Short Vector SIMD Code Generation for DSP Algorithms

Abstract Short vector SIMD instructions on recent general purpose microprocessors, such as SSE on Pentium III and 4, offer a high potential speed-up but require a very high level of programming expertise. We present a compiler that generates vectorized code for digital signal processing algorithms such as the fast Fourier transform (FFT). The input to our compiler is a mathematical description of the algorithm in the language SPL. SPL is used in the SPIRAL system to automatically generate and optimize code for a given computing,platform. The output of our compiler is a C function enhanced,with vector instructions. Interfacing our compiler with SPIRAL yields speed-ups of more than a factor of 2 in several important cases including the FFT and the DCT used in the JPEG compression,standard. For the FFT, our automatically generated code is competitive with the hand-coded Intel Math Kernel Library. Short Vector SIMD Instructions