Repository navigation
HPR-LP v0.1.2
Highlights of HPR-LP v0.1.2 (September 27, 2025):
- SpMV rewrites. Add a preprocessing step and buffer preallocation to avoid redundant work between iterations.
- Kernel rewrites. Several CUDA kernels were refactored to reduce memory traffic and improve occupancy.
- In terms of SGM10 (1e-8 accuracy), 11% faster for Mittelmann's LP benchmark set and 7% faster for MIP2017 large-scale LP relaxations (compared to v0.1.1).