alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - https://gitlab.freedesktop.org/panfrost - Logs https://freenode.irclog.whitequark.org/panfrost - Transientification is terminating. Memory reductions in progress.
NeuroScr has quit [Quit: NeuroScr]
NeuroScr has joined #panfrost
chewitt has joined #panfrost
NeuroScr has quit [Quit: NeuroScr]
stikonas has quit [Read error: Connection reset by peer]
<alyssa> HdkR: Hm
<alyssa> ?
<alyssa> I mean, literally, like,
<alyssa> varying vec4 whatever;
<alyssa> whatever = vec4(1.0, 0.0, 0.0, 1.0);
<alyssa> Super dumb but it's legal :p
<alyssa> Not, like, a special "constant varying" or anything
<alyssa> Just.... a constant, written to a varying
<alyssa> Which is a special case for us
<alyssa> Psst constant varying is an oxymoron :p
<HdkR> alyssa: a constant varying in newer GLSL terms effectively means the "flat" qualifier. So it appears as if it is constant from the fragment side
<alyssa> I see
<alyssa> Not what I meant in the MR but ok
<HdkR> :P
<HdkR> Yea, bit different
<alyssa> Regardless uh
<alyssa> The MR still fixes an actual bug :p
<HdkR> There we go
<alyssa> Merci
<alyssa> ^ Got the other one for ya
<alyssa> Oh hey, we don't fail all of dEQP anymore
<HdkR> =O
<alyssa> ...only most of it
chewitt has quit [Max SendQ exceeded]
chewitt has joined #panfrost
anarsoul has quit [Remote host closed the connection]
anarsoul has joined #panfrost
anarsoul_ has joined #panfrost
anarsoul has quit [Client Quit]
<alyssa> Eyyyy
<alyssa> What do you know? I tried a bunch of random stuff and now all of the Panfrost bugs went away!
<alyssa> :P
<alyssa> So either I actually fixed something among the random changes, or I somehow installed llvmpipe :P
<anarsoul_> can you run q3arena now?
<alyssa> :p
<alyssa> By "all", well, it was all the same bug, but
<alyssa> - Missing triangles in -bshadow
<alyssa> - Missing parts in various renders
<alyssa> - Black rectangles in Weston
<alyssa> - Flashing in Kodi
<alyssa> - Missing draws in single-frame apps
<alyssa> - Failing all of dEQP
<alyssa> Speaking of, hey, let's retry dEQP
<HdkR> But does the Quake 2 ray tracing game work now? :P
* alyssa blinks
<HdkR> hehe
anarsoul_ has quit [Read error: Connection reset by peer]
<alyssa> ^ I'm satisfied
<alyssa> ---Why does [REDACTED] get stuck on the loading screen now?
<chewitt> oooh.. shiny
<alyssa> chewitt: Screen saver still broken, and video playback is wacked but you said that doesn't matter on Amlogic... besides that we seem to be working well :)
* alyssa tries to figure out when [REDACTED] regressed
<chewitt> lrusak asked me to test a patch earlier that results in text but no background GUI
<HdkR> oooo
<alyssa> chewitt: Blugh, I forgot Kodi was a moving target :p
<chewitt> slowly moving target for most GBM things
<chewitt> that's the first change in a few months
<alyssa> Touche
<HdkR> alyssa: I'm 100% fine with this
<alyssa> HdkR: Hm?
<alyssa> Update; [REDACTED] didn't regress, I'm just bad at testing :p
<chewitt> @alyssa what board are you testing Kodi on now? .. playback should be okay on most things
<alyssa> chewitt: Board?
<alyssa> It's just my Panfrost dev laptop (RK3399)
<chewitt> ahh, ok
<alyssa> No VPU acceleration setup, since mainline kernel
<alyssa> [Okay, apparently there's work to get mainline VPU accel. But I don't have that yet :P]
<chewitt> ahh, ok, yeah, that's still wip
<chewitt> rebuilt/updated and I still have the black screen with white text (due to the lrusak patch) but flickering is definitely gone :)
<chewitt> I'll drop the kodi patch and have a play..
<alyssa> Wee
<alyssa> Is that patch pushed to kodi master (so if I pull/rebuild things will replicate nicely?)
<chewitt> no that's not PR'd/merged yet
<alyssa> Hey!
<alyssa> We're passing actual dEQP tests!!
