17 Sep
2013
17 Sep
'13
2:42 p.m.
On 17/09/13 16:11, Andrey Semashev wrote:
It depends on data processing pattern. If this container performs AoS <-> SoA transformation, it can be useful for parallel/vectorized data processing. However, I don't see how this container could be implemented generically for arbitrary types. I would be interested to see such implementation.
The library I work on, NT2, already does this for vectorization purposes. It works for any Fusion Sequence (i.e. any type that offers a tuple-like interface to iterate the members).