SIMD

Classes
class	ryujin::MultiComponentVector< Number, n_comp, simd_length >
struct	ryujin::get_value_type< T >

Functions
std::shared_ptr< const dealii::Utilities::MPI::Partitioner >	ryujin::create_vector_partitioner (const std::shared_ptr< const dealii::Utilities::MPI::Partitioner > &scalar_partitioner, const unsigned int n_components)

Type traits and packed index handling
template<typename T >
unsigned int	ryujin::get_stride_size = 1
template<unsigned int length, typename Functor >
DEAL_II_ALWAYS_INLINE auto	ryujin::generate_iterators (Functor f) -> std::array< auto, length >
template<typename T >
DEAL_II_ALWAYS_INLINE void	ryujin::increment_iterators (T &iterators)

Transcendental and other mathematical operations
template<typename Number >
DEAL_II_ALWAYS_INLINE Number	ryujin::positive_part (const Number number)
template<typename Number >
DEAL_II_ALWAYS_INLINE Number	ryujin::negative_part (const Number number)
template<int N, typename T >
T	ryujin::fixed_power (const T x)
template<typename T >
T	ryujin::pow (const T x, const T b)
template<typename T , std::size_t width>
dealii::VectorizedArray< T, width >	ryujin::pow (const dealii::VectorizedArray< T, width > x, const T b)
template<typename T , std::size_t width>
dealii::VectorizedArray< T, width >	ryujin::pow (const dealii::VectorizedArray< T, width > x, const dealii::VectorizedArray< T, width > b)
template<typename T >
T	ryujin::fast_pow (const T x, const T b, const Bias bias=Bias::none)
template<typename T , std::size_t width>
dealii::VectorizedArray< T, width >	ryujin::fast_pow (const dealii::VectorizedArray< T, width > x, const T b, const Bias bias=Bias::none)
template<typename T , std::size_t width>
dealii::VectorizedArray< T, width >	ryujin::fast_pow (const dealii::VectorizedArray< T, width > x, const dealii::VectorizedArray< T, width > b, const Bias bias=Bias::none)

SIMD based access to vectors and arrays of vectors
template<typename T , typename V >
DEAL_II_ALWAYS_INLINE T	ryujin::get_entry (const V &vector, unsigned int i)
template<typename T , typename T2 >
DEAL_II_ALWAYS_INLINE T	ryujin::get_entry (const std::vector< T2 > &vector, unsigned int i)
template<typename T , typename V >
DEAL_II_ALWAYS_INLINE T	ryujin::get_entry (const V &vector, const unsigned int *js)
template<typename T , typename T2 >
DEAL_II_ALWAYS_INLINE T	ryujin::get_entry (const std::vector< T2 > &vector, const unsigned int *js)
template<typename T , typename V >
DEAL_II_ALWAYS_INLINE void	ryujin::write_entry (V &vector, const T &values, unsigned int i)
template<typename T , typename T2 >
DEAL_II_ALWAYS_INLINE void	ryujin::write_entry (std::vector< T2 > &vector, const T &values, unsigned int i)
template<int rank, int dim, std::size_t width, typename Number >
DEAL_II_ALWAYS_INLINE dealii::Tensor< rank, dim, Number >	ryujin::serialize_tensor (const dealii::Tensor< rank, dim, dealii::VectorizedArray< Number, width > > &vectorized, const unsigned int k)
template<int rank, int dim, typename Number >
DEAL_II_ALWAYS_INLINE dealii::Tensor< rank, dim, Number >	ryujin::serialize_tensor (const dealii::Tensor< rank, dim, Number > &serial, const unsigned int k)
template<int rank, int dim, std::size_t width, typename Number >
DEAL_II_ALWAYS_INLINE void	ryujin::assign_serial_tensor (dealii::Tensor< rank, dim, dealii::VectorizedArray< Number, width > > &result, const dealii::Tensor< rank, dim, Number > &serial, const unsigned int k)
template<int rank, int dim, typename Number >
DEAL_II_ALWAYS_INLINE void	ryujin::assign_serial_tensor (dealii::Tensor< rank, dim, Number > &result, const dealii::Tensor< rank, dim, Number > &serial, const unsigned int k)

Detailed Description

SIMD related functions and classes.

