00:51 <robertfoss> heya

00:52 <HdkR> Welcome back robertfoss!

00:52 <robertfoss> i just got a rockpi4, rk3399 sbc, up and running using panfrost

00:52 <HdkR> woop woop

00:52 <robertfoss> kmscube runs

00:52 <robertfoss> but glmark2-es2 segfaults

00:54 <HdkR> alyssa: ^ ring any bells? :)

00:54 <robertfoss> I'm also seeing some screenwide artifacts, while doing 3d rendering or not: https://photos.app.goo.gl/fQDwZBvtpWn61HHb6

00:54 <robertfoss> sharp lines are jagged and rough

00:54 <HdkR> Whoa neat

02:07 <alyssa> robertfoss: :eyes:

02:08 <alyssa> robertfoss: So, glmark2-es2-drm has some weird something or other. I have a reeeeally ugly patch I can send, although it -might- be more pleasant just to run weston so you can get glmark2-es2-wayland which should work OOB

02:09 <alyssa> As for the artefacts, I've never seen anything like that and I'd rather like to blame the display driver (etc); that's not the "type" of artefact a 3D driver bug would cause in my experience (and I'd've seen it by now if it were, I think..)

02:29 <alyssa> Edges are ugly but that's not what you're reported, lemme try forcing MSAA

02:30 <alyssa> Yeah, good on our end

02:30 <alyssa> (If you do want to force MSAA on, set "enabled = true" at the beginning of set_framebuffer_msaa in mesa/src/gallium/drivers/pan_context.c)

02:32 <alyssa> Aside: phong shading is twice as fast as cel shading in glmark. wonder why

02:34 <alyssa> Oh, I guess cel shading is a lot more complex of a shader. Good opportunity for optimization when I get the compiler bug :)

03:24 <alyssa> Alright, well, finished the homework I told myself I would get done tonight, so onto much more interesting matters... investigating the tiler magic

04:05 <alyssa> So if I just make misc_0 massive, all the problems disappear..

04:05 <alyssa> ...Honestly, let's do that for now. It goes from a CPU usage issue to a memory usage issue, but that's a lot easier to deal with :)

09:08 <cyrozap> You know you've been watching too much anime when you read "MALI_NO_ALPHA_TO_COVERAGE" as "ALPHA and COVERAGE of MALI" ("MALIのALPHAとCOVERAGE")...

09:31 <raster> へんあ がいじん

09:31 <raster> :)

15:31 <robertfoss> alyssa: thanks for all of the help

15:31 <robertfoss> alyssa: I think the issue is in the display driver

15:33 <HdkR> Oof

16:13 <alyssa> robertfoss: Mm, I love tossing blame at other parts of the stack ;P

18:49 <robertfoss> alyssa: it is one of the most treasured FOSS feelings I believe

18:49 <robertfoss> alyssa: like a more wholesome schadenfreude

18:50 <alyssa> robertfoss: ^_^

21:07 <robher> kmscube working with panfrost drm driver! https://usercontent.irccloud-cdn.com/file/xNXXwKno/irccloudcapture64237708105929396.jpg

21:08 <robher> thanks to tomeu.

21:09 <anarsoul> robher: congrats :)

21:10 <robher> hopefully, that's what it is supposed to look like... :)

21:11 <HdkR> woo hoo

21:11 <daniels> robher, tomeu: :-o

21:17 <Lyude> :D

21:17 <Lyude> hell yeah

21:20 <alyssa> robher: !!!!!!!!!!!!!!!!

21:20 <alyssa> Best news I heard all day!

21:20 <alyssa> Aaaaaa!!!!!!! :D :D :D

21:21 <robher> I'm surprised it works given there's no locking or BO tracking.

21:21 <alyssa> robher: ^_^

21:22 <HdkR> ack. People are pestering me about my Tensilica Xtesna LX6 backend for LLVM that I stopped working on late last year. People /really/ want to not use the downstream GCC fork for that thing

21:23 <HdkR> (Can't blame them, it generates some terrible code and doesn't give you the flexibility of LLVM)

21:25 <alyssa> Xtensa...?

21:25 <HdkR> Silly embedded CPU core that has a sliding register file view. It's cute in its own way

21:27 <HdkR> Backend was working well enough that it would generate .o files that you could use to link using the rest of their toolchain :P

21:31 <Lyude> iirc wasn't xtensa used in the 360 as well

21:32 <HdkR> Oh, it used a quad core design for its DSP. Neat

21:33 <HdkR> Which is also apparetly what AMD Trueaudio used

22:03 <narmstrong> I must try the driver now :-D

22:04 <narmstrong> Great job tomeu & robher, hope t820 support won’t be too painful

22:14 <alyssa> narmstrong: Ought to be a one-line change, just adding a "compatible"... we hope :)

22:34 <mmind00> congrats ... that is really great to see

23:02 <robher> narmstrong: shouldn't be too hard. We need to add in the 'issues' flags and and there's some config based on that which I hard code ATM. Shouldn't be too different for the 820. Patches welcome! :)

23:13 <robher> drm driver changes pushed into panfrost/linux panfrost-5.0-rc4 branch. And you need tomeu's mesa tree to go with it.

23:52 <alyssa> robher: Aaaa reading the code, this is super cool to see <3

23:52 <alyssa> robher: "This is for debugging because we should never page fault." Eh-heh-heh should I tell him? :P