ChanServ changed the topic of #dri-devel to: <ajax> nothing involved with X should ever be unable to find a bar
<jekstrand> karolherbst: I could believe that radeonsi expect variables for images and samplers.
<karolherbst> yeah, it does, but the question was why explicit_io lowers samplers, no?
<karolherbst> but anyway, yeah, that's the thing lowering our samplers to constants
<karolherbst> txl still has sampler_deref things
<jekstrand> I don't remember exactly how I wrote that pass but maybe it did to through lower_explicit_io?
<karolherbst> sooo.. lower_explicit_io turnst the deref_load things into load_const
<jekstrand> When I wrote it, I was assuming everyone wanted indices, not derefs. :0/
<karolherbst> and lower_cl_images turns sampler_deref sources of txl into sampler
<karolherbst> or sampler_offset
<karolherbst> yeah...
<karolherbst> but I think long term zink also wants derefs as it already requires derefs for images
<zmike> leave zink alone
<karolherbst> anyway.. I need a solution and I don't like my current one 🙃
<karolherbst> zmike: too late for that, don't you think?
<zmike> 😬
glennk has quit [Ping timeout: 480 seconds]
<DavidHeidelberg[m]> Gallium Nine seems to be passing a lot tests on freedreno robclark btw. sorry I said hi, I was a feeling bit sick at XDC22 and I didn't want to interrupt aftertalk discussion :)
jewins has quit [Ping timeout: 480 seconds]
heat has quit [Ping timeout: 480 seconds]
<DavidHeidelberg[m]> daniels: btw. I looked into piglit script integration, `.gitlab-ci/piglit/piglit-runner.sh` seems to be place to put it, but it look extremely hacky. I would have to check for fractional runs, probably introduce variable to control it (since some drivers doesn't have wired nine support).. currently I tried only check for HWCI_START_XORG but that's not enough. I'll think about it bit tomorrow, I still don't see better option than
<DavidHeidelberg[m]> piglit jobs (one unlucky thing is that mine machine is TGL, but sadly tgl doesn't run piglit w/ iris, only VK)
<DavidHeidelberg[m]> * I didn't said Hi
ngcortes_ has quit []
FireBurn has quit [Ping timeout: 480 seconds]
<LaserEyess> when I'm adding a drm property with a blob, if I set the blob_id to 0, should that "reset" the blob to the "defaults"?
<LaserEyess> in particular, I want to reset my screen to SDR after using HDR_OUTPUT_METADATA to display HDR
<LaserEyess> drmModeAtomicProperty(...., 0) doesn't seem to work
columbarius has joined #dri-devel
co1umbarius has quit [Ping timeout: 480 seconds]
slattann has joined #dri-devel
Emantor has quit [Quit: ZNC - http://znc.in]
Emantor has joined #dri-devel
<robclark> DavidHeidelberg[m]: kinda cool that anything that uses nine runs on arm things :-)
<DavidHeidelberg[m]> robclark: Box86! :) https://www.youtube.com/watch?v=DfRpY9zfVdY
<DavidHeidelberg[m]> I saw some more "benchmark like" video where someone was using Nine on RPi4 and getting like +20fps extra. I think even here it's visible 20-23 -> 26-30 which is nice on low-power devices
<robclark> nice.. I need to find a 128GB device to switch over to dev mode and move my linux setup since too much of my disk is filled up with deqp ;-)
<DavidHeidelberg[m]> it copies reason why we work on it in 2014. Difference between 20 and 30 fps is not big, but playable vs unplayable :D
<DavidHeidelberg[m]> deqp is ... everywhere :D
<robclark> and also, debugging deqp issues under linux is far less unpleasant than debugging them under cros or android ;-)
<DavidHeidelberg[m]> robclark: this guy showing up Nine on RPi4 https://www.youtube.com/watch?v=O0XNsom1kdE
<robclark> so, where's the version of that video w/ freedreno ;-)
<robclark> I guess it should run pretty well on a618 chromebook if it runs like that on rpi4
<robclark> fwiw, tu should have vk 1.3 which I think should be enough for dxvk
