AccelerateHS / AccelerateHS/accelerate
New-pipeline: improvements to code generation on PTX
Open
- Dominant language
- Haskell
- Stars
- 1k
- Forks
- 135
- PR merge metrics
- No merged PRs in 30d
Description
After implementing one-dimensional folds on PTX (#567) on the new-pipeline, we should make several other improvements to the generated code:
- [ ] Multi-dimensional folds (row per threadblock)
- [ ] 1-dimensional scan (chained scan is implemented, should be extended to chained scan *with decoupled look-back*)
- [ ] Multi-dimensional scans (row per threadblock)
- [ ] Permute: lock-free for simple operations
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.