[vector] Some thoughts on alternative vector loop shapes

[Vladimir Ivanov]

> It can represent vectors of a larger size hardware supports:
> 
>    class Int1028Vector {
>      final Int512Vector lo;
>      final Int512Vector hi;
>    }
> 
> or provide an alternative (more optimal) representation of a vector 
> which perfectly fits into a vector register:

By "more optimal" here I meant there could be cases where performing an 
operation on smaller vectors can be faster: either due to more efficient 
hardware implementation or absence of hardware support (e.g., lack of 
256-bit integral support in AVX).

Best regards,
Vladimir Ivanov