Transprecision Computing (Micro-benchmarks)

The micro-benchmarks involve multiple Floating-Point variables whose precision, that is the number of bits used for the mantissa and the exponent, can be tuned using the FlexFloat SW library (see "A Transprecision Floating-Point Platform for Ultra-Low Power Computing", by G. Tagliavini et al., in 2018 Design, Automation & Test in Europe Conference & Exhibition (DATE) 2018 Mar 19 (pp. 1051-1056). IEEE). The data set comprises 7 different micro-benchmarks run under different precision configurations; this data set was created during the 2020 FET project OPRECOMP (g.a. 732631). The data set is fully described in the research paper: "Combining learning and optimization for transprecision computing", by authors Borghesi A, Tagliavini G, Lombardi M, Benini L, Milano M. In Proceedings of the 17th ACM International Conference on Computing Frontiers 2020 May 11 (pp. 10-18). See here the full paper: https://arxiv.org/pdf/2002.10890.pdf