<alyssa>
tomeu: "wonder why the blob takes so long to run deqp-gles2 on the t720" T720 or S912? Remember that dEQP is heavily CPU-bound, generally speaking
<tomeu>
actually, ended up taking more or less the same time
<tomeu>
but running the whole deqp on t720 and t820 takes much longer than on t760 and t860
<alyssa>
Interesting.
<tomeu>
guess because of running in parallel and there being several CPU cores, the GPU ends up being the bottleneck
<alyssa>
Oh, with one of the parallelizing deqp runners, i guess that's possible
<alyssa>
Single-threaded dEQP is heavily CPU bound anyway
guillaume_g has quit [Quit: Konversation terminated!]
nerdboy has joined #panfrost
enunes has quit [Ping timeout: 246 seconds]
* alyssa
is trying to derive a heuristic for the work/uniform split problem
<alyssa>
Making some coarse assumptions, it *ought* to be doable but we'll see
cyrozap has joined #panfrost
enunes has joined #panfrost
davidlt has quit [Ping timeout: 248 seconds]
jolan has joined #panfrost
icecream95 has joined #panfrost
mearon_ is now known as mearon
lmcloughlin has joined #panfrost
<lmcloughlin>
probably a dumb question, but rockchip_dri.so is the nonfree mali binary drivers right?
<lmcloughlin>
I'm trying to get Chromium running with accelerated graphics on a Pinebook Pro (RK3399), but finding it picks up the llvmpipe driver according to chrome://gpu
<icecream95>
For me, rockchip_dri.so and panfrost_dri.so are hardlinked to each other, i.e. the same file
<lmcloughlin>
glxgears -info shows panfrost as I'd expect, but `xdriinfo` shows screen 0 is using "rockchip"
<lmcloughlin>
hmm, they're not symlinks here
<icecream95>
About Chromium, go to about:flags and enable the one to override the hardware rendering blacklist.
<lmcloughlin>
yeah, that's set, and I've tried --ignore-gpu-blacklist on the commandline too
<icecream95>
I said *hardlinks*. If you run `du panfrost_dri.so rockchip_dri.so` it should only list one file.
<lmcloughlin>
rockchip_dri.so is being used, if I move it and restart X, I get sw rendering in glxgears
<lmcloughlin>
oh, you're right, sorry, my mistake
<mmind00>
lmcloughlin: –use-gl=egl for chromium perhaps?
<lmcloughlin>
mmind00: doesn't appear to make any difference
<lmcloughlin>
if i set LIBGL_DEBUG=verbose I see: "libGL error: failed to load driver: rockchip"
<lmcloughlin>
and then it proceeds to load swrast_dri.so
<lmcloughlin>
this is on Debian bullseye, btw
<lmcloughlin>
mesa 19.2.6
<lmcloughlin>
kernel 5.4.2
<mmind00>
lmcloughlin: I'd guess mesa might be a tag old
<mmind00>
tag -> tad
<lmcloughlin>
hmm, I can try building from source, but I thought 19.2 was new enough
<lmcloughlin>
and it does appear to work for glxgears
TheKit has joined #panfrost
<lmcloughlin>
are there more debug options I can set to diagnose that "failed to load driver" error log?
<icecream95>
I just tried Chromium, and I'm getting that error too...
<lmcloughlin>
there's quite a large delay in characters I type appearing in the address bar, which aligns with what I'd expect to see of SW rendering
raster has quit [Quit: Gettin' stinky!]
<lmcloughlin>
firefox correctly reports panfrost being used and is far faster
<lmcloughlin>
ah, I installed chromium from a different PPA and that doesn't have the same issue
<alyssa>
robmur01_: chromium's still broken at least with desktop GL + panfrost
<icecream95>
That would explain why chromium doesn't work but other chromium-based browsers do (but with broken rendering)...
stikonas has quit [Remote host closed the connection]
<endrift>
alyssa: thanks
stikonas has joined #panfrost
Stenzek has quit [Ping timeout: 250 seconds]
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
Stenzek has joined #panfrost
stikonas has quit [Remote host closed the connection]
<lmcloughlin>
alyssa: got it, didn't realise the display out HW came into play in userspace, I thought it was all on the kernel side of the equation. thanks for the explanation!
<lmcloughlin>
is the Chromium issue tracked anywhere?
<lmcloughlin>
i guess it's a collection of different issues affecting it, so maybe not :)
icecream95 has quit [Ping timeout: 252 seconds]
icecream95 has joined #panfrost
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
<icecream95>
With the shadow stuff merged, I tried enabling shadows in OpenMW. It "worked", but then I went outside and everything hanged, and I got a pile of errors in dmesg such as [drm:drm_mm_takedown] *ERROR* node [00002074 + 00000001]: inserted at panfrost_gem_open+0xd0/0x194 [panfrost]
<icecream95>
Here's another kernel message I haven't seen before: panfrost ffa30000.gpu: matching BO is not heap type (GPU VA = 18345000)
icecream95 has quit [Ping timeout: 268 seconds]
<alyssa>
icecream95: Yes, I'm afraid it would.. I can add other browsers to the blacklist if you can give a lsit
icecream95 has joined #panfrost
<alyssa>
lmcloughlin: Just the name of the display hw
<alyssa>
icecream95: That's a new message even for me *blink*
<alyssa>
At any rate, Firefox works great! :)
<robmur01_>
alyssa: that's just the foreword to an unexpected page fault splat these days
<alyssa>
robmur01_: :c
<alyssa>
At any rate, it's time to do witchcraft!
<alyssa>
I mean uhm actually this time it's regular magic, not witchcraft.
<alyssa>
Have 3 pages of handwritetn mathematics from this morning, let's get implementin'
<robmur01_>
(specifically the "this address seems to match to some non-growable BO" case)
<alyssa>
OIC
<robmur01_>
so chances are it's unmapped, but still cached