Cleanup

[yasahi-hpc]

Following modifications have been made.

1. Use templates for most of the tests to reduce lines
2. Some interfaces for FFT are suppressed which are more or less the same
3. Avoid normalization if unnecessary

The remaining issues before release are as follows:

• Unuse double underscore identifiers like __KOKKOSFFT_NORMALIZATION_HPP__
• Rename FFT_Normalization to Normalization because they are already under KokkosFFT namespace
• Support transforms over odd number. The number of points for FFT direction is limited to be even number (so that the size after transform would be n/2+1)
• More static assertions. Particularly, we assume that the In and Out Views have same Layout and ranks, which should be checked.
• Introduce ExecSpace as a template argument as well as DDC. This allows the coexistence of fft helpers for both host and device:

KokkosFFT::fft<Kokkos::OpenMP>(a, out); // a and out are on Host or Device is CPU
KokkosFFT::fft<Kokkos::Cuda>(a, out); // a and out are on Device

KokkosFFT::fft(Kokkos::OpenMP, a, out); // a and out are on Host or Device is CPU
KokkosFFT::fft(Kokkos::Cuda, a, out); // a and out are on Device

We can also make a check for memory and exec space consistency by

static_assert(
  Kokkos::SpaceAccessibility<ExecSpace, MemorySpace>::accessible,
  "MemorySpace has to be accessible for ExecutionSpace."
);

• Add capability to reuse plans. This is particularly important for NVIDIA GPUs, which has some overheads to create plans.
• Management of default arguments. As well as numpy, I would like to allow the function calls with minimum overloading

KokkosFFT::fft(a, out);
KokkosFFT::fft(a, out, /*axis=*/-1);
KokkosFFT::fft(a, out, /*n=*/n0);
KokkosFFT::fft(a, out, KokkosFFT::FFT_Normalization::BACKWARD);
KokkosFFT::fft(a, out, KokkosFFT::FFT_Normalization::BACKWARD, /*axis=*/*-1);
KokkosFFT::fft2(a, out, KokkosFFT::FFT_Normalization::BACKWARD, /*axes=*/axes_type{-2, -1});
KokkosFFT::fft(a, out, plan, KokkosFFT::FFT_Normalization::BACKWARD, /*axis=*/-1);

Particularly, it is difficult to distinguish n and axis, where n is a size_t type and axis is an int type.

• Introduce the Impl namespace to disallow users to access implementation details
• Add missing functions such as hfft
• Add capability to allow optimal argument n. (See fft)
• Googletest should be parameterized over types using ::testing::Types
• Add docs to explain what are our interfaces and how to use them
• introduce a clang format file ideally from Kokkos repo
• Adding perf_test with googlebenchmark.
• Improve docker files and add nvidia-hpc-sdk env

The remaining issues which may be important in the future

• Allow to access from Views with strided Layouts. Internally, the input and/or output buffers with LayoutRight are created and the strided Views are copied into these buffers.
• Add Intel GPU support based on oneMKL
• Adding 4D to 8D capabilities
• Seek for non-contiguous fat plans which may be available in some cases
• Add more advanced examples. Derivatives in Fourier space, simulations with spectral method?

[yasahi-hpc]

@jbigot OK to merge? Others are supposed to be resolved based on issues

[jbigot]

please merge, it improves the code-base a lot