Problems with "__select_nocancel" when using the P

I write a CUDA Fortran program and when I use the PGPROF tool to test the program performance. I find there is a function named “__select_nocancel” in the function list, and it take over 52 percent of the time.

Can someone tell me what the function “__select_nocancel” is?

Hi OceanCloud,

That’s the CPU waiting for the GPU kernels to finish.

  • Mat