Function Documentation

create_vector_partitioner()

std::shared_ptr< const dealii::Utilities::MPI::Partitioner > ryujin::create_vector_partitioner	(	const std::shared_ptr< const dealii::Utilities::MPI::Partitioner > &	scalar_partitioner,
		const unsigned int	n_components
	)

This function takes a scalar MPI partitioner scalar_partitioner as argument and returns a shared pointer to a new "vector" multicomponent partitioner that defines storage and MPI synchronization for a vector consisting of n_comp components. The vector partitioner is intended to efficiently store non-scalar vectors such as the state vectors U. Let (U_i)_k denote the k-th component of a state vector element U_i, we then store

\begin{align} (U_0)_0, (U_0)_1, (U_0)_2, (U_0)_3, (U_0)_4, (U_1)_0, (U_1)_1, (U_1)_2, (U_1)_3, (U_1)_4, \ldots \end{align}

Note

This function is used to efficiently set up a single vector partitioner in OfflineData used in all MultiComponentVector instances.

generate_iterators()

template<unsigned int length, typename Functor >

DEAL_II_ALWAYS_INLINE auto ryujin::generate_iterators ( Functor  f ) -> std::array<auto, length>

inline

Given a callable object f(k), this function creates a std::array with elements initialized as follows:

{ f(0) , f(1) , ... , f(length - 1) }

We use this function to create an array of sparsity iterators that cannot be default initialized.

Definition at line 81 of file simd.h.

increment_iterators()

template<typename T >

DEAL_II_ALWAYS_INLINE void ryujin::increment_iterators ( T &  iterators )

inline

Increment all iterators in an std::array simultaneously.

Definition at line 94 of file simd.h.

positive_part()

template<typename Number >

DEAL_II_ALWAYS_INLINE Number ryujin::positive_part ( const Number  number )

inline

Return the positive part of a number.

Definition at line 112 of file simd.h.

Referenced by ryujin::Euler::RiemannSolver< dim, Number >::compute_gap(), ryujin::ShallowWater::RiemannSolver< dim, Number >::compute_h_star(), ryujin::ShallowWater::RiemannSolver< dim, Number >::compute_lambda(), ryujin::Euler::RiemannSolver< dim, Number >::compute_lambda(), ryujin::EulerAEOS::RiemannSolver< dim, Number >::compute_lambda(), ryujin::ShallowWater::HyperbolicSystemView< dim, Number >::inverse_water_depth_mollified(), ryujin::ShallowWater::RiemannSolver< dim, Number >::lambda1_minus(), ryujin::Euler::RiemannSolver< dim, Number >::lambda1_minus(), ryujin::EulerAEOS::RiemannSolver< dim, Number >::lambda1_minus(), ryujin::Euler::RiemannSolver< dim, Number >::lambda3_plus(), ryujin::EulerAEOS::RiemannSolver< dim, Number >::lambda3_plus(), ryujin::ShallowWater::RiemannSolver< dim, Number >::lambda3_plus(), ryujin::Euler::Limiter< dim, Number >::limit(), ryujin::EulerAEOS::Limiter< dim, Number >::limit(), ryujin::ScalarConservation::Limiter< dim, Number >::limit(), ryujin::ShallowWater::Limiter< dim, Number >::limit(), ryujin::EulerAEOS::RiemannSolver< dim, Number >::p_star_interpolated(), ryujin::EulerAEOS::RiemannSolver< dim, Number >::p_star_RS_full(), ryujin::EulerAEOS::RiemannSolver< dim, Number >::p_star_SS_full(), and ryujin::Euler::RiemannSolver< dim, Number >::p_star_two_rarefaction().

negative_part()

template<typename Number >

DEAL_II_ALWAYS_INLINE Number ryujin::negative_part ( const Number  number )

inline

Return the negative part of a number.

Definition at line 124 of file simd.h.

