12:42 <chadmed> im gonna add a new category to the broken software page for stuff that explodes with m1 optimisations. theres actually a ton of stuff that builds and works with clang just fine but then breaks if you add -mcpu=apple-m1

12:42 <chadmed> i wonder if it's trying to use AMX or something like that...

12:54 <jannau> no AMX in non-Apple llvm, see https://github.com/llvm/llvm-project/blob/main/llvm/lib/Target/AArch64/AArch64.td for llvm's model of the cpu. CycloneModel, ProcessorFeatures.AppleA14 and TuneAppleA14

13:01 <chadmed> yeah i just made clang emit the LLVM attributes, it just enables a shitload of crypto extensions on top of neon and outline-atomics

13:01 <chadmed> i wonder which one is causing issues

13:07 <jannau> what's the smallest and easiest to reproduce example? kmod with zstd compressed modules?

13:11 <j`ey> outline atomics hmmm

13:15 <chadmed> j`ey: that's enabled for the generic/non-specified aarch64 target too so its not that

13:16 <chadmed> jannau: mesa is probably the easiest to investigate since it fails rather spectacularly

13:16 <chadmed> id rather not screw around with kmod again, i like my system bootable :P

13:17 <chadmed> libinput also seems to dislike the optimisations

13:17 <chadmed> i have some time to kill tomorrow so ill add the whole target features string manually to my cflags and remove them one at a time until something starts working

13:18 <jannau> chadmed: I would have tested a local kmod build and not the system one

13:22 <chadmed> might do that too

20:34 <kettenis> back from a week in the mountains

20:35 <kettenis> rebased the u-boot asahi branch on u-boot master once more since u-boot 2022.10 is around the corner

20:35 <kettenis> although I don't think there are any changes that are significant for us