Copy if map is identical in transpose helper

[yasahi-hpc]

This PR aims at allowing transpose function to perform deep_copy if transpose is not needed.

• Using execution space instance appropriately in transpose unit-tests
• Skip some case when map is identical. I found that the test will result in test failures in fftshift tests in Release build on Cuda backend. This does not happen in CPU backends or Cuda backend with Debug build. This happens only with 4D cases.

[PaulGannay]

This is not related to this PR.
Maybe you should have an array std::size_t idx[7], that you'd use to iterate over idx[0], idx[1], ...
An other array std::size_t dst_idx[7]
And then, you could write

for (int i=0; i<7;++i) {
  dst_idx[i] = idx[map[i]];
}

And maybe you could use a recursive template for the 7 and put the whole loop in a function that could be reused across all the dimensions?

[PaulGannay]

The changes in KokkosFFT_transpose.hpp looks correct, but the test file is so big that it is becoming hard to review, it would be best if you could find a way to factorise some of the code to make the changes easier to track.
It is orthogonal with the work done in this PR though.

[tretre91]

looks good to me

[yasahi-hpc]

Thanks @PaulGannay @tretre91
I will refactor first in another PR
As you said, tests can be simplified

[yasahi-hpc]

@PaulGannay I opened #348 focusing on the execution space instance. Can I have your review please.

[yasahi-hpc]

@PaulGannay I opened #349 to make reference transposed view simpler. Actually, I encountered an issue with this change alone. The failed test is skipped now. Can I have your review please.

[yasahi-hpc]

Hi, I have rebased after PRs #348 and #349
May I have your another review please @PaulGannay

[PaulGannay]

LGTM

[yasahi-hpc]

@PaulGannay

Thanks! I will merge this