Script to parse the profiler output

[mfatica]

If you want to find out the total time recorded by the profiler, this is a simple script that may be useful.

The first step is to enable the profiler and use the comma separated format

export CUDA_PROFILE=1

export CUDA_PROFILE_CSV=1

The script ( total_time) is a simple one line of awk:

awk -F"," ‘NR!=1 { sum += $2} END {print "Total time in trace= " sum}’ $1

Once the log file has been generated (if the CUDA_PROFILE_LOG variable is not set, it will be named cuda_profile.log), you can simply invoke the script

sh total_time cuda_profile.log

Massimiliano

[Mark_Harris]

Note: You should also put your own timers in your code to measure total wall-clock time. The timers in CUDA are inside of the API calls, so they don’t include any app overhead.

Mark