ChanServ changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://oftc.irclog.whitequark.org/panfrost - <macc24> i have been here before it was popular
rkanwal has quit [Quit: rkanwal]
<HdkR> alyssa: gfxbench has worked for quite a long time in FEX.
<alyssa> HdkR: Great, I need a new excuse not to setup an x86 rootfs then :p
<HdkR> FEXRootFSFetcher will fetch a prebuilt rootfs for you
<icecream95> alyssa: I hear there's this new box64 thing which means you don't need a full rootfs
<HdkR> Which I'll be updating to Mesa 22.1 soon as well.
* alyssa sweats
<alyssa> HdkR: how would I go about using my own Mesas?
<alyssa> building for x86_64 and sticking it in the rootfs? building for arm64 and thunking somehow?
<icecream95> thonk
<HdkR> building for x86_64 and inserting in to the rootfs currently. Thunks are being rewritten to make them more stable right now.
<alyssa> icecream95: +1
<alyssa> HdkR: ack
<alyssa> icecream95: I suppose you would know if panfrost works on x86_64? :P
<HdkR> It builds at least, and the ioctls don't have problems as far as I could tell.
<icecream95> Maybe? Possibly I tested it, but I don't remember well. I know that it works on OpenBSD though
<alyssa> ;
<alyssa> :D
<alyssa> HdkR: real question is how do the pain levels compare of cross building vs building inside the FEX root
<icecream95> alyssa: Cross compiling isn't that hard
<HdkR> I've never actually tried building inside the root under FEX
<icecream95> .gitlab-ci/container/create-cross-file.sh might be useful
* HdkR tries building
<alyssa> icecream95: noted, thank you C:
<alyssa> HdkR: lol
<HdkR> Rootfs building happens on an x86-64 host in a VM. Which builds latest mesa without FEX :P
<HdkR> oop, new libdrm version. Good to know that is necessary.
<HdkR> Yea, don't compile in the rootfs with FEX. Painful engagement there.
<HdkR> Need code caching before that becomes better
hexdump0815 has joined #panfrost
hexdump01 has quit [Ping timeout: 480 seconds]
dsotsen has joined #panfrost
dsotsen has quit []
floof58 has quit [Ping timeout: 480 seconds]
floof58 has joined #panfrost
guillaume_g has joined #panfrost
nlhowell has joined #panfrost
nlhowell is now known as Guest985
nlhowell has joined #panfrost
Guest985 has quit [Ping timeout: 480 seconds]
rasterman has joined #panfrost
nlhowell is now known as Guest989
nlhowell has joined #panfrost
Guest989 has quit [Read error: Connection reset by peer]
enunes has quit [Read error: Connection reset by peer]
enunes has joined #panfrost
nlhowell is now known as Guest993
nlhowell has joined #panfrost
Guest993 has quit [Read error: Connection reset by peer]
Rathann has joined #panfrost
rkanwal has joined #panfrost
icecream95 has quit [Ping timeout: 480 seconds]
<macc24> oh!
<HdkR> Nice
<HdkR> alyssa: RootFS images updated with Mesa 22.1 now. Probably not as nice as thunks but that's going to take a bit of time still
<alyssa> HdkR: woof
<HdkR> printf debugging to the rescue? :P
<HdkR> Crashes and asserts can still be fairly easily tracked at least.
camus has quit [Ping timeout: 480 seconds]
nlhowell has quit [Ping timeout: 480 seconds]
guillaume_g has quit []
<macc24> greenjustin: ^^^^^^ i linked patch series that you might find interestg
<greenjustin> macc24: Was just about to tell you about this actually lol. ChenYu is on a sister team, he figured out a good solution for the DVFS stuff on MT8183
<macc24> oh nice
<greenjustin> Apparently there's some weird requirements about keeping the SRAM voltage a certain level above the GPU voltage? So the regulators have to be changed slowly and in lock step
<macc24> wtf
<greenjustin> Might have been the source of some of the instability, in addition to the clock reparenting dance
Rathann has quit [Ping timeout: 480 seconds]
Lyude has quit [Quit: WeeChat 3.4]
bbrezillon has quit [Ping timeout: 480 seconds]
bbrezillon has joined #panfrost
rasterman has quit [Quit: Gettin' stinky!]
camus has joined #panfrost
camus has quit [Ping timeout: 480 seconds]
Lyude has joined #panfrost
Lyude has quit [Read error: Connection reset by peer]
Lyude has joined #panfrost
rasterman has joined #panfrost
rkanwal1 has joined #panfrost
rkanwal has quit [Ping timeout: 480 seconds]
<alyssa> Weee!
