The SCT library was invented with the aim of having an easy to use timer with as little overhead as possible.

The timers (which can also measure hardware performance counters via PAPI interface) are implemented for pure sequential, OpenMP, MPI and hybrid OpenMP-MPI usage with a special emphasis on selectable reduction operations in order to provide a compact presentation of the measurements of parallel applications.

We support writing the measured timers to
* stdout
* ASCII file
* hdf5 file

PLease use the folowwing envVars:
- SCT_OUT = [stdout | stderr | hdf5] (default: stdout)
- SCT_FILENAME = \<filename\> 


The data layout of raw timer data (i.e. no reduction has been used) will always be used in hdf5 files and looks like

timer[0 ... nb_tasks-1][0...nb_threads-1][0 ... nb_timer-1] for time executed in parallel regions
timer[0 ... nb_tasks-1][nb_threads][0 ... nb_timer-1]       for time executed in serial part

hence, the thread dimension is always nb_threads+1

To reduce the amount of reported data when ASCII output is used, we provide the following envVars:

- SCT_PROC_CHOICE = [SCT_SELECT_ALL | SCT_REDUCE_ALL | \<proc_id\>]
- SCT_THREAD_CHOICE = [SCT_SELECT_ALL | SCT_REDUCE_ALL | \<thread_id\>]

Special care needs to be taken when combining serial phase measurement and parallel phase measurement (this only matters for OpenMP):

- SCT_SP_MERGING = 
 - SCT_SP_SERIAL_ONLY: only the serial phase measurement is used
 - SCT_SP_PARALLEL_ONLY: only the thread-parallel measurement is used
 - SCT_SP_MERGE_SIMPLE: the serial phase measurement is added to the master thread of parallel regions
 - SCT_SP_SELECT_ALL: the serial phase measurement and the thread-parallel measurement are reported separately