alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - https://gitlab.freedesktop.org/panfrost - Logs https://freenode.irclog.whitequark.org/panfrost - Discord Discard
raster has quit [Read error: Connection reset by peer]
adjtm has joined #panfrost
BenG83 has joined #panfrost
stikonas has quit [Remote host closed the connection]
NeuroScr has quit [Quit: NeuroScr]
anarsoul|2 has quit [Ping timeout: 240 seconds]
<alyssa> Wallpaper code snails along
<alyssa> (Snails because texturing is slow, also because I'm distracted with real life :P)
<alyssa> Have most of the pipeline rigged up, trying to figure out some layering stuff..
<alyssa> (How does z-order work again? :P)
<Ashy> anyone know a good place to buy an edp cable to use with the rockpro64?
<Ashy> or is everyone here using hdmi?
<alyssa> I'm on HDMI
<alyssa> Demo at XDC was HDMI (IIRC)
NeuroScr has joined #panfrost
<Ashy> ah ok, i've got a spare old thinkpad edp panel sitting around
_whitelogger has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
BenG83 has quit [Ping timeout: 250 seconds]
NeuroScr has quit [Quit: NeuroScr]
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
raster has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
<narmstrong> Hmm, I get OOM killer while running kodi
<narmstrong> And just re-ran kmscube : it uses 1.07GiB of VM !
<narmstrong> tomeu: did you see something similar ?
rhyskidd has quit [Quit: rhyskidd]
<urjaman> only 1GB? that's not a lot for panfrost right now ;P
rhyskidd has joined #panfrost
<urjaman> (yeah i ran my 2GB C201 quite rough into swap when i decided to test panfrost while firefox was running...)
<narmstrong> yeah, a little too much for a simple cube
<urjaman> (i hear it's just allocating too much and not releasing anything so WIP)
<narmstrong> yeah
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
jernej has quit [Ping timeout: 250 seconds]
rhyskidd has quit [Quit: rhyskidd]
<tomeu> yeah, we don't release anything atm
<tomeu> wanted to do a few things before that
rhyskidd has joined #panfrost
robclark has quit [Read error: Connection reset by peer]
robclark has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
<narmstrong> tomeu: is there some simple release I could implement to lower the VM a little ?
<tomeu> narmstrong: I think so, just look at how resource_create is implemented in my branch and undo it in panfrost_resource_destroy
<tomeu> see other gallium drivers for inspiration
<narmstrong> tomeu: thanks for the pointer
<tomeu> hth
<tomeu> I will look at it if I find some extra time to hack on panfrost
<tomeu> please keep me postd :)
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
<narmstrong> sure !
NeuroScr has joined #panfrost
jernej has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
<alyssa> narmstrong: Only 1GB? Fascinating, what'd you do to lower it so much?!
<alyssa> ;)
<narmstrong> lol
<alyssa> narmstrong: One of the big problems is panfrost_setup_hardware
<alyssa> Which basically takes the approach of "allocate as much space as we could possibly ever need on-startup, so that way we can forget about memory management for the rest of the driver"
<alyssa> Bad idea? Probably. We accept pull requests ;)
<narmstrong> yah I saw this
<alyssa> Take those cmdstream rings
<narmstrong> I know, but I need something to push !!
<alyssa> 2 * (8 * 64 * 8 * 8 * 18 * 4096) =
<alyssa> 4.5 gigabytes? That can't be right...
<alyssa> 2 * (8 * 64 * 8 * 16 * 4096) / 1024 / 1024
<alyssa> 512MB of temp cmdstream buffers
<alyssa> 32MB for persistent
<narmstrong> yeah, this explains the 1GiB VM
<alyssa> 256MB for textures
<alyssa> 16MB of shaders
<narmstrong> (I did a short valgrind run http://termbin.com/rexs)
<alyssa> 192MB for internal varying/tiler heap, but that's not mapped so it might not count?
<narmstrong> but yeah most of the unnecessary allocation is done via slab
<alyssa> Seems right
<alyssa> I swear, one of these days I'll learn how to do memory management correctly..
<narmstrong> *sweet dreams*
<alyssa> Pf
<alyssa> It's actually a serious problem for me :p
<tomeu> I was kind of hoping we would take care of that properly once we have our own kernel driver
<alyssa> tomeu: I was as well
<tomeu> same for performance
<alyssa> I'm not really sure what the bottleneck is rn, fwiw
<alyssa> I guess we would have to setup the performance counters but
BenG83 has joined #panfrost
BenG83 has quit [Quit: Leaving]
raster has quit [Ping timeout: 245 seconds]
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
raster has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
anarsoul|2 has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
BenG83 has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
BenG83 has quit [Quit: Leaving]
rhyskidd has joined #panfrost
rhyskidd has joined #panfrost
NeuroScr has quit [Quit: NeuroScr]
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
samueldr has quit [Changing host]
samueldr has joined #panfrost
stikonas has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
NeuroScr has joined #panfrost
<narmstrong> tomeu: in any case, we will need to have proper on-demand coherent memory allocation with the actual and our own kernel driver
<narmstrong> I'll play with it somehow, but don't expect miracles from me :-)
<alyssa> :D
<Lyude> alyssa: it will probably be worth looking at the performance on the new uapi branches as well since we have gotten rid of a lot of the unneeded __packed structs
<alyssa> Lyude: Difference is neglible
<Lyude> alyssa: gotcha
<alyssa> Since we make very few ioctl calls
<alyssa> (One ioctl per frame, basically. Just because we save a few cycles doesn't mean our FPS will jump ;P)
<Lyude> yeah-I figured I should at least mention it though
<alyssa> :)
<Ashy> so what is the deal with ayufan's kernel for rockpro64?
<Ashy> is that work going to be mainlined at some point or does he not attempt that?
<Ashy> "This branch is 112 commits ahead, 168 commits behind torvalds:master. "
<anarsoul|2> Ashy: ask him? :)
<Ashy> good idea, i'll do some trawling of git logs and mailing lists first