You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In CKKS pipeline, aside from handling arith.{add,mul}f, we also have some code handling arith.{add,mul}i in like secret-insert-mgmt-ckks and secret-to-ckks. Technically we can support some kind of integer arithmetic in CKKS, however, we have no example mlir that goes through them to actually test them.
This issue is intended as meta-issue on integer arithmetic support in CKKS.
I think this touches on a deeper issue with the CKKS pipeline, which is that we don't really have a semantic for the types. My intuition would be that any integer "x" should basically be treated as "x.0", so arith.uitofp would become a no-op (just like arith.extui already is). However, we need to somehow communicate this semantic to the developer in a concise way + there's the question of "at what scale will x be encoded?", especially in more complex CKKS settings where we don't have a single scale for everything.
In CKKS pipeline, aside from handling
arith.{add,mul}f
, we also have some code handlingarith.{add,mul}i
in likesecret-insert-mgmt-ckks
andsecret-to-ckks
. Technically we can support some kind of integer arithmetic in CKKS, however, we have no example mlir that goes through them to actually test them.This issue is intended as meta-issue on integer arithmetic support in CKKS.
One sub issue is the
arith.uitofp
one (#1628)The text was updated successfully, but these errors were encountered: