kinkinkijkin has quit [Quit: Leaving]
Cyrinux9474 has joined #panfrost
Cyrinux947 has quit [Ping timeout: 480 seconds]
DVulgaris has quit [Ping timeout: 480 seconds]
rasterman has joined #panfrost
DVulgaris has joined #panfrost
Leopold_ has joined #panfrost
CalebFontenotHaileysCuteNerdyB has quit []
dliviu has quit [Ping timeout: 480 seconds]
dliviu has joined #panfrost
<
PaulFertser>
rasterman: hey :) so you basically reproduced my odd results?
<
rasterman>
i swear i have seen better
<
rasterman>
actually on my amdgpu its reverse...
<
rasterman>
x11 is much better than wl
<
rasterman>
like glmark2 Score: 9840 for x
<
rasterman>
glmark2 Score: 5985 for wl
<
PaulFertser>
rasterman: you do get bad results even with a GL-enabled X11 compositor?
<
PaulFertser>
(on panfrost)
<
rasterman>
for x composited vs wl composited on panfrost - midgard tho
<
rasterman>
307 vs 213
<
rasterman>
so nowhere near as bad as your results
<
rasterman>
same exact wm and compositor for both x and wl
<
rasterman>
so i'm being super fair :)
<
rasterman>
(in this case i know the compositor wont be using hw planes so that will impact results)
<
rasterman>
i didnt check glmark without a compositor tho...
<
PaulFertser>
I still can't understand why I might be getting bad results if I'm not CPU-limited.
<
rasterman>
maybe i should have
<
rasterman>
copies going through cpu?
<
rasterman>
on the xserver side
<
rasterman>
is xorg using like 100% cpu?
<
rasterman>
sorry - im not at home. can't test now
<
PaulFertser>
I showed Linux perf results, it doesn't look like being CPU-limited at all.
<
rasterman>
residential redox flow
<
PaulFertser>
All of the cores are > 85 % idle.
<
rasterman>
well the most efficient compositing pipeline for the x side and for the wl side assuming you dont use hw planes.... will be the same
<
rasterman>
zero copy swap on client side to server (wl or x)
<
rasterman>
compositor uses src buffer as texture and draws to its backbuffer (copy)
<
rasterman>
then zero copy swap of compositor screen buffer
<
rasterman>
in both cases
<
rasterman>
extra ipc bounces of course as discussed
<
rasterman>
it almost certainly will not be found by perf
<
rasterman>
perf does not know what the gpu is doing.
<
PaulFertser>
Yes, my point is that it shows how I'm not CPU-limited at all.
<
rasterman>
well asking if its at 100% cpu was that :)
<
PaulFertser>
So something is different on GPU level but I have no idea how to "perf" that.
<
rasterman>
it's either blocking and waiting on something
<
rasterman>
eg is limtied to 60hz ...
<
rasterman>
because its blocking on vblank
<
rasterman>
or somethnig similar
<
rasterman>
or its doing more copies
<
PaulFertser>
I'm running with vblank_mode=0 and both results are way higher than 60 Hz.
<
rasterman>
yeah - well it being stuck at 60hz would be a clear sign its hz limited :)
nashpa has joined #panfrost
dliviu has quit [Ping timeout: 480 seconds]
simon-perretta-img has joined #panfrost
pbrobinson has quit [Remote host closed the connection]
pbrobinson has joined #panfrost
samuelig has quit [Quit: Bye!]
hanetzer1 has quit [Ping timeout: 480 seconds]
samuelig has joined #panfrost
hanetzer1 has joined #panfrost
alyssa has joined #panfrost
pbrobinson has quit [Ping timeout: 480 seconds]
pbrobinson has joined #panfrost
Cyrinux9474 has quit []
Cyrinux9474 has joined #panfrost
rasterman has quit [Quit: Gettin' stinky!]
pbrobinson has quit [Ping timeout: 480 seconds]
pbrobinson has joined #panfrost
Leopold_ has quit [Remote host closed the connection]
Leopold_ has joined #panfrost
DVulgaris has quit [Read error: Connection reset by peer]
DVulgaris has joined #panfrost
rasterman has joined #panfrost
<
alyssa>
bbrezillon: ^
tchebb has quit [Ping timeout: 480 seconds]
Leopold_ has quit [Ping timeout: 480 seconds]
tchebb has joined #panfrost
rasterman has quit [Quit: Gettin' stinky!]
paulk-bis has joined #panfrost
paulk has quit [Ping timeout: 480 seconds]