<SolidHal>
I realize its likely not a panfrost issue, but still likely a rk graphics stack issue. maybe mesa? maybe rk drm?
<SolidHal>
🐽
<urjaman>
The Unicode works fine, it's just I didn't happen to have a hecking nose in this terminal font :P
<SolidHal>
gotta get a better font then ;)
* urjaman
likes his Terminus
<SolidHal>
Really just hoping someone that knows more about the drm stack has an idea of something I can test next
<SolidHal>
Source Code Pro ftw
<alyssa>
SolidHal: I have an idea
<alyssa>
will I try it or go to sleep?
<SolidHal>
if you go to sleep, please tell me your idea :D
<SolidHal>
actually, I'd love to hear your idea either way lol
<alyssa>
SolidHal: Nope
<alyssa>
For a moment, I thought _cursor planes_ might be the issue.
<alyssa>
Since cursor handling is special cased, handled independent of the GPU/display/etc
<alyssa>
weston has a debug option to disable cursor planes (mod+shift+space c, you can verify it works by looking at gallium_hud=fps)
<alyssa>
But even with that debug option, no dice.
<SolidHal>
darn
<alyssa>
But! That does still say something
<alyssa>
Namely, the "jump" is just two consecutive frames
<alyssa>
I *think* that implies the jump is from something more subtle than just missing flips but dunno
<SolidHal>
I *think* its system wide initial rendering lag. The first keypress after some time seems to be delayed as well but I can't measure that well
<SolidHal>
how did you determine its just two frames?
<alyssa>
not rigorously but gallium_hud and broken old mesa that i need to really upgrade sometime
<SolidHal>
or, wait two consecutive frames? as in it just doesn't render between the first mouse position and the second?
<alyssa>
It doesn't look like it? :|
<SolidHal>
hmm
<alyssa>
But check that yourself, this is late night handwaving from someone who doesn't look at this part of the stack
<alyssa>
I think "gears makes it go away" is crucial, though.
<SolidHal>
really, anything that provides constant video content makes it go away
<alyssa>
yeah. that's a big sign of... something
<SolidHal>
lol
<alyssa>
...Hmm
<SolidHal>
what is the name of "this part of the stack"? I'm not familiar with how things stack up here
<alyssa>
display drivers
<alyssa>
as opposed to 3D
<alyssa>
SolidHal: Also for a data point
<alyssa>
# echo 0 | sudo tee /sys/devices/.../fbcon/cursor_blink
<alyssa>
(while at fbcon, no X/wayland)
<alyssa>
hexdump /dev/input/mice
<alyssa>
(also possibly hexdump /dev/urandom in another pane of tmux to simulate constant video content or not)
<alyssa>
rk doesn't have a text mode so anything that's actually strictly display should be reproducible at fbcon instead of x/wayland..
<SolidHal>
There doesn't seem to be any lag with that setup
<SolidHal>
tried with and without the second pane of random
<alyssa>
right.. so what's the difference?
<SolidHal>
uhhhh wayland and xorg?
<alyssa>
right but... what are they doing that fbcon isn't
<SolidHal>
lots
<alyssa>
to the hw, it's all just pixels
<alyssa>
*display
<alyssa>
SolidHal: "Note how the graphs barely update, especially the fps one"
<SolidHal>
so the possible culprits are: wayland/xorg, mesa, panfrost, rockchip_drm ?
<alyssa>
FWIW that means cursor planes are working -correctly-, bypassing the gpu entirely
<SolidHal>
>FWIW that means cursor planes are working -correctly-, bypassing the gpu entirely
<SolidHal>
makes sense
<alyssa>
we've both reproduced this with no panfrost, take mesa out of this :P
<SolidHal>
isn't mesa involved with and without panfrost?
<alyssa>
nope
<alyssa>
I mean
<alyssa>
what is your !panfrost test?
<SolidHal>
mesa doesn't do anything with the llvmpipe?
<alyssa>
oh.
