AFFORDe: Automatic Allocation and Floorplanning for SPMD Architecture - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2016

AFFORDe: Automatic Allocation and Floorplanning for SPMD Architecture

Résumé

Nowadays, Field Programmable Gate Arrays (FPGAs) platforms offer a high density to allow designing Multi Processor-based System on Chip. SPMD (Single Program Multiple Data) is a massively parallel execution model based on the assembly of a given number of homogeneous Processing Elements (PEs). This model is often relaying on Master/Slaves architecture composed by a Master PE that manages the parallel execution of a set of identical slave PEs. Furthermore, Dynamic Partial Reconfiguration (DPR) feature allows such computing system to be reconfigured on the fly for a given application requirement. Given the growing number of PEs in Master/Slaves architecture, it is difficult to estimate the time of specification and design during the phase of allocation and floorplanning of Partial Reconfigurable Regions (PRRs) because it is still performed manually. In this work, we present AFFORDe, a tool enable to automate the Xilinx DPR flow for SPMD architecture that allows parsing the resource requirements of the static and the dynamically reconfigurable parts to perform an automatic floorplanning. The floorplanning is based on a Heuristic Algorithm for Automatic Floorplanning in SPMD Architectures (HAAFSA). This tool is used to generate a configuration file that allows floorplanning of reconfigurable regions in an automatic way of a given Master/Slaves configuration. Experimental results show the effectiveness of our tool to increase the design productivity for dynamically reconfigurable SPMD-based architecture.
Fichier non déposé

Dates et versions

hal-01757158 , version 1 (03-04-2018)

Identifiants

Citer

Wissem Chouchene, Rabie Ben Atitallah, Jean-Luc Dekeyser. AFFORDe: Automatic Allocation and Floorplanning for SPMD Architecture. MCSOC 2016 - 10th International Symposium on Embedded Multicore/Many-core Systems-on-Chip, Sep 2016, Lyon, France. ⟨10.1109/MCSoC.2016.38⟩. ⟨hal-01757158⟩
129 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More