ThreadSpotter for Cray XE
Rogue Wave Software, which provides cross-platform HPC software development, has released ThreadSpotter 2012.1 offering support for the Cray XE supercomputers.
Several leading labs recently adopted previous versions of ThreadSpotter, including the UK Atomic Weapons Establishment (AWE), the US Lawrence Livermore National Laboratory (LLNL), the French Alternative Energies and Atomic Energy Commission (CEA), and the German Jülich Supercomputing Centre. AWE significantly improved code efficiencies and the productivity and performance of its developers. Jülich Supercomputing Centre, which cooperates on the European research projects (H4H und HOPSA), was able to optimise large-scale, real-world codes with ThreadSpotter. For LLNL, ThreadSpotter pointed out loop fusion and data structure transformation tuning opportunities within the critical Livermore Unstructured Lagrangian Explicit Shock Hydrodynamics (LULESH) application that resulted in a 25 per cent increase to serial run time performance and a doubling of parallel performance.
With the new release of ThreadSpotter 2012.1, Cray XE users can now improve performance and gain better efficiencies. The new version of ThreadSpotter has been modified to provide better support on ALPS and to work with schedulers/resource managers PBS Pro, MOAB/Torque, and SLURM. In addition, this version of ThreadSpotter also provides an updated CPU database that now includes AMD Bulldozer processors, Intel Sandy Bridge and Ivy Bridge processors, and IBM Power and BlueGene for Cross-Platform analysis.
‘Scientists and engineers expressed to Rogue Wave their need to employ ThreadSpotter to optimise the performance of parallel applications on the Cray XE platform. As Cray supercomputers are strategic platforms for our customers, we are happy to quickly respond to these requests so our products stay at the cutting-edge of the HPC market,’ said Chris Gottbrath, product manager at Rogue Wave.
ThreadSpotter helps eliminate performance issues by identifying problematic sites in the code where a change could make the program more efficient. Failure to use cache memory efficiently is a frequent cause of poor performance, because the processor has to stall for many cycles waiting for data to be fetched from main memory. ThreadSpotter provides specific guidance on performance issues by identifying them, estimating each issue’s importance and rank ordering them. ThreadSpotter then guides the developer to the location in the source code where the issues are located.