<DavidHeidelberg[m]> robclark: hehe, I have to compare panfrost w/ freedreno supported features (nine-tests), but I think freedreno will not be at this moment run complex stuff (just guessing, haven't tried anything on my phone yet)
<DavidHeidelberg[m]> robclark: dxvk has high cpu and memory costs on embedded, it may not be best option for that reason
<robclark> pre-a6xx, you might be right as far as features/etc.. a6xx it should not be a problem.. vk 1.3 and with an mr I pushed today gl4.3 (and the associated gallium driver features for nine)
<DavidHeidelberg[m]> my first test of my freedreno enablement patch was nine-tests (as I wrote, 10000 passes, ~85 fails), while on Intel which it's like 20000 passes
<DavidHeidelberg[m]> cool
ana has quit [Remote host closed the connection]
yang_ has joined #dri-devel
milek7_ has joined #dri-devel
<DavidHeidelberg[m]> robclark: wow, that looks like huge pile of work. More traces will work! :D
bbrezillon has quit [Ping timeout: 480 seconds]
eloy__ has joined #dri-devel
yang has quit [Ping timeout: 480 seconds]
<DavidHeidelberg[m]> oh viewports! d3d9 feel happy about your MR. I'll test it against nine-tests
milek7 has quit [Ping timeout: 480 seconds]
eloy_ has quit [Ping timeout: 480 seconds]
<robclark> tbh recent gl4 and gl43 work was pretty small.. for most part polishing off features that already existed for gles or switching on existing lowering passes
bbrezillon has joined #dri-devel
<robclark> fwiw, zink on tu should get you gl46 (but at some layering overhead cost)
ana has joined #dri-devel
<HdkR> Indeed it does
Company has quit [Quit: Leaving]
* robclark should push the patch to turn on GL_ARB_texture_mirror_clamp_to_edge since it is already supported
<DavidHeidelberg[m]> robclark: hmm, I'm afraid the MR broke two tests nine-tests, one fails , one flakes (maybe unrelated, have to run more times)
<robclark> patches welcome to get nine in CI ;-)
<DavidHeidelberg[m]> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19246 :D (but only DRAFT, since it runs < 5 minutes, I think I'll squash it with piglit as Daniel proposed)
<robclark> was it pass->fail or skip->fail.. turning on a bunch of functionality could ofc expose bugs that we didn't hit before
<DavidHeidelberg[m]> it it's pass -> fail
<DavidHeidelberg[m]> ok, the flake was probably random something, but the fail persist
<robclark> hmm, well I can see if I can find a device with more disk space that I can move over too.. but won't be this weekend..
<DavidHeidelberg[m]> robclark: btw. if you want try nine-tests, they have like 1kB and repo is like 100kB..
<robclark> I mean if I can get the whole setup with dependencies to repro in 100k, that I can manage
<DavidHeidelberg[m]> yeah, I think you have all deps
<robclark> was that main vs fd/viewport-array?
fxkamd has joined #dri-devel
<DavidHeidelberg[m]> into mesa CI I had to add just libxcb-dri3-dev libxcb-present-dev
<DavidHeidelberg[m]> robclark: this is description how to build it in fact: https://gitlab.freedesktop.org/okias/mesa/-/commit/c06c17b0754bd08d843a9ff426ee2a5ce5d389af
<DavidHeidelberg[m]> just without the sed for D3DADAPTER9_LOCATION, are u on Debian?
yuq825 has joined #dri-devel
slattann has quit []
<robclark> no, fedora
<DavidHeidelberg[m]> robclark: then you have to put path to your d3dadapter.so by hand into Makefile
<DavidHeidelberg[m]> never mind, you have built mesa locally.. so you have to do that anyway (since without the patch to meson it won't link freedreno against nine)
<robclark> first step is to enable nine in my build.. but I probably won't do that tonight ;-)
<DavidHeidelberg[m]> robclark: `curl -L https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19244.patch | git am` :D
<DavidHeidelberg[m]> oh right, even in US is night. I got to used to UTC-7 timezone and now I'm in UTC+2
<DavidHeidelberg[m]> * UTC-5
<DavidHeidelberg[m]> I just did one testing build without MR commits and no fail again. So 99% confirmed.
