>I don't think a compiler could ever reliably transpose an array-of-structs representation to a struct-of-arrays one. An interesting question, but a SoA isn't

>I don't think a compiler could ever reliably transpose an array-of-structs representation to a struct-of-arrays one. An interesting question, but a SoA isn't necessarily better than an AoS: it depends on the access pattern, so the question is whether this optimisation could be added as a part of PGO..