Title :
Program partitioning for multithreaded dataflow computers
Author :
Lee, Ben ; Kavi, Krishna
Author_Institution :
Dept. of Electr. & Comput. Eng., Oregon State Univ., Corvallis, OR, USA
Abstract :
A compile-time partitioning strategy for multithreaded dataflow computers is presented. The proposed scheme is based on the minimum granularity required to maximally utilize a given processor pipeline. To determine such a partition, architectural and program characteristics, such as the frequency and the delays of split-phase loads and their effect on the parallelism required, are considered. Once these parameters are known, a partition can be chosen that best matches the capability of a given multithreaded dataflow multiprocessor
Keywords :
delays; parallel processing; program compilers; architectural characteristics; delays; minimum granularity; multithreaded dataflow computers; processor pipeline; program characteristics; program partitioning; split-phase loads; Computational modeling; Computer architecture; Concurrent computing; Delay effects; Frequency synchronization; Multiprocessing systems; Parallel processing; Pipelines; Proposals; Switches;
Conference_Titel :
System Sciences, 1993, Proceeding of the Twenty-Sixth Hawaii International Conference on
Conference_Location :
Wailea, HI
Print_ISBN :
0-8186-3230-5
DOI :
10.1109/HICSS.1993.284077
