Support in place transform

[yasahi-hpc]

Resolves #61

This PR aims at supporting in-place transform.
Following changes are made.

• Add a function are_aliasing to check if the input and output views are alias
• Add a test function for are_aliasing
• Improve get_extents function to compute extents for in-place case
• Use in-place options for ROCM and SYCL in case of in-place transforms
• Add tests for 1D and 2D in-place transforms

[tpadioleau]

Can you confirm that fftw, cufft and hipfft internally do a check for an in-place execution ? No need to modify anything on our side ?

[yasahi-hpc]

Can you confirm that fftw, cufft and hipfft internally do a check for an in-place execution ? No need to modify anything on our side ?

Thank you for reviews

I expect that fftw3 and cufft are doing checks internally, but it is not very clear from documentations.
If I understand, all we need is just passing same pointers as input and output
hipfft in turn checks internally (see https://github.com/ROCm/hipFFT/blob/f9a5de3d5bfaa08b52d964bf6501478ee71398f3/clients/hipfft_params.h#L512-L516)

[tpadioleau]

I expect that fftw3 and cufft are doing checks internally, but it is not very clear from documentations. If I understand, all we need is just passing same pointers as input and output hipfft in turn checks internally (see https://github.com/ROCm/hipFFT/blob/f9a5de3d5bfaa08b52d964bf6501478ee71398f3/clients/hipfft_params.h#L512-L516)

That is also my understanding reading cufft.

[yasahi-hpc]

@tpadioleau
May I have your second review please

[yasahi-hpc]

@tpadioleau
It seems that we need to keep pi in double precision.

[tpadioleau]

@tpadioleau It seems that we need to keep pi in double precision.

Not precise enough otherwise ?

[yasahi-hpc]

@tpadioleau It seems that we need to keep pi in double precision.

Not precise enough otherwise ?

Maybe.
If they compute twiddles with double precision and cast them into single precision, we need to use pi in double precision

[yasahi-hpc]

@tpadioleau Seems OK now.
If you feel good, I will merge this

[tpadioleau]

Seems ok to me

[yasahi-hpc]

@tpadioleau Thank you for review. I think I have fixed accordingly

[tpadioleau]

ok for me

[yasahi-hpc]

I will merge this. Thank you again for your review