ChanServ changed the topic of #zink to: official development channel for the mesa3d zink driver || https://docs.mesa3d.org/drivers/zink.html
Ristovski has quit [Remote host closed the connection]
Ristovski has joined #zink
<ajax> how am i still hitting that stupid wait_for_event deadlock
fahien has joined #zink
<kusma> zmike: Nice work on the optimal shader keys :)
<kusma> I've just done my first shader-key variant thing in Zink in a looong time, and just had to rebase on top of it, and that MR actually fixed a bug for me due to some cleanups, nice ;)
<kusma> (lowering line stippling \o/)
<kusma> (no, don't ask about the stipple-counter. It doesn't work, and can't work without unreasonable hacks in a portable way. driver
<kusma> ...s will just get incorrect stippling for loops and non-gs emitted strips unless they do this in hw. luckily, the cts doesn't care about this)
Soroush has quit []
Soroush has joined #zink
fahien has quit []
<zmike> that sounds awful
* zmike still has ptsd from line stipple cts
<kusma> Yeah, it is...
<kusma> Maybe we should change the member to unnormalized_coords so it has a more sane default?
<zmike> eh
<zmike> could be
<kusma> What, you don't want to touch a gazillion lines of code?!
<kusma> (patch is r-b)
<zmike> 😬
<zmike> ajax: I'd guess you're still hitting the deadlock because it still exists ?
<zmike> I don't remember where we were with it
<ajax> i'm trying to think of a reasonable way to fix it without blowing up the universe
<ajax> hm. just glx though, this should be fairly easy...
<zmike> ajax: you might be an expert in this: I've got the zink driconf loading in pipe_loader_sw.c
<zmike> but it doesn't seem to actually be reading driinfo_zink.h?
<zmike> like, I can just fill that header file with garbage and it'll still compile
<zmike> weirdly, if I remove the static from const driOptionDescription zink_driconf[] in pipe_loader_sw.c then it seems to sort of read it
<zmike> in that it has trailing garbage at the end which causes crashes
<zmike> okay, I think this is a compiler bug
<zmike> what the entire
<zmike> hm
<zmike> supertuxkart:
<zmike> iris min 59.799858 max 74.338440 av 68.249969
<zmike> zink min 67.953178 max 96.431969 av 78.616646
<zmike> seems fine
<anholt_> hmm. zink+anv is bad at gfxbench gl_driver2 compared to iris, while zink+radv and zink+tu are fine.
<anholt_> -72% perf
<anholt_> intel_gpu_top says 88% GPU busy
<anholt_> (iris is just 93%)
<zmike> weird
<anholt_> it's suspiciously close to 60fps, but radv and tu don't get vblank synced, and the GPU does say it's busy.
<zmike> 🤔
<zmike> nothing preventing ccs usage anymore?
<anholt_> not sure how I check easily these days
eukara has quit []
<zmike> I thought there was an INTEL_DEBUG for it, but maybe I'm remembering turnip or something
cheako has quit [Quit: Connection closed for inactivity]