alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://freenode.irclog.whitequark.org/panfrost - <daniels> avoiding X is a huge feature
alpernebbi has quit [Quit: alpernebbi]
raster has joined #panfrost
kinkinkijkin has quit [Remote host closed the connection]
stikonas has quit [Ping timeout: 272 seconds]
kaspter has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
vstehle has quit [Ping timeout: 272 seconds]
archetech has quit [Quit: Leaving]
archetech has joined #panfrost
icecream95 has joined #panfrost
tlwoerner has joined #panfrost
archetech has quit [Quit: Konversation terminated!]
davidlt has joined #panfrost
archetech has joined #panfrost
chewitt has quit [Read error: Connection reset by peer]
chewitt has joined #panfrost
vstehle has joined #panfrost
<dstzd> has anyone here tried to get gamma control working on rk3399?
nlhowell has quit [Ping timeout: 256 seconds]
chewitt has quit [Read error: Connection reset by peer]
chewitt has joined #panfrost
chewitt has quit [Quit: Adios!]
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 260 seconds]
camus1 is now known as kaspter
nlhowell has joined #panfrost
archetech has quit [Quit: Konversation terminated!]
archetech has joined #panfrost
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 246 seconds]
camus1 is now known as kaspter
<macc24> HdkR: i got terraria to run on my arm laptop :D
<macc24> i have to close everything else to free up the ram and it runs at 5fps but it's just a minor detail
<macc24> need for speed carbon too
tomboy64 has quit [Ping timeout: 240 seconds]
tomboy64 has joined #panfrost
davidlt has quit [Remote host closed the connection]
tomboy64 has quit [Ping timeout: 240 seconds]
tomboy64 has joined #panfrost
icecream95 has quit [Ping timeout: 272 seconds]
cowsay has quit [Ping timeout: 272 seconds]
tomboy64 has quit [Remote host closed the connection]
tomboy64 has joined #panfrost
cowsay has joined #panfrost
stikonas has joined #panfrost
camus1 has joined #panfrost
kaspter has quit [Remote host closed the connection]
camus1 is now known as kaspter
raster has joined #panfrost
<tomeu> anybody knows if the kernel can get to know when the GPU started and finished executing a given vertex or fragment job?
<tomeu> for profiling
<tomeu> robmur01: maybe you would know?
<stepri01> tomeu: The 'best' method is to insert "write value jobs" with either the cycle counter or system timestamp and then try to correlate the values back to real time
<stepri01> the easiest option is to just look at the registers - and that might be sufficient for Panfrost for now because we're not queuing jobs on the hardware
<stepri01> the blob makes use of write value jobs and has a ioctl to ask the kernel to provide both the cycle counter/system timestamp and real time at an instant for use in correlating back to real time
<stepri01> cycle counter suffers from DVFS changing the speed. system timestamp is broken on some platforms. So neither is great
yann has quit [Ping timeout: 260 seconds]
yann has joined #panfrost
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 244 seconds]
camus1 has quit [Read error: Connection reset by peer]
kaspter has joined #panfrost
tomboy64 has quit [Write error: Connection reset by peer]
tomboy64 has joined #panfrost
<tomeu> stepri01: makes sense, thanks!
nlhowell has quit [Ping timeout: 256 seconds]
davidlt has joined #panfrost
alyssa has joined #panfrost
<alyssa> warpme_: pong
<warpme_> hello alyssa !
<warpme_> probably you saw chat exchanges with daniels 6/11 12:25?
italove has joined #panfrost
italove has quit [Client Quit]
italove has joined #panfrost
<macc24> good news, duet keyboard is just regular google hammer usb keyboard lol
_whitelogger has joined #panfrost
alpernebbi has joined #panfrost
<alyssa> Today's task: try to figure out how to even get started on Bifrost scheduling
<alyssa> Task #0 is documenting all the scheduling constraints in the first place..
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 260 seconds]
camus1 is now known as kaspter
<alyssa> The overall strategy is trying to decompose and simplify the problem space, so instead of having a massive list of constraints and a complicated cost function and trying to bruteforce something, we instead reduce to a sequence of "familiar" transforms.
<alyssa> Yay formal methods...?
<alyssa> I wrote up my list of constraints, threw a model checker at it, and found my model was horribly buggy.
<alyssa> I guess I should be happy but why is this off by a factor of two :V
<alyssa> Oh, right, ok, fine.
<macc24> does anyone have working mainline kernel config for duet?
<alyssa> why can't I have nice regular encodings
* macc24 has found patches adding display to mt8183
<bbrezillon> it's not up-to-date, but I had the duet DSI panel working
<macc24> \o/
<macc24> 5.8... i've seen worse
* alyssa is worse
<alyssa> ...Can a message-passing instruction be issued in the same clause as a branch?
<alyssa> +BLEND can't but what about things like +LOA..
<alyssa> +LOAD
kinkinkijkin has joined #panfrost
<macc24> bbrezillon: what did you do extra to have display working?