Referenced by ryujin::NavierStokes::ParabolicSolver< Description, dim, Number >::backward_euler_step(), ryujin::Euler::RiemannSolver< dim, Number >::compute_gap(), ryujin::ShallowWater::RiemannSolver< dim, Number >::compute_lambda(), ryujin::Euler::RiemannSolver< dim, Number >::compute_lambda(), ryujin::EulerAEOS::RiemannSolver< dim, Number >::compute_lambda(), ryujin::Euler::Limiter< dim, Number >::limit(), ryujin::EulerAEOS::Limiter< dim, Number >::limit(), ryujin::ScalarConservation::Limiter< dim, Number >::limit(), and ryujin::ShallowWater::Limiter< dim, Number >::limit().

fixed_power()

template<int N, typename T >

T ryujin::fixed_power ( const T  x )

inline

A wrapper around dealii::Utilities::fixed_power. We use a wrapper instead of calling the function directly so that we can easily change the implementation at one central place.

Definition at line 138 of file simd.h.

pow() [1/3]

template<typename T >

T ryujin::pow	(	const T	x,
		const T	b
	)

Custom serial pow function.

Referenced by ryujin::ShallowWaterInitialStates::SlopingFriction< Description, dim, Number >::compute(), ryujin::EulerInitialStates::Rarefaction< Description, dim, Number >::compute(), ryujin::EulerInitialStates::IsentropicVortex< Description, dim, Number >::compute(), ryujin::EulerInitialStates::LeBlanc< Description, dim, Number >::compute(), ryujin::ShallowWaterInitialStates::FlowOverBump< Description, dim, Number >::compute(), ryujin::TimeLoop< Description, dim, Number >::compute_error(), ryujin::Geometries::Airfoil< dim >::create_triangulation(), ryujin::Euler::RiemannSolver< dim, Number >::df(), ryujin::Euler::RiemannSolver< dim, Number >::f(), ryujin::fast_pow_impl(), ryujin::Euler::HyperbolicSystemView< dim, Number >::harten_entropy(), ryujin::Euler::HyperbolicSystemView< dim, Number >::harten_entropy_derivative(), ryujin::Euler::Limiter< dim, Number >::limit(), ryujin::EulerAEOS::Limiter< dim, Number >::limit(), ryujin::Euler::HyperbolicSystemView< dim, Number >::mathematical_entropy(), ryujin::Euler::HyperbolicSystemView< dim, Number >::mathematical_entropy_derivative(), ryujin::EulerAEOS::RiemannSolver< dim, Number >::p_star_interpolated(), ryujin::EulerAEOS::RiemannSolver< dim, Number >::p_star_RS_full(), ryujin::EulerAEOS::RiemannSolver< dim, Number >::p_star_SS_full(), ryujin::Euler::RiemannSolver< dim, Number >::p_star_two_rarefaction(), ryujin::Euler::HyperbolicSystemView< dim, Number >::precomputation_loop(), ryujin::Euler::HyperbolicSystemView< dim, Number >::prescribe_riemann_characteristic(), ryujin::Manifolds::GradingManifold< dim >::pull_back(), ryujin::Manifolds::GradingManifold< dim >::push_forward(), ryujin::EulerInitialStates::Rarefaction< Description, dim, Number >::Rarefaction(), ryujin::Euler::HyperbolicSystemView< dim, Number >::specific_entropy(), ryujin::EulerAEOS::HyperbolicSystemView< dim, Number >::surrogate_harten_entropy(), ryujin::EulerAEOS::HyperbolicSystemView< dim, Number >::surrogate_harten_entropy_derivative(), and ryujin::EulerAEOS::HyperbolicSystemView< dim, Number >::surrogate_specific_entropy().

pow() [2/3]

template<typename T , std::size_t width>

dealii::VectorizedArray< T, width > ryujin::pow	(	const dealii::VectorizedArray< T, width >	x,
		const T	b
	)

Custom implementation of a vectorized pow function.

Definition at line 238 of file simd.template.h.

