Implement Kokkos::Experimental::{min,max,minmax}(std::initializer_list)

[dalg24]

Rational: reviewed some code downstream that was doing things like min(a, min(b, c)

Make sure you see the drib-by fixing of a defect in operator== for Kokkos::pair<T1, T2> that was missing the constexpr specifier.
About the overloads I am adding, I just followed closely the design of std::{min,max,minmax} (see https://eel.is/c++draft/alg.min.max)

[fnrizzi]

@dalg24 one thing: minmax in the std is supposed to do this: " If several elements are equivalent to the smallest, the leftmost such element is returned. If several elements are equivalent to the largest, the rightmost such element is returned."
I saw that in your minmax tests you don't check for this but only that the values found are the smallest and max.
I think this is fine because minmax returns the values not iterators.

[masterleinad]

Looks good to me.

[dalg24]

@dalg24 one thing: minmax in the std is supposed to do this: " If several elements are equivalent to the smallest, the leftmost such element is returned. If several elements are equivalent to the largest, the rightmost such element is returned."
I saw that in your minmax tests you don't check for this but only that the values found are the smallest and max.
I think this is fine because minmax returns the values not iterators.

I overlooked it. I think actually this means that my max and my minmax do not follow the standard specs.

[fnrizzi]

as discussed, i think we need to verify/fix behavior

[masterleinad]

Surprisingly, a (different) OpenMPTarget test is failing:

[ RUN      ] openmptarget.unique_token_global
4: /var/jenkins/workspace/Kokkos/core/unit_test/TestUniqueToken.hpp:151: Failure
4: Expected equality of these values:
4:   sum
4:     Which is: 9999990
4:   int64_t(N) * R
4:     Which is: 10000000
4: [  FAILED  ] openmptarget.unique_token_global (261 ms)

[dalg24]

For reference it looks like implementations of the standard library compute the same thing as I did https://godbolt.org/z/qP5Tv3jds

[fnrizzi]

@dalg24
There is a difference in how max behaves when it is by itself vs when it is part of minmax.

• std::min(std::initializer_list):

  • If several values are equivalent to the smallest, returns the LEFTMOST one

• std::max(std::initializer_list):

  • If several values are equivalent to the greatest, returns the LEFTMOST one

• std::minmax(std::initializer_list):

  • If several elements are equivalent to the SMALLEST, the leftmost is returned.
  • If several elements are equivalent to the LARGEST, the rightmost such element is returned.

so min does not change, but max does. See this: https://godbolt.org/z/9EohjbsGT

[fnrizzi]

thanks for getting it through :)