Dear NVIDIA,
I was wondering if it is possible to print out the parameters of memory-related APIs such as cudaMemcpy that are invoked for a given program that I do not have access to the source code. To be more specific, I want to see if two cudaMemcpys that I can identify with the nvprof actually copies from a consecutive memory location or not.
Would this be possible with the existing tools?
If not, I tried to hook the memory-related API such as cudaMemcpy so that it prints out the parameters before invoking the libcudart library function. Can I get some help with this?
Thank you.
Sincerely,
YJ