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
pH5 has quit [Ping timeout: 250 seconds]
robink has joined #panfrost
megi has quit [Ping timeout: 240 seconds]
adjtm has quit [Ping timeout: 240 seconds]
warpme_ has joined #panfrost
adjtm has joined #panfrost
jernej has quit [Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net]
jernej has joined #panfrost
jernej has quit [Client Quit]
jernej has joined #panfrost
yann has joined #panfrost
<narmstrong> daniels: I've reconfugured the runner to get 4 concurrent jobs
<narmstrong> daniels: and I've added a second meson-gxl-s905x-libretech-cc board
<narmstrong> daniels: I've asked more h3 libretech boards
<narmstrong> tomeu: do you plan adding the same lima boards in your lab for redundancy ?
raster has joined #panfrost
megi has joined #panfrost
<daniels> narmstrong: that's great, thanks! :)
vstehle has quit [Ping timeout: 265 seconds]
vstehle has joined #panfrost
<tomeu> narmstrong: no plans for now, as we don't do any work with those boards
<tomeu> but we can talk about it if the current setup proves too unstable
<tomeu> narmstrong: we plan to add a bunch of rockpi4 boards though, and use them instead of the kevins
<tomeu> do you have plans with the rockpi4 as well?
<tomeu> alyssa: how is opencl looking atm? will it be fun?
<tomeu> alyssa: guess I should be looking at updating the kernel
<tomeu> robher: has the fix for the NULL dereference in drm_sched_increase_karma made into a rc already?
<robher> tomeu: That's Steven's timeout rework, right?
<tomeu> robher: I think so
<robher> tomeu: humm, I guess i should have committed that to drm-misc-fixes instead of next.
<robmur01> export failed: Too many open files
<robmur01> kmscube: ../src/gallium/drivers/panfrost/pan_screen.c:648: panfrost_fence_create: Assertion `fd != -1' failed.
<robmur01> that's new :)
<tomeu> bbrezillon: ^
<tomeu> we seem to be leaking FDs heavily
<tomeu> wonder if fences or buffers
<robmur01> is there an easy way to tell?
adjtm has quit [Ping timeout: 265 seconds]
<robmur01> /proc/<blah>/fd fills up with links to "anon_inode:sync_file", so I'm guessing fences?
yann has quit [Ping timeout: 265 seconds]
<tomeu> robmur01: probably, is that the case?
<paulk-leonov> out of curiosity, anyone seeing "mmc0: error -110 whilst initialising SD card" (external sdcard slot) on kevin with 5.3?
<paulk-leonov> (sorry for asking here, I know it's not directly related)
krh has joined #panfrost
<paulk-leonov> mhh a different card works
<anarsoul> tomeu: out of curiosity, does rockpi4 work with 5v power supply or I have to use one that supports PD?
<tomeu> anarsoul: haven't used it myself, but I think it has a barrel socket
<anarsoul> tomeu: it's usb-c
raster has quit [Quit: Gettin' stinky!]
stikonas has joined #panfrost
<alyssa> tomeu: Haha, not sure on OpenCL yet
<alyssa> Currently working on rewriting the RA to cope with mixed types, which is a prereq for doing any sort of interesting compute
<alyssa> (As a more immediate win : it will allow us to do fp16 once Igalia's mediump work lands)
<anarsoul> :)
<alyssa> Okay, uh where was I it's ben a whole two days
<alyssa> Ah k
<alyssa> Mask reduction on the fritz.
<alyssa> Fixed, more bugs now
<alyssa> OK, change of plans
<alyssa> I'm getting a clean master, and I'm going to add a mask reduction pass (against the old RA) and get that to zero regressions
<alyssa> Lighten the diff.
<alyssa> Also, move the derivations of LCRA to a dedicated document (like, in LaTeX) ... currently they live within the source code of my dev branch
<alyssa> Not for any high-level academic reason, but again, lighten the diff
<alyssa> robmur01: I just got the fence thing too for the first time
<alyssa> I haven't updated kernel so something in userspace must have regressed
* alyssa is inclined to assume point a finger at pipelining
<alyssa> bbrezillon: ^^
<alyssa> Maybe we're not garbage collecting fences correctly?
<alyssa> okay what
<alyssa> This is
<alyssa> baffling.
<alyssa> Er no
<alyssa> Consider me unbaffled, it was a scheduling thing
<alyssa> It is worth noting that the TEXW and TEXR really are totally distinct
<alyssa> You *can*, however, write to TEXR from a texture instruction
<alyssa> Or, actually,
<alyssa> You can *read* from TEXW by setting unknown4=0x1 on the reading instruction
<anarsoul> what is texw and texr?
<alyssa> anarsoul: texw - registers written from texture pipeline
<alyssa> erm
<alyssa> texw - output from texture pipeline
<alyssa> texr - input to texture pipeline
<alyssa> neither are really registers but they're addressed as such
<anarsoul> so texr are coords, sampler #, etc?
<alyssa> Ya
<alyssa> Actually that's a good point
adjtm has joined #panfrost
<urjaman> oh great, now i'm getting that panfrost_fence_create thing (it's killing my xfwm4)
enunes has quit [Remote host closed the connection]
enunes has joined #panfrost
stikonas_ has joined #panfrost
stikonas has quit [Ping timeout: 276 seconds]
yann has joined #panfrost
raster has joined #panfrost
raster has quit [Remote host closed the connection]
stikonas_ is now known as stikonas