- author = "Wenhao Jia and Kelly A. Shaw and Margaret Martonosi",
- title = "Starchart: Hardware and software optimization using recursive partitioning regression trees",
- booktitle = "Proceedings of the 22nd International Conference on Parallel Architectures and Compilation Techniques",
- year = "2013",
- editor = "Christian Fensch and Michael F. P. O'Boyle and Andre Seznec and Francois Bodin",
- pages = "257--267",
- address = "Edinburgh, United Kingdom",
- month = sep # " 7-11",
- publisher = "IEEE Computer Society",
- keywords = "genetic algorithms, genetic programming, genetic improvement, GPU, SBSE, energy optimisation, auto-tuning, design space exploration, regression tree, decision tree",
- isbn13 = "978-1-4799-1018-2",
- timestamp = "Wed, 16 Oct 2019 14:14:52 +0200",
-
biburl = "
https://dblp.org/rec/conf/IEEEpact/JiaSM13.bib",
- bibsource = "dblp computer science bibliography, https://dblp.org",
-
URL = "
http://jiawenhao.com/starchart.pdf",
-
DOI = "
doi:10.1109/PACT.2013.6618822",
- size = "11 pages",
-
abstract = "Graphics processing units (GPUs) are in increasingly
wide use, but significant hurdles lie in selecting the
appropriate algorithms, runtime parameter settings, and
hardware configurations to achieve power and
performance goals with them. Exploring hardware and
software choices requires time-consuming simulations or
extensive real-system measurements. While some
auto-tuning support has been proposed, it is often
narrow in scope and heuristic in operation.
This paper proposes and evaluates a statistical analysis technique, Starchart, that partitions the GPU hardware/software tuning space by automatically discerning important inflection points in design parameter values. Unlike prior methods, Starchart can identify the best parameter choices within different regions of the space. Our tool is efficient, evaluating at most 0.3 percent of the tuning space, and often much less, and is robust enough to analyze highly variable real-system measurements, not just simulation. In one case study, we use it to automatically find platform-specific parameter settings that are 6.3 fold faster (for AMD) and 1.3 fold faster (for NVIDIA) than a single general setting. We also show how power-optimized parameter settings can save 47 Watts (26 percent of total GPU power) with little performance loss. Overall, Starchart can serve as a foundation for a range of GPU compiler optimisations, auto-tuners, and programmer tools. Furthermore, because Starchart does not rely on specific GPU features, we expect it to be useful for broader CPU/GPU studies as well.",
- notes = "Not GP? Parameter Selection",
