Numerical reproducibility for the parallel reduction on multi- and many-core architectures

•A parallel algorithm to compute correctly-rounded floating-point sums•Highly-optimized implementations for modern CPUs, GPUs and Xeon Phi•As fast as memory bandwidth allows for large sums with moderate dynamic range•Scales well with the problem size and resources used on a cluster of compute nodes...

Ausführliche Beschreibung

Bibliographische Detailangaben

Veröffentlicht in:	Parallel computing 2015-11, Vol.49, p.83-97
Hauptverfasser:	Collange, Caroline, Defour, David, Graillat, Stef, Iakymchuk, Roman
Format:	Artikel
Sprache:	eng
Schlagworte:	Accuracy Computer Arithmetic Computer Science Error-free transformations Hardware Architecture Long accumulator Multi- and many-core architectures Parallel floating-point summation Reproducibility
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!