The long and winding road to high-performance image processing with MMX/SSE

Recently introduced “SIMD on registers” extensions to commodity microprocessors instruction sets promise (according to the makers) high potential speed-ups for multimedia processing tasks. This paper introduces the complex programming model of MMX/SSE extensions and discusses how the achievement of an effective performance increase over sequential code is no easy task also due to a poor software support. Experiences gained on two real-time image processing projects helped defining a programming methodology and evaluating the tradeoffs to obtain substantial speed-ups with MMX/SSE code