Automatic Tools for Developing Fine-Grained Signal Processing Programs onMulticomputers

摘要

During the course of this contract we developed the first system that integratestask and data parallelism in a uniform compiler framework. The compiler which is called Fx, translates a dialect of High Performance Fortran into parallel code that runs on distributed memory computer systems such as the Intel Paragon, the Lintel iWarp, the IBM SP/2, and networks of workstations. We demonstrated the effectiveness of our technIque on a wide variety of applications, includIng spotlight synthetic aperture radar (SAR) , multidimensional fast Fourier transform (FFT), narrowband tracking, radar, air quality modeling, earthquake ground motion modeling, and multibase stereo vision. The system is in daily use at Carnegie Mellon. The Carnegie Mellon vision group uses the Fx compiler to develop their codes for the Intel Paragon. The Mechanical Engineering air quality modeling group at Carnegie Mellon is using Fx to develop their airshed model on the Paragon. A Seismologist at the University of Southern California used the Fx system to develop an application based on the method of boundary elements for predicting ground motion during strong earthquakes.