BoxLoops Namespace Reference

Namespace for encapsulating loop types. More...

Functions
template<int Si, int Sj, int Sk, typename Functor >
ALWAYS_INLINE void	loop (const Box &a_computeBox, Functor &&kernel)
	Launch a C++ kernel over a regular grid with compile-time per-dimension strides.
template<typename Functor >
ALWAYS_INLINE void	loop (const IntVectSet &a_ivs, Functor &&a_kernel)
	Launch a C++ kernel over a subset of cells.
template<typename Functor >
ALWAYS_INLINE void	loop (const DenseIntVectSet &a_ivs, Functor &&a_kernel)
	Launch a C++ kernel over a subset of cells.
template<typename Functor >
ALWAYS_INLINE void	loop (VoFIterator &a_iter, Functor &&a_kernel)
	Launch a C++ kernel over cut-cells.
template<typename Functor >
ALWAYS_INLINE void	loop (FaceIterator &a_iter, Functor &&a_kernel)
	Launch a C++ kernel over faces.
template<typename T , typename Functor >
ALWAYS_INLINE void	loop (const Vector< T > &a_subset, Functor &&a_kernel)
	Launch a C++ kernel over a specified subset of something.

Detailed Description

Namespace for encapsulating loop types.

BoxLoops provides several overloads of loop for iterating over structured and unstructured grid data. The general pattern is:

BoxLoops::loop(domain, kernel);

where kernel is a callable (lambda or functor) taking the element type for that overload.

Box loop — compile-time stride: Strides are non-type template parameters, making their value visible to the compiler so the inner loop can be auto-vectorized. All strides must be >= 1 (enforced by static_assert). Use D_DECL for dimension-independent call sites. Unit-stride iteration is D_DECL(1,1,1):

BoxLoops::loop<D_DECL(1, 1, 1)>(box, [&](const IntVect& iv) {
    phi(iv, 0) = 0.0;
});
 
BoxLoops::loop<D_DECL(2, 2, 2)>(colorBox, [&](const IntVect& iv) {
    phi(iv, 0) = 0.0;
});

VoF, face, and set iterators:

BoxLoops::loop(vofIt,  [&](const VolIndex&  vof)  { ... });
BoxLoops::loop(faceIt, [&](const FaceIndex& face) { ... });
BoxLoops::loop(ivs,    [&](const IntVect&   iv)   { ... });

Function Documentation

loop() [1/6]

template<int Si, int Sj, int Sk, typename Functor >

ALWAYS_INLINE void BoxLoops::loop	(	const Box &	a_computeBox,
		Functor &&	kernel
	)

Launch a C++ kernel over a regular grid with compile-time per-dimension strides.

Strides are encoded as non-type template parameters, making them visible to the compiler as constants and enabling auto-vectorization of the innermost loop. All strides must be >= 1 (enforced by static_assert).

Note: IntVect cannot be a non-type template parameter in C++14, so per-dimension int parameters are used instead. Call sites can use D_DECL for dimension independence, e.g.:

BoxLoops::loop<D_DECL(2, 2, 2)>(box, kernel);

Parameters

[in]	a_computeBox	Computation box
[in]	kernel	Kernel

loop() [2/6]

template<typename Functor >

ALWAYS_INLINE void BoxLoops::loop	(	const DenseIntVectSet &	a_ivs,
		Functor &&	a_kernel
	)

Launch a C++ kernel over a subset of cells.

Parameters

[in,out]	a_kernel	Kernel to launch
[in]	a_ivs	Ivs

loop() [3/6]

template<typename Functor >

ALWAYS_INLINE void BoxLoops::loop	(	const IntVectSet &	a_ivs,
		Functor &&	a_kernel
	)

Launch a C++ kernel over a subset of cells.

Parameters

[in,out]	a_kernel	Kernel to launch
[in]	a_ivs	Ivs

loop() [4/6]

template<typename T , typename Functor >

ALWAYS_INLINE void BoxLoops::loop	(	const Vector< T > &	a_subset,
		Functor &&	a_kernel
	)

Launch a C++ kernel over a specified subset of something.

Parameters

[in,out]	a_subset	Subset. E.g. a std::vector<VolIndex>
[in,out]	a_kernel	Kernel to launch

loop() [5/6]

template<typename Functor >

ALWAYS_INLINE void BoxLoops::loop	(	FaceIterator &	a_iter,
		Functor &&	a_kernel
	)

Launch a C++ kernel over faces.

Parameters

[in,out]	a_kernel	Kernel to launch
[in]	a_iter	Iter

loop() [6/6]

template<typename Functor >

ALWAYS_INLINE void BoxLoops::loop	(	VoFIterator &	a_iter,
		Functor &&	a_kernel
	)

Launch a C++ kernel over cut-cells.

Parameters

[in,out]	a_kernel	Kernel to launch
[in]	a_iter	Iter