Abstract
|
|
---|---|
Modern automatic analytical methods for studying range and accuracy in ?xed-point systems are gradually replacing the traditional bit-true ?xed-point simulations used inWord-Length Optimization (WLO) problems. But these models have several limitations that must be overcome if they are going to be used in real world applications. When targeting large systems, the mathematical expressions quickly become too large to be handled in reasonable times by numerical engines. This paper proposes adapting the classical Fiduccia-Mattheyses partitioning algorithm to the WLO domain to automatically generate hierarchical partitions of the systems to quantize. This is the ?rst time this type of algorithms are used for this purpose. The algorithm has been successfully applied to large problems that could not be addressed before. It generates, in the order of minutes, maneuverable sub-problems where state-of-the-art models can be applied. Thus, scalability is achieved and the impact of the problem size as a constraint is minimized. | |
International
|
Si |
Congress
|
10th International Symposyum on Applied Reconfigurable Computing, ARC'14 |
|
960 |
Place
|
Vilamoura (Portugal) |
Reviewers
|
Si |
ISBN/ISSN
|
978-3-319-05959-4 |
|
10.1007/978-3-319-05960-0_12 |
Start Date
|
14/04/2014 |
End Date
|
16/04/2014 |
From page
|
133 |
To page
|
143 |
|
Reconfigurable Computing: Architectures, Tools, and Applications. Proceedings of the 10th International Symposyum on Applied Reconfigurable Computing, ARC'14. Lecture Notes in Computer Science Volume 8405, 2014 |