Abstract
In this paper, we develop a static global scheduling scheme for mapping computer vision and image processing (CVIP) operations on distributed-memory multiprocessors. Unlike most current parallel image processing research which focuses on parallelizing individual processing algorithms on a particular parallel architecture, our scheduler is for optimizing processor assignment and data partition for an entire image processing pipeline. The scheduler operates on task graphs specified by conventional visual languages such as Khoros and Explorer. A task graph is assumed to be a linear chain of operations with any number of nested loops. The task chain is first decomposed into simpler subchains; each a linear sequence of tasks without loops. The communication and computation costs of the component tasks in the subchains are determined by a taxonomy of CVIP operations. Data redistribution overheads in between tasks can also be tabulated in advance for many popular data partitioning schemes. The scheduler then employs a shortest path algorithm to optimize the parallel time, taking into consideration possible variation in the task and resource parameters (such as the image size and number of processors used), and both the intra-operation and the inter-operation computation and communication times. In this paper, we present the scheduling scheme, and provide analyses and experimental results to verify our approach.
Submitted Version
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
