Fast color image processing using quantized color instruction set

The paper describes the Quantized Color Pack eXtension (QCPX) instruction set that can accelerate color-image processing applications. QCPX employs heterogeneous-subword-parallel instructions, which utilize the microprocessor's 16-bit wide datapaths to process a packed, quantized 16-bit color data type in YCbCr (Y: luminance, Cr and Cb: chrominance) format in parallel. Unlike typical multimedia instruction set extensions (e.g. MMX, SSE, ALTIVEC), QCPX obtains performance and code density improvements through implicit support for color pixel processing rather than depending solely upon generic subword parallelism. Five time-critical color image processing algorithms are coded with and without QCPX to fully measure its impact upon programming techniques. Simulation results indicate that the applications using QCPX achieve significant speedups in execution time over non-QCPX. In addition, QCPX results in higher system utilization (in excess of 94% in all cases) due to significant reduction of conditional instructions.