<macc24> OH MY GOD PLEASE NO
* macc24 stumbled upon the cursed land of rotated displays
<alyssa> yupyup
<bbrezillon> macc24: I don't remember, but it's all in the branch :P
<bbrezillon> and yes, rotated display
<bbrezillon> but weston seems to handle it correctly :)
<macc24> yeah i think it should work from now on https://i.imgur.com/RqAPSlC.jpg
<bbrezillon> alyssa: we have a precision issue on f16 blend shaders
<bbrezillon> I wonder why it works on midgard
<alyssa> bbrezillon: Oh?
<alyssa> (Could you elaborate?)
<alyssa> As you are well aware the whole blend shader stack is ... delicate ... so
<bbrezillon> alyssa: I'm not sure what happens yet, but if I force f32 dEQP-GLES2.functional.fragment_ops.blend.rgb_func_alpha_func.src.constant_color_dst_alpha passes, but if I don't, it fails
<alyssa> bbrezillon: Interesting. Could you send the DEBUG=shaders output? (NIR + disasm)
<alyssa> thanks
<bbrezillon> the assembly LGTM
<bbrezillon> error is on the alpha chan BTW (max deviation of 13 on an 8bit channel)
<alyssa> hmm
<alyssa> Assembly lgtm as well
<alyssa> bbrezillon: Unrelated but we should probably lower load_blend_const_color_*_float at the NIR level if we want constant folding
<alyssa> wait, here
<alyssa> Register allocation failure
<alyssa> mov.u32 br10, #0x3ecccccd, _, _, _,
<alyssa> ..
<alyssa> convert.f16 br9, br10.f32.x, br10.f32.x, _, _,
<alyssa> er
<alyssa> yeah
<alyssa> ..
<alyssa> fma.f16 br7, br9.f16.yy, br7.f16.xx, #0, _,
<alyssa> ..
<alyssa> select.u32 br5, br6.u16.x, br7.u16.x, _, _,
<alyssa> vs the pre-RA IR:
<alyssa> load_tile.v4.f16 7, #0xff000000, br60.u32.x, #0xed000, _,
<alyssa> ..
<alyssa> fma.f16 17, 7.f16.ww, 6.f16.xx, #0, _,
<alyssa> R9 is getting clobbered
<alyssa> Can't say I'm surprised, the entire bifrost RA is a hack that needs a proper, top-down designed replacement sooner than later
<alyssa> Ugh...
* HdkR stacks hacks
<alyssa> well, I hve been nerdsniped
<alyssa> let's see..
<alyssa> ---Oh, rip
<alyssa> d'oh
<bbrezillon> alyssa: lowering of blend consts => I agree, but that means not caching the NIR for such shaders, and that's the reason I didn't do it initially
<alyssa> mh, cache the NIR just before lowering blend consts, so all the expensive stuff (algebraic, etc) is cached and all that remains is constant folding?
<alyssa> ^^ fixes the test for m
<alyssa> and actually fixes the rest of fragment_ops.*, nice
<alyssa> thanks for tracking that down :)
<bbrezillon> alyssa: yep, works here too
<bbrezillon> thanks for fixing it ;)
<bbrezillon> and you can add my R-b
<bbrezillon> and T-b
didntread has joined #panfrost
archetech has quit [Quit: Leaving]
raster has quit [Quit: Gettin' stinky!]
archetech has joined #panfrost
<archetech> does 5.10rc3 have latest patches in it?
<macc24> archetech: define "latest patches"
<archetech> for bifrost G52
<macc24> amlogic?
<archetech> yes
<macc24> then the answer is a definitive 'perhaps'
<macc24> my g31 board works fine without panfrost patches on 5.10
<macc24> no idea about g52
davidlt has quit [Ping timeout: 240 seconds]
italove has quit [Ping timeout: 256 seconds]
raster has joined #panfrost
kinkinkijkin has quit [Remote host closed the connection]
<alyssa> bbrezillon: collabora-tion
<HdkR> You've been infected with puns
<narmstrong> archetech: no the latest patches for G52 are in drm-misc-next
<alyssa> HdkR: it's all pun and games
<HdkR> Next step is becoming a cat parent and whispering them to your child
<macc24> so i got debian to boot on my duet with bbrezillon's kernel
<HdkR> woo
<macc24> hopefully i can bump the kernel to 5.10 without serial
<archetech> narmstrong: sry im not famil with that location
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 256 seconds]
camus1 is now known as kaspter
<alyssa> $\LaTeX$ in my graphics driver? Likelier than you would think.
raster has quit [Quit: Gettin' stinky!]
<HdkR> Mono in my graphics driver?
* macc24 stops tracking git master
<alyssa> wha?
<archetech> <narmstrong> archetech: no the latest patches for G52 are in drm-misc-next
<archetech> somebody give a link pls?
Green has quit [Ping timeout: 244 seconds]
alyssa has quit [Quit: \o]
_whitelogger has joined #panfrost
alpernebbi has quit [Ping timeout: 240 seconds]
italove has joined #panfrost
italove has quit [Ping timeout: 264 seconds]