In the real-time embedded computing environment, applications demand multiprocessor hardware solutions to achieve their performance goals. These advanced computing architectures increase the complexity of development exponentially over that which is found in single processor solutions. As a result, developing a multiprocessor application is seldom a simple task. Advanced vectorizing compilers, low-level math libraries coded in assembly, and new, high-level specifications, such as VSIPL, all combine forces to provide performance and portability across hardware platforms. Compilers and assembly coding are geared toward extracting high performance from the hardware.

This paper introduces the Vector, Signal, and Image Processing Library (VSIPL), a new math library specification that addresses the need for portability in the mathematical applications that are executed primarily in embedded systems.