- Address space cast of nullptr in local_space into a generic_space for the CUDA backend. The reason for this cast was having invalid local memory base address for the associated variable. - In the context of AMD GPU, assigns a NULL value as ~0 for the address spaces of sycl_local and sycl_private to match the ones for opencl_local and opencl_private.