Hello,
I am spreading a lot of process in a node using MPI. I want one of those processes launching kernels and other one monitoring activities through its callback mechanism. Is it possible? Maybe using correlation ids o something? Or will it remain silent?
CUPTI doesn’t provide the support for profiling the child or the other processes.
For profiling all the processes, user can inject the CUPTI based shared library into the target application, e.g. using LD_PRELOAD or Detours, or by modifying the target application itself if applicable. From that library, user can initialize CUPTI for the whole target process.
Refer the related thread CUPTI activity API and child processes - CUDA Profiler Tools Interface (CUPTI) - NVIDIA Developer Forums