<DavidHeidelberg[m]> Anyway, good night and thanks! gtg sleep, It's like 4:30 am here :)
<robclark> if main is green but mr is not sounds like nine hits some case that piglit and other CI tests do not.. I'll look later but I guess there might be some different code paths involved with new pipe caps and glsl version
<robclark> anyways gn ;-)
<mareko> karolherbst: we want to lower sampler and image derefs eventually
sdutt_ has joined #dri-devel
sdutt has quit [Ping timeout: 480 seconds]
pendingchaos_ has joined #dri-devel
pendingchaos has quit [Ping timeout: 480 seconds]
sdutt_ has quit [Read error: Connection reset by peer]
sdutt has joined #dri-devel
Danct12 has joined #dri-devel
<Frogging101> Hm, how do I run kwin_wayland from a command line? Unlike kwin_x11 one can't simply replace the running instance because the wl compositor is the session.
YuGiOhJCJ has joined #dri-devel
kts has joined #dri-devel
kts has quit [Quit: Leaving]
kts has joined #dri-devel
fab has joined #dri-devel
sdutt has quit []
sdutt has joined #dri-devel
sdutt_ has joined #dri-devel
kts has quit [Quit: Leaving]
sdutt_ has quit []
sdutt has quit []
chipxxx has quit [Remote host closed the connection]
fxkamd has quit []
kts has joined #dri-devel
chipxxx has joined #dri-devel
<ishitatsuyuki> Frogging101, you might want to ask at #kde-devel or #kwin @ libera
Akari has quit [Ping timeout: 480 seconds]
kts has quit [Quit: Leaving]
alanc has quit [Remote host closed the connection]
alanc has joined #dri-devel
Duke`` has joined #dri-devel
robink has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
robink has joined #dri-devel
Danct12 has quit [Remote host closed the connection]
rasterman has joined #dri-devel
rgallaispou has joined #dri-devel
rgallaispou1 has quit [Ping timeout: 480 seconds]
flibit has joined #dri-devel
glennk has joined #dri-devel
flibitijibibo has quit [Ping timeout: 480 seconds]
Akari has joined #dri-devel
psykose has quit [Remote host closed the connection]
psykose has joined #dri-devel
pcercuei has joined #dri-devel
Jeremy_Rand_Talos has quit [Remote host closed the connection]
Jeremy_Rand_Talos has joined #dri-devel
Jeremy_Rand_Talos has quit [Remote host closed the connection]
Jeremy_Rand_Talos has joined #dri-devel
Duke`` has quit [Ping timeout: 480 seconds]
Company has joined #dri-devel
CounterPillow has quit [Quit: Bye.]
CounterPillow has joined #dri-devel
gouchi has joined #dri-devel
gouchi has quit []
linkmauve has left #dri-devel [#dri-devel]
linkmauve has joined #dri-devel
Akari has quit [Ping timeout: 480 seconds]
gouchi has joined #dri-devel
Namarrgon has quit [Ping timeout: 480 seconds]
gouchi has quit [Remote host closed the connection]
Haaninjo has joined #dri-devel
FireBurn has joined #dri-devel
Net147 has quit [Quit: Quit]
Net147 has joined #dri-devel
Namarrgon has joined #dri-devel
danvet has joined #dri-devel
chip_x has joined #dri-devel
chipxxx has quit [Ping timeout: 480 seconds]
kts has joined #dri-devel
kts has quit [Quit: Leaving]
YuGiOhJCJ has quit [Quit: YuGiOhJCJ]
digetx has quit [Ping timeout: 480 seconds]
<karolherbst> nice.. our implementation of SPIR-V FRem and FMod is probably wrong :(
<zamundaaa[m]> Frogging101: you can run most compositors directly from the tty. That includes kwin_wayland
<Frogging101> zamundaaa[m]: When I do that, it replaces the tty I ran it from. Is there an env var or something?
