Skip to Main content Skip to Navigation
Theses

Composition de squelettes algorithmiques : application au prototypage rapide d'applications de vision

Abstract : Our work deals with parallel programming environments which are developed in order to help the user of ''Multiple Instruction stream Multiple Data stream - Distributed Memory'' (MIMD-DM) computers to easily program such a machine. Image processing is our target application field.
The SKiPPER (SKeleton-based Parallel Programming EnviRonment) toolset was developed for this purpose. It is designed to address the problem of rapid prototyping of image processing applications, and to get a fully operational application in a very short time with a few stages within the design process. The methodology used in SKiPPER is based on algorithmic skeletons. Algorithmic skeletons are higher-order program constructs encapsulating common and recurrent forms of parallelism to make them readily available for the application programmer.
More precisely our work concerns the composition problem of algorithmic skeletons.
Our goal is to increase the number of image processing applications which can be written with this methodology and therefore the SKiPPER tool. The version of SKiPPER described here was specifically designed to support arbitrary nesting of skeletons, which is the hardest form of composition we have to deal with.
This version relies on a fully dynamic execution model whereas the previous versions relied on mostly static execution models. The main idea was to design a generic new skeleton which is able to dynamically emulate any skeleton of the SKiPPER's library. In this way the new version of the SKiPPER's kernel is able to manage the skeleton composition ''on the fly''.
In order to evaluate our approach, some benchmark programs with a real case study of a face-tracking algorithm are given. The interest of this example is to cast a light on the need for skeleton nesting in realistic image processing applications.
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00003350
Contributor : Rémi Coudarcher <>
Submitted on : Thursday, September 11, 2003 - 2:58:47 PM
Last modification on : Friday, October 23, 2020 - 4:52:18 PM
Long-term archiving on: : Friday, April 2, 2010 - 7:27:47 PM

Identifiers

  • HAL Id : tel-00003350, version 1

Citation

Rémi Coudarcher. Composition de squelettes algorithmiques : application au prototypage rapide d'applications de vision. Interface homme-machine [cs.HC]. Université Blaise Pascal - Clermont-Ferrand II, 2002. Français. ⟨tel-00003350⟩

Share

Metrics

Record views

330

Files downloads

447