05:01tiredchiku[d]: so
05:01tiredchiku[d]: NVK-git isn't busted
05:01tiredchiku[d]: it was probably a change in the PKGBUILD
05:05tiredchiku[d]: no wait nvm I spoke too soon
05:05tiredchiku[d]: <a:ahh:1022261668148940810>
05:05tiredchiku[d]: rebisect time ig
05:13tiredchiku[d]: guess I have to reboot between each bisect step
05:13tiredchiku[d]: :why:
08:21orowith2os[d]: tiredchiku[d]: what's going on?
08:21tiredchiku[d]: tiredchiku[d]: see ^
08:22orowith2os[d]: Joy
08:22orowith2os[d]: Can you not build nvk and point sddm to a local build like you would anything else?
08:34tiredchiku[d]: that's what I was doing, but it didn't do much
15:29gfxstrand[d]: When the problem is plasma blowing up, there's a limit as to what local installs will do.
15:31gfxstrand[d]: tiredchiku[d]: I'm gonna take a wild stab and guess that the bug is somewhere in https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35660/commits
15:31gfxstrand[d]: Maybe that'll save you some steps?
15:32tiredchiku[d]: I was in the middle of bisecting when life happened ๐
15:32tiredchiku[d]: had 3 steps left
16:07snowycoder[d]: KeplerB instruction scheduling MR done: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35821 gfxstrand[d]
16:28gfxstrand[d]: snowycoder[d]: Awesome! Blackwell is top billing next week but I'll try to give it a look if I can. Or maybe Mel can.
16:29gfxstrand[d]: I also need to review the data flow stuff
16:34karolherbst[d]: 2.5x perf boost in piano... nice
16:36gfxstrand[d]: Did you try Talos with it? Did you see a bump there, too?
16:44snowycoder[d]: gfxstrand[d]: I have the worst video card ever, a GT 710. I think it will go from unplayable to still unplayable๐
16:45karolherbst[d]: oh wow
16:45karolherbst[d]: so did pixmark piano increase from 2 to 5 fps or what?
16:45snowycoder[d]: I've seen a good bump in furmark (still behind nvc0 driver, but I haven't tested with texdepbar)
16:45karolherbst[d]: mhhh furmark...
16:45snowycoder[d]: karolherbst[d]: 1 fps to 2fps๐
16:45karolherbst[d]: nice
16:46karolherbst[d]: ~~at 320p~~
16:46karolherbst[d]: at that fps you might not notice a 0.2% perf improvement tho ๐
16:47gfxstrand[d]: snowycoder[d]: I've got a 790 Ti, I think. I can probably test something eventually.
16:47gfxstrand[d]: I bought some stupidly large old GPUs.
16:49karolherbst[d]: snowycoder[d]: did you increase the clocks through the `pstate` file?
16:50snowycoder[d]: I have a beefy gt 770 i can test on but it's keplerA, no instruction latencies :/
16:50karolherbst[d]: huh?
16:50karolherbst[d]: it's the same
16:50snowycoder[d]: karolherbst[d]: Wait, isn't it automatc on kepler?
16:50karolherbst[d]: it's not
16:50snowycoder[d]: Oh then I should try it
16:51karolherbst[d]: fermis is where it's automatic
16:51karolherbst[d]: which is wild because it's the same isa
16:51karolherbst[d]: ๐
16:51karolherbst[d]: but yeah
16:51karolherbst[d]: ohh wait
16:51karolherbst[d]: pstate being automatic
16:52snowycoder[d]: karolherbst[d]: I'm not sure but I think instruction latencies are only from SM32
16:52karolherbst[d]: it's all manual
16:52karolherbst[d]: snowycoder[d]: nah
16:52karolherbst[d]: all kepler need the latencies
16:52snowycoder[d]: Then they're not plumbed through yet
16:53karolherbst[d]: fermi was SM20/21 and kepler is SM30+
16:53karolherbst[d]: snowycoder[d]: in nak?
16:55snowycoder[d]: karolherbst[d]: Yes, I've only encoded them in sm32, I didn't see them in sm30.
16:56karolherbst[d]: ohh is there even sm30 support in NAK?
16:57karolherbst[d]: mhh it's sm20...
16:57snowycoder[d]: Yes, it's all written by Faith
16:57karolherbst[d]: mhhh
16:57karolherbst[d]: nah it's all there it seems
16:58karolherbst[d]: well.. maybe not used? weird..
16:58karolherbst[d]: maybe it's optional?
16:59snowycoder[d]: I'm not at the computer to check but I remember seeing nvcc not writing instruction latencies for SM30
16:59karolherbst[d]: well codegen does
17:04karolherbst[d]: I suspect it's just all 0
17:04karolherbst[d]: and the encoder just skips over them
17:08gfxstrand[d]: It's not used at the moment but they code to break things into groups of instructions is all there. I just don't have them encoded. It should be easy if you've already done it for Kepler B.
17:08karolherbst[d]: I'm just confused because it's not obvious looking at the code where that happens exactly ๐
17:09snowycoder[d]: Ohhh, then I should dust off my GT 770!
17:09snowycoder[d]: P.s. I rewrote the code to pack basic blocks more tightly together, it should remove a lot of nops
19:48tiredchiku[d]: tiredchiku[d]: tl;dr I was bisecting but then the girls took me out clubbing
19:48tiredchiku[d]: just got home, gonna continue :3
20:38airlied[d]: karolherbst[d]: were you seeing some vector/scalar stuff with the parallel-rdp shaders?
20:38karolherbst[d]: airlied[d]: it's doing some 8x4 stuff, yeah
20:38karolherbst[d]: but that's not the issue really
20:38airlied[d]: there was a comment about those in a radv mr,
20:38karolherbst[d]: the issue is, that scalar phi allow for some deeper control flow to get nuked
20:39karolherbst[d]: which MR?
20:40airlied[d]: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35784
20:41karolherbst[d]: ahh interesting
20:50karolherbst[d]: airlied[d]: though not sure how much that would help given that keeping the 8x4 phis to be vectors actually hurts the shader...
20:51karolherbst[d]: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35784/diffs?commit_id=ba62ca59d1438dfe92eec48f68f7603a2964f09e#note_2979743
21:09tiredchiku[d]: `# first bad commit: [e99446fc5e6b773cd2a77f1315a08b5e200ae5b9] egl: Add EGL_EXT_device_query_name and EGL_EXT_device_persistent_id`
21:32tiredchiku[d]: I wonder why that breaks things though
21:33mhenning[d]: Yeah, that's odd. Do you get anything in dmesg?
21:34mhenning[d]: Also, if you could file a bug report that would be helpful, it isn't anything obvious looking at that commit
21:34tiredchiku[d]: I can't check D:
21:34tiredchiku[d]: well
21:34tiredchiku[d]: I guess I can
21:34tiredchiku[d]: but that's for tomorrow
21:35mhenning[d]: if you have systemd you can `journalctl -b -1` to show logs from the previous boot
21:35mhenning[d]: but also yes it can wait for tomorrow
21:36tiredchiku[d]: yeah that's what I was thinking too
21:36tiredchiku[d]: either that, or try to ssh in
21:36mhenning[d]: yeah
22:38gfxstrand[d]: tiredchiku[d]: Ugh... I bet Zink doesn't know what to do with those. Or that it's returning something odd that screws up apps.
22:49gfxstrand[d]: Shouldn't be too bad to fix if there's a good way to see what Zink is returning.
22:49gfxstrand[d]: zmike[d]: ^^