Abstract:
In this presentation, we address the challenging problem of deploying complex astronomical applications on efficient target architectures. The allocation of resources in such scenarios significantly impacts performance outcomes. To automate and expedite this process, we propose a method that leverages the Synchronous Dataflow (SDF) model and the Scaling up of Cluster of Actor on the Processing Element (SCAPE) technique. Our approach, implemented in the PREESM rapid prototyping tool, facilitates automated deployment on CPU, GPU, and FPGA architectures. We showcase the advanced implementation of CPU-based resource allocation in our tool, which serves as a foundation for other target architectures. By providing a fast and easy resolution to this NP-hard problem, our method greatly enhances the efficiency of deploying astronomical applications.
|