Partitioning a finite difference code for a local memory multiprocessor
The TSAR code, a 3d finite difference EM model, was partitioned into parallel modules where each processor computed a subset of the 3d mesh. The multiprocessor was arranged in a square mesh of processors with the full range of one dimension in each processor and sub ranges in the other dimensions. Finding the range and checking for independence of array elements over those sub ranges is the main problem of partitioning. In addition with local memory multiprocessors, the array elements are distributed over the processors' local memory and a communications structure must be available which allows non-local elements to be assessed. Element communications must be fast enough or few enough to allow good utilization of the processor compute power. Automated partitioning was not available, but recent work provides some hope in this area. TSAR stands for Time-Domain Scattering and Response software that has been used for many years to model electric and magnetic waves in a three dimensional box. This code was partitioned to run on the SPRINT processor. SPRINT is a Systolic Processor with a Reconfigurable Interconnection Network of Transputers. It incorporates 64 floating point transputers. Results of this partitioning resulted in performance nearly equal to a CRAY XMP for identical problems. 4 refs., 3 figs.
- Research Organization:
- Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 6138620
- Report Number(s):
- UCRL-99893; CONF-890372-8; ON: DE89010223
- Resource Relation:
- Conference: 4. hypercube, concurrent computers and applications, Monterey, CA, USA, 6 Mar 1989; Other Information: Portions of this document are illegible in microfiche products
- Country of Publication:
- United States
- Language:
- English
Similar Records
A systolic array for efficient execution of the Faddeev Algorithm
A systolic array for efficient execution of the radon and inverse radon transforms