Software Support for Non-Numerical Computing on Multi-Core Chips

Multi-core chips present a new computing environment that can benefit from software support for non-numerical applications. Heterogeneous cores will allow efficient sophisticated multi-level parallel processing. Techniques are described that enable the association of the elements of related heterogeneous SIMD vectors. These techniques can be used for processing arrays of records, dynamic allocation of a core’s memory and for imposing a multitasking layer of parallelism over a data parallel layer. Some of the operating system and hardware modifications needed to support these techniques are discussed. A version of the Smith Waterman algorithm for DNA sequence comparison was investigated briefly to study some of the advantages of an associative bit-serial core.