High Performance Object Oriented Software for Parallel Embedded Systems

The hardware used to perform vector, signal, and image-processing computations is becoming increasing parallel. This parallelism is present both at the low level (in the form of Single Instruction Multiple Data instructions in the microprocessors) and atthe high level (in the form of multiple processors in a single machine.) We suggest making modifications and enhancements to VSIPL++ (a highly abstract C++ library being developed by the DoD) to take advantage of both forms of parallelism. Our objectiveis to demonstrate that VSIPL++ can run as fast as VSIPL, despite its heavy use of high-level abstractions and object-oriented technologies.