gromero commented on pull request #8761: URL: https://github.com/apache/tvm/pull/8761#issuecomment-899951536
> regardless i think we should still ensure that write() conforms to the API properly. @areusch yep, I agree. > can you let me know which test case you're running? It's pretty much the sine model in tutorials/micro_tflite.py. I have prepared a branch which reproduces it (after several runs, you know): https://github.com/gromero/tvm/commits/8728 Just clone && cd 8728 && `./run.sh`. The MLF sine.tar is the same model as in micro_tflite.py but already compiled. The run must crash with: ``` [...] INFO:__main__:b'/tmp/x20/qemu-hack/qemu-system-arm -cpu cortex-m33 -machine mps2-an521 -nographic -m 16 -vga none -net none -pidfile qemu.pid -chardev pipe,id=con,mux=on,path=/tmp/tmpq4aigpet/fifo -serial chardev:con -mon chardev=con,mode=readline -icount shift=7,align=off,sleep=off -rtc clock=vm -kernel /tmp/x20/build/zephyr/zephyr.elf\n' qemu-system-arm: warning: nic lan9118.0 has no peer [02:19:49] /home/gromero/git/tvm/src/runtime/micro/micro_session.cc:367: remote: microTVM Zephyr runtime - running B INFO:__main__:b"make[3]: Leaving directory '/tmp/x20/build'\n" INFO:__main__:b'[100%] Built target run\n' INFO:__main__:b"make[2]: Leaving directory '/tmp/x20/build'\n" INFO:__main__:b'/usr/bin/cmake -E cmake_progress_start /tmp/x20/build/CMakeFiles 0\n' INFO:__main__:b"make[1]: Leaving directory '/tmp/x20/build'\n" Traceback (most recent call last): File "/home/gromero/git/tvm/8728/./an.py", line 38, in <module> syslib = session.get_system_lib() File "/home/gromero/git/tvm/python/tvm/micro/session.py", line 88, in get_system_lib return self._rpc.get_function("runtime.SystemLib")() File "/home/gromero/git/tvm/python/tvm/rpc/client.py", line 73, in get_function return self._sess.get_function(name) File "/home/gromero/git/tvm/python/tvm/runtime/module.py", line 85, in get_function check_call( File "/home/gromero/git/tvm/python/tvm/_ffi/base.py", line 348, in check_call raise get_last_ffi_error() tvm._ffi.base.TVMError: MicroSessionTimeoutError: failed to read reply message after timeout 5s FAIL ``` Which points to the timeout happening at https://github.com/gromero/tvm/blob/8728/8728/an.py#L38 , i.e. just after the handshake as you mentioned I believe. So I don't believe it's related to the model itself. I'm also with that hypothesis in mind that you raised about something fishy in mps2_an521 drivers ... But can't make sense of it yet. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
