Is there any cast operation from int to float applied to the arguments of the texture fetch function in the kernel? For example:
int i=0;
tex2D(texture, i,i);
The question is whether “i” is cast to float here in execution time?
The reason for my question is that I noticed that texture fetches from the same address by all the threads in a block work slightly slower than similar constant memory fetches. Or maybe its just an artifact of something else.