__builtin_cpu_support("sse4.1"); // Return 128 if 128-bit register allowed
__builtin_cpu_support("sse4.2"); // Return 256 if 256-bit register allowed
__builtin_cpu_support("avx"); // Return 512 if 512-bit register allowed
__builtin_cpu_support("avx2"); // Return 1024 if 1024-bit (!!!) register allowed