<alyssa>
i mean the bug reproduces without GL at all
<kinkinkijkin>
could be a Bug Unique to Hardware, or BUH for short
<alyssa>
i.e. throw mesa out entirely
<alyssa>
(that means no GNOME)
<alyssa>
x11 with i3 and pure sw, no glamor or llvmpipe in sight
<alyssa>
mesa is never involved, the mali neither ... still same bug
<kinkinkijkin>
lemme read back and see whats been tested and what the bug is
<kinkinkijkin>
cursor stopping/starting?
<kinkinkijkin>
verify your mouse is plugged directly into the device before continuing to hunt this, or if its a built-in touchpad verify that it's clean
<kinkinkijkin>
just in case
<SolidHal>
kinkinkijkin this occurs on multiple gru-kevins :)
<SolidHal>
with multiple input devices
<kinkinkijkin>
ah
<kinkinkijkin>
good godd
<kinkinkijkin>
good*
<SolidHal>
alyssa i3 does it too
<alyssa>
SolidHal: re boot splash, stock plymouth works fine with depthcharge
<alyssa>
I used it briefly to get MLP fanart as a bootsplash. Friends were not impressed. (This was prepandemic)
<SolidHal>
lol, random, but appreciated :)
<SolidHal>
I'll not that on the PR
<SolidHal>
*note
<alyssa>
(I stopped because it slowed down boot noticably.)
<SolidHal>
decrypting is by far the longest part of boot right now for prawn
<kinkinkijkin>
so, correct me if im wrong, llvmpipe is not the same thing as softpipe and swrast
<kinkinkijkin>
so if youre seeing llvmpipe, its not necessarily fully cpu-based software
<alyssa>
uhhh
<kinkinkijkin>
is this true or not
<alyssa>
swrast can mean any of { classic swrast, softpipe, llvmpipe }
<alyssa>
softpipe is the reference Gallium swrast
<kinkinkijkin>
ah
<alyssa>
llvmpipe is an optimized version of softpipe using LLVM
<alyssa>
all three are fully cpu-based
<alyssa>
bu llvmpipe is by far the fastest of the bunch
<kinkinkijkin>
ok ok, wasnt fully aware on this
<alyssa>
(swrast/softpipe are interpreters, llvmpipe is a JIT compiler)
<SolidHal>
TIL
<HdkR>
Now if you were running SWR, you'd be in real trouble ;)
<alyssa>
(swrast usually means the classic one which you should never really use)
<alyssa>
HdkR: shh
<SolidHal>
so, how can I ensure the cursor in i3 is not doing anything with panfrost, llvmpipe?
<SolidHal>
for previous panfrost-free tests i just removed panfrost from the kernel lol
<kinkinkijkin>
so cursor jumping is something i have only seen in my testing with any of my devices on softpipe and llvmpipe, where it happens consistently
<kinkinkijkin>
it's how i quickly test if im rendering with gpu or not on my xu4
<SolidHal>
heh
<kinkinkijkin>
you've had this happen using panfrost, with cursor planes enabled and disabled, correct?
<kinkinkijkin>
if so then i have no idea at all
<SolidHal>
pretty sure I'm rendering with my gpu using panfrost, I can confirm with the string of panfrost dmesg lines while doing something intense ;)
<kinkinkijkin>
in a wayland compositor, use the command glxinfo -B, that's a better way of confirming this
<urjaman>
hmm, at some point I remember there being a thing (not sure if it went away or not...) where using kicad arrow keys to move the cursor, the cursor onscreen would be one update late from where the action happened if you pressed a key (kicad uses keys to do things under the cursor too)
<urjaman>
my guess was that the rockchip DRM was updating the location of the cursor plane one flip late or something but I didnt look into it at all
<SolidHal>
oh yeah, kinkinkijkin I have confirmed with glxinfo -B as well :)
<kinkinkijkin>
ive seen it before where you can have your compositor do *something* wrong and not open an actual gpu context on the display, but then everything else opens a gpu context when rendering to the screen as long as the screen exists
<kinkinkijkin>
ive never seen anyone mention this though, and i dont know how to test for it
<SolidHal>
confirmed on weston that it occurs with cursor planes disabled
<kinkinkijkin>
there might be a small chance that's happening
<kinkinkijkin>
a compositor dev or someone more intimately involved with the EGL side of things (i just write data control and libraries) of the driver might know what im talking about and could get you to test if it's that
<kinkinkijkin>
if you can test for it there's probably a single line you can write to know instantly
<SolidHal>
How do I locate such a dev? Any hang around here?
