Classes
class	ryujin::SynchronizationDispatch

class	ryujin::Scope

Functions
template<int dim>
void	ryujin::Checkpointing::load_mesh (Discretization< dim > &discretization, const std::string &base_name)

template<int dim, typename Number , int n_comp, int simd_length>
void	ryujin::Checkpointing::load_state_vector (const OfflineData< dim, Number > &offline_data, const std::string &base_name, MultiComponentVector< Number, n_comp, simd_length > &U, Number &t, unsigned int &output_cycle, const MPI_Comm &mpi_communicator)

template<int dim, typename Number , int n_comp, int simd_length>
void	ryujin::Checkpointing::write_checkpoint (const OfflineData< dim, Number > &offline_data, const std::string &base_name, const MultiComponentVector< Number, n_comp, simd_length > &U, const Number t, const unsigned int output_cycle, const MPI_Comm &mpi_communicator)

template<int dim, typename Number , typename Callable >
ToFunction< dim, Number, Callable >	ryujin::to_function (const Callable &callable, const unsigned int k)

Various convenience functions and macros
#define	AssertThrowSIMD(variable, condition, exception)

#define	ACCESSOR_READ_ONLY(member)

#define	ACCESSOR(member)

#define	ACCESSOR_READ_ONLY_NO_DEREFERENCE(member)

#define	ASM_LABEL(label) asm("#" label);

OpenMP parallel for macros
Intended use: // serial work RYUJIN_PARALLEL_REGION_BEGIN // per thread work and thread-local storage declarations RYUJIN_OMP_FOR for (unsigned int i = 0; i < size_internal; i += simd_length) { // parallel for loop that is statically distributed on all available // worker threads by slicing the interval [0,size_internal) } RYUJIN_PARALLEL_REGION_END RYUJIN_PARALLEL_REGION_BEGIN #define RYUJIN_PARALLEL_REGION_BEGIN Definition: openmp.h:54 RYUJIN_OMP_FOR #define RYUJIN_OMP_FOR Definition: openmp.h:70 RYUJIN_PARALLEL_REGION_END #define RYUJIN_PARALLEL_REGION_END Definition: openmp.h:63
#define	RYUJIN_PRAGMA(x) _Pragma(#x)

#define	RYUJIN_PARALLEL_REGION_BEGIN

#define	RYUJIN_PARALLEL_REGION_END }

#define	RYUJIN_OMP_FOR RYUJIN_PRAGMA(omp for)

#define	RYUJIN_OMP_FOR_NOWAIT RYUJIN_PRAGMA(omp for nowait)

#define	RYUJIN_OMP_BARRIER RYUJIN_PRAGMA(omp barrier)

#define	RYUJIN_OMP_CRITICAL RYUJIN_PRAGMA(omp critical)

#define	RYUJIN_OMP_SINGLE RYUJIN_PRAGMA(omp single)

#define	RYUJIN_LIKELY(x) (__builtin_expect(!!(x), 1))

#define	RYUJIN_UNLIKELY(x) (__builtin_expect(!!(x), 0))

Detailed Description

Miscellaneous helper functions, macros and classes.

Macro Definition Documentation

◆ AssertThrowSIMD

#define AssertThrowSIMD	(	variable,
		condition,
		exception
	)

Value:

  if constexpr (std::is_same<                                                  \
                    typename std::remove_const<decltype(variable)>::type,      \
                    double>::value ||                                          \
                std::is_same<                                                  \
                    typename std::remove_const<decltype(variable)>::type,      \
                    float>::value) {                                           \
    AssertThrow(condition(variable), exception);                               \
  } else {                                                                     \
    for (unsigned int k = 0; k < decltype(variable)::size(); ++k) {            \
      AssertThrow(condition((variable)[k]), exception);                        \
    }                                                                          \
  }

Mixed serial/SIMD variant of the dealii AssertThrow macro. If variable is just a plain double or float, then this macro defaults to a simple call to dealii::AssertThrow(condition(variable), exception). Otherwise (if decltype(variable) has a subscript operator operator[], the dealii::AssertThrow macro is expanded for all components of the variable.

Definition at line 119 of file convenience_macros.h.

