#asahi-gpu · 2023-03-04 04:27
alyssaif there's an export, that doesn't trigger a flush so it needs to use the existing syncobj#asahi-gpu · 2023-03-04 04:26
alyssathe only dicey thing is that newfangled BO syncobj#asahi-gpu · 2023-03-04 04:26
alyssain terms of waiting for a BO to be ready, there's now a new writer to wait on so that's right#asahi-gpu · 2023-03-04 04:26
alyssaand ... in terms of batch ordering, the implicit in-order guarantee from the kernel says that's totally fine#asahi-gpu · 2023-03-04 04:25
alyssabecause agx_writer_add asserts there's no existing writer, but if there's something submitted but not finished there's a writer there#asahi-gpu · 2023-03-04 04:25
alyssa"i'm not convinced" see the "XXX: is this kosher"#asahi-gpu · 2023-03-04 04:23
alyssa(i've seen that hash table be a big chunk %wise of CPU overhead for the driver)#asahi-gpu · 2023-03-04 04:23
alyssaotoh the dynarray approach is almost certainly lighterweight than the hash table so not inclined to drop the commit#asahi-gpu · 2023-03-04 04:22
alyssaI'm starting to think there's a much dumber solution I should've gone with, actually..#asahi-gpu · 2023-03-04 04:22
alyssabut it does fix the use after free#asahi-gpu · 2023-03-04 04:22
alyssai'm not convinced I did it right#asahi-gpu · 2023-03-04 04:22
alyssaand then the batch tracking code I rebased on top of it (easy conflicts)#asahi-gpu · 2023-03-04 04:21
alyssatowards the top there is a commit "asahi: Use a dynarray for writers" which I'm on the fence about no pun intended#asahi-gpu · 2023-03-04 04:21
alyssalina: I pushed an agx/next#asahi-gpu · 2023-03-04 01:18
alyssathat shouldve been asahi-dev#asahi-gpu · 2023-03-04 01:18
alyssaer#asahi-gpu · 2023-03-04 01:18
alyssa(and also geometry + tessellation shaders)#asahi-gpu · 2023-03-04 01:18
alyssafor mesh shaders#asahi-gpu · 2023-03-04 01:18
alyssachadmed: down the line I think I'm going to want 13.x firmware for AGX #asahi-gpu · 2023-03-03 21:48
alyssaand i keep reading about people saying the gpu support is flawless, I want to be sure it stays that way ^^#asahi-gpu · 2023-03-03 21:48
alyssathere's a LOT of changes this time around.. I don't want regressions :)#asahi-gpu · 2023-03-03 21:46
alyssa^last)#asahi-gpu · 2023-03-03 21:46
alyssaso piglit is the next (and hopefully snafu before we can release#asahi-gpu · 2023-03-03 21:46
alyssaat any rate I have a wip branch that gets deqp{2,3,31} where they should be#asahi-gpu · 2023-03-03 21:45
alyssa(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)#asahi-gpu · 2023-03-03 21:44
alyssabut it's kinda hard to find/fix the mesa regression when the gpu crashes#asahi-gpu · 2023-03-03 21:44
alyssadon't think this is related to explicit sync, I think the mesa side regressed a few months back and exposed the kernel bug#asahi-gpu · 2023-03-03 21:43
alyssakernel bugs? firmware bugs? both?#asahi-gpu · 2023-03-03 21:43
alyssalina: trying to run piglit is crashing the gpu within minutes#asahi-gpu · 2023-03-03 21:43
alyssa:D#asahi-gpu · 2023-03-03 21:33
Tramtristyou have gained +5 follows from yesterday in case you havent checked :p#asahi-gpu · 2023-03-03 20:17
alyssatoo much late night code for me :p#asahi-gpu · 2023-03-03 20:17
alyssafixed the packing issue#asahi-gpu · 2023-03-03 20:05
alyssaoddly deqp-gles3 results look great#asahi-gpu · 2023-03-03 19:59
alyssa(It's a broken Mesa branch but that shouldn't be able to provoke.. that)#asahi-gpu · 2023-03-03 19:56
alyssa(pushed mesa branch "splat", run deqp-gles31 with deqp-runner and it crashed in a few minutes)#asahi-gpu · 2023-03-03 19:56
alyssado I win a prize#asahi-gpu · 2023-03-03 19:56
alyssalina: https://rosenzweig.io/splat-03-03.txt#asahi-gpu · 2023-03-03 19:50
alyssayeet#asahi-gpu · 2023-03-03 19:50
alyssa"GPU is crashed cannot submit"#asahi-gpu · 2023-03-03 19:34
alyssaoh ugh this looks like an actual issue#asahi-gpu · 2023-03-03 19:32
alyssai love assert#asahi-gpu · 2023-03-03 19:32
alyssaweeee#asahi-gpu · 2023-03-03 19:32
alyssaglcts: ../src/gallium/drivers/asahi/agx_state.h:399: agx_writer_add: Assertion `(*value) == 0 && "there should be no existing writer"' failed.#asahi-gpu · 2023-03-03 19:32
linanini~!#asahi-gpu · 2023-03-03 19:32
alyssanini#asahi-gpu · 2023-03-03 19:32
alyssato avoid reference count hell#asahi-gpu · 2023-03-03 19:32
linaAnd I should get some sleep#asahi-gpu · 2023-03-03 19:32
linaGot it ^^#asahi-gpu · 2023-03-03 19:31
alyssajust changing the data structures to something with a lot less pointers