Danct12 has quit [Remote host closed the connection]
camus has joined #panfrost
rasterman has joined #panfrost
Lyude has quit [Remote host closed the connection]
Lyude has joined #panfrost
floof58 is now known as Guest4106
Guest4106 has quit [Ping timeout: 480 seconds]
rasterman has quit [Quit: Gettin' stinky!]
Mangy_Dog has joined #panfrost
hanetzer2 has joined #panfrost
hanetzer1 has quit [Ping timeout: 480 seconds]
alarumbe has quit [Quit: ZNC 1.8.2+deb2 - https://znc.in]
alarumbe has joined #panfrost
Putti has joined #panfrost
Putti has quit [Ping timeout: 480 seconds]
Putti has joined #panfrost
hanetzer3 has joined #panfrost
hanetzer2 has quit [Ping timeout: 480 seconds]
userish has joined #panfrost
<userish>
Back online, curious about my RK3588 question with VESA driver
<CounterPillow>
Personally not sure what you mean by "VESA driver", video output is a separate matter from GPU acceleration.
<userish>
CounterPillow: as far as I was aware the VESA driver gives output with *no* acceleration
<userish>
That is the point of it
<CounterPillow>
You don't need the GPU at all for that
<CounterPillow>
just VOP2
<userish>
What is VOP2?
<CounterPillow>
Rockchip's video output IP on the RK3588
<CounterPillow>
it's what handles video output
<CounterPillow>
You don't need a GPU driver at all to get video output
<userish>
Oh okay does this work even with dual 4K@60p and 8K@60p displays?
<CounterPillow>
Sure, but 8K only if you implement the awful cluster splicing stuff RK3588's VOP2 has
<userish>
I am unfamiliar with the term 'cluster splicing'
<CounterPillow>
Well I should be more precise, 8K60p is impossible, the hardware doesn't support that
<CounterPillow>
the way it does 8k30p is that it splices together the outputs of two 4k60p capable clusters
<CounterPillow>
Either way, this is unrelated to Panfrost, so off-topic for this channel
<CounterPillow>
oh actually it does say 8k@60 in the TRM
<CounterPillow>
I wonder how they do that with 4 times the area but two clusters
<userish>
Yeah I saw an SBC saying it has 8k60p
<userish>
I was thinking the board can do all 3 at once, with two 4k60p's at the same time as the 8k60p
<CounterPillow>
Maybe? The TRM isn't super clear on this
atler has quit [Quit: atler]
<CounterPillow>
Signs point to yes it can do that
atler has joined #panfrost
rasterman has joined #panfrost
<userish>
To be clear: Panfrost cannot work at all on the RK3588 without a blob, even for 2D? (this is why I ask about VESA driver)
<Mangy_Dog>
i just wish rockchip would release there blobs for mainline, even if they dont want to release source... at least allow us to use the hardware to its fullest without having to resort to (no offence) community endevours
<CounterPillow>
Yes the mainline driver will require the GPU firmware blob, as that is the only stable interface as per ARM. It's ARM's blob, not Rockchip's.
<CounterPillow>
However, you can have video output without GPU (and thus without the firmware blob), as that's a completely separate IP.
<Mangy_Dog>
oh well in any case... why cant arm just release a public blob?
<Mangy_Dog>
not like its giving away IP
<Mangy_Dog>
its just a driver so people can use their own hardware
<CounterPillow>
Ask ARM
<CounterPillow>
Also, it's not a driver, we're talking about the firmware running on the GPU. ARM's proprietary drivers are a separate matter and not on-topic here.
<Mangy_Dog>
in any case..... i just wish the gpu was running at its full capacity when i run a program that requires it
<Mangy_Dog>
because i want to use this chip in a gaming handheld im designing
<daniels>
yep, it’s WIP
<CounterPillow>
daniels: "it" referring to VOP2?
<userish>
CounterPillow: You say it is not a driver, but the firmware is PART of the driver IMHO
<daniels>
the only Arm blobs are for Mali; VOP is RK
<userish>
I thought the RK3588 blob was for 3D and video acceleration, etc. and CounterPillow says the VOP2 logic block is for video output bypassing most of the GPU?
<CounterPillow>
just 3d, video acceleration does not require firmware
<CounterPillow>
and "bypassing most of the GPU" isn't really accurate, it has nothing to do with the GPU. It just reads from a memory address; that memory address can be populated with data by the CPU, GPU, or by RGA, or hell even by a video decoder
<CounterPillow>
on an aside note, video decode/encode acceleration also has nothing to do with the GPU
<Mangy_Dog>
you sure?
<Mangy_Dog>
im pretty certain video decoding is a gpu feature
<CounterPillow>
It's not.
<Mangy_Dog>
and its been shown in youtube play back tests
<CounterPillow>
the confusion about video output and hardware video coding being somehow related to the GPU is probably that on systems with dedicated GPUs, these functions are usually all done by the graphics card, in order to avoid the round trip through the PCIe bus/CPU memory (since there is a physical split between GPU and CPU memory on systems with dGPUs)
<Mangy_Dog>
for instance, in android where the gpu driver is fully implimented 4k playback has worked flawlessly
<Mangy_Dog>
on the rk3588
<Mangy_Dog>
but in linux where there basically isnt much gpu drivers running... its a mess
<CounterPillow>
Both hardware video decode and gpu drivers missing at the moment is not evidence of them being somehow the same IP, they're not. Read the RK3588 TRM.
<Mangy_Dog>
the trm?
<Mangy_Dog>
technical reference manual?
<CounterPillow>
yes
<Mangy_Dog>
no offence as im not at all in the legue for making gpu api drivers... im not going to nose dive into hundreds of pages to fully understand it right now :p
<Mangy_Dog>
im an embedded developer... but bpu apis
<Mangy_Dog>
gpu
<Mangy_Dog>
is not me
<Mangy_Dog>
:p
<Mangy_Dog>
in any case
<Mangy_Dog>
its still shitty that arm or rockchip... or whoever makes the hardware/designs doesnt just release the driver files for it to run properly
<Mangy_Dog>
i honestly dont understand teh mindset behind not wanting to do so
<Mangy_Dog>
even nvidia managed to do it and still protect their ip :p
<Mangy_Dog>
i admit they took some convincing
<Mangy_Dog>
but it was done
rasterman has quit [Remote host closed the connection]