00:41 karolherbst: we might want to optimize our any% CI runs as well 🙃
10:08 ishitatsuyuki: maybe someone also want to approve https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14168 for probably a fraction percent of speedup (it's hot in hashmaps)
10:50 dj-death: alyssa: starting to wondering if depending on LLVM is a good idea : https://paste.debian.net/1297414/
10:50 dj-death: just a function call can trigger a segfault...
10:51 karolherbst: dj-death: well.. we can get rid of clang/llvm if somebody writes a full C to spir-v compiler we could use instead
10:51 HdkR: 😬
10:52 karolherbst: or directly to nir or whatever
10:54 karolherbst: dj-death: a llvm build with assertions enabled could probably point out what's wrong there, but that usually only happens if something invalid has happened
10:54 karolherbst: or well.. software bugs
10:55 karolherbst: the amount of people throwing raw host code at this is kinda small, so random bugs are expected
10:58 dj-death: ==812701== Address 0x8 is not stack'd, malloc'd or (recently) free'd
11:01 dj-death: karolherbst: can you see anything wrong? : https://paste.debian.net/1297415/
11:01 dj-death: removing the call to that empty function works
11:05 dj-death: removing all arguments to the function also works
11:05 dj-death: but just leaving that first global pointer crashes
11:11 dj-death: ah figured it out
11:11 dj-death: just using dst instead of dst_ptr
11:11 dj-death: apparently that must be another variable :)
11:24 karolherbst: mhh weird
11:24 karolherbst: I don't see why `dst_ptr` shouldn't work here
11:25 karolherbst: but yeah.. I kinda expect us running into random bugs here
11:36 alyssa: dj-death: sigh.
11:36 alyssa: ~~intelclc did it first~~
11:41 dj-death: alyssa: you can blame gfxstrand ;)
11:43 alyssa: dj-death: meh, i blame her enough for things already ;p
12:26 alyssa: q