Assembly is mostly used to optimize rarely maintained vector/matrix SIMD operations and even then it's mostly wrapped in intrinsics.