CharmLU is a high-performance dense LU solver library that is highly scalable and follows the HPCC specification. In 2011, it won an HPCC award for best performance as part of a set of benchmarks.
The implementation is designed to be composable with user-level application code.
In addition, the chare array mapping flexibility built into Charm++ helps to minimize coupling with control flow code and allows the user or library developer to explore various mappings of block to processor easily.
Papers / Talks
-
11-512011
PaperMapping Dense LU Factorization on Multicore Supercomputer Nodes
- Jonathan Lifflander
- Phil Miller
- Ramprasad Venkataraman
- Anshu Arya
- Terry Jones
- Laxmikant Vasudeo Kale
-
11-492011
PaperCharm++ for Productivity and Performance: A Submission to the 2011 HPC Class II Challenge
- Laxmikant Vasudeo Kale
- Anshu Arya
- Abhinav Bhatele
- Abhishek Gupta
- Nikhil Jain
- Pritish Jetley
- Jonathan Lifflander
- Phil Miller
- Yanhua Sun
- Ramprasad Venkataraman
- Lukasz Wesolowski
- Gengbin Zheng
-
11-342011
PaperExploring Partial Synchrony in an Asynchronous Environment Using Dense LU
- Jonathan Lifflander
- Phil Miller
- Ramprasad Venkataraman
- Anshu Arya
- Terry Jones
- Laxmikant Vasudeo Kale