Redefine pitches

[bernhardmgruber]

This PR shifts the values returned from getPitchesInBytes to be consistent with std::mdspan (except in bytes).

Example: the pitch vector for the extent {42, 10, 2} changes:

Before: {4, 3360, 80, 8}
After: {80, 8, 4}

The new meaning is that the pitch value is the number of bytes to jump from one element to the next in the given dimension.

Fixes: #2083

Waiting on #2092 to be merged before rebase.

This is a silent breaking change. However, to a function that was only introduced a few days ago with #2092.

[fwyzard]

why the change from

struct Action {
  static auto action(...) { ... }
};

to

struct Action {
  auto operator()(...) const { ... }
};

?

[bernhardmgruber]

why the change from

struct Action {
  static auto action(...) { ... }
};

to

struct Action {
  auto operator()(...) const { ... }
};

?

I tried hard to find the corresponding issue, but probably it was just a long discussion on a PR somewhere. I argued for using operator() consistently to avoid repreating the name of the functionality in the customization point struct name (trait) and the member function. I think we concluded to move to operator() in the future. All math customizations are already using this scheme btw. But the current situation in alpaka is inconsistent.

I don't care too much what we do. operator() is what I would prefer, but I won't lose any sleep over something else. What do other people think?

[fwyzard]

I don't care too much what we do. operator() is what I would prefer, but I won't lose any sleep over something else. What do other people think?

If the reason is to avoid duplicating the name you could use something like static auto apply() for each class.

But I don't really mind, I was only curious about the change.

[bernhardmgruber]

This should be swapped back.

[bernhardmgruber]

I really appreciate how a MD-subscript operation into a buffer is now the dot product between index and pitch vector <3. This is also consistent with some MD array talk I have seen.

[psychocoderHPC]

only one question but nothing against the PR at all.

[bernhardmgruber]

Hmm. Something got stuck with the readthedocs build. The error saids: "This build was terminated due to inactivity. If you continue to encounter this error, file a support request with and reference this build id (21768482)."