1 July 1990 Massively parallel processors in real-time applications
Author Affiliations +
In previous work, we have discussed the issue of real-time control, which is important for the effective application of massively parallel architectures in real-time, or near real-time, signal processing. We first briefly review our work on how to incorporate adaptive programming decisions in systems accommodating massive data parallelism without losing machine efficiency. This involves re-examination of functions to be performed by the software versus those performed by the hardware. Our discussion is based on a Multiple Instruction Multiple Data (MIMD) controller that we have constructed for a Geometric Single Instruction Single Data (GSIMD) array. We then address two more issues facing a GSIMD architecture for its efficient use, especially in vision applications. These are: how to treat the overlap in small neighborhood operations when a large image array is segmented for processing on a smaller GSIMD array, and how to transmit small amounts of information between the array processor and external machines for adaptive signal processing. We describe the architecture of our hardware solutions.
© (1990) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Ken K. Jung, Ken K. Jung, H. T. Nguyen, H. T. Nguyen, Raghu Raghavan, Raghu Raghavan, Hoa D. Truong, Hoa D. Truong, } "Massively parallel processors in real-time applications", Proc. SPIE 1246, Parallel Architectures for Image Processing, (1 July 1990); doi: 10.1117/12.19573; https://doi.org/10.1117/12.19573

Back to Top