# Automatic code generation for manycore embedded platform _by Gageot Dylan (IETR-VAADER) - 2019.08.22_ ###### tags: `VAADER` `Seminar` ![](https://i.imgur.com/sBmPpkW.png) ## Abstract When developing applications for manycore architecture, it’s easier to consider a cluster of cores rather than individual core. This abstraction layer can be implemented by using an API such as OpenMP. In this context, applications described with Synchronous Dataflow (SDF) should be synthetize more rapidly by generating appropriate OpenMP code for clusters instead of scheduling actors on individual core. A proof of concept was already running under PREESM : graph’s actors were regrouped into an atomic one and caused code generation to print OpenMP’s directive over nested loop. The goal of my internship was to enhance the clustering mechanism by adding different algorithm to control memory space and parallelism. During this presentation, the concept of clustering will be introduced, followed by the result of different clustering algorithms over memory and parallelism. ## [Slides](https://mycore.core-cloud.net/index.php/s/mCeX17TqLPzofv1)