannauKHR-GLES31.core.draw_indirect.advanced-twoPass-transformFeedback-elements
17:41jannaufor various configs
17:42jannaufor all configs
17:43alyssaInteresting
17:43alyssadefinitely passes on M1
17:44jannauerror is in all cases: "(x,y)= (0,0). Color RGBA(0,0,0,1) is different than expected RGBA(0.1,0.2,0.3,1)"
17:44alyssajannau: The potentially "spicy" part of those tests is that they feed transform feedback results into the inidrect draw
17:45alyssasince we dispatch xfb with the VDM (i.e. as vertex shaders with no fragment output), that's a forward VDM->VDM dependency, that does not require a full flush of the batch, but it does require a memory barrier
17:45alyssaI'm wondering if this is morally the same issue as the atomics
17:46alyssathe barrier we're using is strong enough to flush a cluster but not the whole system
17:46alyssaSee line 2892 of agx_state.c
17:47alyssatry setting more bits (this will require adding extra fields in asahi/lib/cmdbuf.xml)
17:47alyssaIf someone has the ability to run wrap.dylib + agxdecode against t600x this corresponds to a metal memory barrier, that's where I got those magic bits in the first place
17:48alyssabut in this case since we know what we're looking for bruteforcing might be faster anyway lol
18:10alyssajannau: any luck?
18:15jannaualyssa: no. I think we're looking at different agx_state.c files. is line 2892 'cfg.unknown_30 = frag_tex_count >= 4;'?
18:17alyssaam i wrong the wrong branch
18:18alyssaok i was definitely way off
18:19alyssahopefully that makes more sense X_X
18:19alyssaso sorry about that X_X
18:20jannauadding more bits at the end or in the gaps? I suppose the answer is yes
18:36ella-0alyssa: tests i'm using are VK.api.copy_and_blit.core.image_to_image.simple_tests.*, branch is https://gitlab.freedesktop.org/Ella-0/mesa/-/tree/wip/agxv/meta
18:37alyssaella-0: thx
18:41jannaualyssa: becomes a flake with bit (4 and) 24 set
18:42alyssathat's still setting all the other bits too?
18:43jannauyes, those are in addition
18:43alyssainteresting
18:43lena6 has quit [Remote host closed the connection]
18:43alyssamay be necessary to see what metal does after all :|
18:45alyssa[encoder memoryBarrier: MTLBarrierScopeBuffers afterStages: MTLRenderStageVertex beforeStages: MTLRenderStageVertex]
18:46alyssaor maybe need even more bits set. IDK
18:47jannauall good with bits 4, 21-26 in addition
18:50alyssayou going to write the patch then?
18:50jannauif bit 27 would not break it I'd say bit 20 to 27 are for each cluster
18:50jannaulet me first test if I need all bits
18:52alyssaI could easily believe that
18:54jannauminimal set seems to be 4, 24, 26
18:56alyssaare we sure it doesn't flake like that, though?
19:03jannaureasonably sure. no fail in 1000 repeats and if omit any of the bits it fails in 50% +/- 10
19:05* alyssaeagerly awaits the patch.. and the qpas ;)
19:11jannaulet me test if that breaks something on m1/m2
19:26alyssatangentially related, I don't think G13J exists
19:26alyssaIf I'm not mistaken m1 ultra is G13D
19:27alyssaI may be mistaken of course
19:30jannauyou're correct. I think I mixed it up with the SoC family codename
19:33alyssaapple refers to the collection as G13X which isn't really right either :-D
19:42systwi has quit [Ping timeout: 480 seconds]
20:03jannau"56/56 sessions passed, conformance test PASSED"
20:28systwi has joined #asahi-gpu
20:36systwi_ has joined #asahi-gpu
20:37systwi has quit [Ping timeout: 480 seconds]
20:41c10l has quit [Quit: Bye o/]
20:42c10l has joined #asahi-gpu
20:47systwi has joined #asahi-gpu
20:51systwi_ has quit [Ping timeout: 480 seconds]
20:57jannauand the next target where fractional scaling breaks the x11_egl CTS. let's hope my fix gets merged soon so we can run under native wayland
21:05mkurz has joined #asahi-gpu
21:25systwi_ has joined #asahi-gpu
21:25systwi has quit [Remote host closed the connection]
21:54cylm_ has joined #asahi-gpu
21:56cylm has quit [Ping timeout: 480 seconds]
22:23systwi has joined #asahi-gpu
22:27systwi_ has quit [Ping timeout: 480 seconds]
22:46compassion178 has joined #asahi-gpu
22:49compassion17 has quit [Ping timeout: 480 seconds]
22:49compassion178 is now known as compassion17
23:06pyropeter1 has joined #asahi-gpu
23:08PyroPeter_ has quit [Ping timeout: 480 seconds]
23:35nsklaus has quit [Remote host closed the connection]
…