ChanServ changed the topic of #zink to: official development channel for the mesa3d zink driver || https://docs.mesa3d.org/drivers/zink.html
omegatron has quit [Quit: What happened? You quit!]
omegatron has joined #zink
<kusma> zmike: I decided to send out https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11795 separately, and I prefer if we merge that one before the Lavapipe line-rasterization one.
<kusma> The reason I prefer that, is to avoid breaking tests in between, while not piling even more onto that already big enough MR ;)
<kusma> I also want to fix up Zink to respect lines_rectangular flag for correct Nine rendering in the Lavapipe MR, but I can't do that until we've landed the other one.
<zmike> kusma: cool, I'll have some followup for this I guess to beat it into shape for other stuff I'm doing
<zmike> glad we're making progress on all this line stuff!
<kusma> Yeah, me too :)
<kusma> @zmike: OK, that went faster than I feared. So are you OK if I merge that one now, then do my little tweaks to the other MR and merge that once the new zink-patch has gotten your stamp of approval? :)
<zmike> kusma: yes
<kusma> Awesome. So I'm not causing issues for your bit-crunching in the shader-key? :)
<zmike> it helps that I've already implemented this extension myself so I have context already
<zmike> oh you totally are
<kusma> Hehe :)
<zmike> but I'm just gonna rewrite your patch after to avoid that
<kusma> That's fair.
<zmike> seems like the easiest solution
<kusma> Did you see the dates on my commits there? :)
<kusma> These are ~1 year old patches ;)
<zmike> I did
<zmike> slacker
<zmike> we could've had great lines a year ago!
<kusma> Yeah, except I had issues with this on ANV, which was the only driver that I could test it on at the time.
<zmike> they still have some issues with lines today
<kusma> Yeah, I bet.
<zmike> but I'd never mention Sachiel the person working on such things to see how that was going
<kusma> But we don't fail CI because of that, so I'm happy with passing on Lavapipe ;)
<omegatron> just out of curiosity: is the Zink driver development your main work (meaning: you get paid for it) or are you doing it in your spare free time?
<zmike> yes
<kusma> Yeah here too. I mean, I work on multiple things, but Zink is one of them.
<kusma> Heh, I just noticed this: https://github.com/Think-Silicon/GLOVE
<kusma> 404!
<zmike> you hate to see it
<kusma> Seems there's still some stale forks around: https://github.com/de2nis/GLOVE
<kusma> I wonder what happened... Did they switch to Zink or ANGLE or something?
<kusma> I guess the moment Android no longer mandated vendor GLES support, they didn't need to work on this any more?
<zmike> prob
<kusma> @zmike: I've assigned the lavapipe storage-image MR to marge
<zmike> nice
<zmike> uh ohhhhhhhhhhhhhhh
<zmike> drumroll please
<kusma> Even though it seems to cause some havoc
<zmike> MESA_LOADER_DRIVER_OVERRIDE=zink ./glcts -n 'KHR-GL46.blend_equation_advanced*'
<zmike> Test run totals:
<zmike> Passed: 87/88 (98.9%)
<zmike> Failed: 0/88 (0.0%)
<kusma> WOOOO!
<zmike> Not supported: 1/88 (1.1%)
<kusma> I mean, I think most of those failures are just because of GL 4.5, some of them seems to be kinda related. But I can't find any issues in Lavapipe here, so I'm guessing Zink or the GL frontend stumbles.
<zmike> I vaguely recall looking into a bunch of them
<zmike> hard to remember now though
<kusma> I say: let's deal with it in the glorious future!
<zmike> amen!
<kusma> I heard we'll have flying cars then!
<kusma> And jetpacks.
<zmike> and self-lacing shoes
<zmike> I'm still holding off on doing any merging until after the branchpoint btw, so blend will have to wait till next cycle
<kusma> Oh, maybe I should have done that for these changes too?
<kusma> Or do I feel brave?
<zmike> eh it's just lines
<zmike> I figured it was ok
<kusma> Yeah, fair. Who cares about lines, right?
<zmike> haha
<zmike> I meant it wasn't likely to impact anything else
<kusma> Oh, I actually kinda meant it.
<zmike> I didn't want to denigrate your hard work
<kusma> GL lines are garbage, and anyone who actually cares about proper line-rendering will use something better.
<zmike> certainly it hasn't been on my list of priorities
<zmike> you'll be pleased to know I managed to wrangle fbfetch without making any wild ntv changes
<kusma> Holy macarony!
<zmike> yeah it actually ended up being super reasonable
<zmike> I blame jekstrand for this
<zmike> if only he'd told me "why don't you just do it?" sooner
<kusma> That's 500 internet glory credits to you and 500 internet glory credits to jekstrand
<zmike> gotta get this up on the blog to make it official
<jekstrand> \o/
* jekstrand questions zmike's latest MR
<zmike> it's not a productive week for me if I don't find at least one years-old jekstrand bug
<jekstrand> zmike: Ooh, more validation. I'm rolling bits of that into !11775
<zmike> uh oh
<jekstrand> ?
<zmike> mo validation mo problems
<jekstrand> Well, I did find a BLORP bug with my validation. :)
<kusma> Yes and that was a PROBLEM. See, what zmike said is true.
<zmike> sweatytowelguy.jpg
Ristovski has joined #zink
hch12907 has quit [Ping timeout: 480 seconds]
* airlied would like to lands line pre-branch if we can
<airlied> since that would only leave minor changes to get 1.1 cts passing which can go post branch if necessary
<kusma> airlied:: sure, there's just the CTS issue remaining, dunno if we want to disable that feature or just live with the failures until resolved
<kusma> (or who knows, maybe there's a real issue there and I just don't get it)
<airlied> kusma: the fact that the CTS logs the wrong image as reference is annoying
<airlied> kusma: still a fix for that whether it be in the CTS or on our end is probably small, so I'm happy to land with documented fails
<airlied> it's no worse than what we have now
<kusma> It's more than that, as I've pointed out. It's verifying against bresenham rules regardless when single sampling.
<airlied> the CTS has like 3 verifying algorithm of varying specificity I thought, each slightly more forgiving than the last
<airlied> and it seems to only log the reference image from the first if I read the issue you filed correctly
<airlied> would be nice if we could at least see the reference image it was failing against
<airlied> also swiftshader does pass CTS by whatever means
<airlied> but I didn't dig in on that side
nushoin has joined #zink
nushoin has quit []
Dragon has joined #zink
Dragon has quit []
hch12907 has joined #zink