direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Kernel Analysis for High Performance Computing



By 2020, we will have computing systems capable of processing one exaFLOPS. However, current software systems are not yet capable of efficiently exploiting such computational power. Runtime systems are recognized as a key software component to improve the software productivity and performance of such systems. A recent, novel approach to improve the performance of run-time systems is to enhance runtime efficiency with compiler-based static analysis of the computational kernel. Related work shows that such information can be of great help in order to improve the scalability of HPC large-scale compute clusters.
The student should implement an LLVM compiler pass able to extract such information from an OpenCL kernel, and its intermediate representation (SPIR). The resulting compiler pass will be integrated with an existing runtime system and tested on large-scale HPC infrastructures (e.g., the Leibniz Supercomputing Centre).

Keywords: LLVM, SPIR, HPC, compilers, analysis, runtime systems

Required Skills


Contact Persons


  1. Ivan Grasso, Simone Pellegrini, Biagio Cosenza, Thomas Fahringer, "A uniform approach for programming distributed heterogeneous computing systems," Journal of Parallel and Distributed Computing, 2014
  2. Ivan Grasso, Simone Pellegrini, Biagio Cosenza, Thomas Fahringer, "libWater: Heterogeneous Distributed Computing Made Easy," ACM International Conference on Supercomputing 2013
  3. Klaus Kofler, Ivan Grasso, Biagio Cosenza, Thomas Fahringer, "An Automatic Input-Sensitive Approach for Heterogeneous Task Partitioning," ACM International Conference on Supercomputing 2013



Zusatzinformationen / Extras


Schnellnavigation zur Seite über Nummerneingabe