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