Determine which kernel launch causing "invalid configuration argument"

I’m trying to track down an “invalid configuration argument” error. Is there a way to determine which kernel launch is causing the error? The tools I have available are cuda-gdb, cuda-memcheck, nsys, anything other than ncu. (I don’t have the computer set up correctly to run ncu, nor do I have permission to change the computer.) I’m attempting track down an issue using a third party library, so I can’t just add calls to cudaGetError. Knowing which kernel is being called incorrectly might show me what I’m doing wrong when calling the library.

Edit: running on CUDA 11.4.x, in a docker container, on Linux.

Ah found it. Use cuda-gdb, then “set cuda api_failures stop” before running the program.

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