monolish
MONOlithic LIner equation Solvers for Highly-parallel architecture
monolish is a linear equation solver library that monolithically fuses variable data type, matrix structures, matrix data format, vendor specific data transfer APIs, and vendor specific numerical algebra libraries.
- Documentation: https://ricosjp.github.io/monolish/
- Source code: https://github.com/ricosjp/monolish/
- Contribution guide : https://ricosjp.github.io/monolish/da/dce/contribution.html
- Bug reports: https://github.com/ricosjp/monolish/issues
- Releases: https://github.com/ricosjp/monolish/releases
- Changelog: https://github.com/ricosjp/monolish/blob/master/CHANGELOG.md
- monolish log viewer: https://pypi.org/project/monolish-log-viewer/
- monolish benchmark result: https://ricosjp.github.io/monolish_benchmark_result/
monolish let developer forget about:
- Performance tuning
- Processor differences which execute library (Intel CPU, NVIDIA GPU, AMD CPU, ARM CPU, NEC SX-Aurora TSUBASA, etc.)
- Vendor specific data transfer APIs (host RAM to Device RAM)
- Finding bottlenecks and performance benchmarks
- The argument data type of matrix/vector operations
- Matrix structures and storage formats
- Cumbersome package dependency