Add a batched operator() to SplineEvaluator without the coordinates argument

[EmilyBourne]

The batched operator() in SplineEvaluator takes the coordinates for the evaluation as argument:

ddc/include/ddc/kernels/splines/spline_evaluator.hpp

Lines 263 to 273 in 98c1fc3

	template <class Layout1, class Layout2, class Layout3, class... CoordsDims>
	void operator()(
	ddc::ChunkSpan<double, batched_evaluation_domain_type, Layout1, memory_space> const
	spline_eval,
	ddc::ChunkSpan<
	ddc::Coordinate<CoordsDims...> const,
	batched_evaluation_domain_type,
	Layout2,
	memory_space> const coords_eval,
	ddc::ChunkSpan<double const, batched_spline_domain_type, Layout3, memory_space> const
	spline_coef) const

This is useful for example for the Semi-Lagrangian scheme. However it is more common to use the operator() to evaluate the spline at the coordinates associated with the function. It is therefore common to call the function with the following code:

ddc::ChunkSpan<double const, batched_spline_domain_type> const spline_coef(...);
ddc::Chunk<double, batched_evaluation_domain_type> const spline_eval(eval_domain);
ddc::Chunk<ddc::Coordinate<CoordsDims...>, batched_evaluation_domain_type> coords_eval_alloc(
        eval_domain);
ddc::ChunkSpan<ddc::Coordinate<CoordsDims...>, batched_evaluation_domain_type> coords_eval
        = coords_eval_alloc.span_view();
ddc::parallel_for_each(
        eval_domain,
        KOKKOS_LAMBDA(batched_evaluation_element_type idx) {
            coords_eval(idx) = ddc::coordinate(idx);
        });
spline_evaluator(spline_eval.span_view(), coords_eval.span_cview(), spline_coef);

The lines:

ddc::Chunk<ddc::Coordinate<CoordsDims...>, batched_evaluation_domain_type> coords_eval_alloc(
        eval_domain);
ddc::ChunkSpan<ddc::Coordinate<CoordsDims...>, batched_evaluation_domain_type> coords_eval
        = coords_eval_alloc.span_view();
ddc::parallel_for_each(
        eval_domain,
        KOKKOS_LAMBDA(batched_evaluation_element_type idx) {
            coords_eval(idx) = ddc::coordinate(idx);
        });

lead to excessive code repetition and unnecessary memory allocation.

The addition of a new operator() defined as:

 template <class Layout1, class Layout2> 
 void operator()( 
         ddc::ChunkSpan<double, batched_evaluation_domain_type, Layout1, memory_space> const 
                 spline_eval, 
         ddc::ChunkSpan<double const, batched_spline_domain_type, Layout2, memory_space> const 
                 spline_coef) const 

would avoid this.

[tpadioleau]

Seems reasonable yes