Implementation of CUDA kernels with cppPosit and BFLOAT16

Implementation of CUDA kernels with cppPosit and BFLOAT16

This report details adding CUDA support to the cppPosit library for floating point numbers representation and comparing its performance against BFLOAT16 and IEEE floats. It's been developed in 2022 at the university of Pisa.

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ibrahim, Ahmed Salah Tawfik, Del Turco, Francesco
Format: Report
Sprache:eng
Schlagworte:
bfloat
cppPosit
cuda
floating point
gpu
hardware acceleration
posit
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:This report details adding CUDA support to the cppPosit library for floating point numbers representation and comparing its performance against BFLOAT16 and IEEE floats. It's been developed in 2022 at the university of Pisa.
DOI:10.5281/zenodo.8087492