<kinkinkijkin>
probably yeah
<kinkinkijkin>
i would guess that alyssa probably would know how to test for this, if it's testible
<SolidHal>
heh
<urjaman>
*testable
<urjaman>
i usually don't correct others but i just lol'd at that one
<kinkinkijkin>
?
<kinkinkijkin>
oh because it looks like
<kinkinkijkin>
right
<SolidHal>
hahaha
<kinkinkijkin>
i speak two languages natively off my back ;-;
<SolidHal>
alyssa must have gone to bed like she said she was going to while back
<SolidHal>
For my understanding, this almost definitely a bug in the compositor as all other layers have been ruled out? Including rockchip_drm? tbh I'm not really sure where rockchip_drm sits in the stack.
<kinkinkijkin>
no, it might be rockchip_drm
<kinkinkijkin>
there's a lot of things it could be
<kinkinkijkin>
you said it happened in multiple compositors and in x11
<kinkinkijkin>
so its likely not a compositor bug
davidlt has joined #panfrost
<SolidHal>
hm, well thanks for your help kinkinkijkin :)
<SolidHal>
maybe I'll go blindly poke at rockchip_drm
davidlt has quit [Ping timeout: 240 seconds]
camus has joined #panfrost
kaspter has quit [Ping timeout: 240 seconds]
camus is now known as kaspter
cowsay_ has joined #panfrost
cowsay has quit [Ping timeout: 260 seconds]
_whitelogger has joined #panfrost
vstehle has joined #panfrost
camus has joined #panfrost
kaspter has quit [Remote host closed the connection]
archetech has quit [Quit: Konversation terminated!]
stikonas has joined #panfrost
camus has joined #panfrost
kaspter has quit [Ping timeout: 260 seconds]
camus is now known as kaspter
rak-zero has quit [Ping timeout: 272 seconds]
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
camus has joined #panfrost
camus has quit [Client Quit]
kaspter has quit [Ping timeout: 260 seconds]
popolon has joined #panfrost
gcl has quit [Ping timeout: 260 seconds]
gcl has joined #panfrost
<daniels>
SolidHal: if I were you I'd start by running Weston, pressing win-shift-space followed by 't', reproducing the issue, then sticking the resulting weston-timeline-*.json into https://github.com/ppaalanen/wesgr
<daniels>
that would let us at least reason about the results
<kinkinkijkin>
macc24: i caved and went to shopper's drug mart to get a usb-c to usb-a cable
<kinkinkijkin>
they only had apple's adapters for this
<kinkinkijkin>
so i just spent $33 on a $2 adapter because im moving before the next mail day
<alyssa>
---they sell usb adapters at shoppers? :P
<kinkinkijkin>
for $33? yes
<kinkinkijkin>
no less
<kinkinkijkin>
maybe more
<kinkinkijkin>
oh joy it's double-boxed
<kinkinkijkin>
... and then stuck to one of those paper card things
<alyssa>
this is why i only buy usb adapters at bulkbarn, mmk?
<alyssa>
(/sarcasm)
<robmur01>
if it's the USB-C to USB-A+HDMI+PD one, congratumiserations on your pretty but frustratingly flaky new tool :)
* robmur01
uses one with this Envy X2 and it's... less than perfect
<kinkinkijkin>
just usb-c to usb-a
<kinkinkijkin>
wait how am i going to make a cadmium usb if i dont have another computer macc24
<kinkinkijkin>
actually... nvm i have my odroid-xu4
<macc24>
kinkinkijkin: bug me about that so much that i make it work like that ;D
<macc24>
wait wtf why would you pay $33 for that
<macc24>
robmur01: i've got one of these, but without PD (:
<macc24>
kinkinkijkin: building cadmium on armhf host will be super slow because arm64 debootstrap needs qemu to run on armhf, unless you want armhf rootfs on it too
<kinkinkijkin>
oh it's not gonna work anyways
<kinkinkijkin>
cant install crossbuild packages on armbian apparently
<macc24>
ill make a 2/4gb release image
<alyssa>
🙉
<kinkinkijkin>
i might grab one of my laptops just for this
<tlwoerner>
kinkinkijkin: nice! another canadian :-)
<alyssa>
tlwoerner: there are dozens of us! dozens!
