Thur 1/15/09: A Cross-Input Adaptive Framework for Optimizing GPUPrograms - JLab Technical Computing Seminar

[Sandy Philpott <sandy.philpott@jlab.org>]

See you tomorrow!

JLab Technical Computing Seminar
Thursday, January 15, 2009, 10:30am
CEBAF Center L104

"A Cross-Input Adaptive Framework for Optimizing GPU Programs"

Xipeng Shen
Assistant Professor, The College of William and Mary
http://www.cs.wm.edu/~xshen/

   Recent years have seen a trend in using graphic processing units
(GPU) as accelerators for general-purpose computing.  The inexpensive,
single-chip, massively parallel architecture of GPU has evidentially
brought factors of speedup to many numerical applications. However,
the development of a high-quality GPU application is challenging, due
to the large optimization space and complex unpredictable effects of
optimizations on GPU program performance.

   Recently, several studies have attempted to use empirical search to
help the optimization. Although those studies have shown promising
results, one important factor---program inputs---in the optimization
has remained unexplored. In this work, we initiate the exploration in
this new dimension. By conducting a series of measurement, we find
that the ability to adapt to program inputs is important for some
applications to achieve their best performance on GPU. In light of the
findings, we develop an input-adaptive optimization framework, namely
G-ADAPT, to address the influence by constructing cross-input
predictive models for automatically predicting the (near-)optimal
configurations for an arbitrary input to a GPU program. Experimental
results demonstrate the promise of the framework in serving as a tool
for alleviating the productivity bottleneck in GPU programming.


Xipeng Shen has been an assistant professor at The College of William
and Mary since 2006. He received the Ph.D. and Master degree in
Computer Science from University of Rochester in 2006 and 2003
respectively. He received the M.S. degree in Pattern Recognition from
Chinese Academy of Sciences in 2001, and the B.S. degree from The
North China University of Technology.

Xipeng Shen's main research lies in the area of Compiler Technology and
Programming Systems, covering Optimizing Compilers, Parallel Computing,
GPU Computing, Program Behavior Analysis, etc.  He leads the Compilers
and Adaptive Programming Systems research group at The College of
William and Mary. The group have been focusing on integrating automatic
learning, adaptation, and evolvement into different computing layers to
form a whole-system synergy.