This title appears in the Scientific Report : 2014 

Binary Instrumentation for Scalable Performance Measurement of OpenMP Applications
Jaeger, Julien (Corresponding Author)
Philippen, Peter / Petit, Eric / Rubial, Andres Charif / Rössel, Christian / Jalby, William / Mohr, Bernd
Jülich Supercomputing Center; JSC
Parallel Computing: Accelerating Computational Science and Engineering (CSE)
IOS Press 2014
783 - 792
International Conference on Parallel Computing, Munich (Germany), 2013-09-10 - 2013-09-13
Contribution to a book
Contribution to a conference proceedings
Computational Science and Mathematical Methods
Advances in Parallel Computing 25
Please use the identifier: in citations.
In this paper we present a binary instrumentation methodology to monitor runtime events. We demonstrate our approach on OpenMP constructs for the Intel and GNU compilers. A binary-level static analysis detects the compiler patterns and the runtime function calls corresponding to OpenMP regions. To this effect we integrate the software tool MAQAO with the scalable measurement infrastructure Score-P. We design a new interface and modify both tools to support the new events. The main advantages of using binary instrumentation are the possibility to retrieve implicit runtime events, to instrument without recompilation, to be independent from the language, and not to interact with compiler optimization. Our validation experiments and first results shows that binary instrumentation has not introduced any additional overhead.