Update and simplify cmake

[DiamonDinoia]

Dear all,

I made some changes on cmake to remove the _static to finufft and cleanup other things in the process.

Changelog:

• -march=native like on msvc
• static libraries by default only (shared replaces static)
• flags are detected by cmake and all the supported ones are enabled

Known issues:

• move msvc march=native to a separate file
• build GPU library does not depend on CPU anymore
• GPU should not depend on FFTW
• clang-cl declspec issue

[DiamonDinoia]

@lu1and10 do you know why windows clang-cl is not building?

[lu1and10]

@lu1and10 do you know why windows clang-cl is not building?

I'm not sure, it seems too much going on with the cmake changes, need to take a look in detail.

[DiamonDinoia]

I think I fixed everything. I think we should use the matrix I use for testing python to test cmake-ci too. This way we can test all os and all compilers.

[janden]

To reduce issues with CMakeLists style, etc., would it make sense to use a tool like cmake-format (https://github.com/cheshirekow/cmake_format) in our pre-commit hook?

[lu1and10]

One note for future: @DiamonDinoia finds that on windows with msvc cl, ducc0 has to compile with /fp:fast, otherwise some tests(run_finufft3d_test_float, run_finufft3dmany_test_float) may fail because of the resulting error is larger than the tolerance.

On the other hand, finufft on windows with msvc cl should not compile with flag /fp:fast, with /fp:fast the test run_dumbinputs_double will result in segfault, because /fp:fast makes values (NaN, +infinity, -infinity, -0.0) may not be propagated or behave strictly according to the IEEE-754 standard

We may need to add this note in the installation doc.

[ahbarnett]

Thanks for summarizing - can you add to the Issue on installation/docs. Thanks! Alex

…

On Wed, Jul 31, 2024 at 6:29 PM Libin Lu ***@***.***> wrote: One note for future: @DiamonDinoia <https://github.com/DiamonDinoia> finds that on windows with msvc cl, ducc0 *has to* compile with */fp:fast*, otherwise some tests(run_finufft3d_test_float, run_finufft3dmany_test_float) may fail because of the resulting error is larger than the tolerance. On the other hand, finufft on windows with msvc cl *should not* compile with flag */fp:fast*, with /fp:fast the test run_dumbinputs_double will result in segfault, because /fp:fast makes values (NaN, +infinity, -infinity, -0.0) may not be propagated or behave strictly according to the IEEE-754 standard <https://learn.microsoft.com/en-us/cpp/build/reference/fp-specify-floating-point-behavior?view=msvc-170#fast> We may need to add this note in the installation doc. — Reply to this email directly, view it on GitHub <#495 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACNZRSWQUG2AXVIDATGP5YDZPFQKHAVCNFSM6AAAAABLC3GJIKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENRRGU3DIMJYGE> . You are receiving this because you commented.Message ID: ***@***.***>

-- *-------------------------------------------------------------------~^`^~._.~' |\ Alex Barnett Center for Computational Mathematics, Flatiron Institute | \ http://users.flatironinstitute.org/~ahb 646-876-5942

[DiamonDinoia]

I added a warning in the docs.