Conversation
Hmmm possibly. But it would add a bit of complexity (not literally -- its just a bool -- but by giving the caller a choice). To give some background here: When we decided to extend it to cupynumeric, we moved it to this repo and made it project agnostic (though it already mostly was, as I figured this move would eventually happen). But it was only as agnostic as Legate or cupynumeric needed it to be. All this is to say: let's add this when we need it. But I am open to having my mind changed, just don't want to complicate things apriori. WDYT @tisaac @bryevdv? |
I agree. |
This fixes the problem of not exposing
--with-cudacto cupynumeric by default.