An overview for the PTRAN analysis system for multiprocessing
Journal of Parallel and Distributed Computing - Special Issue on Languages, Compilers and environments for Parallel Programming
Run-Time Parallelization and Scheduling of Loops
IEEE Transactions on Computers
ICS '94 Proceedings of the 8th international conference on Supercomputing
PLDI '95 Proceedings of the ACM SIGPLAN 1995 conference on Programming language design and implementation
The range test: a dependence test for symbolic, non-linear expressions
Proceedings of the 1994 ACM/IEEE conference on Supercomputing
Parallel Programming with Polaris
Computer
WOMPAT '01 Proceedings of the International Workshop on OpenMP Applications and Tools: OpenMP Shared Memory Parallel Programming
SPEComp: A New Benchmark Suite for Measuring Parallel Computer Performance
WOMPAT '01 Proceedings of the International Workshop on OpenMP Applications and Tools: OpenMP Shared Memory Parallel Programming
Automatic scoping of variables in parallel regions of an OpenMP program
WOMPAT'04 Proceedings of the 5th international conference on OpenMP Applications and Tools: shared Memory Parallel Programming with OpenMP
IWOMP'12 Proceedings of the 8th international conference on OpenMP in a Heterogeneous World
Hi-index | 0.00 |
In [1], Dieter an Mey proposes the addition of an AUTO attribute to the OpenMP DEFAULT clause. A DEFAULT(AUTO) clause would cause the compiler to automatically determine the scoping of variables that are not explicitly classified as shared, private or reduction. While this new feature would be useful and powerful, its implementation would rely on automatic parallelization technology, which has been shown to have significant limitations. In this paper, we implement support for the DEFAULT(AUTO) clause in the Polaris parallelizing compiler. Our modified version of the compiler will translate regions with DEFAULT(AUTO) clauses into regions that have explicit scopings for all variables. An evaluation of our implementation on a subset of the SPEC OpenMP Benchmark Suite shows that with current automatic parallelization technologies, a number of important regions cannot be statically scoped, resulting in a significant loss of speedup. We also compare our compiler's performance to that of an Early Access version of the Sun Studio 9 Fortran 95 compiler [2].