Title/Author/Abstract
- Title:
Compiler and Language Issues for the Blue Gene Massively Parallel
Systems
- Author:
Jose E. Moreira, Sam P. Midkiff and Manish Gupta
IBM Thomas J. Watson Research Center
- Full Paper(.pdf version)
- Abstract:
Blue Gene is a massively parallel system currently under
development at the IBM Thomas J. Watson Research Center. Blue Gene will
achieve a hundred-fold improvement in performance, when compared to the
fastest supercomputers of today, through the exploitation of unprecedented
levels of parallelism. More than 10 million concurrent threads of execution
are supported by a Petaflop-scale Blue Gene machine. Effective exploitation
of the parallelism and computational power offered by Blue Gene represents
a new challenge in Computer Science. Significant compiler work exists in
optimizing conventional programming languages (Fortran and C) for execution
on Blue Gene. At the same time, there is a strong motivation to pursue new
programming models and new programming languages for Blue Gene, if they can
facilitate the job of application programmers in achieving high
performance. In this article, we describe our explorations in the area of
languages and compilers for Blue Gene. We start by identifying the main
optimization issues to be addressed when compiling multithreaded code for
Blue Gene. We then discuss some language constructs that extend C and
Fortran to facilitate the job of expressing application parallelism in a
manner suitable for execution on Blue Gene. Finally, we discuss a path for
exploring higher-level and domain specific programming languages for Blue
Gene.
Please contact our
webadmin with any comments or changes.
Unless explicitly stated otherwise, all material is
copyright © The University of Edinburgh.