IIR parallelization on multi-datapath SIMD architecture
Yingying Liu, Dake Liu, Wei Wang · 2016
The recursive filter (IIR) parallel programming on SIMD is more difficult than that of nonrecursive algorithms due to data dependency. Several transformation methods for parallel coding of IIR filter on SIMD have already been proposed to deal with data dependency. However, the inherent prologue and epilogue in these methods obviously increase the complexity of control structures and induce extra hardware overhead. Hence, this paper presents a new transformation method for IIR filter parallel coding on multi-datapath SIMD DSP architecture. By this method, data dependency can thus be removed without introducing epilogue and prologue, the extra execution time of prologue and epilogue can be minimized, and the extra control induced silicon cost is eliminated. Therefore, coding for IIR filter on any normal SIMD can be in parallel.