Runtime coupling of data-parallel programs
ICS '96 Proceedings of the 10th international conference on Supercomputing
Building programs in the network of tasks model
SAC '00 Proceedings of the 2000 ACM symposium on Applied computing - Volume 1
Detection of Implicit Parallelisms in the Task Parallel Language
HPC-ASIA '97 Proceedings of the High-Performance Computing on the Information Superhighway, HPC-Asia '97
Opus: A Coordination Language for Multidisciplinary Applications
Scientific Programming
| Hi-index | 0.03 |
Many important applications have a heterogeneous structure, and can be implemented efficiently only with task parallelism. This paper presents a set of extensions to Fortran to build task parallel programs for multicomputers consisting of distinct nodes: either private memory parallel machines, or autonomous computers connected by a high speed network. The design of these extensions is driven by the following objectives: The compiler should handle inter-node communication, not the programmer. The tasking extensions should be integrated into an existing parallel programming language, so that existing programs can benefit from task parallelism, existing libraries can be used, and the users do not have to learn a new language. Current compiler technology should be able to produce efficient parallel programs that are competitive with hand parallelized codes. Our implementation of task parallelism is integrated with an HPF like data parallel Fortran compiler (Fx) developed at Carnegie Mellon University. The design, implementation, and experimental results from Fx are presented in several related publications. In this paper, we describe the programming model and the set of Fortran extensions for task parallelism and discuss the basis for the design decisions.