cuda kernel parameter

say I have a cuda kernel

__global__ foo (int a, int b)
{
    ... ...
}

where a and b are stored. Does this takes register space fo开发者_Python百科r each thread?

No this doesn't take register space for each thread, instead a and b are allocated once in a constant space - a read only space - from which all thread can read.

Note that this space is cached by constant registers and shared by all threads:

A read-only constant cache that is shared by all scalar processor cores and speeds up reads from the constant memory space, which is a read-only region of device memory [PTX ISA Version 2.1 Chapter 3].