2011 | OriginalPaper | Buchkapitel
A Memory Access Pattern Based Data Distribution Technique for Array Processors
verfasst von : Doosan Cho
Erschienen in: Convergence and Hybrid Information Technology
Verlag: Springer Berlin Heidelberg
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
The successful use of array-type multiprocessors strongly depends on how to efficiently distribute data and code in target applications. However, producing efficient code for such multiprocessors is a difficult task. Hand-coding efficient parallel programs for these processors can be extremely difficult, time consuming and error-prone, so developers have turned to automatic compiler techniques to ease the task.
To efficiently execute an multimedia application on array processors, a good data partitioning is required. Default partitioning such as row-wise or column-wise may not produce a good data distribution and, thus, will prolong the total execution time of an application.
The two main approaches to this are using compilers that 1) generate data transfer code, and/or 2) generate code for data distribution. Neither has been completely successful for all types of programs.
In this paper, we propose efficient methods for finding the data distribution for array processors while minimizing the total execution time. The experimental results indicate that our noble technique would be able to support many user programs efficiently.