Update reduction
This merge request has many changes related to parallel reductions. These changes have now been separated to create this new merge request.
The main changes are:
- Move reduction code out of
QuantizeOutput.cu
intoreduction.cuh
- Make the
reduce_sum
function more robust (e.g. correct regardless of block size) - Add
tReduction
to test the functionality provided inreduction.cuh
Edited by Bram Veenboer