PR 43256 was going to re-write norm to be careful about overflow only when necessary. Used in the style of norm2diff3 this is 5x faster. Not sure about simd on arrays, perhaps it could be further improved.
1 Like