DDT 3.2 parallel debugger (a product of Allinea Software ») is now available at CSCS. DDT provides source-level debugging of Fortran, C, and C++ codes. It can be used for debugging serial, multi-threaded (OpenMP), multi-process (MPI) and accelerator based (CUDA, OpenACC, HMPP) programs running on production and research systems, including the CRAY XE6 Monte Rosa and CRAY XK6 Todi.
Here you can find more information on how to get started with DDT :
Today August 16th, Tödi is going to be under maintenance from 7:00 AM until 7:00 PM.
The system will be upgraded to the pre-release of SMW7.0, CLE4. 1 and Cuda5.0.
Be aware that you will need to recompile your applications for the new environment. Find below a more detailed description of the upgrade.
Details Tödi Upgrade
custom output of ” apstat ”
users can parse the output of ” apstat ” the way they want with ” -f ” option i.e: apstat -no -f ” NID,Pl,Rv,Hw”
GPU support in CLE4.1:
* CUDA 5.0 API
* Kernel Driver + CUDA library + CUDA Toolkit
* GPU “Modes” exclusive, persistent
* Simultaneous Execution contexts
. Only one CUDA context ( i,e MPI rank ) per node by default
. Multiple contexts support throught CUDA proxy
. set ENV variable CRAY-CUDA_PROXY=1 before aprun
. Limitations include debugging, profiling memory protection
. Feature support is evolving…..
* Memory scrubbing
. Zero GPU memory between jobs
. Automatic, transparent
. Not delivered by NVIDIA as of Aug 01, 2012
. Expect in future update