#asahi-gpu · 2023-03-03 18:57
alyssas/explicit//#asahi-gpu · 2023-03-03 18:54
alyssain panfrost#asahi-gpu · 2023-03-03 18:54
alyssaoh ugh ok this is very broken#asahi-gpu · 2023-03-03 18:52
alyssaget some sleep, sis :)#asahi-gpu · 2023-03-03 18:52
alyssahaha alright :)#asahi-gpu · 2023-03-03 18:50
alyssahow did this work before#asahi-gpu · 2023-03-03 18:46
alyssaso this is busted#asahi-gpu · 2023-03-03 18:45
alyssaoh because we're reference counting BOs, but not resources#asahi-gpu · 2023-03-03 18:44
alyssause-after-free, yeet0#asahi-gpu · 2023-03-03 18:44
alyssa==53478== by 0x5FC1687: ??? (in /home/alyssa/mesa/build/src/gallium/targets/dri/asahi_dri.so)#asahi-gpu · 2023-03-03 18:44
alyssa==53478== at 0x4889F94: calloc (in /usr/libexec/valgrind/vgpreload_memcheck-arm64-linux.so)#asahi-gpu · 2023-03-03 18:44
alyssa==53478== Block was alloc'd at#asahi-gpu · 2023-03-03 18:44
alyssa==53478== by 0x5DDD3DF: ??? (in /home/alyssa/mesa/build/src/gallium/targets/dri/asahi_dri.so)#asahi-gpu · 2023-03-03 18:44
alyssa==53478== at 0x4887B40: free (in /usr/libexec/valgrind/vgpreload_memcheck-arm64-linux.so)#asahi-gpu · 2023-03-03 18:43
alyssa==53478== Address 0x7b45180 is 128 bytes inside a block of size 480 free'd#asahi-gpu · 2023-03-03 18:43
alyssa==53478== at 0x5FBD6F0: ??? (in /home/alyssa/mesa/build/src/gallium/targets/dri/asahi_dri.so)#asahi-gpu · 2023-03-03 18:43
alyssa==53478== Invalid read of size 8#asahi-gpu · 2023-03-03 18:43
alyssasigh, C#asahi-gpu · 2023-03-03 18:43
alyssahappens running dEQP-GLES31.functional.compute.*#asahi-gpu · 2023-03-03 18:43
alyssalina: rsrc->bo can be corrupted in agx_batch_cleanup#asahi-gpu · 2023-03-03 18:31
alyssaThat's not what we're looking for but it gives a sense that AGX might have an analogous reg#asahi-gpu · 2023-03-03 18:30
alyssathe addressmode. When set to 0, HW will be seemless, ie, ignore addressmode, filter between faces at the edges/corners#asahi-gpu · 2023-03-03 18:30
alyssaPixel data master. Disable filtering over edges/corners for CEM. When set to 1, HW will be seemfull, ie, always stay in the current map, always honour#asahi-gpu · 2023-03-03 18:30
alyssaTAG_CEMEDGE_DONTFILTER#asahi-gpu · 2023-03-03 18:30
alyssaRGX has this bit#asahi-gpu · 2023-03-03 18:27
alyssaI can implement the workarounds if needed but I'd much rather not#asahi-gpu · 2023-03-03 18:27
alyssareally hoping AGX has a TPU control register that can toggle this behaviour tbh#asahi-gpu · 2023-03-03 18:20
alyssaRGX has a TPU control register, maybe AGX does too, idk#asahi-gpu · 2023-03-03 18:19
alyssaIDK#asahi-gpu · 2023-03-03 18:19
alyssait's entirely possible that there's, like, a bit in some control register to toggle this behaviour#asahi-gpu · 2023-03-03 18:18
alyssaso it's hard to say what's actually happening here#asahi-gpu · 2023-03-03 18:18
alyssaand I don't see the Apple Metal blob fixing this up#asahi-gpu · 2023-03-03 18:18
alyssaHowever.. I don't see any references to this issue in Metal docs nor MoltenVK nor ANGLE#asahi-gpu · 2023-03-03 18:17
alyssawhich is very much not great but not catastrophic#asahi-gpu · 2023-03-03 18:17
alyssathe only solution I can come up with for Vulkan is duplicating samplers to have a regular version and a cube map version, and toggle between the two in the shader depending on whether loading from a cube map or not#asahi-gpu · 2023-03-03 18:16
alyssafor VK it'll be a bit more annoying because we have no way of knowing whether a given sampler is for a cube map or not#asahi-gpu · 2023-03-03 18:16
alyssa'll be ok#asahi-gpu · 2023-03-03 18:15
alyssaadd a hack in mesa/st to fix up the wrap mode and it#asahi-gpu · 2023-03-03 18:15
alyssaif this is indeed the issue -- I haven't confirmed it, it's basically just from staring at the fails/pass list -- then for GL it shouldn't be too bad to fix#asahi-gpu · 2023-03-03 18:15
alyssawhich is all kinds of wrong#asahi-gpu · 2023-03-03 18:14
alyssabut I think for some reason AGX (and Metal?) honour the repeat wrap mode even for seamless cube maps#asahi-gpu · 2023-03-03 18:14
alyssaseamless cube maps are supposed to ignore the wrap mode in GL and VK#asahi-gpu · 2023-03-03 18:14
alyssaI *think* I figured out what the deqp-gles3 cube maps fails are about, *maybe*#asahi-gpu · 2023-03-03 18:14
alyssalina: btw on the subject of known fails#asahi-gpu · 2023-03-03 18:13
alyssaDoing things better than the industry? :~)#asahi-gpu · 2023-03-03 18:13
alyssalina: but isn't that our thing?#asahi-gpu · 2023-03-03 18:13
alyssaUh oh!#asahi-gpu · 2023-03-03 18:12
alyssasome regressions for me to fix but back to a world of sanity at least#asahi-gpu · 2023-03-03 18:12
alyssa^31#asahi-gpu · 2023-03-03 18:12
alyssadeqp results look good again