tum-ei-eda / mlonmcu

Tool for the deployment and analysis of TinyML applications on TFLM and MicroTVM backends
Apache License 2.0
29 stars 12 forks source link

Support more targets #3

Closed PhilippvK closed 2 years ago

PhilippvK commented 2 years ago

Currently we support two targets:

In the future we might support further architectures/simulator/devices e.g.:

PhilippvK commented 2 years ago

The main question is how we get useful metrics on these platforms. Getting an estimation of Cycle counts as well as tracing dynamic memory accesses would be optimal but is not feasible for most platforms.

PhilippvK commented 2 years ago

Here is a progress update:

Targets:

PhilippvK commented 2 years ago

Now. That we have also non-ETISS targets available, it might be feasible to use a different simulator for the MicroTVM Autotuning flow. The Spike Proxy Kernel seems to also offer a UART interface which could be used for this. OVPsim seems to be the fastest simulator right now which would be desirable as autotuning is already very time consuming, but I have not yet found out if it supports a virtual UART.

PhilippvK commented 2 years ago

Here is an update regarding real-hardware support: