Demian,
The way you propose to do it (setting the duration longer then yours program execution) will work.
Just take into account that you'll get so called "calibrations" runs (see more about the callibration in help).
The number of calibration runs depends on events and its number you chose, as well on the processor you use.
So if you program takes a long time to run you might prefer to switch to the mode without calibration.
To run sampling collection on the whole session of your program and eliminate calibration runs set the duration of the activity to zero (use the switch "-d 0") and "-cal no" option to the sampling collector.
See the example of the command line:
That command will run sampling collection with default settings (except the calibration) for the default events on Netscape application, that will open the html file with help for VTune Performance Environment. (Here you can find more info about the calibration and sample after). In this activity run the default values for sample after will be used. Generally speaking you have to provide the sample after values, when you do not use the calibration.
will show your the total count of events you used in the last activity run.
To see what are the default setting for sampling collector are dump them to some file (e.g. /tmp/dss):
and investigate it.
You can prepare your own setting files (use the example of the dumped one). To use the prepared setting files use the "-of" flag for activity:
Daniel



Using VTune to count (rather than sample)
Let's say I am interested in counting the overall number of L1 I- and D-cache misses experienced by my application, and let's assume that my platform is a "standard" PIII (733MHz variety).
Can I use VTune to do this accurately? For example, would setting the sampling rate longer than my program's execution do the trick?
Thanks for your help,
Demian M. Nave
Pittsburgh Supercomputing Center
Carnegie Mellon University
Pittsburgh, PA 15213
dnave@psc.edu