The problem I am facing is:
I have a table like this
In my kernel, I need to look up ( search ) for some values in this table. For example, I may be given the key value 12.105, need to get the corresponding data value.
I think I should save this table to constant memory, and do binary searching for the the values I needed and do the interpolation. Then the searching time will be log(N),
this means I may need log(N) times global memory access----------think it is going to kill my performance.
It is not a issue in CPU but seems it is a problem for CUDA…
wondering if anybody have good idea about this?