<HdkR> :D
<alyssa> Like
<alyssa> We pass 24/24 of dEQP-GLES2.functional.draw.draw_arrays.*
<HdkR> Nice!
<alyssa> Cherry picked I will admit
<alyssa> But still! We were at 0/24 yesterday!
<HdkR> Time to upstream
<HdkR> :P
<alyssa> HdkR: Pending on kernel, sure :P
<chewitt> with latest changes (and without lrusak's patch) Kodi GUI is rather nice
<chewitt> it still memleaks badly.. but all the glitches in the GUI are gone
<alyssa> Woop
<alyssa> I can work on memleaks next burst of energy I get, bugsquashing is only fun for so long ;)
<alyssa> Oh, dear, it appears I've substantially regressed [REDACTED]
<alyssa> HdkR: Don't merge the thing I just opened
<HdkR> lol
<alyssa> The greatest thing about the phrase "[REDACTED]" is that it can expand to two totally separate things
<alyssa> HdkR: OK now it's good
anarsoul has joined #panfrost
chewitt has quit [Quit: Adios!]
NeuroScr has joined #panfrost
NeuroScr has quit [Quit: NeuroScr]
sphalerite has quit [Quit: rebooting!]
sphalerite has joined #panfrost
NeuroScr has joined #panfrost
stikonas has joined #panfrost
NeuroScr has quit [Quit: NeuroScr]
_whitelogger has joined #panfrost
stikonas has quit [Remote host closed the connection]
_whitelogger has joined #panfrost
Elpaulo has quit [Quit: Elpaulo]
pH5 has quit [Ping timeout: 245 seconds]
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
BenG83 has quit [Read error: Connection reset by peer]
Elpaulo has joined #panfrost
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
<alyssa> Oh hey, [REDACTED} gets a teensy bit further now
<sphalerite> supertuxkart? :D
<alyssa> sphalerite: Other [REDACTED]
<sphalerite> and why is it redacted?
<Lyude> yeah
<Lyude> Can we just say the name of the thing?
<Lyude> it feels really weird and unusual we can't talk about hardware being worked on in here
<alyssa> Well, I guess I can let the cat out since there's no bag :(
<alyssa> [REDACTED} alternately refers to SuperTuxKart or GNOME Shell. Both are severely broken so there's that :p
<sphalerite> OpenClonk when?
<alyssa> OpenClonk?
<alyssa> Oh hey
<alyssa> If I cherrypick aggressively, I can find apps that work right with 'frost ;P
<sphalerite> openclonk.org
<sphalerite> hedgewars!
<alyssa> I was actually going to say Neverball, but sure :p
<alyssa> Actually, re Neverball... it has one critical bug, and it's.... exactly the same as STK's
<alyssa> How..?
<alyssa> :P
<alyssa> (Background/grass/etc is TOTALLY corrupted and has random massive tris in random places and it's whaaa?)
<alyssa> Guessing something isn't happy about these big tris
<alyssa> I wonder if it's depth near/far
<alyssa> Eh nope, it's still being set to (0, 1) like we hardcode as is. Can't be that
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
<alyssa> d
<rtp> alyssa: oh, I'm surprised you didnt notice the fnctl(F_DUPFD) thing. it was quite obvious to me when looking at the strace output :P
BenG83 has joined #panfrost
<alyssa> rtp: I can't be amazing at _everything_ ;)
<alyssa> Collaboration! Teamwork! Friendship!
<alyssa> :)
<anarsoul> hey rtp
<anarsoul> long time no see :)
<alyssa> You guys know each other?
<anarsoul> well, it's a small world
<alyssa> Fair enough
stikonas has joined #panfrost
<alyssa> Looking into gl_PointSize
<alyssa> Okay, reading panwrap, when gl_PointSize is written in the vertex shader, we have:
<alyssa> - varying_count incresed by two
<alyssa> - uniform_count increased by one but I think that's spurious
<alyssa> - Some new stuff in the shader I'll read in a moment
<alyssa> - Two new varyings as expected
<alyssa> - New uniform, again as expected
<alyssa> - line_width field is set to -0.00f (negative zero?!)
