buzzmarshall has quit [Remote host closed the connection]
davidlt has joined #panfrost
karolherbst has quit [Ping timeout: 255 seconds]
enunes has quit [Ping timeout: 260 seconds]
karolherbst has joined #panfrost
enunes has joined #panfrost
vstehle has joined #panfrost
tgall_foo has quit [Read error: Connection reset by peer]
<tomeu>
robmur01: got to reproduce the random failures while watching the frequency, and indeed it starts going wrong after a switch to the highest freq
<nlhowell>
aside from the framerate, it looks fine
<nlhowell>
says panfrost is being used
<urjaman>
the server glx vendor string is always SGI (i guess if you have modern Xorg atleast)
<urjaman>
like even on my AMD graphics PC
<nlhowell>
urjaman: ah, ok; thanks, I didn't know this
<nlhowell>
so it seems, then, that everything is fine, except that performance seems pretty slow
<urjaman>
i guess it's how every browser user agent is Mozilla :P
<nlhowell>
:D
<urjaman>
(similar to how...)
<robmur01>
tomeu: I guess that's "good" in the sense that at least it's probably not some other *new* problem :)
<tomeu>
robmur01: it's *very* good in that sense :)
<robmur01>
if I can make the time I'll try to resurrect my first attempt at registering the regulators for a quick fix (and think about coming back to the bigger rework later)
<robmur01>
(from what I saw, the whole API stinks)
<alyssa>
nlhowell: if you could install perf and run `sudo perf top`, use the desktop doing something slow and screenshot that -- it might help?
JaceAlvejetti has quit [Ping timeout: 240 seconds]
JaceAlvejetti has joined #panfrost
<nlhowell>
alyssa: ok, i will give it a go
<nlhowell>
thanks!
nlhowell has quit [Read error: Connection reset by peer]
<nlhowell>
it seems some spinlocks are the major offenders
<nlhowell>
i tried to get some results from forcing urxvt to redraw, but i wasn't able to time it right
<robmur01>
nlhowell: if you're feeling brave, it's usually possible to infer which locks time is being spent under by looking at the call chains for where the unlocks happen (perf record -g/perf report)
<alyssa>
(Also, compiling mesa master might help. If nothing else, it'll get us symbols.)
<robmur01>
although it's a pretty safe bet that it's mostly monkeying about with the pagetables under as_lock
* robmur01
contemplates digging out and reviving the RK3288 box for the sake of 32-bit kernel gubbins
<daniels>
robmur01: should be pretty straightforward, the firefly + rock2-square boards have worked well with mainline since forever
<robmur01>
daniels: yeah, I had it running mainline kernel + u-boot ~2 years ago, it's just not been powered up since
<robmur01>
the only annoyance is that it has the weirdo RK1000 audio/video codec, which has no mainline driver and gets disconcertingly hot