Measure number of individual instructions executed


I would like to profile my applications on the NVIDIA Drive AGX in a way that I see how many floating point instructons, integer, etc. were exectued after passing the DCO. Is there any way to do that?

Thank you &

Dear markus.kohout,
Nvidia visual profiler can count certain events and metrics. Please check nvprof --query-events and nvprof --query-metrics to see if it helps.

Thank you for your response. This seems only to work for CUDA applications. What if I just want to profile these metrics for a CPU application?

–> number of floating point operations etc. on the CPU? Utilization of the ARM NEON units?

Dear markus.kohout,
Tegra stats can show the utilization of CPU cores and Tegra system profiler can be used to know performance metric counters. Please check for more details.