parsetiming
Parse timing data from CPPTRAJ output.
parsetiming ... [out ] [name ] [sortby {time|cores|filename}] [includebad] [showdetails] [type {trajproc|trajread|actframe}] [reverse] [groupout [grouptype {prefix|name|kind}]]
<filename args>
Files containing CPPTRAJ output to get timing data from.
[out <file>]
Write total sorted timing sets to .
[name <setname>]
Set name for timing data sets.
[sortby {time | cores | filename}]
Sort timing data sets by either time, number of cores (MPI processes * OpenMP threads), or file name.
[includebad]
If specified, include run output for which timing data cannot be extracted (e.g. an incomplete/failed run).
[showdetails]
If specified, details about each run will be printed to STDOUT.
[type {trajproc | trajread | actframe}
If specified, report time other than the total time (which is the default):
trajproc
Total trajectory processing time (trajectory I/O plus action frame time).
trajread
Trajectory read time (requires compiling with ‘-timer’ configure option/-DTIMER compiler define).
actframe
Action frame time (requires compiling with ‘-timer’ configure option/-DTIMER compiler define). [reverse] Instead of longest time to shortest time, sort shortest time to longest time.
[groupout <file>]
Group run output by a property and write to a file. Additional details will be written like speedup and efficiency (relative to the slowest run).
[grouptype {prefix|name|kind}]
Property to group runs by in group output file.
prefix
Group by directory prefix.
name
Group by run type name (see <setname>[name]
below).
kind
Group by run type name; OpenMP runs are separated by number of OpenMP threads.
DataSets Created:
<setname>
Set containing sorted run times.
<setname>[name]
Set containing shortedhand run names. Consists of a prefix (S for serial, O for OpenMP, M for MPI, H for hybrid MPI/OpenMP) followed by numbers indicating number of MPI processes ‘x’ OpenMP threads; e.g., H16x4 means a hybrid run consisting of 16 MPI processes and 4 OpenMP threads per process. If CUDA is active, the name will be wrapped in ‘G()’, e.g. G(H16x4).
<setname>[dir]
Set containing output directory prefixes. If no directory prefix, just contains output file name.
The parsedata
command can be used to extract timing data from CPPTRAJ output.