search

hollance / neural-engine

Everything we actually know about the Apple Neural Engine (ANE)
MIT License
1.97k stars 72 forks source link

Added explanation for multiprecision #7

Closed Piratach closed 3 years ago

Piratach commented 3 years ago

I'm not entirely certain if this is what Apple meant by multiprecision or not, but I based the explanation off these two links: https://en.wikipedia.org/wiki/Arbitrary-precision_arithmetic https://stackoverflow.com/questions/12088857/what-is-the-precise-meaning-of-arbitrary-precision

Please feel free to edit or decline if this is not correct!

hollance commented 3 years ago

Thanks for the suggestion but I'm not 100% sure that is what Apple actually means here, or if they do what part of the number has arbitrary precision. Maybe all they mean is that it supports float32, float16, and int8 for certain operations instead of just float16.