oh very cool information, thank you sooooo much for sharing the colab.
Seems i can just stack my 2d arrays and execute all 2d arrays at the same time. Exactly what i did want to know :-) seems its also possible to define the vector of the direction of calculation.
Yup, as long as it is valid mathematical operations you can process any vector, matrices, tensors depend on your need. The only problem usually is how to write it in valid python code, so we need more example :)
on python i would use first numba with a custom kernel to try it out, improve it... after that switch to C
