alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - https://gitlab.freedesktop.org/panfrost - Logs https://freenode.irclog.whitequark.org/panfrost - <daniels> avoiding X is a huge feature
vstehle has quit [Ping timeout: 258 seconds]
sravn has quit [Ping timeout: 246 seconds]
sravn has joined #panfrost
<Lyude> alyssa: nearly have my t820 system running, just trying to figure out some scpi issues that came up on 5.2.1 (if I don't, all of the CPUs will be stuck at min freq iirc ;p). might take until thursday for me to get this working, or hopefully tommorrow
vstehle has joined #panfrost
davidlt has joined #panfrost
davidlt has quit [Ping timeout: 245 seconds]
_whitelogger has joined #panfrost
jernej has quit [*.net *.split]
MoeIcenowy has quit [*.net *.split]
xHire has quit [*.net *.split]
memeka has quit [*.net *.split]
chrisf has quit [*.net *.split]
afaerber has quit [*.net *.split]
lvrp16 has quit [*.net *.split]
mearon has quit [*.net *.split]
robertfoss has quit [*.net *.split]
lvrp16 has joined #panfrost
afaerber has joined #panfrost
mearon has joined #panfrost
robertfoss has joined #panfrost
MoeIcenowy has joined #panfrost
chrisf has joined #panfrost
jernej has joined #panfrost
xHire has joined #panfrost
memeka has joined #panfrost
davidlt has joined #panfrost
yann has quit [Ping timeout: 245 seconds]
chewitt has quit [Quit: Zzz..]
chewitt has joined #panfrost
afaerber has quit [*.net *.split]
lvrp16 has quit [*.net *.split]
robertfoss has quit [*.net *.split]
mearon has quit [*.net *.split]
xHire has quit [*.net *.split]
jernej has quit [*.net *.split]
MoeIcenowy has quit [*.net *.split]
memeka has quit [*.net *.split]
chrisf has quit [*.net *.split]
yann has joined #panfrost
xHire has joined #panfrost
MoeIcenowy has joined #panfrost
chrisf has joined #panfrost
robertfoss has joined #panfrost
lvrp16 has joined #panfrost
mearon has joined #panfrost
raster has joined #panfrost
jcureton has quit [Ping timeout: 260 seconds]
adjtm has quit [Remote host closed the connection]
adjtm has joined #panfrost
chewitt has quit [Quit: Zzz..]
chewitt has joined #panfrost
davidlt has quit [Remote host closed the connection]
davidlt has joined #panfrost
<alyssa> Lyude: :tada:
<alyssa> Thank you!
JaceAlvejetti has joined #panfrost
guillaume_g has joined #panfrost
<guillaume_g> Hi! Trying to launch kmscube on t604 mali, I get the following output: https://pastebin.com/03kkyaKg and this kernel message: [ 68.441051] exynos-drm exynos-drm: [drm:exynos_plane_atomic_check [exynosdrm]] *ERROR* unsupported pixel format modifier
<guillaume_g> How could I move forward?
<guillaume_g> daniels: ^ (I used kmsro in Mesa as you suggested, and panfrost is properly detected, now, with kmscube)
<guillaume_g> rtp: thanks for this info, but how could we fix it? On the panfrost side, I guess?
<rtp> guillaume_g: no idea but I guess it's on panfrost side.
<guillaume_g> rtp: ok
<guillaume_g> Maybe some panfrost guys could guide me on this?
<alyssa> guillaume_g: Did you try making the fix rtp suggested? :)
<guillaume_g> alyssa: rtp pointed the problem, not a fix. ;)
<alyssa> I mean
<alyssa> What if you changed it to LINEAR?
<guillaume_g> alyssa: I can give it a try :)
<guillaume_g> alyssa: thanks for the suggestion. But maybe it was there on purpose. :D
<guillaume_g> alyssa: just found https://gitlab.freedesktop.org/lyudess/panfrost-mesa/commit/120b40ff02c3abaaa551f943d6ad4d052c4c8cde so it breaks rockchip, let's see if it fixes exynos
buzzmarshall has joined #panfrost
jolan has quit [Quit: leaving]
jolan has joined #panfrost
jcureton has joined #panfrost
afaerber has joined #panfrost
guillaume_g has quit [Quit: Konversation terminated!]
<alyssa> Oh, interesting
<alyssa> tomeu: ^ This is definitely your department :)
<alyssa> Alright, BO caching is up next
empty_string has quit [Ping timeout: 258 seconds]
empty_string has joined #panfrost
yann has quit [Ping timeout: 268 seconds]
<alyssa> Implement a first draft of a BO cache. Initial results with STK are very promising
<alyssa> There is some memory leak somewhere that's hindering the effectiveness of the cache and also causing us to leak like pigs.
