#
✕ nick:alyssa
#asahi-gpu · 2022-11-25 22:16
alyssafor the fillModeNonSolid feature in Vulkan too
#asahi-gpu · 2022-11-25 22:15
alyssaalso, gl_GL_POLYGONS
#asahi-gpu · 2022-11-25 22:14
alyssaif we even can.
#asahi-gpu · 2022-11-25 22:14
alyssatodo: figure out how to GL_QUADS with glPolyMode(front=GL_LINE, back=GL_FILL)
#asahi-gpu · 2022-11-25 03:10
alyssanice~
#asahi-gpu · 2022-11-24 23:15
alyssa+1, thanks!
#asahi-gpu · 2022-11-24 23:03
alyssa"I just tried this random DirectX game and it doesn't work very good, here is an apitrace, please fix, cc @someone_banned_for_code_of_conduct_violations" -- not good
#asahi-gpu · 2022-11-24 23:02
alyssa"This workload used to work on the stable tree, but it no longer works on the stable tree. I've bisected the failure down to commit sha 123456. If I revert that commit on top of main, it works again." -- good
#asahi-gpu · 2022-11-24 23:02
alyssa"Reproducible segfault in this workload, here is the backtrace, here is the line number that's failing, I am using the stable tree with no patches" -- good
#asahi-gpu · 2022-11-24 23:01
alyssareally appreciates bug reports that she can actually do something about
#asahi-gpu · 2022-11-24 23:01
alyssathank you for the helpful bug report :-)
#asahi-gpu · 2022-11-24 23:01
alyssabind_sampler_states doesn't zero [nr_samplers, PIPE_MAX_SAMPLERS) so can get non-null garbage samplers leading to a use-after-free (segfault derefencing sampler) or a buffer overflow (writing samplers[] out of bounds, we only allocated room for nr_samplers of them) or both
#asahi-gpu · 2022-11-24 23:00
alyssachange PIPE_MAX_SAMPLERS to nr_samplers in line 1351
#asahi-gpu · 2022-11-24 22:59
alyssaok yes that should fix it
#asahi-gpu · 2022-11-24 22:59
alyssalmk if that fixes it
#asahi-gpu · 2022-11-24 22:59
alyssatry replace PIPE_MAX_SAMPLERS a few lines above with ctx->stage[shader].sampler_count
#asahi-gpu · 2022-11-24 22:58
alyssaoh I think I see the problem
#asahi-gpu · 2022-11-24 22:57
alyssadcow: Odd... is that reproducible?
#asahi-gpu · 2022-11-24 14:46
alyssai saw real x11 working once and have since forgotten how i did it :p
#asahi-gpu · 2022-11-24 14:46
alyssajannau: +1
#asahi-gpu · 2022-11-24 04:03
alyssa(Seeing as 2 of those were written today and 1 of those I found big issues with as recently as this afternoon, they'll be a bit longer. But they're coming)
#asahi-gpu · 2022-11-24 04:02
alyssa[My unstable Linux integration branch has asahi/main + NIR UBO/VBO lowering + jmp_exec_none + framebuffer compression + DISCARD_WHOLE_RESOURCE]
#asahi-gpu · 2022-11-24 04:00
alyssare performance, I have a LOT more patches where that came from but letting them trickle out to give time to test everything
#asahi-gpu · 2022-11-24 04:00
alyssajannau: yuka: You may wish to upgrade ^^
#asahi-gpu · 2022-11-24 04:00
alyssabut maybe other people will find other issues
#asahi-gpu · 2022-11-24 04:00
alyssaI tested my current set of { GNOME, neverball, supertuxkart, quake, xonotic }
#asahi-gpu · 2022-11-24 03:59
alyssathe latter should help performance noticeably, and it's pretty well tested now
#asahi-gpu · 2022-11-24 03:59
alyssaUAPI glue + https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19971
#asahi-gpu · 2022-11-24 03:59
alyssaPushed a new version to https://gitlab.freedesktop.org/asahi/mesa
#asahi-gpu · 2022-11-24 01:19
alyssalina: https://rosenzweig.io/more-splat.txt
#asahi-gpu · 2022-11-24 01:13
alyssagles3 is still going but seems about as broken as usual
#asahi-gpu · 2022-11-24 01:12
alyssaok, bunch of fixes and we're back to no gles2 regressions
#asahi-gpu · 2022-11-24 01:06
alyssait's worth seeing if Apple uses this compression for funny stuff that shouldn't be renderable :p
#asahi-gpu · 2022-11-24 01:05
alyssain principle all formats are renderable if we want them to be, but..
#asahi-gpu · 2022-11-24 01:05
alyssasimilarly, compression to non-renderable formats doesn't work
#asahi-gpu · 2022-11-24 01:00
alyssai've just disabled non-2D compression for now because that's nontrivial to workaround (though certainly doable)
#asahi-gpu · 2022-11-24 00:57
alyssacube maps + compression is botched because we can't do linear staging 2D Array resources
#asahi-gpu · 2022-11-24 00:53
alyssaTramtrist: i've fixed plenty of stuff remotely :-p
#asahi-gpu · 2022-11-24 00:49
alyssafixes stuff locally
#asahi-gpu · 2022-11-24 00:47
alyssaleaking staging resources, just fixed that locally
#asahi-gpu · 2022-11-24 00:46
alyssaalso just hit the oom killer yikes
#asahi-gpu · 2022-11-24 00:42
alyssait should not be possible to provoke an rtkit crash from userspace so this makes me profoundly uncomfortable
#asahi-gpu · 2022-11-24 00:41
alyssasplat for you
#asahi-gpu · 2022-11-24 00:41
alyssalina: https://rosenzweig.io/splat.txt
#asahi-gpu · 2022-11-24 00:38
alyssawhich are uncompressed twiddled and the alignment matters
#asahi-gpu · 2022-11-24 00:38
alyssaWorth noting that t-rex uses (IIRC) ETC2 or ASTC cube maps
#asahi-gpu · 2022-11-24 00:37
alyssaok did the obvious fix and it's happy
#asahi-gpu · 2022-11-24 00:36
alyssaalthough my unti test failing no
#asahi-gpu · 2022-11-24 00:36
alyssaif I align for uncompressed but not for twiddled, it seems happy
#asahi-gpu · 2022-11-24 00:34
alyssaI am unsure where you got the logic from

plus ›