allow constructor for density free formular functor

[psychocoderHPC]

FreeFormulaImpl allows the user functor to define a constructor to move additional information into the profile.
This allow e.g. filtering by cell position relative to the global volume

example

//! example implementation is not supporting moving window
struct RelativeRangeProfile
{
    static constexpr float_X start_ratio = 0.1_X;
    static constexpr float_X end_ratio = 0.3_X;
    static constexpr uint32_t usedDim = 1u;

    float_64 first_cell = 0.0_X;
    float_64 last_cell = 0.0_X;
    RelativeRangeProfile()
    {
        const SubGrid< simDim >& subGrid = Environment< simDim >::get().SubGrid();

        auto globalSize = subGrid.getGlobalDomain().size;
        auto deviceOffset = subGrid.getLocalDomain().offset;
        first_cell = ( static_cast< float_64 >( globalSize[usedDim] ) * start_ratio ) -
            static_cast< float_64 >( deviceOffset[ usedDim ] );
        last_cell = ( static_cast< float_64 >( globalSize[ usedDim ] ) * end_ratio ) -
            static_cast< float_64 >( deviceOffset[ usedDim ] );
    }

    //! homogenous density between [start_ratio;end_ratio) of the global domain
    HDINLINE float_X
    operator()(
        floatD_64 const & position_SI,
        float3_64 const & cellSize_SI
    )
    {
        return position_SI[usedDim] >= first_cell && position_SI[usedDim] < last_cell ? 1.0_X : 0.0_X;
    }
};

// definition of free formula profile
using HomogenousRange = FreeFormulaImpl< RelativeRangeProfile >;

[PrometheusPi]

Great work - I will test it right away.

[sbastrakov]

Looks good to me, not sure if @PrometheusPi has tried this out.

[PrometheusPi]

@psychocoderHPC and @sbastrakov sorry I was busy with creating videos and did not test it yet - will do this tomorrow

[PrometheusPi]

I am testing it right now.

[ax3l]

@PrometheusPi how did the test go? :)

[PrometheusPi]

Just an inbetween remark:
Where do these strange velocity fluctuations on the shear surface come from?

[PrometheusPi]

Just to illustrate this a bit further

Blue is the velocity in one x-direction - red in the other. This strange overlap at the shear surface should not look like this).

This has so far nothing to do with this pull request, but is the current status of the dev to check this pull request.

[PrometheusPi]

The issue seen in the isaac visualization was caused by an error on my input set and the strange behavior of the automatic range setting (see ComputationalRadiationPhysics/isaac#99).

[PrometheusPi]

Setting density only in (..., [0.25, 0.75], ...) works fine:

[PrometheusPi]

Looks good to me.
Is the only algorithmic difference, that the constructor is now called by the template constructor?

[PrometheusPi]

Works like a charm:

[sbastrakov]

@PrometheusPi on your constructor question - yes, the idea is that with this PR one can pass arbitrary parameters to a constructor of the user-provided functor through the constructor of FreeFormulaImpl class.

Edit: upon a further look, it is not actually arbitrary but the choice is only between the default constructor and the one taking a uint32_t currentStep. However, it looks like support for arbitrary parameters can be added rather easily when needed.

[PrometheusPi]

@sbastrakov Thank you for the explanation.

[PrometheusPi]

@ComputationalRadiationPhysics/picongpu-maintainers after this pull request has been added - should the KHI example be refactored to use 4 instead of 2 species?

[PrometheusPi]

@sbastrakov Since @psychocoderHPC is on vacation, I quickly fixed the typo.

[PrometheusPi]

discussed offline with @sbastrakov:
The documentation of these "template concepts" needs to be done in a wider and more general scope. Thus we will need to postpone this to a future pull request.