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
vstehle has quit [Ping timeout: 246 seconds]
nroberts has quit [Ping timeout: 240 seconds]
<robclark> we could probably run GALLIUM_HUD w/ one of the trace-replay tests.. I guess it might not be easy to validate result image in a sane way but it could at least validate that things don't crash
<robclark> tomeu: ^^^
<robclark> (that said, pretty sure I've used GALLIUM_HUD within the last week without issue)
<robclark> (maybe it was last couple weeks.. but recentlyish)
atler is now known as Guest9230
Guest9230 has quit [Killed (tolkien.freenode.net (Nickname regained by services))]
atler has joined #panfrost
bbrezillon has quit [Ping timeout: 240 seconds]
camus has joined #panfrost
kaspter has quit [Ping timeout: 240 seconds]
camus is now known as kaspter
<alyssa> robclark: lower_uniforms_to_ubo bug that only affected midgard due to silly compiler things, see #dri-devel or latest gitlab MR
camus has joined #panfrost
kaspter has quit [Ping timeout: 260 seconds]
camus is now known as kaspter
davidlt has joined #panfrost
kaspter has quit [Ping timeout: 240 seconds]
kaspter has joined #panfrost
<macc24> fun fact: increasing dmc frequency on rk3326 bumps gpu performance quite a bit
<alyssa> yep..
vstehle has joined #panfrost
camus has joined #panfrost
kaspter has quit [Remote host closed the connection]
camus is now known as kaspter
cowsay has quit [Ping timeout: 260 seconds]
nlhowell has joined #panfrost
cowsay has joined #panfrost
nlhowell has quit [Ping timeout: 252 seconds]
<tomeu> robclark: what would be the problem when validating the result image?
bbrezillon has joined #panfrost
raster has joined #panfrost
orkid has quit [Ping timeout: 258 seconds]
pendingchaos has quit [Ping timeout: 252 seconds]
pendingchaos has joined #panfrost
icecream95 has quit [Ping timeout: 245 seconds]
icecream95 has joined #panfrost
raster has quit [Ping timeout: 240 seconds]
raster has joined #panfrost
stikonas has joined #panfrost
warpme_ has joined #panfrost
raster has quit [Ping timeout: 252 seconds]
Net147 has quit [Ping timeout: 252 seconds]
Net147 has joined #panfrost
Net147 has quit [Read error: Connection reset by peer]
Net147 has joined #panfrost
paulk-leonov has quit [Ping timeout: 260 seconds]
popolon has joined #panfrost
raster has joined #panfrost
raster has quit [Ping timeout: 252 seconds]
raster has joined #panfrost
raster has quit [Ping timeout: 240 seconds]
raster has joined #panfrost
raster- has joined #panfrost
raster has quit [Ping timeout: 240 seconds]
raster- has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
unoccupied has joined #panfrost
unoccupied is now known as leah
Elpaulo has quit [Remote host closed the connection]
Elpaulo has joined #panfrost
thecycoone has quit [Quit: Bridge terminating on SIGTERM]
thecycoone has joined #panfrost
archetech has joined #panfrost
<robclark> tomeu: that it would change from run to run.. unless you managed to clip out the part of the screen where the HUD is
<tomeu> oh, because the data would be coming from the GPU?
<robclark> right
kaspter has quit [Ping timeout: 240 seconds]
kaspter has joined #panfrost
leah has quit [Quit: WeeChat 2.8]
unoccupied has joined #panfrost
unoccupied is now known as leah
leah has quit [Client Quit]
kaspter has quit [Quit: kaspter]
<alyssa> robclark: samples-passed "should" be consistent
<robclark> perhaps
<alyssa> robclark: worst case could add a GALLIUM_HUD=dummy that just returns constant value, to check if the HUD is rendered correctly at all
megi has quit [Quit: WeeChat 3.1]
megi has joined #panfrost
<robclark> maybe something like # of draws per frame?
<alyssa> that'd do it
<alyssa> deceptively complicated though because of indirect multidraw
<robclark> or maybe "# of draw *calls*" per frame?
<cphealy> Regarding the HUD and draw calls, I notice that etnaviv exposes draw-calls with the GALLIUM_HUD here: https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/gallium/drivers/etnaviv/etnaviv_query_sw.c#L121
<cphealy> Is draw call count reporting supported with Panfrost?
<cphealy> I've found this very useful when optimizing application performance.
<alyssa> robclark: Heh, sure
<alyssa> cphealy: Nobody's typed it out but obviously that's a sw-only feature
<robclark> I heard panfrost also has some perfcntrs that may be supported by perfetto at some point :-P
<alyssa> robclark: Hush you :-P
stikonas has quit [Remote host closed the connection]
<cphealy> alyssa: as it's a sw-only feature, I wonder if there's a way to make it generic to all gallium drivers instead of adding it to each driver explicitly.
stikonas has joined #panfrost
<robclark> like alyssa mentioned, things like glMultiDrawElementsIndirect() make it not really a sw-only feature for a lot of drivers
<cphealy> ahh, I missed that part, tnx
warpme_ has quit [Quit: Connection closed for inactivity]
<alyssa> cphealy: draw call count is sw only, draw count is not
<cphealy> Yep, makes sense. With etnaviv I used the draw-calls HUD entry to get a better understanding of how well Qt draw call batching was working when dealing with a bunch of movie posters on the screen.
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
archetech has quit [Quit: Konversation terminated!]
warpme_ has joined #panfrost
danboid has joined #panfrost
<danboid> Has anyone got a panfrost mesa Arch PKGBUILD?
