Skip to Main content Skip to Navigation
Theses

Design Space Exploration for data-dominated image applications with non-affine array references

Abstract : In this dissertation we present a method able to run a Design Space Exploration oriented to the optimization of the data transfer and storage management. The corresponding developed tool has been used as a front-end of HLS in order to help the user to find an optimized memory micro- architecture. Our method is able to handle image processing applications with non- affine array refer- ences. It is able to apply a paving which, on one hand, is based on a run-time dependence analysis and, on the other hand, uses disjoint and equal-by- translation blocks to parti- tion the data and instruction sets. The non- affinity of the array references is taken into account by projecting the instruction paving on the data paving. This method leads to a memory micro-architecture that is, at the same time, adapted to the non- affinity of the array references of the application and has a cheap control on the data transfer because of the invariability of the size of transferred data blocks.
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00456577
Contributor : Rosilde Corvino <>
Submitted on : Wednesday, February 24, 2010 - 5:56:16 PM
Last modification on : Thursday, November 19, 2020 - 12:59:40 PM
Long-term archiving on: : Thursday, October 18, 2012 - 3:50:11 PM

Identifiers

  • HAL Id : tel-00456577, version 1

Collections

Citation

Rosilde Corvino. Design Space Exploration for data-dominated image applications with non-affine array references. Computer Science [cs]. Université Joseph-Fourier - Grenoble I, 2009. English. ⟨tel-00456577⟩

Share

Metrics

Record views

628

Files downloads

385