Chapter 10: Experimental Evaluation of Graphics Algorithms

Abstract

Performance evaluation techniques for fundamental graphics algorithms and for algorithms to be used in multimedia and embedded systems are investigated. Models of computation considering only arithmetic and logic operations taken on input data are regarded as inadequate for processors with instruction-level parallelism. For experimental evaluation of graphics algorithms clock-cycle counting is found more accurate than elapsed-time functions provided by system software. Some difficulties and anomalies with clock-cycle counting on Pentium processors are also reported. An evaluation technique based on best sustained performance is proposed. The paper concludes that it may not even be possible to tell the exact number of clock cycles taken by operations of processors based on asynchronous circuits.