Demo of constexpr/nonconstexpr bench

[fredroy]

Obviously, if the constexpr function compiles, then the result is instantaneous

BM_Matrix_typemat3x3f_noconstexpr_bench/512      0.443 us        0.445 us      1544828
BM_Matrix_typemat3x3f_constexpr_bench/512        0.002 us        0.002 us    373333333

+ add a constexpr random generator (from https://www.youtube.com/watch?v=rpn_5Mrrxf8 😎 of course)

This branch needs SOFA's Mat.h to be fixed, as std::abs is not constexpr (officially)

[alxbilger]

Could you make the seed fixed so that benchmarks are reproducible from a compilation to another?

[alxbilger]

@fredroy there is a conflict to solve.

Also, could you check that the random numbers are always the same?

[fredroy]

@fredroy there is a conflict to solve.

Also, could you check that the random numbers are always the same?

same between what?
in any case it wont be the same between each compilation as the seed is based on TIME (which I suppose is the time at the moment of the compilation)

[alxbilger]

That's what I mean. If you compile twice the benchmark you won't have the same numbers, so the benchmarks are not rigorously reproducible. If the seed is, let's say, 12, the numbers will stay the same between two compilations

[fredroy]

That's what I mean. If you compile twice the benchmark you won't have the same numbers, so the benchmarks are not rigorously reproducible. If the seed is, let's say, 12, the numbers will stay the same between two compilations

I see, so we should add a fixed seed or something like that

[alxbilger]

Yes, we should be able to provide a seed to the random numbers generator. And in the benchmarks, use a constant seed.