Segmented Arithmetic Operators for Graphics Processing

Graphics processing relies on executing similar instructions repeatedly on a large data set. This parallelism in the data gives rise to the Single-Instruction Multiple-Data (SIMD) paradigm which is used in modern processors. This paper explores several techniques that exploit the parallelism in the SIMD execution functional units and proposes several new SIMD methods. The methods discussed in this paper cover SIMD Addition and SIMD Matrix Multiplication.

PDF

Leave a Reply