Comparative debugging using TotalView scripting

Comparative debugging makes use of direct comparisons between two different runs of the same or similar programs to diagnose unexpected and often erroneous behaviour in the application or applications. Comparative debugging is an important technique for STFC and for computational scientists at other labs around the world. As a six month project at STFC, in collaboration with Rogue Wave Software Inc., an automated comparative debugging script (CDTV) has been created that makes use of the command interface of RogueWave’s TotalView debugger. To demonstrate this script, example scenarios are shown and discussed within the report, highlighting the key features, current limitations and possible improvements that could guide any future work. The CDTV script has been shown to be capable of locating differences caused by an artificial bug introduces into a serial program written in C or in Fortran as well as parallel programs using MPI and OpenMP. As well as simple test codes, the script has been demonstrated using a real application, the UK Turbulence Consortium’s Shock/Boundary Layer Interaction code. We believe that similar large scale applications could benefit from comparative debugging without making major changes to the underlying CDTV script.