#
✕ nick:alyssa
#asahi-gpu · 2025-03-04 19:14
alyssaKHR-GLES31.core.shader_storage_buffer_object.advanced-write-fragment-cs
#asahi-gpu · 2025-03-04 19:14
alyssathe GL case is super bizarre
#asahi-gpu · 2025-03-04 19:14
alyssa(on g14x)
#asahi-gpu · 2025-03-04 19:14
alyssaincorrectly setting this flag breaks a couple of cases in both the gl & vk cts when vertex clustering is enabled
#asahi-gpu · 2025-03-04 19:13
alyssa`no_clear_pipeline_textures` is definitely not that
#asahi-gpu · 2025-03-04 19:13
alyssainvestigating some of the weird uapi bits
#asahi-gpu · 2025-02-26 13:30
alyssajannau: Nice!
#asahi-gpu · 2025-02-26 13:30
alyssasven: depends, is the soc/apple maintainer amenable to rust? ;)
#asahi-gpu · 2025-02-25 22:23
alyssaYay!
#asahi-gpu · 2025-02-25 21:47
alyssajannau: hey progress is progress =D
#asahi-gpu · 2025-02-25 21:42
alyssajannau: amazing work :)
#asahi-gpu · 2025-02-25 21:39
alyssaI'm now confident in my understanding of the different tilings so https://lore.kernel.org/asahi/20250225-apple-twiddled-modifiers-v2-1-cf69729e87f6@rosenzweig.io/T/#u sent out
#asahi-gpu · 2025-02-25 21:39
alyssaok, after that... significant diversion
#asahi-gpu · 2025-02-25 21:38
alyssayay!
#asahi-gpu · 2025-02-25 19:43
alyssavk, you're up next
#asahi-gpu · 2025-02-25 19:43
alyssaok, gl cts passing now with twiddled images
#asahi-gpu · 2025-02-25 19:35
alyssa /* AGX: Even the exceptions have exceptions */
#asahi-gpu · 2025-02-25 19:34
alyssacts likes this now
#asahi-gpu · 2025-02-25 19:34
alyssaOK yes here we go
#asahi-gpu · 2025-02-25 19:34
alyssa...except for compressed?
#asahi-gpu · 2025-02-25 19:20
alyssabut that doesn't matter for perf since levels are cache line padded anyway
#asahi-gpu · 2025-02-25 19:20
alyssabut worse then divide-then-round for RAM usage since there's extra padding sometimes
#asahi-gpu · 2025-02-25 19:20
alyssa(perf is the same as divide-then-round, the "correct" approach)
#asahi-gpu · 2025-02-25 19:20
alyssathis is better than always doing round-then-divide, because at least within a level stuff is nice and packed so you access fewer cache lines --> faster perf
#asahi-gpu · 2025-02-25 19:19
alyssabut divide-then-round to figure out the actual tile sizes in the levels
#asahi-gpu · 2025-02-25 19:19
alyssaI *think* it does round-then-divide to figure out where levels start
#asahi-gpu · 2025-02-25 19:18
alyssaohhhh
#asahi-gpu · 2025-02-25 19:12
alyssai guess maybe major vs minor axis is the actual issue here
#asahi-gpu · 2025-02-25 19:11
alyssaer, 65x63
#asahi-gpu · 2025-02-25 19:09
alyssaso why is 65x65 different than 3x7?
#asahi-gpu · 2025-02-25 19:06
alyssafound a fault for a 65x65x6 cubemap case with r32f .. ok
#asahi-gpu · 2025-02-25 19:01
alyssaserious wat
#asahi-gpu · 2025-02-25 18:55
alyssathen why did we break dEQP-GLES31.functional.copy_image.non_compressed.viewclass_128_bits.rgba32f_rgba32ui.texture2d_to_cubemap
#asahi-gpu · 2025-02-25 18:55
alyssawait what
#asahi-gpu · 2025-02-25 18:54
alyssathat makes sense, I guess
#asahi-gpu · 2025-02-25 18:54
alyssaso it shares the same weird special case as gpu-tiled
#asahi-gpu · 2025-02-25 18:54
alyssa...except for compressed
#asahi-gpu · 2025-02-25 18:45
alyssathere we go
#asahi-gpu · 2025-02-25 18:39
alyssanice
#asahi-gpu · 2025-02-25 18:38
alyssacool
#asahi-gpu · 2025-02-25 18:38
alyssaso the hw is a bit more efficient than my original guess.
#asahi-gpu · 2025-02-25 18:38
alyssainstead of rounding first then dividng
#asahi-gpu · 2025-02-25 18:38
alyssamipmapping first divides, then rounds
#asahi-gpu · 2025-02-25 18:38
alyssaoh, sure
#asahi-gpu · 2025-02-25 18:33
alyssaoh dEQP-GLES3.functional.texture.filtering.2d.sizes.3x7_linear_mipmap_linear fails too
#asahi-gpu · 2025-02-25 18:19
alyssaunk mipmappe dis set
#asahi-gpu · 2025-02-25 18:19
alyssafirst,last level = 0..2
#asahi-gpu · 2025-02-25 18:19
alyssaditto for the texture descriptor
#asahi-gpu · 2025-02-25 18:18
alyssaeven though obviously they are
#asahi-gpu · 2025-02-25 18:18
alyssaPBE descriptors looks almost like we expect. only weird thing is page-aligned layers isn't set?!

plus ›