<alyssa>
sway + panfrost + 4k seems a lot faster/stable
<alyssa>
wonder what gnome-shell is doing silly
<alyssa>
although deqp wayland doesn't seem happy on sway
hifi has joined #panfrost
<hifi>
hi, I've been looking into a performance issue with urxvf on rk3399 and panfrost with glamor, it seems to be it's flooding glamor with draw commands which then end up flooding panfrost, this is somehow optimized out on intel/amdgpu but I was wondering if anyone else on any other hardware that panfrost runs on could verify it's not rk3399 specific?
<hifi>
urxvt*
<alyssa>
known issue with tiler gpus
<alyssa>
afaiu, lima has the same issue
<hifi>
is it possible to replace the jobs in the job queue with newer ones that affect the same thing as an optimization or is it too late at that point?
<hifi>
I'm not an expert in this area, full disclosure
<alyssa>
it's.. sort of too late
<alyssa>
we do some tricks like that but apparently not enough
<alyssa>
x11 hasn't been a priority
<hifi>
could this be solved before it hits panfrost with a shadow framebuffer?
<alyssa>
afaiu the issue was that glamor glFlush's a _lot_
<hifi>
the thing is with glamor disabled there's excellent 2d performance on x11 but we lose all gpu acceleration for the important stuff
<alyssa>
and that it sort of has to for conformance with xrender
<hifi>
(okay rockchipdrm needs a patch for non-accelerated glamor to work fast but close enough)
<hifi>
tricky situation, x11 is fairly important on a laptop like device like the pinebook pro which I'm on compared to other kinds of mobile devices where this kind of optimimzation is not needed
<daniels>
there's no future where glamor doesn't spam glFlush
<hifi>
it's not a huge issue, I don't know any other x11 program (that is currently relevant) that behaves so badly as urxvt
<alyssa>
how do I get ssh to remember my passphrases in sway/i3
<alyssa>
uhh
macc24 has joined #panfrost
<alyssa>
nvm, swapped identities
<amonakov>
daniels: how so? why it cannot setup a 16ms timer to coalesce flushes?
<daniels>
amonakov: it's more about how the server is structured to process X11 rendering commands
<daniels>
fixing that is ... non-trivial
* macc24
is back
<hifi>
anyway, thanks for the information so I don't send more time trying to fix the unfixable