Does Nsight Compute profile into the library call? and regarding obtaining series of executed functions

  1. The application calls the library function from inside the kernel, similar to cublas library call
    However, Nsight Compute does not seem to profile into the library call. Is this possible?

  2. Another question. Is it possible to obtain the dynamically executed function names? including into the library?

Thank you

In general, Nsight Compute profiles all of the device code executed as part of the kernel, including device functions and functions in libraries that are running on the GPU. In the “Details” page, the metrics will be an aggregate of all those functions. What are you seeing that is leading you to believe it is not profiling library calls?

For your second question, Nsight Compute doesn’t collect any dynamic information about what functions are running like a callstack or anything like that. If you have symbol information, you can see metrics at the source level and infer which functions ran by which lines of code have metrics, but there isn’t an explicit list of functions that were called.

1 Like

One other thing that did come to mind - if you have the SASS view open on the source page, the dropdown will list functions that were executed. That is an option to see some of the data you’re looking for.

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.