<alyssa> danboid: mesa-git is in AUR iirc
<danboid> alyssa, Its horribly broken and outdated. Plus no panfrost
<danboid> I've almost finished tweaked the regular one, I think
<danboid> tweaking
<danboid> not twerking
<alyssa> danboid: panfrost is just setting -Dgallium-drivers=panfrost
<alyssa> to meson
<alyssa> and -Dllvm=false if you have a vendetta against LLVM, i guess
<danboid> alyssa, Yep, thats what I thought. It should've just been that, removing the drivers I don't need and adding aarch64 to the supported arch but it didn't work out like that
<danboid> Its building, its just the install bit I need to fix now tbh
<danboid> alyssa, Are you mostly working on the Apple GPU now? What is the nearest GPU, spec wise?
<danboid> Its not that beefy is it?
<alyssa> danboid: Panfrost is my full time job, Apple is just (one of several...) hobby on nights/weekends
<danboid> alyssa, Oh! I didn't know you were paid to work on panfrost. Who sponsors that?
<alyssa> and uhhhh if you took the highest end Mali and put dozens of them in a cluster, you might get close to the M1's GPU ...
<alyssa> danboid: Collabora.
<danboid> Oh wow! Its more powerful than I thought then. I expected it to be like an updated PowerVR thing, Applised
<amonakov> alyssa: but Collabora is not seeking that for itself, someone is sponsoring Collabora in turn?
<danboid> Collabora do a lot of cool stuff. Do they have physical offices for the devs?
<danboid> Not anymore most likely
<alyssa> amonakov: /* no comment */
<alyssa> danboid: We used to have an office in Montreal.
<alyssa> I mean, the building is still there but... ;)
* alyssa coughs
<danboid> I'll upload my PKGBUILD when I fix it up and hae tested it then
<danboid> Ohh don't cough! ;)
* alyssa goes and gets a vaccine dose so she's allowed to cough again
<alyssa> Don't you miss randos coughing on you in the winter in enclosed spaces?
<alyssa> 2019 was such a great time.
<danboid> Don't think you can cough just because you've been jabbed!
<danboid> I can't remember the last time I heard someone cough
<alyssa> I'm not allowed to infect you with viruses other than SARS-CoV-2?
<danboid> That's the only infectant in town!
<robmur01> amonakov: as a consultancy, I'd imagine Collabora don't do much "for themselves" (apart from fulfilling their contracts, if you look at it that way) ;)
<danboid> Have any of you ever tried icecc with a mixture of amd64 and aarch64 nodes?
<danboid> Its on my todo list for tonight
<danboid> can't be sditting around waitind an hour to build mesa :/
<robmur01> an hour? pff, it's about 10 minutes on my RK3399
<danboid> icecc is super simple to set up vs distcc, which I never got to work, and apparently it does support heterogeneous, cross compiling configs
<danboid> robmur01, I'm upgrading from a 20 MB/s sd card to a NvME disk right now so that should help a bit
<danboid> Although I'm using it via an USB 3 adapter so I only get a fraction of the speed, it's still a damn sight better
<danboid> Its a USB C adapter but my ARM only has USB 3 so
<danboid> robmur01, What is your ARM device of choice this hour?
<robmur01> well, the machine I'm typing on is a shiny new Surface Pro X, but I guess that doesn't really count :P
<danboid> Sure doesn't!
<danboid> Try again
<robmur01> my Avantek eMAG workstation in the office?
<robmur01> (although that's not actually of this hour, because some git unplugged it from the network a while ago and I've not sorted it out yet)
<danboid> Oooh! I didn't even know they existed!
<danboid> It EVEN has a serial port!
<danboid> Yow!
<danboid> How much for one of those bad boys? Its the polar opposite of my wee TV box toy
<danboid> They don't want to tell you on the site.
<robmur01> for Panfrost, my NanoPC-T4 sees the most action, and when I find the time to tinker on work hours I've got an Arm Juno board with its hard T620 plus various Midgard and Bifrost FPGA bitfiles
<danboid> Those are acceptable answers indeed
<robmur01> one day I might un-break the RK3288 box as well, but, increasingly, meh
<danboid> robmur01, Do you know how much one of those eMAGs costs? Is that the most powerful ARM workstation out there?
<danboid> alyssa, Have you got anything like an eMAG?
<robmur01> not sure, but I think it's of the order of "not much change from £5k"
<danboid> Sounds about right
<robmur01> at this point you'd probably be better off holding out for Altra-based systems to start showing up
<danboid> Thats a new thing to me
<danboid> Is this the next gen ARM CPU?
<robmur01> eMAG is OK for day-to-day, but it's still basically v8.0 X-Gene, and in single-threaded stuff it really shows
<robmur01> Altra is Ampere's Neoverse-N1-based offering - similar to AWS Graviton2 but with higher clocks, and (eventually) actually ownable
<danboid> Do you use a Radeon with that?
<danboid> Do the AMD mesa drivers work on aarch64?
<alyssa> danboid: Uh, an M1 mini :p
<danboid> alyssa, It could be worse
<alyssa> Don't use it for Mesa dev yet, holding out until I can put proper Linux on it
<danboid> Looks like Hector is making amazing progress. That man knows all things tech it seems
davidlt has quit [Ping timeout: 260 seconds]
thecycoone has quit [Quit: Bridge terminating on SIGTERM]
thecycoone has joined #panfrost
<macc24> is g31 broken on wayland on main branch only for me?
raster has quit [Quit: Gettin' stinky!]
<alyssa> broken how?
<macc24> after pulling main branch it now works
<macc24> ppsspp works with panblit fyi
<alyssa> I like the mane branch.
mixfix41 has quit [Ping timeout: 258 seconds]
stikonas has quit [Remote host closed the connection]
pendingchaos has quit [Ping timeout: 260 seconds]
pendingchaos has joined #panfrost
stikonas has joined #panfrost
raster has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
robmur01 has quit [Read error: Connection reset by peer]
robmur01 has joined #panfrost