#asahi-gpu · 2023-07-20 23:33
alyssathere's still a pile of vk_meta work to go to pass the relevant dEQP-VK.api.* tests#asahi-gpu · 2023-07-20 23:33
alyssa- i guess that was most of it, I did other non-AGXV stuff today idk#asahi-gpu · 2023-07-20 23:32
alyssa- fixed silly index buffer bug, which makes a big chunk of dEQP-VK.glsl.* start passing#asahi-gpu · 2023-07-20 23:32
alyssa- vkCmdCopyBufferToImage stuff. or maybe it was ImageToBuffer. regardless, the non-array tests are passing now#asahi-gpu · 2023-07-20 23:32
alyssaToday in AGXV:#asahi-gpu · 2023-07-19 22:44
alyssaso not yet in a spot for .. productive fullspeed ahead development, if that makes sense#asahi-gpu · 2023-07-19 22:43
alyssastill a huge pile of tests failing for reasons completely obvious to what they're testing#asahi-gpu · 2023-07-19 22:40
alyssa- primitive restart implemented so dEQP-VK.pipeline.monolithic.input_assembly.* is passing now#asahi-gpu · 2023-07-19 22:39
alyssa- line width fixed#asahi-gpu · 2023-07-19 22:38
alyssa- misc deqp fixes .. dEQP-VK.glsl.operator.* is passing now, as is dEQP-VK.pipeline.monolithic.logic_op.*#asahi-gpu · 2023-07-19 22:37
alyssa- STORAGE_TEXEL_BUFFER support (used in that implementation)#asahi-gpu · 2023-07-19 22:37
alyssa- WIP vk_meta implementation of vkCmdCopyImageToBuffer#asahi-gpu · 2023-07-19 22:37
alyssaToday in AGXV:#asahi-gpu · 2023-07-19 11:35
alyssabut I would definitely like us to know the r/e side at least :)#asahi-gpu · 2023-07-19 11:35
alyssathe conversion part#asahi-gpu · 2023-07-19 11:35
alyssaFWIW I don't care to support that in Mesa most likely#asahi-gpu · 2023-07-19 11:34
alyssaEasiest way will be if you watch a film with whole-system hv on, I think#asahi-gpu · 2023-07-19 11:33
alyssaMetal doesn't support that publicly#asahi-gpu · 2023-07-19 11:33
alyssaNo#asahi-gpu · 2023-07-19 11:18
alyssathough none of this seems that weird#asahi-gpu · 2023-07-19 11:17
alyssaI'll look when I'm awake in a few hours#asahi-gpu · 2023-07-19 11:16
alyssaIDK#asahi-gpu · 2023-07-19 11:16
alyssaok#asahi-gpu · 2023-07-19 11:16
alyssais this with Mesa patches?#asahi-gpu · 2023-07-19 11:16
alyssabut it says Layout Compressed?#asahi-gpu · 2023-07-19 11:16
alyssaI see#asahi-gpu · 2023-07-19 11:16
alyssaooooooooh#asahi-gpu · 2023-07-19 11:16
alyssaunkB1 is a PBE block write, the descriptor is the first one (the PBE descriptor)#asahi-gpu · 2023-07-19 11:15
alyssawhat are you talking about compression=0#asahi-gpu · 2023-07-19 11:14
alyssato disassemble TODO.unkB1#asahi-gpu · 2023-07-19 11:14
alyssahttps://github.com/dougallj/applegpu/pull/41#asahi-gpu · 2023-07-19 11:14
alyssaSpicy#asahi-gpu · 2023-07-19 02:50
alyssa(I hope.)#asahi-gpu · 2023-07-19 02:50
alyssathis code has passed All The Tests and all the important stuff was discussed on stream so any review at this point would likely be finding local style things and not Big Problems#asahi-gpu · 2023-07-19 02:49
alyssalina: i'm about to go to sleep. i'll merge tomorrow my time barring discovered issues. that gives you all of today JST to review if you want. or to do nothing if you want.#asahi-gpu · 2023-07-18 20:44
alyssawhat value of $TIME?#asahi-gpu · 2023-07-18 20:43
alyssashould I wait $TIME to let you review?#asahi-gpu · 2023-07-18 20:43
alyssalina: should I merge https://gitlab.freedesktop.org/asahi/mesa/-/merge_requests/108 ?#asahi-gpu · 2023-07-17 20:32
alyssawhy can't mesa just do gettimeofday and call it a day?#asahi-gpu · 2023-07-17 20:32
alyssalina: silly question but, if CPU and GPU time are synched, why do we need anything asahi specific at all?#asahi-gpu · 2023-07-17 15:25
alyssakarolherbst: shrug.#asahi-gpu · 2023-07-17 15:22
alyssayep#asahi-gpu · 2023-07-15 03:06
alyssaI think we might be talking about differnet things#asahi-gpu · 2023-07-15 03:05
alyssalina: hm?#asahi-gpu · 2023-07-15 02:06
alyssa(and separate system values in the shader for "dispatched wg size" and "actual current wg size")#asahi-gpu · 2023-07-15 02:06
alyssakarolherbst: native support for non-uniform workgroup sizes on m1 :~)#asahi-gpu · 2023-07-15 01:25
alyssabut what it's doing I.. cannot understnad#asahi-gpu · 2023-07-15 01:24
alyssaand then with concurrent mode, https://developer.apple.com/documentation/metal/mtlcomputecommandencoder/2966553-memorybarrierwithscope?language=objc becomes non-nop#asahi-gpu · 2023-07-15 01:24
alyssamight be worth poking at sometime, maybe get some more uapi unknowns discovered#asahi-gpu · 2023-07-15 01:23
alyssait seems the default is Serial, which is sensible, but MTLDispatchTypeConcurrent generates some... genuinely bizarre stuff