
Passing parameters by value
If you recall, in the gpuAdd program, the syntax for calling the kernel was as follows:
gpuAdd << <1,1> >>(1,4,d_c)
On the other hand, the signature of the gpuAdd function in definition was as follows:
__global__ gpuAdd(int d_a, int d_b, int *d_c)
So, you can see that we are passing values of d_a and d_b while calling the kernel. First, parameter 1 will be copied to d_a and then parameter 4 will be copied to d_b while calling the kernal. The answer after addition will be stored at the address pointed by d_c on device memory. Instead of directly passing values 1 and 4 as inputs to the kernel, we can also write the following:
gpuAdd << <1,1> >>(a,b,d_c)
Here, a and b are integer variables that can contain any integer values. Passing parameters by values is not recommended, as it creates unnecessary confusion and complications in programs. It is better to pass parameters by reference.