DMCA
Quasi-Static Scheduling for SCORE (2004)
Cached
Download Links
- [brass.cs.berkeley.edu]
- [www.cs.berkeley.edu]
- DBLP
Other Repositories/Bibliography
| Citations: | 1 - 0 self |
BibTeX
@MISC{Markovskiy04quasi-staticscheduling,
author = {Yury Markovskiy},
title = {Quasi-Static Scheduling for SCORE},
year = {2004}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Previous work introduced a dynamic compute model aimed at eliminating existing barriers to the widespread efficient exploitation of reconfigurable devices. Among other achievements this model decoupled application design-time decisions from the run-time physical resource bindings. The compute model uses graphs of compute pages and memory blocks connected by stream links to capture the definition of a computation abstracted from the detailed hardware size. An automatic runtime scheduler is a required component in this compute model in that it selects the temporal sequencing of virtual resources onto the physical device, allocates hardware resources, and configures the device. Although such a scheduler could be computationally expensive, this work describes a quasi-static scheduling strategy that dramatically reduces run-time overhead without restricting the full semantic power of the dynamic dataflow graphs. This work describes the quasi-static scheduling system, analyzes the trade-offs involved in selecting a scheduler implementation, and highlights critical algorithms. It pays particular attention to the
Keyphrases
quasi-static scheduling compute model temporal sequencing physical device reconfigurable device dynamic compute model automatic runtime scheduler run-time physical resource binding application design-time decision previous work virtual resource highlight critical algorithm scheduler implementation memory block detailed hardware size quasi-static scheduling strategy widespread efficient exploitation particular attention quasi-static scheduling system run-time overhead full semantic power dynamic dataflow graph required component hardware resource stream link compute page
