Controlling Granularity of Data Collection and Reports
You can control the granularity of data collection and reports. If you want finer granularity (that is, more samples), use the
If your program is likely to have repeating patterns, such as loops, those patterns might coincide with the fixed sampling rate, affecting the results. For example, some functions might not be sampled because the IP is never in the function on the measurement cycle. A solution for this is to vary the sampling rate by some number of cycles.
For more information, see
Specifying Processes to Measure
The
•Measure the selected processes
•Track the processes without measuring them in order to identify interesting child processes
•Ignore specific processes
By default, HP Caliper measures and reports on the parent process and all child processes, including native Integrity
To measure the parent process alone, use the
By default, HP Caliper produces separate reports for each process and concatenates them to stdout or the output file specified in the
If you want a separate file for each report, use the
basename.executablename
For example, the following command on
$ caliper fprof
produces these files:
COUT | contains an overview of the entire collection run |
COUT.cc contains report data for all cc processes
COUT.ecom contains report data for the ccom process
COUT.ld contains report data for the ld process
HP Caliper can measure shell script files. By default, HP Caliper measures the shell program and the programs that the script invokes.
For Fortran MPI programs, by default, HP Caliper measures the mpirun controlling process and the real application.
Process Tree Report
With multiprocess reports, HP Caliper creates a summary report of the process tree to help you navigate the reports. The summary includes:
•The process name indented to show lineage.
•The argv0 argument.
•The process ID (PID) and parent process ID (PPID).
Configuring Data Collection 97