Generative communication in Linda
ACM Transactions on Programming Languages and Systems (TOPLAS)
ICS '88 Proceedings of the 2nd international conference on Supercomputing
I-structures: data structures for parallel computing
ACM Transactions on Programming Languages and Systems (TOPLAS)
Statistically rigorous java performance evaluation
Proceedings of the 22nd annual ACM SIGPLAN conference on Object-oriented programming systems and applications
Engineering A Compiler
Revisiting Out-of-SSA Translation for Correctness, Code Quality and Efficiency
Proceedings of the 7th annual IEEE/ACM International Symposium on Code Generation and Optimization
Scientific Programming - Exploring Languages for Expressing Medium to Massive On-Chip Parallelism
LVars: lattice-based data structures for deterministic parallelism
Proceedings of the 2nd ACM SIGPLAN workshop on Functional high-performance computing
| Hi-index | 0.00 |
The dynamic-single-assignment property for shared data accesses can establish data race freedom and determinism in parallel programs. However, memory management is a well known challenge in making dynamic-single-assignment practical, especially when objects can be accessed through tags that can be computed by any step. In this paper, we propose a new memory management approach based on user-specified folding functions that map logical dynamic-single -assignment (DSA) tags into dynamic-multiple-assignment (DMA) tags. We also compare folding with get-counts, an approach in which the user specifies a reference count for each single-assignment value. The context for our work is parallel programming models in which shared data accesses are coordinated by put/get operations on tagged DSA data structures. These models include dataflow programs with I-structures, functional subsets of parallel programs based on tuple spaces (notably, Linda), and programs written in the Concurrent Collections (CnC) coordination language. Our conclusion, based on experimental evaluation of five CnC programs, is that folding and get-counts can offer significant memory efficiency improvements, and that folding can handle cases that the get-counts cannot.