ChanServ changed the topic of #zink to: official development channel for the mesa3d zink driver || https://docs.mesa3d.org/drivers/zink.html
* zmike pushes the button
<zmike> first try \o/
<ajax> aw yiss
<zmike> and I've already got a fixup MR up
<zmike> ajax: great work on this project
<zmike> couldn't have done it without you
cheako has quit [Quit: Connection closed for inactivity]
<ajax> right back atcha
<zmike> 💪
<airlied> yay kopper lands!
<zmike> this is a good day.
eukara_ has quit []
LexSfX has quit [Ping timeout: 480 seconds]
LexSfX has joined #zink
<zmike> kusma: have you had any time to look at cubes lately?
<kusma> I had a long look the other day, and it's a mystery to me still...
<kusma> Everything looks correct... Well, I can't manage to convince myself that the -1 bias is actually correct, but removing that just makes things worse...
<zmike> increasing the bias improves the mipmap accuracy somehow
<zmike> I'm considering maybe just landing as-is and then we can try and figure it out over time
<zmike> seamed cubes are probably more common than seamed+mipmapped cubes
<kusma> But I actually wonder if this whole thing is a bit of a red herring, and what you really want is a shadow-cubemap specific fix.
<zmike> I'm not sure how that helps?
<zmike> the problem cases aren't shadow
<kusma> Well, the use-case you listed was shadows?
<zmike> huhuhhh
<zmike> no
<zmike> SOME of the cases were shadow
<zmike> it's just ALL mipmap cases
<zmike> not shadow-specific
<zmike> and the base mipmap case does work at least a bit, since cts uses this
<kusma> The only one I saw was shadow... Do you have some links to other cases?
<kusma> Yeah, the CTS is a bit of a different matter. We're going to need this to pass the CTS.
<zmike> yes, and it does pass cts
<zmike> so that's something
<kusma> Oh, I see. So it's just piglit who complains?
<zmike> yes, exactly
<kusma> Ah, that worries me much less, then :)
<zmike> yes, exactly
<zmike> :)
<kusma> I'm fine with landing this with some broken piglit test-cases
<zmike> and in those cases, increasing the bias increases the % of iterations that pass
<zmike> so I think it must be bias-related somehow
<kusma> Yeah, seems like that to me.
<kusma> Although it does fail in some cases where the bias is zero, which is... surprising.
<zmike> I don't think it's that surprising? it's just mipmap edges that fail
<kusma> I'm not sure I follow...
<zmike> the tests that fail (I think, I'm out of context) are just failing along the edges of miplevels
<zmike> it's the implicit lod calc that's failing
<zmike> whether the explicitly-specified bias is zero is not relevant I don't think
<kusma> Hmm, I see. I can't say I understood that well what the test actually did.
<kusma> That could indeed explain.
<zmike> I looked at it for a day or two before I gave up and that was my conclusion
<kusma> Gotta go pick up in kindergarden now...
<kusma> You can probably stick my r-b on it as-is.
<zmike> cool
<zmike> I'll open a tracker
omegatron has quit [Quit: What happened? You quit!]
ced117 has quit [Ping timeout: 480 seconds]
eukara has joined #zink
LexSfX has quit [Ping timeout: 480 seconds]
LexSfX has joined #zink
<ajax> donk
<zmike> ajax: I'm looking into these ext_image_dma_buf fails now and I think the fix is going to be very, very gross
<ajax> oh goody
<zmike> maybe not that gross actually, but it's in dri so still gross
<ajax> do we really need that KILL message
<zmike> which one
<ajax> in kill_swapchain()
<zmike> oh
<zmike> probably not haha
<zmike> had it for debug and forgot to remove
<zmike> ajax: what do you think about adding an abort-y error if LIBGL_ALWAYS_SOFTWARE=1 and MESA_LOADER_DRIVER_OVERRIDE=zink are specified
<zmike> this used to be required behavior but now it just crashes
<zmike> later
Sachiel has quit [Quit: WeeChat 3.4]
Sachiel has joined #zink
<zmike> Assertion `!"assumption failed"' failed.
<ajax> i am so confused
<ajax> it's only possible to set swap interval after the thing is created
<ajax> if i take out the if (ptex) and just do it unconditionally, i crash, but none of the set-swap-interval code get called
<ajax> i crash in MakeCurrent of all things
<zmike> what the
<ajax> which makes me wonder how i decided to put that if where i did!
<zmike> I've just spent an hour trying to figure out how to deal with framebuffer srgb without mutable swapchain and my best answer is pain
<airlied> is that for GTF-GL45.gtf30.GL3Tests.framebuffer_srgb.framebuffer_srgb_draw?
* airlied sees regression
<zmike> no, piglit
<zmike> wtf
<zmike> ohhhhhhhhhhhhhhhhhhhh
<zmike> I always run with release builds
<zmike> which seems to imply that lavapipe can support mutable swapchain without much/any work
<zmike> indeed
<zmike> hooray
* zmike opens a ticket and puts non-mutable on the hamster wheel
<zmike> ajax: I just made a kopper label in gitlab
<zmike> pls subscribe