I am new to VTune and trying to figure out how to relate the time displayed in VTune to the wall time clocked using the time command in Linux. I recently added $OMP calls to one of my Fortran applications and I have been seeing a discrepancy between the effective time listed in Vtune (under Elapsed time) and the wall time obtained via time command in Linux, for OMP_NUM_THREADS > 1
for OMP_NUM_THREADS=1, the times obtained using time command in Linux and the effective time are 15.4 and 15.6 seconds respectively.
for OMP_NUM_THREADS=4, the times obtained using time command in Linux and the effective time are 15.3 and 23 seconds respectively.
I am unable to comprehend the reason behind this discrepancy. My understanding of effective time is the time spent in the application excluding the spin and overhead time. So shouldn't this match the time command in Linux ? Any information that could help me understand these observations better are greatly appreciated.
For reference purposes, I am using the following command to generate the Vtune results -
amplxe-cl -collect hotspots -knob analyze-openmp=true -result-dir vtune_profile <target application>