fix: cuda atomicMin and atomicMax for <double>s

[Moelf]

fix #3528

the root cause is that the trick for float does not work for double due to, eh, IEEE754 ... My understanding is you can't cast double into long long and still have a total order like you would have for float -> int. See also: https://stackoverflow.com/questions/55140908/can-anybody-help-me-with-atomicmin-function-syntax-for-cuda/55145948#55145948

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 82.41%. Comparing base (b749e49) to head (548ecd8). Report is 366 commits behind head on main.

Additional details and impacted files

see 195 files with indirect coverage changes

:rocket: New features to boost your workflow:

• :snowflake: Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Moelf]

related question, how is the CUDA backend tested? If there's CI I would add a test too, but I don't see any... yet

[ikrommyd]

It’s not tested in ci yet. @ariostas has an open PR for that. Typically @ianna runs the tests offline. I could sometimes also do it but I’m not in the US. I don’t have access to my desktop.

[ianna]

related question, how is the CUDA backend tested? If there's CI I would add a test too, but I don't see any... yet

@ariostas has opened a PR for CUDA integration testing

[ariostas]

The GPU CI is live now!

[ianna]

The GPU CI is live now!

@ariostas - many thanks! I think, we need to add tests-cuda-kernels and tests-cuda-kernels-explicit - they are generated.

[ariostas]

I think, we need to add tests-cuda-kernels and tests-cuda-kernels-explicit - they are generated.

Oh okay, I missed that part. I assumed that it only generated cpu kernels

[ianna]

@all-contributors please add @Moelf for code

[allcontributors[bot]]

@ianna

I've put up a pull request to add @Moelf! :tada: