Suggestion to add instruction offsets in cuobjdump's output


I’d like to do source code line mappings from runtime pc to source code.
I think that could be much easier if each line in the output of cuobjdump/nvdisasm could contains the instruction offset.

Is there any future development plan for this support?

You can do this with nsight compute.

according to the following post, the first column of the cuobjdump/nvdisasm output is the instruction offset in the current kernel. That may solve part of my problem.

Unless I’m misunderstanding what you mean here, as Robert mentioned, this is a feature available in Nsight Compute:

Sorry, maybe I didn’t make my question clear.

I’m writing a tool using sanitizer API which could help me get the PC of the executed instructions, and I’d like to add the source code mapping feature to my tool.

I know ncu has the source code mapping but it is only for when you have done the profiling, and the pc->source code future is not exposed.

1 Like

OK. If you don’t already have it, you may find Chapter 2 and the Appendix of opcodes helpful here:

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