Hi im trying to compile Tensorflow Lite for the Cortex-M4 on Colibri imx7 emmc 1gb,
the problem is that i don’t know if it is possible.
I compiled all using MBED software but the .elf file does not run on the M4.
Is there a simple way to compile tensorflow lite for the Cortex-M4 ?
We newer tried to compile a Tensorflow Lite for the Cortex-M4.
Which U_Boot version do you have on module and how did you try to run an elf file?
U_boot supplied with our latest BSPs has some issue loading elf files. Could you try to compile it to bin ?
Please doublecheck your linkers settings to use proper memory areas.
Hi Alex, im answering your questions:
Which U_Boot version do you have on module and how did you try to run an elf file? : Im using U-Boot 2019.07-3.0.4+g26d926eda0
Could you try to compile it to bin ? : The compiler created a .elf and a .bin file, and they do not run
This is what i tried: (2 methods)
Compile for Mbed Software : By using this method i created a Mbed Project for the Software to detect and compile, (like Eclipse .project file). ---->The problem is tha when i compile it ask me to specify a TARGET MCU, so my best guess was to search for an MCU with the same Cortex-M4F. in the end i tried all of them and none of them worked. (Probably for memory areas reasons as you mentioned).
Compile for cortex_m4_generic : With this method i created a file called libtensorflow-microlite.a, so the ouput of the compilation of Tensorflow was a static library for g++.---->The Problem is that i don’t understand how to use it; I true to give it as and argument for the compiler but it does not compile because the compiler does not know where to find all .h files related to Tensorflow.
So this is what i tried, it would be very interesting to see tensorflow lite running on the M4.
Thank you for the help, i will investigate more, and if i succeed i will let you know.
Sorry but you can just chose generic M4 model and make working binary file. We don’t support neither Mbed project nor Tensor Flow lite for M4 core on iMX7. Though we do have a Tensorflow lite example running on A core(s).
Please try an NXP community for help with your original question.