NVIDIA / NVIDIA/cccl

[CUB]: Support processing multiple segments per block in agent_segment_scan

Open
#6,783 1 comment 1 reaction 1 assignee Claimed by @oleksandr-pavlyk View on GitHub
Dominant language
C++
Stars
2.5k
Forks
486
Avg merge
2d 6h
Merged PRs (30d)
295

Description

Currently, `cub::detail::segmented_scan::agent_segmented_scan` processes one segment per block.

This approach does not perform well when segments are short.

This issue is to support processing statically requested number of segments per block though using Schwarz reformulation of segmented scan as an ordinary scan over augmented type and using augmented scan operator.

Having such a support would allow `DeviceSegmentedScan` to process small size segments more efficiently.

This issue can be closed by merged implementation. The interface for enabling it will come as a follow up work after we have parameter guarantees framework.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.