◆ ACCESSOR_READ_ONLY

#define ACCESSOR_READ_ONLY ( member )

Value:

  inline auto &member() const                                                  \
  {                                                                            \
    return dereference(member##_);                                             \
  }

A convenience macro that automatically writes out an accessor (or getter) function:

const Foo& bar() const { return bar_; }

or

const Foo& bar() const { return *bar_; }

depending on whether bar_ can be dereferenced, or not.

Definition at line 185 of file convenience_macros.h.

◆ ACCESSOR

#define ACCESSOR ( member )

Value:

  inline auto &member()                                                        \
  {                                                                            \
    return dereference(member##_);                                             \
  }

Variant of the macro above that returns a mutable reference.

Definition at line 197 of file convenience_macros.h.

◆ ACCESSOR_READ_ONLY_NO_DEREFERENCE

#define ACCESSOR_READ_ONLY_NO_DEREFERENCE ( member )

Value:

  inline const auto &member() const                                            \
  {                                                                            \
    return member##_;                                                          \
  }

Variant of the macro above that does not attempt to dereference the underlying object.

Definition at line 210 of file convenience_macros.h.

◆ ASM_LABEL

#define ASM_LABEL ( label ) asm("#" label);

Injects a label into the generated assembly.

Definition at line 222 of file convenience_macros.h.

◆ RYUJIN_PRAGMA

#define RYUJIN_PRAGMA ( x ) _Pragma(#x)

Macro expanding to a #pragma directive that can be used in other preprocessor macro definitions.

Definition at line 47 of file openmp.h.

◆ RYUJIN_PARALLEL_REGION_BEGIN

#define RYUJIN_PARALLEL_REGION_BEGIN

Value:

RYUJIN_PRAGMA(omp parallel default(shared)) \

{

RYUJIN_PRAGMA

#define RYUJIN_PRAGMA(x)

Definition: openmp.h:47

Begin an openmp parallel region.

Definition at line 54 of file openmp.h.

◆ RYUJIN_PARALLEL_REGION_END

#define RYUJIN_PARALLEL_REGION_END }

End an openmp parallel region.

Definition at line 63 of file openmp.h.

◆ RYUJIN_OMP_FOR

#define RYUJIN_OMP_FOR RYUJIN_PRAGMA(omp for)

Enter a parallel for loop.

Definition at line 70 of file openmp.h.

◆ RYUJIN_OMP_FOR_NOWAIT

#define RYUJIN_OMP_FOR_NOWAIT RYUJIN_PRAGMA(omp for nowait)

Enter a parallel for loop with "nowait" declaration, i.e., the end of the for loop does not include an implicit thread synchronization barrier.

Definition at line 79 of file openmp.h.

◆ RYUJIN_OMP_BARRIER

#define RYUJIN_OMP_BARRIER RYUJIN_PRAGMA(omp barrier)

Declare an explicit Thread synchronization barrier.

Definition at line 86 of file openmp.h.

◆ RYUJIN_OMP_CRITICAL

#define RYUJIN_OMP_CRITICAL RYUJIN_PRAGMA(omp critical)

Annotate a critical section that has to be accessed sequentially.

Definition at line 93 of file openmp.h.

◆ RYUJIN_OMP_SINGLE

#define RYUJIN_OMP_SINGLE RYUJIN_PRAGMA(omp single)

Annotate a section that has to be executed on one thread only.

Definition at line 100 of file openmp.h.

◆ RYUJIN_LIKELY

#define RYUJIN_LIKELY ( x ) (__builtin_expect(!!(x), 1))

Compiler hint annotating a boolean to be likely true.

Intended use:

if (RYUJIN_LIKELY(thread_ready == true)) {
  // likely branch
}

Note: The performance penalty of incorrectly marking a condition as likely is severe. Use only if the condition is almost always true.

Definition at line 116 of file openmp.h.

◆ RYUJIN_UNLIKELY

#define RYUJIN_UNLIKELY ( x ) (__builtin_expect(!!(x), 0))

Compiler hint annotating a boolean expression to be likely false.

Intended use:

if (RYUJIN_UNLIKELY(thread_ready == false)) {
  // unlikely branch
}

Note: The performance penalty of incorrectly marking a condition as unlikely is severe. Use only if the condition is almost always false.

Definition at line 132 of file openmp.h.

