02:04alyssa: bbrezillon: jenatali: "we need to skip all primitive-restart entries when turning the triangle fan into a triangle list, which implies serializing the index buffer rewrite procedure (at least I didn't find any clever way to parallelize things)"
02:04alyssa: allow me to introduce you to the magic of parallel prefix sums :~)
02:04jenatali: If you have a common meta pass for it, I'll take it
02:05jenatali: But also we added triangle fans to D3D so we don't have to run that emulation path often
02:05alyssa: jenatali: Oh, I don't think I'll be implementing that particular case, no
02:05alyssa: we have native triangle fans & restart
02:05alyssa: though I have similar problems for my GS-on-compute implementation
02:06alyssa: I just think that it's a really fun puzzle of how to parallelize a kernel like that
02:06alyssa: and parallel prefix sums are basically magic :3
02:08DemiMarie: alyssa: will GS-on-compute make Vulkan bringup easier for future drivers?
09:03alkisg: Hi, `echo on >/sys/kernel/debug/dri/0/DP-3/force` forces that output to on. How can I set it back to unspecified? It accepts on, off but I can't find anything that would set it to the initial unspecified state...
10:12jani: apparently gone already, but the answer would've been "unspecified"
10:12jani: (super obvious...)
10:22vsyrjala_: isn't it "detect"? or is that an alias?
10:28vsyrjala: it's using a different string than the sysfs counterpart? apparently so. amazing
10:30vsyrjala: and looks like the ON_DIGITAL string is also different :/
11:21jani: why not, the more the merrier :p
12:05zmike: eric_engestrom: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24925
12:05zmike: ugh cargo errors?
12:08karolherbst: jenatali: you think you have some time to think about this MR? https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24905 I want to reduce the size of two sysvalls to 32 bit to make it easier for drivers to enable CL (all drivers just upcast to 64 bit and it was always pointless to not let nir optimize against it generally). I don't think it will regress clon12, but still wanted to let you know about it.
12:13jenatali: karolherbst: I'm on parental leave until November. If CI is fine with it then it shouldn't impact us
12:17karolherbst: ohh right.. I should check ci..
12:17karolherbst: thanks for pointing that out :D
12:17karolherbst: and have fun with your parental leave and everything
12:35hakzsam: anholt: airlied: FYI, I'm trying to update CTS in Mesa CI
14:34alkisg: jani: sorry my thunderbird irc client got "internal server error" and disconnected :/
14:34alkisg: This is what I'm getting while trying it: root@alkis:/sys/kernel/debug/dri/0/DP-2# echo unspecified >force
14:34alkisg: -bash: echo: write error: Invalid argument
14:43jani: alkisg: gah. try 'echo -n unspecified > force'
14:44alkisg: @jani thank you very much, that works!
14:45jani: alkisg: np, I'm just disgusted at the code
14:46jani: there's a char buf[12], and "unspecified\n\0" overflows that
14:47jani: yet the code does otherwise accept \n in there, and it should work without -n for the shorter values
14:57alkisg: Yes echo on / echo off work fine, but of course also echo -n on..
15:20youmukonpaku1337: hi guys
15:29jani:was going to link to some #nohello site but turns out there are at least six different dedicated no hello sites nowadays
16:59MrCooper: jani: FYI, 6.5 final is released
17:01jani: MrCooper: *facepalm*
17:09anholt: alyssa: how do you feel about load_frag_coord_zw becoming separate z and w, so we could have separate sysvals for them?
17:24alyssa: anholt: that's fine with me, I think I even tried that before settling on the merged intrinsic
17:25alyssa: the merged form is marginally more convenient for the two users, but if you're going to port another driver to using them and you need it split out, by all means
17:28anholt: I'm cleaning up intel fs interp spaghetti using your new bits, and it'll end up giving them less thread payload.
17:33ayaka: emersion, may I knew the email of James Jones. I am going to send my thought about allocation constraint sets. Or may you forward it?
18:06alyssa: anholt: wfm
18:28orogor: hi
18:29orogor: i got graphic glitches on a radeon, not sure where to ask for suggestions
18:32orogor: i use ubuntu with oibaf , and i think something broke sometime ago ... not sure where ...
18:32orogor: not all ways of decoding videos have an issue
18:34kisak: dcbaker: friendly nudge on mesa 23.2.0. It's been a month since we saw the last weekly release candidate. Want/need to hand it off to someone else?
18:34mattst88: eric_engestrom: ^
18:35orogor: xine on auto , so i gues vaapi has a full blue video, kodi has greenscale video output, i guess it also use vaapi, xine on opengl is ok
18:36orogor: if that may be the issue , i use have mesa-va-drivers:amd64 23.3 installed ( mesa-va-drivers:amd64 (23.3~git2308290600.639bda~oibaf~l)
18:44orogor: ... purging the repo .. so going back to 23.02
18:45orogor: that fixes the issue
18:45orogor: .. trying a reboot
20:52DemiMarie: Do upstreamed GPU drivers have a better security track record than Nvidia’s proprietary drivers?
21:01zmike: mareko: how to determine INTERP_MODE_NOPERSPECTIVE with lowered io?
21:10anholt: zmike: do you have bary intrinsics?
21:11anholt: interp mode of that, in that case.
21:12zmike: oh
21:12zmike: hm
21:12zmike: thx
21:22alyssa: idr: ;-) https://gitlab.freedesktop.org/asahi/mesa/-/merge_requests/179
21:24idr: alyssa: Right... I need to get back to that one of these days.
21:25alyssa: I might give it a shot myself. TBD
21:25alyssa: Need to think about how this interacts with continue constructs
21:26idr: I seem to recall that I only did it in the most conservative set of cases.
21:26idr: Basically... loops that would have been unrolled, but were too big or had too many iterations.
21:27idr:tries to remember...
21:28alyssa: Sure
21:28alyssa: The juicy optimization involves duplicating instrucitons
21:28alyssa: I think
21:28alyssa: > After this lands we might want helpers to detect a loop-with-terminating-if pattern but IDK what that would look like yet.
21:29alyssa: I wrote this helper this morning, then realized it's a Bad Idea because backends can't do anything with it without introducing critical edges
21:29alyssa: which is a no-go for SSA backends, without doing more patch up
21:31alyssa: > Due to the way branch prediction works on CPUs, it should always be an improvement, but I don't have the same level of certainty about branching on GPUs
21:31alyssa: I don't know that I followed this ... executing half the # of branches should be a win even with no branch predictor
22:21airlied: alyssa, gfxstrand : care to get a nir eye on 24803
22:23alyssa: nir eye for the reply?
22:24alyssa: I don't do variables, sorry
22:26airlied: maybe you should, challenge to get rid of compact :-P
22:27karolherbst: the only purpose of variables is to get rid of them :P
22:29alyssa: karol gets it
22:30airlied: the only incentive I have to ack that patch is it helps remove more TGSI :-)
22:34karolherbst: though to be frank, I keep my uniform variables around, because I can't see what's inside an ubo in the nir shader 🙃
22:47alyssa: airlied: I would love to ack, I just dont understand it
23:00zmike: that's the best time to ack
23:06airlied: yeah all of the usefulness, none of the responsibility :)
23:10emersion: ayaka: jajones@nvidia.com