#asahi-gpu · 2023-07-05 13:53
alyssa(blocking CTS)#asahi-gpu · 2023-07-05 13:53
alyssait's my understanding the atomics with clustering isssue and the heisenbug are the hot items#asahi-gpu · 2023-07-05 13:52
alyssaI heh#asahi-gpu · 2023-07-05 13:39
linaalyssa: Well the one jannau ran into with the TVB size at least... and I'm pretty sure I fixed other stuff too, yes ^^#asahi-gpu · 2023-07-05 12:38
alyssa:-D#asahi-gpu · 2023-07-05 12:38
alyssaany progress on our bugs? or is this strategy of "well, it definitely fixed SOMETHING!"#asahi-gpu · 2023-07-05 12:38
alyssawoot#asahi-gpu · 2023-07-05 12:19
linaI 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 ^^#asahi-gpu · 2023-07-05 12:19
linaTL;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).#asahi-gpu · 2023-07-05 12:17
alyssacorrect#asahi-gpu · 2023-07-05 11:52
jannaulooks familiar, Alyssa looked at some/all of them and said they use atomics#asahi-gpu · 2023-07-05 11:51
linadEQP-GLES31.functional.image_load_store.early_fragment_tests.no_early_fragment_tests_depth,Fail#asahi-gpu · 2023-07-05 11:51
linadEQP-GLES31.functional.image_load_store.early_fragment_tests.early_fragment_tests_stencil,Fail#asahi-gpu · 2023-07-05 11:51
linadEQP-GLES31.functional.compute.indirect_dispatch.upload_buffer.multiple_groups,Fail#asahi-gpu · 2023-07-05 11:51
linadEQP-GLES31.functional.compute.indirect_dispatch.gen_in_compute.multiple_groups,Fail#asahi-gpu · 2023-07-05 11:51
linadEQP-GLES31.functional.compute.basic.ssbo_cmd_barrier_multiple,Fail#asahi-gpu · 2023-07-05 11:51
linadEQP-GLES31.functional.compute.basic.image_barrier_multiple,Fail#asahi-gpu · 2023-07-05 11:50
linaThese are the tests that failed without "atomic" in the name but I think they're related?#asahi-gpu · 2023-07-05 11:50
jannauI suspect some atomics tests might succeed spuriously#asahi-gpu · 2023-07-05 11:49
jannauyes, that looks like atomics tests#asahi-gpu · 2023-07-05 11:45
jannauyes, es31. the top 2 commits were unfinished tests for the heisenbug#asahi-gpu · 2023-07-05 11:43
jannaulina: with asahi.initial_tvb_size=64 and alyssa's es31 branch only atomic tests failed#asahi-gpu · 2023-07-05 11:38
jannaulina: deqp-gles3/functional/shaders/builtin_functions/precision/matrixcompmult/highp_fragment/mat3#asahi-gpu · 2023-07-05 11:38
linaDo you remember if it was dEQP-GLES3 or dEQP-GLES31 or something else?#asahi-gpu · 2023-07-05 11:37
jannauat least not out of my head#asahi-gpu · 2023-07-05 11:37
jannaulina: no#asahi-gpu · 2023-07-05 11:36
linajannau: Do you remember what test was failing with the TVB thing?#asahi-gpu · 2023-07-05 03:50
linaLooks like my previous kernel fix didn't fix the Firefox hangs, it just made them not spin a CPU core... ;;#asahi-gpu · 2023-07-05 02:55
alyssalina: https://gitlab.freedesktop.org/asahi/mesa/-/merge_requests/95 <-- this one you definitely want to look @#asahi-gpu · 2023-07-05 02:53
alyssaOpened some more "hopefully not at all scary" MRs to whittle things down#asahi-gpu · 2023-07-05 02:42
alyssaeh, we'll get through it :)#asahi-gpu · 2023-07-05 02:13
alyssawow, sure whittled down my patch stack a lot T_T#asahi-gpu · 2023-07-05 02:12
alyssalina: 42 files changed, 2624 insertions(+), 469 deletions(-)#asahi-gpu · 2023-07-04 18:58
alyssaof course Rust's default sort is stable :~)#asahi-gpu · 2023-07-04 18:49
alyssastares at qsort_r calls that are not stable.#asahi-gpu · 2023-07-04 18:49
alyssaI wonder if that's causing our evil flake.#asahi-gpu · 2023-07-04 18:49
alyssajannau: I've just made the horrible discovery that the compiler is not deterministic. ugh ugh ugh.#asahi-gpu · 2023-07-04 16:35
alyssahappy to look into that as well#asahi-gpu · 2023-07-04 16:35
alyssaadmittedly not sure how to best express that with vk_meta#asahi-gpu · 2023-07-04 16:35
alyssa2. heh, you should be able to do raw copies into the tilebuffer and never need to convert#asahi-gpu · 2023-07-04 16:34
alyssa1. I can take a look if you push the branch + give me the test caselist#asahi-gpu · 2023-07-04 16:23
ella-0Some copies that aren't tile aligned are resulting in faults, and I've done nothing about properly converting values in shaders and am simply relying on images having mutable formats which isn't correct.#asahi-gpu · 2023-07-04 16:08
alyssa[I have a ~rough idea of how it 'should' work but you -- actually working on it -- can tell me why I'm wrong ;) ]#asahi-gpu · 2023-07-04 16:07
alyssawhich parts of meta copies are you finding tricky?#asahi-gpu · 2023-07-04 16:07
alyssaella-0: heh. If you pass it on I can work on it. Not /immediately/ but hey#asahi-gpu · 2023-07-04 15:54
ella-0You can fit in whereever you like but this copies stuff is what's really been whittling away at my motivation and I wanna be mean and pass it onto someone else#asahi-gpu · 2023-07-04 15:52
ella-0alyssa: With agxv stuff I've been working on implementing copies in meta. I've not been making the best progress with it.#asahi-gpu · 2023-07-02 17:22
jannausigh, was just a wrong parameter#asahi-gpu · 2023-07-02 17:18
jannauthat doesn't explain why glcts has problems though#asahi-gpu · 2023-07-02 17:16
jannauarch's /usr/bin/eglinfo has RPATH set. sigh