JaceAlvejetti has quit [Remote host closed the connection]
adjtm has quit [Quit: Leaving]
<alyssa> But even so, supertuxkart's fps is up maybe 50%?
<alyssa> Either way, very noticeable improvement
<alyssa> It regresses neverball, oink oink
<alyssa> Never mind, can't reproduce nb regression, maybe I had a bad branch compiled
adjtm has joined #panfrost
<alyssa> Meanwhile, with these fixes, SuperTuxKart is now very comfortably playable
<alyssa> (Tried a few levels. The heaviest levels fullscreen have a 15fps min which is annoying, but an average level at not-quite-so-high-a-resolution is much better)
afaerber_ has joined #panfrost
<alyssa> dEQP-GLES2.functional.flush_finish.wait is unhappy
afaerber has quit [Ping timeout: 276 seconds]
<alyssa> Lovely.. Works in valgrind with no errors, crashes outside
<alyssa> ---Oy. I forgot there's a hard limit of 65k draws.
davidlt has quit [Remote host closed the connection]
davidlt has joined #panfrost
herbmillerjr has quit [Quit: Konversation terminated!]
chewitt has quit [Read error: Connection reset by peer]
yann|work has joined #panfrost
afaerber has joined #panfrost
herbmillerjr has joined #panfrost
afaerber_ has quit [Ping timeout: 276 seconds]
davidlt has quit [Remote host closed the connection]
davidlt has joined #panfrost
stikonas has joined #panfrost
raster has quit [Remote host closed the connection]
janrinze has joined #panfrost
<alyssa> There is still a mysterious memory leak I'm trying to track.
<alyssa> Well, plugged a handful of CPU-side memory leaks so that's a net win
<alyssa> Still not the big one eating up all the everything on heavy STK levels
davidlt has quit [Remote host closed the connection]
davidlt has joined #panfrost
buzzmarshall has quit [Remote host closed the connection]
<alyssa> So the big leak is from oversized transient buffers
anarsoul has quit [Ping timeout: 272 seconds]
<alyssa> Aaaaa I see
anarsoul has joined #panfrost
afaerber has quit [Quit: Leaving]
<alyssa> So, now that scoreboarding (landed) and BO caching (branch) are here, I reprofiled STK
<alyssa> Now the bottlenecks are:
<alyssa> - memcpy in u_vbuf_draw_vbo
<alyssa> - Minor, but _mesa_hash_data called from panfrost_get_job
<alyssa> - vbo_get_minmax_indices called from st_draw_vbo
<alyssa> - memcpy in panfrost_draw_vbo
<alyssa> The u_vbuf_draw_vbo is from user buffers.. a copy there is sort of inevitable, sadly
afaerber has joined #panfrost
sravn has quit [Ping timeout: 258 seconds]
_whitelogger has joined #panfrost
sravn has joined #panfrost
NeuroScr has joined #panfrost
NeuroScr has quit [Quit: NeuroScr]
cphealy has joined #panfrost
NeuroScr has joined #panfrost
jernej has joined #panfrost
<stikonas> panfrost is now added to gentoo
<alyssa> :+1:
stikonas has quit [Remote host closed the connection]
alyssa has quit [Ping timeout: 245 seconds]
alyssa has joined #panfrost
<alyssa> Things that happen when a shader spills:
<alyssa> framebuffer.unk0 |= 0x1 (across all shader stages)
<alyssa> That's it, so presumably we've an incomplete decode :C
<HdkR> Probably a bit to enable local or global loadstore ops
<HdkR> :D
<alyssa> Like, don't I have to give it backing memory.. somewhere?
<HdkR> Depends on hardware. If you're lucky then the hardware has a small spill space it can abuse
<alyssa> ....and if I need a *lot* of registers..?
<HdkR> Will need a memory backing :P
<alyssa> Yea
<alyssa> So it's more likely I'm not printing out memory I should be S:
<alyssa> HdkR: It's worth noting compute shaders use the same ld/st ops but with an explicit address (stuffed in a uniform)
<alyssa> So that's global vs local I guess?
<HdkR> Explicit address + per thread calculated offset in to it I guess?
<alyssa> Sure, but where's the explicit address :/
<HdkR> Probably just something allocated previously and slapped in to GPU memory space
<HdkR> Interleaving spills in memory makes sense
<HdkR> Probably adds new thread local loadstore ops that do it magically