fix: int type in transpose

[yasahi-hpc]

This PR fixes the base int type of map in transpose helper.
Previously, it only worked for std::array<int, DIM>.
Now, it works also for std::array<std::size_t, DIM>

[PaulGannay]

Since the map is holding index value used to access views, why do you need a new type in addition to IndexType?

[yasahi-hpc]

Since the map is holding index value used to access views, why do you need a new type in addition to IndexType?

Map does not hold the index value used to access views.
It holds the permutation array that maps src and dst index arrays.

m_out(src_idx[m_map[Is]]...) = m_in(src_idx[Is]...);

As you said, src_idx must be in IndexType, but int type of the m_map can be different.

[PaulGannay]

Why offer a customisation point for this?
Why not use std::size_t internally?

[yasahi-hpc]

Why offer a customisation point for this? Why not use std::size_t internally?

Actually, I am considering that. Although APIs allow std::array<int, DIM> to specify axes, they should be converted into std::array<std::size_t, DIM> internally
Since it requires more changes, I would like to do that in another PR

[PaulGannay]

Actually, I am considering that. Although APIs allow std::array<int, DIM> to specify axes, they should be converted into std::array<std::size_t, DIM> internally
Since it requires more changes, I would like to do that in another PR

I would close this PR without merge then.
But if you think that it is necessary, in preparation of other work, I haven't seen anything wrong either so I guess that you can merge it.

[yasahi-hpc]

Actually, I am considering that. Although APIs allow std::array<int, DIM> to specify axes, they should be converted into std::array<std::size_t, DIM> internally
Since it requires more changes, I would like to do that in another PR

I would close this PR without merge then. But if you think that it is necessary, in preparation of other work, I haven't seen anything wrong either so I guess that you can merge it.

It is useful for distributed implementation details, wherein all the axes are already converted into std::array<std::size_t, DIM>. If you prefer, I can change this function to work on std::array<std::size_t, DIM> and convert maps into std::array<std::size_t, DIM>