Is there any possibility (some tool) to check if my code does busy waiting (has some conflicts) in CUDA? I've checked nvprof but haven't seen such option (just general information about kernel's execution time, not from kernel itself).
I have some code that works about 2,5sec sequential and about 4,5sec asynchronus and I don't know which part of code can be improved.
via Chebli Mohamed
Aucun commentaire:
Enregistrer un commentaire