jaguilar/pybricks-ev3-ffi-demo

Pybricks EV3 FFI Demo

This project contains a demo of using Micropython's FFI functionality on the Lego EV3 system with Pybricks. This example project takes you through all the steps of getting the ffi module working on pybricks on EV3.

This project is not supported. The current version seems to work fine on a Raspberry Pi 4 build host connected to an EV3 via bluetooth.

Requirements

• SSH/SCP
• Python
• arm-linux-gnueabi-gcc and arm-linux-gnueabi-g++ are on PATH

Tour

• common.sh contains common functions and definitions used throughout the project to build the example code. It also contains hacks for working out of WSL, which I did in a previous version of this work.
• demo_helloworld_ev3.sh takes the previous example a step further and demonstrates the program running on the EV3 brick.
• demo_pyfib_ev3.sh demonstrates running some simple Fibonacci sequence calcuations on the EV3 brick in pybricks-micropython.
• demo_cfib_ev3.sh demonstrates building, loading, and running Fibonacci sequence calculations implemented in C on the EV3 brick, loaded via the ffi module, in pybricks-micropython.

Benchmark

On my EV3 brick:

Implementation	Time
Python	155ms
C	6ms

Given that micropython's Viper code generator is said to be ~4x as fast as plain micropython, this suggests that writing in C can give you approximately an additional 8x speedup compared to Viper. Probably the exact speedup depends on the project.

Next Steps

• C++?
• Optimizer flags?

Conclusion

I hope this is helpful to anyone who wants to use C to speed up their CPU-intensive micropython programs. I suspect this would also work for C++ libraries and I may later enhance this test to show that as well.