<ity>
I am so confused, so where is v9? Why is it not usable?
<sally>
From your experience and knowledge, if you compare between Rock5B and RaspberryPi5, which one will get the fully supported in upstream?
Leopold_ has quit [Remote host closed the connection]
Leopold_ has joined #panfrost
Leopold___ has joined #panfrost
Leopold_ has quit [Remote host closed the connection]
Leopold___ has quit [Remote host closed the connection]
Leopold has joined #panfrost
erlehmann has quit [Quit: Democracy must always be better armed than tyranny.]
Leopold has quit [Remote host closed the connection]
Leopold has joined #panfrost
Leopold has quit [Remote host closed the connection]
Leopold has joined #panfrost
Leopold has quit [Remote host closed the connection]
Leopold has joined #panfrost
krei-se- has joined #panfrost
krei-se has quit [Ping timeout: 480 seconds]
KetilJohnsen has joined #panfrost
erlehmann has joined #panfrost
warpme has joined #panfrost
erlehmann has quit [Quit: Democracy must always be better armed than tyranny.]
erlehmann has joined #panfrost
warpme has quit []
alpernebbi has quit [Ping timeout: 480 seconds]
alpernebbi has joined #panfrost
rasterman has joined #panfrost
<CounterPillow>
ity: what's confusing? It hasn't been written yet.
warpme has joined #panfrost
chewitt has joined #panfrost
warpme has quit []
warpme has joined #panfrost
KREYREN_oftc has quit [Remote host closed the connection]
KREYREN_oftc has joined #panfrost
Dr_Who has joined #panfrost
_41reasons has joined #panfrost
robmur01 has joined #panfrost
erlehmann has quit [Read error: Connection reset by peer]
erlehmann has joined #panfrost
ahmubashshir has quit [Ping timeout: 480 seconds]
ahmubashshir has joined #panfrost
<warpme>
guys: is current mesa (24.0.1) working out-of-box on rpi5 (v3d v7) with mainline 6.8-rc6?
<CounterPillow>
that doesn't seem like a panfrost question
<warpme>
oh sorry ! (is there dedicated irc for v3d?)
hazel_is_nutsasalle[m] has joined #panfrost
chewitt has quit [Quit: Zzz..]
warpme has quit []
chewitt has joined #panfrost
<ity>
CounterPillow: So, v9 support for VK has not been written yet? And what's wrong with panvk?
<CounterPillow>
Yes it does not appear to exist yet, and the current state of panvk is that it is not complete enough for any real world workload (nor would it ever provide performance enhancements on bifrost since bifrost is not made for vulkan on a hardware level)
<daniels>
panvk is being actively worked on, but the work atm is primarily in low-level infrastructure for it, and isn't something you can use as yet
warpme has joined #panfrost
<ity>
Ah, what low-level infra? What doesn't work on PanVK? Is it just that the hardware doesn't support some things, so they have to be emulated in software? Is anyone working on V9 support?
<daniels>
low-level infrastructure for the driver itself - fun stuff like synchronisation, command-buffer building, etc - it's not as simple as just writing a two-line implementation for every command, but there's quite a lot to do to make the driver a reality
<daniels>
gen10+ will probably be the primary focus, but yeah, there will be 7-9 support as well
chewitt has quit [Quit: Zzz..]
<ity>
Doesn't V9 have the hardware capabilities for VK support? Quote from a Mesa commit message: "For whatever it's worth, Valhall is the first Mali hardware that can support Vulkan properly" (commit c0d65398279db02f68b5acc23c3afbf1c34a520441)
<CounterPillow>
Yes it does
<ity>
So I don't fully understand what's the problem with v9?
<daniels>
there is no problem with v9
<daniels>
it is perfectly possible to support vulkan on v9, and vulkan will be supported on v9
<daniels>
it is not supported today because time is a finite resource
<daniels>
it will be supported in future
<ity>
But nobody is working on V9 support? The kernel driver is fully functional, right?
<CounterPillow>
I'm sure if you keep this line of inquiry you will make very good friends with the people who do have v9 vulkan on the roadmap but are swamped with work
<daniels>
ity: yes, it is being worked on!