<zamundaaa[m]> Depends on what you want do to exactly
<jenatali> karolherbst: oh?
<zamundaaa[m]> If you want to see logs, try journalctl
chipxxx has joined #dri-devel
<karolherbst> jenatali: yeah.. read that stuff very very very closely: https://registry.khronos.org/SPIR-V/specs/unified1/SPIRV.html#OpFRem
<karolherbst> I think it was changed to that with 1.4?
<karolherbst> 1.3 actually
<karolherbst> or I think the descrittion on what the sign shall be got added later
<jenatali> Oh about sign matching?
<karolherbst> yeah...
<karolherbst> but I also get not matching values somehow
<karolherbst> doing it all explicit fixes some weird corner cases
<jenatali> Sounds like nir semantics don't necessarily match up with spirv
<karolherbst> I am still trying to figure out why the fmod/frem spirv_new tests fails, it's such a mess and I think both sides are weirdly broken
<karolherbst> the CTS included
<karolherbst> the CTS tries to match those two ops to actual OpenCL C fmod, which... well.. won't work
<karolherbst> or maybe it would? dunno
<karolherbst> though I don't think Kernel SPIR-V should require CL ULPs for SPIR-V ops
<jenatali> Why not?
<karolherbst> because why would it? if you want CL ulps, use the ExtInst stuff
<karolherbst> otherwise mapping CL on VK is a pain
<jenatali> I guess. Makes more sense for things that don't already exist in core SPIR-V though
<jenatali> Mapping CL on DX is the same pain fwiw
<karolherbst> well those are all core spir-v things, no?
<karolherbst> yeah well.. if CL C wants higher precisions if should go through CLC builtins which actually works more or less
<jenatali> Sure where the required precision is defined by the environment
chip_x has quit [Ping timeout: 480 seconds]
<karolherbst> nope
<karolherbst> the only uls the CL SPIR-V env spec defines is FAdd/FSub/FMul and FDiv
<karolherbst> everything else is OpExtInst stuff
<karolherbst> well.. it also defines FMul and FAdd together as correctly rounded fma, but that's a minor detail
<karolherbst> FRem or FMod are not even mentioned
<karolherbst> ahh you wrote "where" not "were"
<karolherbst> anyway, mesa is wrong and the CTS is also wrong :P
<Frogging101> zamundaaa[m]: I want to see console output and I want to add prints to the code which I don't want to flood my journal with. I want to do debugging, basically.
<Frogging101> And I need to run it in a way that gives me the flexibility to play with it
pa- has joined #dri-devel
<zamundaaa[m]> Frogging101: If your debugging doesn't require directly controlling outputs, just run it in a terminal emulator. It'll automatically make a nested session with kwin in a window
<Frogging101> I do need to be able to interact with it and run a game inside it
pa has quit [Ping timeout: 480 seconds]
<Frogging101> So I also need the shell and everything. Running it nested sounds like asking for trouble
<Frogging101> Though correct me if I'm wrong
kts has joined #dri-devel
digetx has joined #dri-devel
yuq825 has quit []
kts has quit [Quit: Leaving]
<zamundaaa[m]> There's no difference in what you can do in a nested session, minus drm leasing perhaps. If you want a full session, iirc you can use
<zamundaaa[m]> `dbus-run-session startplasma-wayland`
camus1 has quit []
gouchi has joined #dri-devel
gouchi has quit [Remote host closed the connection]
djbw has quit [Read error: Connection reset by peer]
chipxxx has quit [Ping timeout: 480 seconds]
Akari has joined #dri-devel
<Frogging101> zamundaaa[m]: I will try it then, thanks :)
soreau_ has joined #dri-devel
<Frogging101> Although if it is nested then we have the input system of the host session to contend with. And the input system is what I'm trying to debug. But I will try it and see.
