Precision Tuning and Internet of Things
Dorra Ben Khalifa, Matthieu Martel · 2019
Numerical precision / memory / energy trade-off is a more and more important concern in the design of future computing systems at all scales. In practice, programmers tend to use the highest precision available in hardware (IEEE754 double precision) on most current processors which can be too costly in terms of computing time, memory transfer and energy consumption. To overcome this difficulty, we present a floating-point precision tuning tool called POP: Precision OPtimizer, which integrates a static forward and backward program analysis, done by abstract interpretation, to determine the minimal precision on the inputs and the intermediary results of a program in order to ensure a desired accuracy on the outputs. Our analysis is expressed as a set of linear constraints easily checked by an SMT solver. The main contribution in this article is to experiment POP in a new application domain, namely for Internet of Things. We present the experimental results on numerical computations performed by an accelerometer comparable to what we can find in smartphones to convert brute sensor data into movements.