How is the run-time reported by nvprof --analysis-metrics
calculated when a certain kernel is replayed to obtain information on events and metrics ? Is it the time take by the first/last call or overall average or … ?
How is the run-time reported by nvprof --analysis-metrics
calculated when a certain kernel is replayed to obtain information on events and metrics ? Is it the time take by the first/last call or overall average or … ?