Determining Optimal Grain Size for Efficient Vector Processing on SIMD Image Processing Architectures

摘要

Adaptable silicon area usage within an integrated pixel processing array is a key issue for embedded single instruction, multiple data (SIMD) image processing architectures due to limited chip resources and varying application requirements. In this regard, this paper explores the effects of varying the number of vector (multichannel) pixels mapped to each processing element (VPPE) within a SIMD architecture. The VPPE ratio has a significant impact on the overall area and energy efficiency of the computational array. Moreover, this paper evaluates the impact of our color-aware instruction set (CAX) on each VPPE configuration to identify ideal grain size for a given SIMD system extended with CAX. CAX supports parallel operations on two-packed 16-bit (6:5:5) YCbCr (luminance-chrominance) data in a 32-bit datapath processor, providing greater concurrency and efficiency for vector processing of color image sequences. Experimental results for 3-D vector quantization indicate that high processing performance with the lowest cost is achieved at VPPE = 16 with CAX.