←back to thread

Finite Field Assembly: A Language for Emulating GPUs on CPU

(leetarxiv.substack.com)
40 points muragekibicho | 1 comments | 14 Jan 25 09:35 UTC | HN request time: 0s | source
Show context
hashxyz ◴[18 Jan 25 00:12 UTC] No.42744594[source]▶
Pretty sure this is just vectorization. You can pack some 8bit ints into a machine-length 32bit int and add them together, that is vectorization.
replies(1): >>42744863 #
1. Conscat ◴[18 Jan 25 01:09 UTC] No.42744863[source]▶
I don't think that's true when the add overflows. You wouldn't want a lane's overflow to carry into an adjacent lane.