Practical fast asymmetric DCT algorithm and its SIMD and VLIW oriented architecture

Discrete cosine transforms (DCT) have been the mainstay of image signal processing. In this paper a practical, efficient asymmetric DCT (ADCT) is proposed. For 2D-DCT, ADCT implements the row computation differently in concept from column computation. A SIMD and VLIW oriented architecture for ADCT is further proposed to achieve about 20% speed improvement over the fastest known VLIW oriented DCT. Compared with Loeffler's algorithm, the architecture saves more than half the arithmetic calculations. When applied to VLIW, it is more efficient and regular than Feig's algorithm.