Sarthak

hello Aritra (@ariG23498 ), thanks a lot for this series!
one doubt which i had: how were you able to know for sure that Memcpy (memory copy) operation in GPU was indeed because of out of place masked_fill?

one way i'm able to think it is because that occured exactly before the masked_fill kernel op? but how did you actually validated that is something i'm not able to understand.

edit:
i figured it out while i was tinkering - but keeping this comment here incase someone else has the same doubt. the arrow mapping in perfetto maps exactly that which op in cpu invoked the kernel at gpu.