<alyssa> Turns out clocks were completely hosed on my MT8192 device tree, so the GPU was stuck at boot frequency which was slow
<alyssa> with some help from a wonderful colleague, I got a hacky devfreq working, letting me pin the max frequency used in chromeos
<alyssa> (866 MHz. MediaTek claims the GPU goes up to 950 MHz, but ChromeOS doesn't use it, and when I tried the GPU was unstable.)
<alyssa> Here I was wondering why my 2022 chromebook was only getting slightly better perf than my 2015 one
<alyssa> now the new shiny is going a *lot* faster ;)
<alyssa> gfxbench numbers still suck though :V
<alyssa> supertuxkart's deferred renderer isn't hitting 60fps but is definitely playable now
<alyssa> neverball is of course vsyncing at 60fps, as it was at base frequency, honestly I'd have to try to make neverball slow :-p
<cphealy> That's great to hear! Hopefully panfrost with working devfreq on MTK SoCs gets fully worked out soon.
<alyssa> [ 91.808183] panfrost 13000000.gpu: gpu sched timeout, js=1, config=0x0, status=0x0, head=0x0, tail=0x0, sched_job=000000003f0fb704
<alyssa> [ 91.830818] panfrost 13000000.gpu: gpu soft reset timed out
<alyssa> [ 91.856486] panfrost 13000000.gpu: error powering up gpu L2
<alyssa> [ 91.882215] panfrost 13000000.gpu: error powering up gpu shader
<alyssa> [ 91.889242] panfrost 13000000.gpu: error powering up gpu tiler
<alyssa> uhoh
icecream95 has joined #panfrost
<alyssa> Rendered 1335 frames in 56.153 secs, average of 23.7743 fps
<alyssa> manhattan, yo
<alyssa> just need to make it... 3x faster? :p
<alyssa> 92% gpu active
<alyssa> 7.5% read stall cycles
<icecream95> alyssa: How many glmarks?
<alyssa> icecream95: 7 glmarks
<alyssa> glmark2 -bbuild at 2427fps, if that means anything? :-p
<alyssa> more usefully, terrain at 81fps
<icecream95> alyssa: So you are running one of the mir variants, otherwise there's only {drm,wayland,x11}-gl{,esv2}?
<alyssa> desktop at 443fps
<alyssa> either glmark2-gl or glmark2-glesv2, don't remember
<alyssa> er
<alyssa> x11-gl or x11-glesv2
<icecream95> Whoa... is X11 rendering that fast? Impressive. Under Xwayland then?
<alyssa> yeah
<alyssa> though window surface is u-interleaved, not AFBC
<alyssa> probably would be better if we could decide what order of colour channels to paint the AFBC bikeshed
<alyssa> (as cphealy has been complaining about for a year now? :p)
<icecream95> YUV!
<alyssa> I yuv that idea!
<cphealy> alyssa: That's my specialty..
<alyssa> cphealy: Complaining when I do dumb things? I thought that's icecream95's specialty, get your own specialty ;-p
<icecream95> But I think that ARM has four different single-plane formats for 2x2 subsampled YUV?
<cphealy> Is the lack of this part of the problem? https://gitlab.freedesktop.org/-/snippets/3177
<icecream95> alyssa: Oh I've got more things to complain about, if you'd only bother to read your emails :P
<alyssa> No.
<alyssa> icecream95: *sweats*
<alyssa> Blend shader calls (call_blend_shader): 7622219
<alyssa> See, that's funny, because I happen to know there are no blend shaders anywhere here.......
<alyssa> 16-bit interpolation active (vary_slot_16): 521456
<alyssa> and I haven't wired up 16bit interp
<alyssa> which makes me wonder if *any* of these counters are legit
<icecream95> alyssa: Probably the banks are switched round again?
<alyssa> maybe?
<alyssa> let's see
<icecream95> So Valhall is almost twice as fast as T760 (at glmark) already?
<cphealy> with the same frequency and core count?
<alyssa> this is Mali-G57 with 5 cores
<alyssa> similar clock rate
<cphealy> ack
<alyssa> icecream95: "Probably the banks are switched round again?"
<alyssa> Effectively
<alyssa> Actually, src/panfrost/perf/ is just totally broken
<alyssa> it is extremely friday night for me to fix it though :p
rasterman has quit [Quit: Gettin' stinky!]
<alyssa> oh man that is magic, vim
<alyssa> ctrl-v 10jll 192 ctrl-x
<alyssa> "subtract 192 from this column"
<alyssa> ^each number in the column
wolfshappen has joined #panfrost
<alyssa> Blend shader calls (call_blend_shader): 0
<alyssa> there.
<alyssa> Tiles (frag_ptiles): 889868630
<alyssa> uh maybe no
<alyssa> that's a monday problema nyway
<alyssa> night!~
<icecream95> g'night
icecream95 has left #panfrost [rcirc on GNU Emacs 28.1]
wolfshappen has quit []