serif::utils::parallel Namespace Reference

Namespaces
namespace	detail

Functions
int	worker_count () noexcept
	Workers an element loop may use: OpenMP's thread limit (OMP_NUM_THREADS), or 1 without OpenMP or inside an enclosing parallel region.
template<typename BodyT>
void	for_each_index (const int count, BodyT &&body)
	Call body(index, worker) once for every index in [0, count), possibly concurrently.
void	element_transformation (const ::mfem::Mesh &mesh, int element, ::mfem::IsoparametricTransformation &transformation)
	Mesh::GetElementTransformation(element, transformation), safe in concurrent element loops.

Function Documentation

element_transformation()

void serif::utils::parallel::element_transformation	(	const ::mfem::Mesh &	mesh,
		int	element,
		::mfem::IsoparametricTransformation &	transformation )

Mesh::GetElementTransformation(element, transformation), safe in concurrent element loops.

Note

MFEM's version gathers curved-mesh nodes through FiniteElementSpace::GetElementVDofs(int, Array&), which writes the space's shared mutable DofTransformation. This copy performs the same steps with a local DofTransformation, so the transformation and every value derived from it are identical.

for_each_index()

template<typename BodyT>

void serif::utils::parallel::for_each_index	(	const int	count,
		BodyT &&	body )

Call body(index, worker) once for every index in [0, count), possibly concurrently.

Note

Results do not depend on the thread count when each call writes only index-owned results and scratch owned by its worker (0 <= worker < worker_count()). Every index runs; the exception of the lowest throwing index is then rethrown, which is the one a serial loop would have raised.

worker_count()

int serif::utils::parallel::worker_count ( )

nodiscardnoexcept

Workers an element loop may use: OpenMP's thread limit (OMP_NUM_THREADS), or 1 without OpenMP or inside an enclosing parallel region.