This way, SIMD-optimized functions don't have to sign-extend their stride argument manually to be able to do pointer arithmetic.