This SBIR Phase I project will develop a compiler for a portable vector assembly language. This tool will allow portable low-level programs to be written for signal-processing applications, and will produce efficient code for modern general-purpose processors (GPP), taking full advantage of their vector (SIMD) instruction sets such as SSE on the Intel Pentium and Altivec on the Motorola PowerPC. This tool will reduce the software development time and cost, and reduce the hardware parts cost, for a variety of BMDO and commercial signal processing systems. GPPs with SIMD offer superior price/performance and MIPS/Watt/Volume compared to any other COTS processors, but currently, programs that exploit SIMD instructions must be explicitly targeted to a single processor and are not portable. By restricting the problem domain to signal-processing algorithms, and by restricting the portable source language carefully, we expect the compiler developed in this project to output code competitive with hand-written SIMD assembly code. The large software engineering improvements that come from eliminating hand-written assembly code, and the reduction in hardware costs due to running efficient code on GPPs, give this compiler significant commercial potential for use by developers of radio communications systems, medical devices, audio systems, and other signal processing applications. Anticipated Benefits/Commercial Applications: BMDO applications that could potentially benefit from this compiler include signal processing of pulsed doppler radar signals for discrimination, acquisition and tracking; pattern recognition; and target identification. Commercial applications include reducing the time-to-market and reducing the parts cost of a variety of signal processing applications, including radio communications, medical devices, and audio systems.
Keywords: General Purpose Processor, SIMD Vector Instruction Sets, Compiler
