Non-generic floating-point software support for embedded media processing

Abstract

This paper presents some work in progress on the design and implementation of efficient floating-point software support for embedded integer processors. We provide quantitative evidence of the benefits of supporting various non-generic (that is, fused, specialized, or paired) operations in addition to the five basic arithmetic operations: for individual calls, speedups range from 1.12 to 4.86, while on DSP kernels and benchmarks, our approach allows us to be up to 1.59x faster.