[FEA] chunked_pack should allow variable sized spans in the next call #13332
Labels
0 - Backlog
In queue waiting for assignment
feature request
New feature or request
libcudf
Affects libcudf (C++/CUDA) code.
Spark
Functionality that helps Spark RAPIDS
As brought up here #13260 (comment),
chunked_pack
requires that it be configured with a single size for the user's bounce buffer, and further requires the same size of device memory made available in every call tochunked_pack::next
.This is overkill in the case where we have actually very little work to be done (the source table is small). So the proposal is to allow for a
get_next_size()
api that lets us know the size of buffer that should be made available tonext
The text was updated successfully, but these errors were encountered: