Programming considerations in the design and use of a SIMD image computer

It is demonstrated that with a modicum of compilation complexity a scan line array processor (SLAP) (a single-instruction multiple-data (SIMD) architecture designed for image computation and similar applications) becomes a versatile tool, efficiently supporting a number of useful programming models: position-independent (low-level image-processing operations), scan-line (intermediate-level image processing and graphics), and systolic. The impact of programming issues on overall system architecture is discussed with respect to hierarchical control structures, the handling of concurrent I/O streams, and the importance of considering whole applications. Such considerations are critical to the success of highly parallel systems, particularly those designed for imbedded applications, yet often are treated as afterthoughts. The high-level SLAP programming language Slang is introduced