stikonas has quit [Remote host closed the connection]
cyrozap has joined #panfrost
thecycoone has joined #panfrost
vstehle has quit [Ping timeout: 260 seconds]
<thecycoone>
tomeu:
<thecycoone>
the kernel patch that adjusts the lowers the voltage you sent me months ago is still required afaik. Do you know what I should be looking for to follow any progress on mainlining a fix?
karolherbst has quit [Ping timeout: 246 seconds]
kherbst has joined #panfrost
rhyskidd has quit [Remote host closed the connection]
robmur01 has joined #panfrost
rhyskidd has joined #panfrost
robmur01 has quit [Ping timeout: 272 seconds]
nlhowell has quit [Read error: Connection reset by peer]
nlhowell has joined #panfrost
_whitelogger has joined #panfrost
<thecycoone>
to clarify I mean the gpu opp table patch
nerdboy has quit [Ping timeout: 258 seconds]
cwabbott has quit [Quit: cwabbott]
cwabbott has joined #panfrost
endrift has joined #panfrost
<endrift>
is that sRGB blanching bug fix in mesa or the panfrost kernel module? I'm hitting it on my PBP and I'm hoping that I can just install a different mesa build instead of a different kernel
<xdarklight>
it will be less hacky if you add PAN_OVERRIDE_ALL_HACKS_ARE_NOW_SANE_CODE=yes ;-)
<daniels>
PAN_YOLO=all
<alyssa>
xdarklight: PAN_MESA_RUN_MY_APP_DAMMIT=y
<alyssa>
....to be clear, it *did* work. I didn't do anything fancy, but it recorded my screen and camera with an overlay and audio and saved it, no crashing. ship it :p
<alyssa>
not sure where the GL 3.3 req comes from
<xdarklight>
...that sounds pretty boring. it's not fun if it's not crashing or at least causing weird behavior *cough*
<urjaman>
where'd you get obs on ARM?
stikonas has quit [Quit: Konversation terminated!]
<alyssa>
xdarklight: I have no idea what you're talking about ;)
<alyssa>
urjaman: Debian repo, once I set magic env flags things seemed to Just Work
<urjaman>
ok i'll take a look at how debian does it, some day when i care that is
<daniels>
xdarklight: if you want fun, run it on the Amlogic platform where the cache is a lot less predictable than you might hope
<alyssa>
Lookslike CPU bottlenecking
<alyssa>
s/w tiling is part of the problem (I really need to get icecream's patch for this merged, I'm really distarcted rn)
<alyssa>
forcing linear helps, but still getting hurt in _mesa_texstore's memcpy
<alyssa>
Likewise X is getting hurt in st_ReadPixel's memcpy
<alyssa>
Now, the fact that we're doing readpixels is in and of itself a problem. but I digress.
<alyssa>
in a double-buffering world this should be zero-copy...
<xdarklight>
daniels: I don't know anything about that - but if you have a link I'd like to read up on it (not saying that I have time or skill to work on it though) :)
<alyssa>
Just doing webcam+audio hits 30fps, it's the screen capture that's no good
<xdarklight>
also I'm not sure if it's the same situation for other SoCs/vendors (Rockchip for example)
<daniels>
it does go through AXI, but it's not an ACE issue - it's that the GPU integration for that SoC has weird specific workarounds for how the GPU's own internal cache behaves after power events
<daniels>
what we have works everywhere else, but we need to implement a workaround for that specific SoC
<milkii>
alyssa: just a note that there's a pop over on the site on mobile that can't be got rid of and goes to LinkedIn when clicked on https://i.imgur.com/LRhFGEY.jpg
<milkii>
Uh I switched back to the FF app and it was gone.. ah well!
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
<alyssa>
milkii: we have a mobile site? cool!
<alyssa>
;-)
<alyssa>
I'm toying with the idea of keeping driver-side accounting of memory bandwidth (in a debug mode)
<alyssa>
This isn't possible to do completely thanks to Turing, but it should be possible to get estimates of about how much bw is going to framebuffers vs textures vs varyings vs attributes vs uniforms, assuming everything we upload is actually read somewhere.
<alyssa>
We have the actual counts from the hw but that doesn't tell you what your offenders are.
<alyssa>
Unfortunately with compression (AFBC in particular) even if we assume everything we pas sis read, we don't get anywhere
<alyssa>
fp16 varying stuff does seem to be a real win at any rate, should get that claned up
mias has quit [Ping timeout: 256 seconds]
icecream95 has joined #panfrost
<milkii>
daniels: no, that just goes direct to their login page. But I realise now it was the is Chrome browser but the FF app that did the funny thing, so idk
raster has quit [Quit: Gettin' stinky!]
Green has quit [Ping timeout: 246 seconds]
raster has joined #panfrost
Green has joined #panfrost
yann has quit [Ping timeout: 246 seconds]
stikonas has quit [Remote host closed the connection]
<alyssa>
Honestly think emit_varying_descriptor needs a good cleanup first... currently at 284 lines ugh
<alyssa>
for a single function not good
stikonas has joined #panfrost
<alyssa>
Lot of bookkeeping for transform feedback and things
<alyssa>
Isn't doing anything unnecessary per se just a lot of special cases..
<alyssa>
To some extent I guess that's life
bnieuwenhuizen has quit [Quit: No Ping reply in 180 seconds.]