ChanServ changed the topic of #etnaviv to: #etnaviv - the home of the reverse-engineered Vivante GPU driver - Logs https://oftc.irclog.whitequark.org/etnaviv
frieder has joined #etnaviv
mvlad has joined #etnaviv
lynxeye has joined #etnaviv
pcercuei has joined #etnaviv
<szemzoa> I try to run the classification.py example on my SOC (Allwinner v851s). It's very constrained, so I use alpine-linux, and compiled tflite delegate and mesa with musllibc.
<szemzoa> without libteflon.so the output is ok, but with -e libteflon.so it has an error:
<szemzoa> File "/usr/lib/python3.12/site-packages/tflite_runtime/interpreter.py", line 516, in __init__
<szemzoa> self._interpreter.ModifyGraphWithDelegate(
<szemzoa> Is there any way to debug further what can be wrong? I'm not sure at all my etnaviv kernel driver is 100% ok, I see the drm debug, and sysfs debug: https://pastebin.com/j3gY338y
<szemzoa> s/driver/port/
<szemzoa> also, if this VIPlitePico is not in mesa hwdb, should I add it to there, or to the kernel's etnaviv hwdb. (now I added that to both places). It's model is 0x9000
<szemzoa> kernel 6.12.1, mesa 24.2.7, tensorlow is 2.18
<austriancoder> szemzoa: kernel wise it looks okay.
<austriancoder> szemzoa: with teflon tomeu can help you out
<austriancoder> tomeu: ^
<tomeu> hmm, I'm not sure of how similar VIPLite is to the non-lite
<tomeu> probably quite different
<tomeu> in any case, the kernel seems to initialize correctly
<tomeu> you can just look into tfl_device.c why the delegate isn't working, some debugging will be needed
<szemzoa> basically It's unmodified driver, I just added a reset line handling to the init
<tomeu> plus the hwdb additions, right?
<szemzoa> yes, I converted a few months ago from the vendor galcore's features .h
<tomeu> cool, so besides debugging the actual failure inside mesa, you can also push your branches in case we can spot something wrong or missing
<szemzoa> but I'm not sure if the clocks needed for the NPU is totally ok in the kernel
<austriancoder> szemzoa: the kernel driver can read some registers .. so not to bad reagarding clocks
<austriancoder> And there is no hang etc
<szemzoa> austriancoder: ok thanks, so kernel vise it's more or less okay
<szemzoa> tomeu: Thanks, I'll try to add debug to the tfl_device.c
<szemzoa> but all of this is quite new for me, so I've to understand what is happening here :)
lynxeye has quit [Quit: Leaving.]
frieder has quit [Remote host closed the connection]
mvlad has quit [Remote host closed the connection]