Data optimization: allocation of arrays to reduce communication on SIMD machines
Journal of Parallel and Distributed Computing - Massively parallel computation
Structured dataflow analysis for arrays and its use in an optimizing complier
Software—Practice & Experience
Efficiently computing static single assignment form and the control dependence graph
ACM Transactions on Programming Languages and Systems (TOPLAS)
Dependence-based program analysis
PLDI '93 Proceedings of the ACM SIGPLAN 1993 conference on Programming language design and implementation
Global optimizations for parallelism and locality on scalable parallel machines
PLDI '93 Proceedings of the ACM SIGPLAN 1993 conference on Programming language design and implementation
IEEE Transactions on Parallel and Distributed Systems
An Array Data Flow Analysis Based Communication Optimizer
LCPC '97 Proceedings of the 10th International Workshop on Languages and Compilers for Parallel Computing
An Overview of a Compiler for Scalable Parallel Machines
Proceedings of the 6th International Workshop on Languages and Compilers for Parallel Computing
Automatic Support for Data Distribution on Distributed Memory Multiprocessor Systems
Proceedings of the 6th International Workshop on Languages and Compilers for Parallel Computing
Array Distribution in Data-Parallel Programs
LCPC '94 Proceedings of the 7th International Workshop on Languages and Compilers for Parallel Computing
Data Redistribution in an Automatic Data Distribution Tool
LCPC '95 Proceedings of the 8th International Workshop on Languages and Compilers for Parallel Computing
Automatic Selection of Dynamic Data Partitioning Schemes for Distributed-Memory Multicomputers
LCPC '95 Proceedings of the 8th International Workshop on Languages and Compilers for Parallel Computing
Efficient Distribution Analysis via Graph Contraction
LCPC '95 Proceedings of the 8th International Workshop on Languages and Compilers for Parallel Computing
Automatic Data Layout Using 0-1 Integer Programming
PACT '94 Proceedings of the IFIP WG10.3 Working Conference on Parallel Architectures and Compilation Techniques
PACT '97 Proceedings of the 1997 International Conference on Parallel Architectures and Compilation Techniques
| Hi-index | 0.00 |
The use of distributed memory architectures as an effective approach to parallel computing brings with it a more complex program development process. Finding a partitioning of program code and data that supports sufficient parallelism without incurring prohibitive communication costs is a challenging and critical step in the development of programs for distributed memory systems. Automatic data distribution techniques have the goal of placing the responsibility of determining a suitable data partitioning into the domain of the compiler. Static program analysis techniques that expose data interrelationships and derive performance estimates are central to the development of automatic data distribution heuristics. In this paper we present a data partitioning heuristic that makes use of array data flow analysis information in the modeling of data interrelationships and the estimation of costs associated with resolving interrelationships via communication. The global view provided by data flow analysis permits consideration of potential communication optimizations before data partitioning decisions are made. Our heuristic uses tiling techniques to determine data partitionings. The resulting data distributions, while still regular, are not limited to the standard BLOCK, CYCLIC and BLOCK-CYCLIC varieties. Preliminary results indicate an overall reduction in communication cost with our technique.