Right now Attributor assumes that if the the target is a GPU is can use a single set of address space numerical values to determine the local address space, but that's not true in general, so add SPIR-V support, which uses different values. This fixes an instruction incorrectly being marked as dead and optimized out for an OpenMP SPIR-V offloading example. --------- Signed-off-by: Nick Sarnie <nick.sarnie@intel.com>