There are 2 implementation, scalar and vector versions. And both implementation are bit complicated, so it's worth to double check both implementations and verify the performance gain of vector version.

## tested on K230-CanMV