Floating-Point Computation with Just Enough Accuracy

摘要: Most mathematical formulae are defined in terms of operations on real numbers, but computers can only operate numeric values with finite precision and range. Using floating-point as numbers does not clearly identify the which each value must be represented. Too little yields inaccurate results; too much wastes computational resources. The popularity multimedia applications has made fast hardware support for low-precision arithmetic common Digital Signal Processors (DSPs), SIMD Within A Register (SWAR) instruction set extensions general purpose processors, Graphics Processing Units (GPUs). In this paper, we describe a simple approach by speed these speculatively employed to meet user-specified accuracy constraints. Where native precision(s) yield insufficient accuracy, technique is used efficiently synthesize enhanced using pairs values.