GLU for Multidimensional Signal Processing

Multidimensional signal processing applications have a great deal of inherent parallelism. Two examples of these applications---MPEG encoding and computed tomography---can be implemented in parallel. With the described methodology these applications were converted into a form usable in GLU and implemented on a network of workstations. Improvements to the GLU programming environment can benefit other inherently parallel applications. 1. Introduction Multidimensional signal processing applications have a great deal of inherent parallelism. Many spatial signal processing applications are also noncausal and shiftinvariant. These applications use various mathematical transformations performed on well-defined data and have well-structured local and global communication. These properties of multidimensional signal processing applications allow most of these transformations to be calculated concurrently. We describe the parallel implementation of two multidimensional signal processing applic..