soreau_ has quit [Read error: Connection reset by peer]
chipxxx has joined #dri-devel
digetx has quit [Remote host closed the connection]
digetx has joined #dri-devel
cmarcelo has quit [Quit: leaving]
cmarcelo has joined #dri-devel
<illwieckz> <alyssa> illwieckz: Any idea if luxmark will build and run on arm64?
<illwieckz> lots of quirks :D
flibit has quit []
srslypascal is now known as Guest3746
srslypascal has joined #dri-devel
Guest3746 has quit [Ping timeout: 480 seconds]
digetx is now known as Guest3748
digetx has joined #dri-devel
Guest3748 has quit [Ping timeout: 480 seconds]
Duke`` has joined #dri-devel
akselmo has joined #dri-devel
heat has joined #dri-devel
Leopold has joined #dri-devel
Akari has quit [Ping timeout: 480 seconds]
<Frogging101> zamundaaa[m]: The bug isn't happening with nested Wayland, sadly. This is the bug I'm looking at. http://www.fastquake.com/files/kwin_wayland-teardown-mouse.mp4
<Frogging101> Basically the wrong pointer position or delta is being reported occasionally and making the map fly to one edge when dragging it
<Frogging101> okay my alt-tab key isn't working anymore, wtf
<karolherbst> illwieckz: yay
<karolherbst> illwieckz: try llvmpipe :D
<karolherbst> sadly I use my pi4 for some infra stuff at home :(
<Frogging101> What might I have done to kill my alt-tab key?
<Frogging101> Somehow my alt-tab key now only works when I have a nested wayland compositor running.
<Frogging101> I bet it the host session gave up the hotkey because of some special-case and now it's gone.
<illwieckz> karolherbst, my big issue is that for now I build on the pi3 😅️
<karolherbst> just use packaged LLVM :P
<karolherbst> mesa should build in like 2 hours
<karolherbst> :D
<illwieckz> so even with 1 core, because there is only 1G of ram, I can only use 1 build job 😁️
<illwieckz> yeah 2 hours is a reasonable estimation
<karolherbst> nah... I build mesa with 4 cores on a jetson nano
<karolherbst> was fine
<illwieckz> ah
<illwieckz> good to know
<illwieckz> I would take any tutorial to make a qemu-based “virtual raspberry pi” anyway
<illwieckz> even through emulation, I assume 20+ threads and dozens of gigabytes of ram would speed up the build.
<linkmauve> illwieckz, I “rent” a free VM at Oracle Cloud, they require you to give them a debit card but afterwards it’s effectively free.
<linkmauve> And for that you get four cores of Neoverse N2 and 24 GiB of RAM.
<linkmauve> I build Mesa in like three minutes on there.
<linkmauve> It became my main build server for ARM programs, so that I don’t have to worry about cross-compilation.
pa has joined #dri-devel
<linkmauve> This CPU has many additional feature flags compared to a SBC (almost all of them are using Cortex-A53, not Neoverse-N2) so you should pass -march=armv8-a+crypto -mtune=cortex-a53 in your CFLAGS.
<linkmauve> -march=native won’t work.
