On x86_64, this is always 128 bits, rendering it incompatible with AVX.
GCC reports the maximum enabled vector width (which is what we want) and Intel reports the maximum supported vector width (which is okay too).