Runtime Coordinated Heterogeneous Tasks in Charm++

Effective utilization of the increasingly heterogeneous hardware in modern supercomputers is a significant challenge. Many applications have seen performance gains by using GPUs, but many implementations leave CPUs sitting idle. In this paper, we describe a runtime managed system for coordinating heterogeneous execution. This system manages data transfers to and from GPU devices and schedules work across the computational resources of the system. The programmer need only tag methods and parameters to enable heterogeneous execution. Using this system, we observe improvements in programmer productivity and application performance. For selected benchmarks, when using heterogeneous execution we observe speedups of up to 3.09x relative to using only the host cores or only the device.

@inproceedings{Robson:2016:RCH:3018814.3018821,
 author = {Robson, Michael P. and Buch, Ronak and Kale, Laxmikant V.},
 title = {Runtime Coordinated Heterogeneous Tasks in Charm++},
 booktitle = {Proceedings of the Second Internationsl Workshop on Extreme Scale Programming Models and Middleware},
 series = {ESPM2},
 year = {2016},
 isbn = {978-1-5090-3858-9},
 location = {Salt Lake City, Utah},
 pages = {40--43},
 numpages = {4},
 url = {https://doi.org/10.1109/ESPM2.2016.7},
 doi = {10.1109/ESPM2.2016.7},
 acmid = {3018821},
 publisher = {IEEE Press},
 address = {Piscataway, NJ, USA},
 keywords = {accelerator architectures, high performance computing, parallel programming, runtime},
}