Yes you can see the broad opportunity but various things can easily break vectorization between extensions of x86 and ARM. The first that comes to mind is use o

Yes you can see the broad opportunity but various things can easily break vectorization between extensions of x86 and ARM. The first that comes to mind is use of doubles instead of floats. Another is too wide of an inner loop, especially on algorithms that must be tuned for cache line sizes. You can also run into accuracy & stability issues between different x86 instruction extensions - it's really a nightmare to debug those.