Pipelined data-parallel algorithms. Part 2; Design

Journal Article · Mon Oct 01 00:00:00 EDT 1990 · IEEE Transactions on Parallel and Distributed Systems; (USA)

DOI:https://doi.org/10.1109/71.80176· OSTI ID:6411035

King, C T ^[1]; Chou, W H; Ni, L M ^[2]

---

1. Dept. of Computer and Information Science, New Jersey Institute of Technology, Newark (US)
2. Michigan State Univ., East Lansing, MI (USA). Dept. of Computer Science

Previously the authors introduced the basic concept of pipelined data parallel algorithms and discussed the modeling of such algorithms on distributed-memory multicomputers. In this paper, a methodology for designing pipelined data-parallel algorithms on multicomputers is studied. The design procedure starts with a sequential algorithm which can be expressed as a nested loop with constant loop-carried dependencies. The main focus of the procedure is on partitioning the loop by grouping related iterations together. Grouping is necessary to balance the communication overhead with the available parallelism and to produce pipelined execution patterns, which result in pipelined data parallel computation. The group should satisfy dependence relationships among the iterations and also allow the granularity to be controlled. Various properties of grouping are studied, and methods for generating communication efficient grouping are given.

• OSTI ID: 6411035
• Journal Information: IEEE Transactions on Parallel and Distributed Systems; (USA), Journal Name: IEEE Transactions on Parallel and Distributed Systems; (USA) Vol. 1:4; ISSN ITDSE; ISSN 1045-9219
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990200* -- Mathematics & Computers
ALGORITHMS
ANALYTICAL SOLUTION
ARRAY PROCESSORS
COMPUTERIZED SIMULATION
DATA PROCESSING
DATA-FLOW PROCESSING
DESIGN
DISTRIBUTED DATA PROCESSING
MATHEMATICAL LOGIC
MEMORY DEVICES
PARALLEL PROCESSING
PERFORMANCE TESTING
PROCESSING
PROGRAMMING
SIMULATION
TESTING