CUDA 2 constant memory access versus 1 access and 1 division and 1 modulo, which is better?

Question

So, in my problem, I need to read a kind of rowOffset and colOffset. The formula is:

index = rowOffset * size + colOffset