<DavidHeidelberg[m]> linkmauve: yeah, love OC for that, less for other their choices :D
<DavidHeidelberg[m]> thou beware, keep your config in script or ansible, since after 1 month trial the machine gets purged, but after that I never lost the content
pa- has quit [Ping timeout: 480 seconds]
<linkmauve> Oh? They never did that to me.
zaratustra has quit [Ping timeout: 480 seconds]
<DavidHeidelberg[m]> after the trial expired they reset all stuff to 0 and after I setup "free tier" it working until today
zaratustra has joined #dri-devel
rasterman has quit [Quit: Gettin' stinky!]
zaratustra has quit [Ping timeout: 480 seconds]
Duke`` has quit [Ping timeout: 480 seconds]
Akari has joined #dri-devel
heat has quit [Read error: No route to host]
heat has joined #dri-devel
fab has quit [Quit: fab]
zaratustra has joined #dri-devel
danvet has quit [Ping timeout: 480 seconds]
mhenning has joined #dri-devel
<LaserEyess> are there any drivers that do not support the drm atomic API? is it required for new drivers to support?
<jannau> illwieckz, karolherbst: a M2 is a little bit faster with PoCL: https://twitter.com/jaguar_9/status/1583950125583724544
<karolherbst> that's quite slow though
<karolherbst> is that emulated?
djbw has joined #dri-devel
<jannau> native with the exception of https://github.com/simd-everywhere/simde for emulating x86 simd intrinsics on arm64
<emersion> LaserEyess: yes there are, yes it is
<jannau> threading seems to have a little trouble with big.LITTLE, only 650% CPU use, but the efficiency cores are not fully used
<LaserEyess> emersion: is there a list somewhere? the context is mpv's drm backend, I need to know if I can drop the legacy interface to clean up some code
<LaserEyess> I have no idea who uses mpv's drm backend, but probably a lot of embedded stuff
<jannau> score for 4 perf cores: 1800, 4 efficiency cores: 850, both with 385% CPU use
<jannau> LaserEyess: it's useful
<LaserEyess> the legacy API?
<jannau> for testing DRM drivers, I use it to test the display controller in apple silicon devices
<LaserEyess> ah
<jannau> sorry, stray 'return', no not the legacy ABI
<jannau> API
<LaserEyess> oh, well, no one is talking about dropping mpv's drm backend
<LaserEyess> I'm talking about improving it
<LaserEyess> I assume the m1 GPU, when it goes in, will support the atomic API, so that won't be an issue
<jannau> yes, it does support the atomic API
<LaserEyess> by the way since I have people here, does anyone know what should happen when you reset the HDR_OUTPUT_METADATA blob to 0
<LaserEyess> I tried to do what was said in this comment https://github.com/mpv-player/mpv/pull/8456#issuecomment-778899198
<LaserEyess> but it does not reset my display to SDR
<LaserEyess> I still have to create a new blob that switches the display to SDR
Jeremy_Rand_Talos has quit [Remote host closed the connection]
Jeremy_Rand_Talos has joined #dri-devel
pcercuei has quit [Quit: dodo]
<emersion> LaserEyess: drmdb
<LaserEyess> oh nice
<LaserEyess> I'm not sure what ast is, but basically older amd/intel cards don't support it
<heat> is there a way to run mmiotrace since early boot? I wanted to trace i915.ko to better understand how i915 initialization works
<LaserEyess> aspeed, so basically not anything mpv will ever really use
<karolherbst> heat: nope, because it will intercept evertyhing loaded after it got enabled
<karolherbst> you probably want to prevent i915 from being loaded automatically and set up the tracing later
<tleydxdy> LaserEyess: unless someone try to play video on their ipmi page
<heat> oh I see, it intercepts every ioremap eh?
<karolherbst> yes
<heat> i see, thanks
mhenning has quit [Quit: mhenning]
<LaserEyess> tleydxdy: they can play it through cage + mpv's wayland backend
reductum has quit [Ping timeout: 480 seconds]
Haaninjo has quit [Quit: Ex-Chat]
Leopold has quit [Remote host closed the connection]
heat has quit [Remote host closed the connection]
alarumbe has quit [Remote host closed the connection]
heat has joined #dri-devel