Sparse tensor ordering
1 : ROMA
(ENS Lyon / CNRS / Inria Grenoble Rhône-Alpes)
-
Website
CNRS : UMR5668, Laboratoire d'informatique du Parallélisme, École Normale Supérieure (ENS) - Lyon, INRIA
Laboratoire de l'Informatique du Parallélisme 46 Allée d'Italie 69364 Lyon -
France
Spare tensors are commonly stored in mode oblivious formats, such as COO and HiCOO. We propose ordering methods for sparse tensors so that operations on their COO and HiCOO representation have improved cache use and hence better performance.
This is joint work with Jiajia Li, Jimeng Sun, Kevin Barker, Rich Vuduc and Umit Catalyurek.
| Subject : | : | oral |
| Topics | : | Numerical Algorithm |
| PDF version | : |  PDF version |
