Communication-model based Embedded Mapping of Dataflow Actors on Heterogeneous MPSoC

International audienceMapping a dataflow application onto a heterogeneous multiprocessor platform cannot longer be static. Ithas to adapt dynamically depending on the data and on thecommunication between the computation cores. This is typicallythe case for mobile devices that run multimedia applications.This paper presents an algorithm fast enough to be executed atrun-time. In addition to computation cost, our approach relieson a communication model to estimate the delay for transmittingdata. The algorithm is compared with METIS tool for randomdataflow graphs and two video decoders, MPEG4-SP and HEVC,considering heterogeneous multiprocessor platforms composed of4 to 8 processors and 6 accelerators. Results on a Zynq platformshow that our algorithm is about 40x faster than METIS tool forthe same throughput (frames per second) on a platform with 8processors and 6 accelerators