[jphilip]

I was once upon a time plumbing together an ARM matrix multiply backend for an on-device neural machine translation engine.

The objective was to get something working for the Mac M1 (which none of us had at the time). So I'd just cross-compile targeting my android phone, download built binary using wget on tmux and then run it to test if it's working. I remember I could build just the matrix multiply library locally on termux as well (after getting cmake and build-essentials via `pkg`).