We use cuMemGetInfo() in our code to track memory use on the cards. I noticed that it takes 32-bit parameters to return the amounts in bytes. I want to future proof our code for Tesla C2070 with 6GB of RAM. What will this call do in that environment. Is the API for it going to change to take size_t as a type instead? Anyone have any ideas?
-Derek
It seems that using the 64-bit version of CUDA, that the size_t variable is 64-bit, however I am unsure whether size_t is signed or unsigned (i.e is size_t a long long or an unsigned long long).
this changed in 3.2, and on 64-bit platforms size_t is generally an unsigned long long