Tailoring Pipeline Bypassing and Functional Unit Mapping to Application in Clustered VLIW Architectures (2001)
| Venue: | In Proc. of CASES |
| Citations: | 7 - 0 self |
BibTeX
@INPROCEEDINGS{Buss01tailoringpipeline,
author = {Marcio Buss and Rodolfo Azevedo and Paulo Centoducatte and Guido Araujo},
title = {Tailoring Pipeline Bypassing and Functional Unit Mapping to Application in Clustered VLIW Architectures},
booktitle = {In Proc. of CASES},
year = {2001},
pages = {141--148}
}
Bookmark
 |
 |
 |
 |
 |
 |
OpenURL
Abstract
paper we des)(3 e ades6N exploration methodology forclus(3$W VLI architectures The central idea of this workis as et of three techniques aimed at reducing the cos of expens( e inter-clus2) copy operations InsR6)WA2) s heduling is performed usR$ alis4)R heduling algorithm thatsatW4 operand chains into thesWR regisN] file. Functionalunits areasWRR$N toclus$(4 bas on the application inter-clusN) communication pattern. Finally, a careful insefulW of pipeline bypas43 is usR to increas the number of data-dependencies that can bes(]436W by pipeline regisW( operands Experimental resN]4N us(( the SPEC95 benchmark and the IMPACT compiler, reveal as ubs$] tial reduction in the number of copies between 1.
