Roadmap: OSP Improvement Plans #6
Description
OSP usage in general:
Versioning for OSP - done
github CI - done in #5
Double-check cmake compatibility with osp ppa libraries & availability in both x86 and ARM - done in #7
Spack module for OSP - done in #18
Adding Python FE to OSP - (4)
GitHub wiki tutorial and blog post - (3)
Integration into Fedora/RHEL via self-hosted DNF/YUM RPM repository (1) - done #19
OSP algorithmics:
Integrate SSP model into new OSP branch - done
Partitioning classes and algorithms in OSP - done in #16
More sophisticated algorithms for SSP/maxBSP (ILPs specifically) - ILP in #21 ; varianceSSP in #13 ; converter, HC, HCcs in #40
Integrate partialILP and csILP into new OSP branch - in #36
More sophisticated algorithms for recomputation - (3)
OSP utility: file conversion interface for hyperDAG DB - #10 + changes in https://github.com/Algebraic-Programming/HyperDAG_DB
Adjust more OSP algorithms to node/core types - (4) Serial as part of #25
Integrate (Co)BaldMiXR into new OSP branch (5)
Single-core scheduling/locality reordering - (collab)
Design/implement acyclic partitioners - (collab)
OSP(^2) interface:
Handling algorithm time limits in OSP^2 - (2)
Handle model extensions in OSP^2 (& annotate OSP algorithms with model extension capabilities, as required)
- node/core types - (2)
- recomp - (2)
- NUMA (warn if ignored) - (2)
- mem constr. - (4)
Add more scheduling algorithms to interface of OSP app / OSP^2
- partialILP - (3)
- allow OSP^2 to freely combine coarsener + scheduler + improver - (4)
Add pebbling algorithms to interface of OSP app / OSP^2 - (3)
Add non-acyclic partitioning algorithms to interface of OSP app / OSP^2 - (5)