ChanServ changed the topic of #lima to: Development channel for open source lima driver for ARM Mali4** GPUs - Kernel has landed in mainline, userspace driver is part of mesa - Logs at https://people.freedesktop.org/~cbrill/dri-log/index.php?channel=lima and https://freenode.irclog.whitequark.org/lima - Contact ARM for binary driver support!
Wizzup has quit [Ping timeout: 240 seconds]
megi has quit [Ping timeout: 276 seconds]
_whitelogger has joined #lima
_whitelogger has joined #lima
<anarsoul> bshah: Tofe: I'd appreciate if you folks test https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2283
<bshah|matrix> Ah right. I'll be able to test this next Monday
<bshah|matrix> Right now I don't have access to device
<anarsoul> OK
<anarsoul> I guess it can wait till Monday :)
<anarsoul> I've been running q3a demo in a loop for more than an hour and so far no pp errors
<bshah|matrix> ping MartijnBraam ^^^ if you want to test this out
<anarsoul> it fixes cursor lag in X11
monstr has joined #lima
<anarsoul> well, I got ppmmu fault on read eventually
<anarsoul> :(
<anarsoul> I'll check if I get the same without my MR
jernej has quit [Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net]
jernej has joined #lima
jernej has quit [Read error: Connection reset by peer]
jernej has joined #lima
adjtm has quit [Read error: Connection reset by peer]
adjtm has joined #lima
adjtm_ has joined #lima
adjtm has quit [Read error: Connection reset by peer]
<Danct12[m]> <bshah|matrix "ping MartijnBraam ^^^ if you wan"> i can test it out
<Danct12[m]> will be compiling it in a few minutes
<Tofe> anarsoul: I'll try it yes
dddddd has quit [Remote host closed the connection]
jernej has quit [Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net]
jernej has joined #lima
jernej has quit [Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net]
jernej has joined #lima
<rellla> anarsoul, MoeIcenowy: http://imkreisrum.de/lima/ideas/
<rellla> i don't know, if it's really complete, because glmark didn't return after the run.
megi has joined #lima
<rellla> please ping me, if you need more information
<rellla> btw, i haven't seen the display output, so i can not say, if it's correct.
<rellla> anarsoul: and yes, i can confirm now, that VS CMD is much larger than in my example yesterday :p
<rellla> used glmark version is commit 71b3274, because since introduction of GLAD i had issues with EGL
chewitt has joined #lima
BenG83 has joined #lima
jailbox has quit [Ping timeout: 265 seconds]
<rellla> fwiw, http://imkreisrum.de/lima/glmark2/ is ~7GB dumps from a full glmark2-drm-es2 standard run without specifying any scene. it's not divided into scenes, but ...
jailbox has joined #lima
<MoeIcenowy> rellla: can I download it with `wget -r` ?
<MoeIcenowy> oh seems that a simple sequence wget works
<rellla> MoeIcenowy: for /glmark you have to wait, since upload hasn't finished yet
<MoeIcenowy> rellla: ah I have no interest
<rellla> :)
<MoeIcenowy> just interested in ideas
<MoeIcenowy> why is the data divided to multiple segments?
<rellla> is the divider
<MoeIcenowy> are they divided by the tracker?
<rellla> yes
<rellla> it's one per frame
<Danct12[m]> <anarsoul "it fixes cursor lag in X11"> doesn't seem to fix cursor lag for me on the pinetab though
<Danct12[m]> still somewhat feeling it
<asriel> I guess the framerate should be better..?
<asriel> Since "optimization"
<Danct12[m]> I haven't tried timedemo yet but appears that quake 3 is still lagging
<Danct12[m]> and that only happens on x11, wlroots doing it just fine
<Danct12[m]> even if the patch isn't applied
paulk-leonov has quit [Ping timeout: 264 seconds]
paulk-leonov has joined #lima
jernej has quit [Ping timeout: 246 seconds]
jernej has joined #lima
wiewo has quit [Ping timeout: 250 seconds]
wiewo has joined #lima
Barada has quit [Quit: Barada]
wiewo has quit [Ping timeout: 252 seconds]
wiewo has joined #lima
dddddd has joined #lima
monstr has quit [Remote host closed the connection]
enunes has joined #lima
<anarsoul> Danct12[m]: are you sure that you're using my branch?
<Danct12[m]> just the MR patch
<anarsoul> Danct12: you need whole branch, not one patch. There're 3 extra commits in this branch
<Danct12[m]> Oh
<anarsoul> yeah
<Danct12[m]> dumb little ol' me, anyway will try again and let you know
<anarsoul> thanks
<Danct12[m]> anarsoul: just tried, can't tell the difference about the cursor lag
<Danct12[m]> and with/without the patch
<Tofe> on my side I'll have to rebuild the image, it'll take a bit of time
<anarsoul> Danct12[m]: I wonder if you have cursor plane patches in your kernel
<anarsoul> :)
<anarsoul> Danct12: I double checked, I can definitely tell the difference
<anarsoul> asriel: there won't be much difference in q3a since it doesn't do partial draws
<anarsoul> but cursor is a lot smoother and I can move terminal window without visible lag
<anarsoul> anyway, I just hit pp task error
<anarsoul> rellla: that's mali450 I guess?
<anarsoul> (plbu_array_address is 0x0)
<rellla> mali450, yes
<anarsoul> darn, I just have no idea why partial update may result in pp error
<anarsoul> looking at the pattern it's clear that only one PP did its job and it was full or almost full redraw
<Danct12[m]> <anarsoul "Danct12: I wonder if you have cu"> if that's the kermel, nope
<Danct12[m]> but would happy to patch it up
<bshah|matrix> Danct12: in pmOS kernel it's included
<Danct12[m]> <bshah|matrix "Danct12: in pmOS kernel it's inc"> ah then i have no idea
<Danct12[m]> on fvwm, i can move the window fine but it'll lag behind
<anarsoul> Danct12: try openbox
<Danct12[m]> in a way which shows delays
<anarsoul> fvwm unlikely to have a patch that submits another draw when previous is complete
<Danct12[m]> i can notice it a little bit, but otherwise moves perfectly fine
<anarsoul> one that's described here: https://anholt.github.io/twivc4/2017/08/07/twiv/
warpme_ has quit [Quit: Connection closed for inactivity]
<anarsoul> rellla: enunes: I checked ideas dump and if it has 2 shaders it sets bit 12 (1 << 12) in aux0 for first shader
<anarsoul> MoeIcenowy: ^^
<anarsoul> also aux1 doesn't have bit 13 set
<anarsoul> enunes: I found the bug. It's trivial one liner.
<anarsoul> I'll send MR tonight
<anarsoul> ideas now render correctly
<anarsoul> I have strong suspicion that it works with flush because stack ends up in cache and with more shaders we have to actually spill
<enunes> anarsoul: aha, nice one
<enunes> anarsoul: flush doesnt fix ideas actually, it was the other reported issue
<enunes> but it might be the same issue
<anarsoul> I also discovered how to change register that's used for storing gl_FragColor
<anarsoul> I tried to adjust some bits according to the dump and accidentally changed wrong one and got interesting effect
<anarsoul> try changing "render->multi_sample = 0x0000F807" to 0x1111F807 :)
<anarsoul> or to 0x0010F807 as I did
<anarsoul> also setting bit 0x1000 in aux0 has another interesting effect
<anarsoul> it slightly misrenders ideas when it's set
<anarsoul> but it yields 2 extra fps
<anarsoul> likely it enables some optimization
<anarsoul> but I have no idea when we can enable it
<anarsoul> blob sets this bit in ideas for some draws but I don't see the pattern