HW-SW optimisation of JPEG2000 wavelet transform for dedicated multimedia processor architectures

The discrete wavelet transform is essential for several image compression applications such as Jpeg2000 encoding. In this area, time-effective execution represents a hot-spot for software realisations targeting microprocessor-based System-on-Chip platforms. The discrete wavelet transform software was redesigned to enhance its performance over a parallel single-instruction-multiple-data (SIMD) architecture built onto very-long-instructions-word processing elements. All major aspects affecting the processing speed were focused on: the penalties deriving from data-cache stalls have been massively eased whereas optimal exploitation of the platform vector processing capabilities has been pursued by instruction-level optimisations. Overall, up to 72% savings was attained in the runtime with respect to a conventional implementation