[DFT][MKLGPU] Use FWD/BWD_STRIDES

[hjabird]

Description

• Intel(R) oneMKL 2024.1 deprecate the INPUT/OUTPUT_STRIDES API
• Using the old API causes error messages to be printed.
• This PR uses the new FWD/BWD_STRIDES API, but maintains support for older Intel(R) oneMKL.
• Fixes a warning
  • sources/src/dft/backends/mklgpu/descriptor.cpp:30:29: warning: a declarative nested name specifier cannot name an alias template [-Walias-template-in-declaration-name]
  • This warning seems to need nightly clang rather than icpx.
• It additionally fixes one-line bug in cuFFT (albeit one that doesn't seem to cause any issues).

Fixes #487

Checklist

All Submissions

• Do all unit tests pass locally? Attach a log.
  • testres.txt
  • Tests only attached for MKLGPU. Other domains have the minor modification to stop the compiler warning. Ctest has been run and passes.
• Have you formatted the code using clang-format?

[hjabird]

Note that we could just hide the deprecation warnings with pragma like here if we just want a quick fix. It would be better to find a long term solution.

MKL prints an warning to std::cerr at runtime, so I don't think this is the fix we're looking for.

[lhuot]

Right, I am still wondering if we should completely remove support for INPUT/OUTPUT STRIDES and require oneAPI 2024.1 once oneMKL Interface supports FWD/BWD STRIDES. I think it would allow to have just one descriptor for MKLGPU and portFFT backends but I don't know the impact on the other backends. I understand this would be a breaking change. I'm curious to know the opinion of @lhuot on the matter.

We should support the latest release as documented in the README so imo it is okay to require oneMKL 2024.1 and cleanup the code to use only fwd/bwd_stride when using the SYCL API from the Intel oneMKL library.

[hjabird]

I'll remove the support for 2024.0.

In the meantime, it looks like I've somehow broken my implementation, so please hold off reviewing!

[hjabird]

I've updated the PR to remove support for earlier than 2024.1, and also to sort out an issue with exception handling. Test results are here: testres.txt

Q: How come it was working earlier and then you said it was broken?
A: I previously fixed this issue, giving the results earlier, but failed push the relavent commit. Then I deleted it locally and and had to redo the fix all over.