Vectorize msgs #17
Vectorize msgs #17
Conversation
| let (commitment, buffer): (Commitment, [u8; 32]) = recv_from(channel, p, "RNG") | ||
| .await? | ||
| .pop() | ||
| .ok_or(Error::EmptyMsg)?; |
There was a problem hiding this comment.
Very nice! I assume this should now be added to each recv_from (whenever it's not a vector that is being sent), do I get it right?
There was a problem hiding this comment.
Yes exactly, we could also add back a method like recv_single_from, but since this behavior isn't needed very often I felt like this way of doing it was fine.
| let c0 = commit(&d0[r].to_be_bytes()); | ||
| let c1 = commit(&d1[r].to_be_bytes()); | ||
| let cm = commit(&dm); | ||
| own_commitments.push((c0, c1, cm, dm)) |
There was a problem hiding this comment.
Nice, and much simpler this way!
| @@ -673,7 +677,7 @@ pub(crate) async fn faand_precomp( | |||
| let triples = transform(xbits, ybits, &zbits, lprime); | |||
|
|
|||
| // Step 2 assign objects to buckets. | |||
| let mut buckets: Vec<Vec<(Share, Share, Share)>> = vec![vec![]; length]; | |||
| let mut buckets: Vec<SmallVec<[(Share, Share, Share); 3]>> = vec![smallvec![]; length]; | |||
There was a problem hiding this comment.
Just out of curiosity and to learn: what is the specific reason here for using SmallVec for buckets? Is there some performance-related reason?
There was a problem hiding this comment.
Exactly. The vector is kept on the stack, bringing down the total runtime of the benchmark from 10 minutes to 7 minutes when I tested it with 250 records.
No description provided.