<alyssa> - unknown_draw ORed with 0x100 (so, 0x3100 instead of 0x3000)
<alyssa> Let's look at the shader
<alyssa> So, we have a st_vary_32 to address 1
<alyssa> Er, no, I think that's gl_Position actually
<alyssa> min(max(a, b), c)
<alyssa> That's, clamp, uh
<alyssa> Mmf
<alyssa> Point is, it's writing the transformed point size to hr53.z and then st_vary_16'ing it
<alyssa> Let's look at the varying descriptor next, I suppose
<alyssa> Of course I don't have the new formats in this panwrap version, annoying
<alyssa> First varying is MALI_R16F, which is what you would expect
<alyssa> xyyy swizzle, for some reason (if I decoded that right -- I have to do this stuff manually since old panwrap :p)
<alyssa> index set to 2, which is bizarre since there's only varying buffers decoded but we might have an off-by-one
<alyssa> Might be easier to just start trying things in a branch, heh "P
<alyssa> HdkR: Actually, don't merge the MR I sent, I'm going to try to do proper PointSize support to get it over with :p
<ezequielg> alyssa: we were right
<ezequielg> nv12 is just a special format
<ezequielg> it's 0xc
<alyssa> Ah?!
<ezequielg> :triumph_emoji:
<alyssa> ^_^
<ezequielg> you know what's funny
<ezequielg> i was about to give u
<ezequielg> then i said, what if it's the next one.
<ezequielg> etc etc
<ezequielg> rocky song climing the stairs, etc
<alyssa> Gambler's fallacy pays off?!
<ezequielg> lol, i didn't know there was a hole fallacy
<alyssa> JHm?
<alyssa> A whole fallacy, yes :)
<ezequielg> right hole, whole
<ezequielg> if it sounds right, it's right - and that's how i learned to speak anglais
<ezequielg> ok, merge request
<alyssa> Anglais? I thought the trick was if it sounds right, it's wrong :P
<alyssa> Source: many years of speaking English
<ezequielg> lol
<alyssa> So kmscube -Bnv12 works now?
<ezequielg> yup
<alyssa> Awesome :)
<alyssa> The harder problem is getting Kodi/mpv/etc to use native NV12... =P
<ezequielg> ...a problem for another day
<ezequielg> \o ttys
<alyssa> \o
<alyssa> ezequielg: (Did you send the MR?)
<ezequielg> i'm on it
<ezequielg> \o
<alyssa> \o
<ezequielg> alyssa: anything special?
<ezequielg> do you want to squash?
<ezequielg> pff, it's just one commit.
<ezequielg> got confused with rebase vs. merge
<ezequielg> anyway
<ezequielg> done, it's #29
<alyssa> Thank you!
<alyssa> Oh, this is interesting
<alyssa> line_width is interpreted as point_size, if point_size is constant (i.e. not written in the shader)
<alyssa> (I'm not sure if that's a case you can get to from GLES, but that's a different issue, haha)
<alyssa> Unpopular opinion: GL_POINTS should've been tossed along with QUADS
<alyssa> What the huh?
<alyssa> ...Okay, so it's shoving a pointer into line_width
stikonas has quit [Remote host closed the connection]
<alyssa> Midgard never ceases to amaze me
stikonas has joined #panfrost
NeuroScr has joined #panfrost
<alyssa> Ooookay, that's kind of terrifying(ly awesome) but I digress
<alyssa> The 0x100 in the unknown_draw means "interpret line_width as a pointer"
<alyssa> And I'm assuming that's how we link the varying in
<stikonas> KDE Plasma bootsplash actually renders nicely now :). Although plasmashell itself is still not working properly
<alyssa> stikonas: Progress is progress! :)
<stikonas> yeah, panfrost team is doing a great job
<alyssa> Thank you :)
<Lyude> robclark: didn't you do a bunch of stuff with 96boards?
<Lyude> robclark: was wondering if you remember what the easiest way to get Fedora onto one would be
<stikonas> by the way I get these errors when I try to run Plasma Wayland (https://paste.kde.org/pyfj3n4lh)
<alyssa> Lovely :(
<alyssa> ...I did not realise gl_PointSize would become an all-day project
<alyssa> (Still not there yet!)
<alyssa> Whatever, I'm having fun and that's what matters :)
<robclark> Lyude, at least a couple 96boards are supported out of the box in fedora.. the ones that have upstream u-boot and kernel support..
<robclark> for the others, dd fedora img to sdcard or usb disk, and make your own kernel arrangements..
<alyssa> robclark: The Embedded Linux Experience (TM)
<robclark> more or less, yup