<daniels>
it is not yet ready
<daniels>
if you want to make it happen faster, please contribute code
<ity>
I would like to, which is why I asked where is the work happening :P
<KREYREN_oftc>
^ ity's a contributor to the development of software stack OLIMEX Teres-I, no need to treat her like a random newb she's decent in kernel development
<ity>
Trying to start, yes, so I am looking into opportunities to contribute if contributions are welcome.
<ity>
daniels: Last commit 3 years ago, and it seems some of the work has been merged into panvk? I am looking but can't seem to find any references to v9/valhall in the fork you linked.
<daniels>
ity: right, you want the panvk-v10-wip branch which is where all the work has been going on - have a look at the MRs in that repo
<ity>
OH
KetilJohnsen has quit [Remote host closed the connection]
KetilJohnsen has joined #panfrost
ndufresne1 has joined #panfrost
nashpa has joined #panfrost
robmur01 has quit [charon.oftc.net helix.oftc.net]
rasterman has quit [charon.oftc.net helix.oftc.net]
krei-se- has quit [charon.oftc.net helix.oftc.net]
f_ has quit [charon.oftc.net helix.oftc.net]
dliviu has quit [charon.oftc.net helix.oftc.net]
mav has quit [charon.oftc.net helix.oftc.net]
Stary has quit [charon.oftc.net helix.oftc.net]
ndufresne has quit [charon.oftc.net helix.oftc.net]
paulk has quit [charon.oftc.net helix.oftc.net]
larunbe has quit [charon.oftc.net helix.oftc.net]
hazel_is_nutsasalle[m] has quit [charon.oftc.net helix.oftc.net]
TheKit[m] has quit [charon.oftc.net helix.oftc.net]
kusma has quit [charon.oftc.net helix.oftc.net]
samuelig has quit [charon.oftc.net helix.oftc.net]
krei-se has joined #panfrost
nashpa is now known as dliviu
TheKit[m] has joined #panfrost
alarumbe has joined #panfrost
hazel_is_nutsasalle[m] has joined #panfrost
robmur01 has joined #panfrost
rasterman has joined #panfrost
f_ has joined #panfrost
Stary has joined #panfrost
paulk has joined #panfrost
mav has joined #panfrost
kusma has joined #panfrost
samuelig has joined #panfrost
f__ has joined #panfrost
f__ is now known as funderscore
f_ has quit [Ping timeout: 480 seconds]
ndufresne1 is now known as ndufresne
erlehmann has quit [Quit: Democracy must always be better armed than tyranny.]
funderscore is now known as f_
<ity>
Hmm, is EnumeratePhysicalDevices unimplementeted? I am not 100% sure if I am reading the code right... There seems to be some device init code though
<bbrezillon>
ity: keep in mind this panvk-v10-wip is not functional, we're progressively adding support for v9 and v10, but we're still weeks away from having something that can run vkcube
<ity>
Ah, what's CSF & JM? I seen it in the codebase & it's mentioned on the MR, but I can't figure out what it stands for
<bbrezillon>
one thing I can think of that would be nice to have is converting <vertex,tiler> job pairs to IDVS jobs
<bbrezillon>
JM => Job Manager
<bbrezillon>
CSF => Command Stream Frontend
<CounterPillow>
v9 is JM only, right?
<bbrezillon>
it's the 2 different HW interfaces used to dispatch jobs to the GPU
<bbrezillon>
JM is v4 -> v9, and CSF is v10+
<ity>
OOOOH, I see, that makes sense
<ity>
Is the kernel driver done, or is that one WIP as well?
<ity>
How is panvk tested btw? Since one cannot use the VK API I presume (as EnumeratePhysicalDevices is not present), are the internal functions called directly in some in-tree testing file?
<bbrezillon>
ity: it's "functional" on Bifrost (v6/v7)
<bbrezillon>
and EnumeratePhysicalDevices is implemented by the vk runtime lib (see vk_common_EnumeratePhysicalDevices)
rasterman has quit [Quit: Gettin' stinky!]
<ity>
OH, I guess I got confused by panvk_entrypoints.c
<bbrezillon>
quite a few vulkan entrypoints are implemented by the vk runtime lib, drivers just overload those if the default implementation is not enough for them