I am trying to profile a kernel that only reads data from a matrix and adds this data to a variable. However, I am not able to view the SASS code of the “For loop” (check picture) using Nsight compute version 2022.2.1.0. The other line of codes are ok. What could be the issue?
I suppose you have compiled the code with nvcc -lineinfo option.
You can try scrolling the SASS code - in case the highlighted SASS lines are outside the view.
Is the CUDA C source file matching the SASS code?
Did Nsight Compute find the source file or you had to browse & select the file?
Did you use the ncu --import-source option? If not you can try if this resolves the issue? (Note that this option is not really needed for your case)
sorry, I think I have found the issue. I wasn’t using the variables elsewehere so I believe the compiler removed these lines for optimisation.