Function Documentation

◆ load_mesh()

template<int dim>

void ryujin::Checkpointing::load_mesh	(	Discretization< dim > &	discretization,
		const std::string &	base_name
	)

Performs a resume operation. Given a base_name the function tries to locate correponding checkpoint files and will read in the saved mesh and reinitializes the Discretization object.

Definition at line 39 of file checkpointing.h.

References ryujin::Discretization< dim >::prepare(), and ryujin::Discretization< dim >::triangulation().

Referenced by ryujin::TimeLoop< Description, dim, Number >::run().

◆ load_state_vector()

template<int dim, typename Number , int n_comp, int simd_length>

void ryujin::Checkpointing::load_state_vector	(	const OfflineData< dim, Number > &	offline_data,
		const std::string &	base_name,
		MultiComponentVector< Number, n_comp, simd_length > &	U,
		Number &	t,
		unsigned int &	output_cycle,
		const MPI_Comm &	mpi_communicator
	)

Performs a resume operation. Given a base_name the function tries to locate correponding checkpoint files and will read in the saved state U at saved time t with saved output cycle output_cycle.

Definition at line 61 of file checkpointing.h.

References ryujin::OfflineData< dim, Number >::dof_handler(), ryujin::MultiComponentVector< Number, n_comp, simd_length >::insert_component(), and ryujin::OfflineData< dim, Number >::scalar_partitioner().

Referenced by ryujin::TimeLoop< Description, dim, Number >::run().

◆ write_checkpoint()

template<int dim, typename Number , int n_comp, int simd_length>

void ryujin::Checkpointing::write_checkpoint	(	const OfflineData< dim, Number > &	offline_data,
		const std::string &	base_name,
		const MultiComponentVector< Number, n_comp, simd_length > &	U,
		const Number	t,
		const unsigned int	output_cycle,
		const MPI_Comm &	mpi_communicator
	)

Writes out a checkpoint to disk. Given a base_name and a current state U at time t and output cycle output_cycle the function writes out the state to disk using boost::archive for serialization.

Todo:: Some day, we should refactor this into a class and do something smarter...

Definition at line 145 of file checkpointing.h.

References ryujin::OfflineData< dim, Number >::discretization(), ryujin::OfflineData< dim, Number >::dof_handler(), ryujin::MultiComponentVector< Number, n_comp, simd_length >::extract_component(), and ryujin::OfflineData< dim, Number >::scalar_partitioner().

Referenced by ryujin::TimeLoop< Description, dim, Number >::output().

◆ to_function()

template<int dim, typename Number , typename Callable >

ToFunction< dim, Number, Callable > ryujin::to_function	(	const Callable &	callable,
		const unsigned int	k
	)

Convenience wrapper that creates a (scalar) dealii::Function object out of a (fairly general) callable object returning array-like values. An example usage is given by the interpolation of initial values performed in InitialValues::interpolate();

for(unsigned int i = 0; i < problem_dimension; ++i)
  dealii::VectorTools::interpolate(
    dof_handler,
    to_function<dim, Number>(callable, i),
    U[i]);

Parameters

callable	A callable object that provides an `operator(const Point<dim> &)` and returns an array or rank-1 tensor. More precisely, the return type must have a subscript operator `operator[]`.
k	Index describing the component that is returned by the function object.

Definition at line 68 of file convenience_macros.h.

Classes

Functions

Various convenience functions and macros

OpenMP parallel for macros

Detailed Description

Macro Definition Documentation

◆ AssertThrowSIMD

◆ ACCESSOR_READ_ONLY

◆ ACCESSOR

◆ ACCESSOR_READ_ONLY_NO_DEREFERENCE

◆ ASM_LABEL

◆ RYUJIN_PRAGMA

◆ RYUJIN_PARALLEL_REGION_BEGIN

◆ RYUJIN_PARALLEL_REGION_END

◆ RYUJIN_OMP_FOR

◆ RYUJIN_OMP_FOR_NOWAIT

◆ RYUJIN_OMP_BARRIER

◆ RYUJIN_OMP_CRITICAL

◆ RYUJIN_OMP_SINGLE

◆ RYUJIN_LIKELY

◆ RYUJIN_UNLIKELY

Function Documentation

◆ load_mesh()

◆ load_state_vector()

◆ write_checkpoint()

◆ to_function()