Search code examples
cudaprofilingnsight-compute

NSight Compute - get total number of samples?


When you use NSight-Compute's Source, PTX or SASS view, you see the number of samples taken on each line. However, you don't see (or it is difficult to find) the total number of samples taken overall, which you would like to normalize by.

Is it listed somewhat inconspicuous or is it just missing?


Solution

  • Is it listed somewhat inconspicuous or is it just missing?

    There is currently no dedicated metric or place in the tool where the total number of samples is shown. We will look into adding this in the future.

    As a workaround, you should be able to collapse the source view using the +/- button, which aggregates the metrics, including samples, per function/file and makes it easier to manually count, should your kernel have multiple functions/files.