<tlwoerner>
w00T! hopefully some of them land in the toronto area ;-)
<tlwoerner>
alyssa: off-topic: unbelievably i'll be getting fibre to the farm in a month or two! it's a dream come true :-D
<alyssa>
gasp
<tlwoerner>
and it'll cost less than what i'm paying for the 3Mb + 10 Mb i'm currently using
<alyssa>
;-;
<macc24>
tlwoerner: oh how much and what speed?
<kinkinkijkin>
if in ontario, definitely bell, who gave me $220 a month for 150/150 that was never met out in the place i was a year ago, out in the middle of nowhere
<macc24>
in poland i have 30/500 for around $15
<kinkinkijkin>
is that U:D?
<macc24>
yeah
<alyssa>
OA
<tlwoerner>
macc24: i've signed up for the "smallest" package (for now) which is 100Mb/50Mb for $100CAD/month
<tlwoerner>
alyssa: in Canada? because (they say) there are too few of us and the distances between are large
<macc24>
at this point my internet speed is probably bottlenecked by write speed to my ssd ,_,
<tlwoerner>
alyssa: luckily for me i've always been in canada, so to me this is dirt cheap. this will be the least i've ever paid for internet, and i'll be getting more than i've ever had before!
<tlwoerner>
perspective, i guess
<macc24>
wtf, they say that no data cap is a feature
<tlwoerner>
macc24: that is a feature! …in canada
<macc24>
no, that's a bug
<tlwoerner>
unlimited data is rarer than a moose with skates
<macc24>
why is internet so expensive in canada
<tlwoerner>
because they can (?)
<alyssa>
tlwoerner: I didn't realize moose with skates are rare.
<macc24>
have you tried powering off the device completely?
<kinkinkijkin>
yes
<kinkinkijkin>
um, I read the script again
<kinkinkijkin>
doesn't install a kernel
<macc24>
it should
<macc24>
it should dd kernel from /dev/sda1 to ${DEV}
<macc24>
it does it around line 65
<kinkinkijkin>
copies the entire fs
<kinkinkijkin>
wait nvm
<macc24>
/dev/sda1 is kernel partition
<SolidHal>
macc24 you should blank the kernel partition first before writing the new kernel
<macc24>
kinkinkijkin: try pressing volume up, volume down and power button at the same time
<SolidHal>
just dd if=/dev/zero of=/dev/sda1 bs=512 count=65536
<macc24>
SolidHal: yeah i learned that after few bricks, but script that installs kernel to emmc writes kernel from sda1 that is known to work
<kinkinkijkin>
oop redid the copy, synced, works now
* alyssa
is doing a s/pan/bi/
<alyssa>
is that, like, wrong?
<macc24>
kinkinkijkin: it works now?
<kinkinkijkin>
yeah
<macc24>
boots to login promp even without any pendrive
<SolidHal>
alyssa sounds like you don't even need a new driver, just use panfrost ;)
<kinkinkijkin>
alyssa that's panfrost erasure :p
<kinkinkijkin>
yeah macc24
<macc24>
oh thank god, i thought for a bit that i soft-bricked your duet
<kinkinkijkin>
should add syncs in a few places in the script
<macc24>
how readable was that script?
<kinkinkijkin>
readable
<kinkinkijkin>
I could read it, I have issues reading code
<SolidHal>
alyssa the weston timeline shows something pretty interesting. on kevin with a cursor jump, there is only a single vblank. On speedy when I do the same "cursor flick" that jumps on kevin there are around 7 more vblanks.
<SolidHal>
so I think you were right, there are a bunch of frames missing?