tree abe088165f907ad84daad98630bc0129db510cde
parent 914b8f8966879c274ca30130a24d502c08f59b6c
author Matthias Boehm <mboehm7@gmail.com> 1605389168 +0100
committer Matthias Boehm <mboehm7@gmail.com> 1605389245 +0100

[SYSTEMDS-2630] Multi-threaded slicing in sliced federated broadcasts

For the case of broadcasting a large, potentially sparse, matrix in a
sliced manner (where every federated partition only received the needed
data) so far we sliced the blocks sequentially. With this patch, we
simply do this independent block slicing in a multi-threaded manner in
order to avoid unnecessary overhead in case of large broadcasts.