Leopold_ has quit [Remote host closed the connection]
Leopold_ has joined #etnaviv
Leopold_ has quit []
frieder has joined #etnaviv
frieder has quit []
frieder has joined #etnaviv
frieder has quit [Quit: Leaving]
frieder has joined #etnaviv
lynxeye has joined #etnaviv
frieder has quit [Quit: Leaving]
frieder has joined #etnaviv
lynxeye has quit [Ping timeout: 480 seconds]
JohnnyonF has joined #etnaviv
lynxeye has joined #etnaviv
lynxeye has quit [Remote host closed the connection]
lynxeye has joined #etnaviv
JohnnyonFlame has quit [Ping timeout: 480 seconds]
pcercuei has joined #etnaviv
<tomeu>
lynxeye: do you think we need a ETNAVIV_PARAM_GPU_SHADER_CORE_COUNT ?
<lynxeye>
tomeu: Do you really mean shader cores or rather NN cores?
<tomeu>
sorry, meant for NN cores
<tomeu>
not sure what userspace would do with that info
<lynxeye>
Yea that's exactly one of the things why I want the NPU support to be marked as experimental. I _feel_ like we may need that info in userspace, but I don't know enough about the NN accelerator yet to be sure.
<lynxeye>
The downstream kernel has even more such properties. From the HWDB there it looks like the NN cores could have different abilities, like the number of cores that support bfloat16. I guess that would also be required info userspace that actually makes use of the NN accelerator.
<lynxeye>
So while I think we need some way to communicate the number of NN cores to userspace I also feel like it may just be a subset of the required info. So I would rather see us add the minimal support for the kernel driver to properly identify NPU cores and then add UAPI once we have actual userspace for the NN accelerator.
<tomeu>
yeah, though downstream drivers tend to have UABIs a few times more complex than mainline
<tomeu>
to me it is not clear even if any silicon was shipped that would make use of some of the properties
<tomeu>
ok, I'm all for adding UAPI only after we have a good use for it
<tomeu>
so won't be exposing it for now
<lynxeye>
I certainly won't argue to replicate the blob UAPI ;) I guess we have a much better picture on what is actually needed once we have some kind of userspace and a bit more RE done.
JohnnyonFlame has joined #etnaviv
JohnnyonF has quit [Ping timeout: 480 seconds]
<tomeu>
yeah, I'm looking now at what would take to run TFLite on it
<tomeu>
that should give us a better idea of what is actually needed
<tomeu>
[ 733.669266] etnaviv-gpu ff100000.npu: etnaviv has been instantiated on a NPU, for which the UAPI is still experimental
<tomeu>
[ 733.680375] [drm] Initialized etnaviv 1.3.0 20151214 for etnaviv on minor 2
<mwalle>
so is that a GPU without an actual graphics output?
<lynxeye>
mwalle: Nope, it misses the 2D/3D pipeline. It just has the same shader clusters, so you can do the OpenCL part on it. Plus a big matrix accelerator for the NN stuff.
<lynxeye>
tomeu: looks good
Leopold_ has joined #etnaviv
Leopold_ has quit [Remote host closed the connection]
Leopold_ has joined #etnaviv
<mwalle>
lynxeye: ah, thanks
Leopold_ has quit [Remote host closed the connection]
Leopold_ has joined #etnaviv
frieder has quit [Remote host closed the connection]
pcercuei has quit [Read error: Connection reset by peer]
pcercuei has joined #etnaviv
pcercuei has quit [Read error: Connection reset by peer]
pcercuei has joined #etnaviv
lynxeye has quit [Quit: Leaving.]
JohnnyonF has joined #etnaviv
JohnnyonFlame has quit [Ping timeout: 480 seconds]
Leopold_ has quit [Remote host closed the connection]
Leopold_ has joined #etnaviv
Leopold_ has quit []
Leopold_ has joined #etnaviv
Leopold_ has quit []
Leopold_ has joined #etnaviv
Leopold_ has quit [Remote host closed the connection]