Using CUB (planned) 2.1.0's batched memcpy
Open
cat:performance
prio:medium
- Dominant language
- Python
- Stars
- 12.3k
- Forks
- 1.1k
- Avg merge
- 1d 20h
- Merged PRs (30d)
- 45
Description
PR ( https://github.com/NVIDIA/cub/pull/359 ) landed in CUB recently. Looks like this will be part of CUB [2.1.0]( https://github.com/NVIDIA/cub/milestone/10 ). This would allow batching multiple memcpy's together.
Potentially would be useful in operations like `concatenate`, `stack`, `block`, etc. Anything involving multiple copies between arrays
Contributor guide
Research direction
Review CUB PR #359 and milestone 2.1.0 first, then inspect the copy paths used by concatenate, stack, and block. Determine which multiple-array copies can use batched memcpy and validate the affected operations; the issue does not name specific files or tests.
Written by the indexing model from the issue text.
Assessment
- Domain
- performance
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100