Catch is a profiler for parallel applications that collects hardware performance counter information for each function called in the program, based on the path that led to the function invocation. It automatically instruments the binary of the target application independently of the programming language. It supports mpi, OpenMP, and hybrid application and integrates the performance data collected for different processes and threads. Functions representing the bodies of OpenMP constructs are also monitored and mapped back to the source code. Performance data is generated in XML for visualization with a graphical user interface that displays the data simultaneously with the source code sections they refer to.
By: Luiz DeRose, Felix Wolf
Published in: Lecture Notes in Computer Science, volume 2400, (no ), pages 167-76 in 2002
Please obtain a copy of this paper from your local library. IBM cannot distribute this paper externally.
Questions about this service can be mailed to reports@us.ibm.com .