| Summary: | 碩士 === 逢甲大學 === 資訊工程所 === 94 === In the process of mapping application to multiprocessors, how to decide multiprocessors architectures, how to modify the software and how to find the corresponding mapping of hardware and software is the core of the whole process. Similar to HW/SW co-design, we usually try different hardware architecture, different software configurations to see if the mapping will better or not. In this paper, we realize a simple flow of mapping application to multiprocessors architecture. We model hardware and software to numbers of processors and data flow and partition the data flow to different grain size. By changing the grain size and number of processors, we get different results after scheduling.
To realize the whole process, we propose a design flow. From the beginning application to final schedule result, we mixed several techniques like Inlining, Loop unrolling, compiler technique, task graph extraction, task graph partition, multiprocessor scheduling etc. All the tasks in the flow are finished by automatic or semi-automatic method, and we use JPEG for example to show how to mapping application to multiprocessor architectures.
|
|---|