References ryujin::pow().

pow() [3/3]

template<typename T , std::size_t width>

dealii::VectorizedArray< T, width > ryujin::pow	(	const dealii::VectorizedArray< T, width >	x,
		const dealii::VectorizedArray< T, width >	b
	)

Custom implementation of a vectorized pow function with vectorized exponent.

Definition at line 247 of file simd.template.h.

References ryujin::pow().

fast_pow() [1/3]

template<typename T >

T ryujin::fast_pow	(	const T	x,
		const T	b,
		const Bias	bias = Bias::none
	)

Custom serial approximate pow function.

fast_pow() [2/3]

template<typename T , std::size_t width>

dealii::VectorizedArray< T, width > ryujin::fast_pow	(	const dealii::VectorizedArray< T, width >	x,
		const T	b,
		const Bias	bias = Bias::none
	)

Custom implementation of an approximate, vectorized pow function.

Definition at line 299 of file simd.template.h.

References ryujin::fast_pow_impl().

fast_pow() [3/3]

template<typename T , std::size_t width>

dealii::VectorizedArray< T, width > ryujin::fast_pow	(	const dealii::VectorizedArray< T, width >	x,
		const dealii::VectorizedArray< T, width >	b,
		const Bias	bias = Bias::none
	)

Custom implementation of an approximate, vectorized pow function with vectorized exponent.

Definition at line 311 of file simd.template.h.

References ryujin::fast_pow_impl().

get_entry() [1/4]

template<typename T , typename V >

DEAL_II_ALWAYS_INLINE T ryujin::get_entry ( const V &  vector, unsigned int  i  )

inline

Return a VectorizedArray with { U[i] , U[i + 1] , ... , U[i + VectorizedArray::size() - 1] }

Definition at line 244 of file simd.h.

get_entry() [2/4]

template<typename T , typename T2 >

DEAL_II_ALWAYS_INLINE T ryujin::get_entry ( const std::vector< T2 > &  vector, unsigned int  i  )

inline

Variant of above function specialized for std::vector.

Definition at line 268 of file simd.h.

get_entry() [3/4]

template<typename T , typename V >

DEAL_II_ALWAYS_INLINE T ryujin::get_entry ( const V &  vector, const unsigned int *  js  )

inline

Return a VectorizedArray with { U[js[0] , U[js[1]] , ... , U[js[VectorizedArray::size() - 1]] }

Definition at line 306 of file simd.h.

get_entry() [4/4]

template<typename T , typename T2 >

DEAL_II_ALWAYS_INLINE T ryujin::get_entry ( const std::vector< T2 > &  vector, const unsigned int *  js  )

inline

Variant of above function specialized for std::vector.

Definition at line 331 of file simd.h.

write_entry() [1/2]

template<typename T , typename V >

DEAL_II_ALWAYS_INLINE void ryujin::write_entry ( V &  vector, const T &  values, unsigned int  i  )

inline

Write out the given VectorizedArray to the vector

Definition at line 357 of file simd.h.

Referenced by ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::read_in().

write_entry() [2/2]

template<typename T , typename T2 >

DEAL_II_ALWAYS_INLINE void ryujin::write_entry ( std::vector< T2 > &  vector, const T &  values, unsigned int  i  )

inline

Variant of above function specialized for std::vector.

Definition at line 379 of file simd.h.

serialize_tensor() [1/2]

template<int rank, int dim, std::size_t width, typename Number >

DEAL_II_ALWAYS_INLINE dealii::Tensor< rank, dim, Number > ryujin::serialize_tensor ( const dealii::Tensor< rank, dim, dealii::VectorizedArray< Number, width > > &  vectorized, const unsigned int  k  )

inline

Return the k-th serialized component of a Tensor of VectorizedArray

Definition at line 412 of file simd.h.

References ryujin::serialize_tensor().

Referenced by ryujin::Postprocessor< ryujin::Euler::Description, dim, double >::compute(), and ryujin::serialize_tensor().

serialize_tensor() [2/2]

template<int rank, int dim, typename Number >

DEAL_II_ALWAYS_INLINE dealii::Tensor< rank, dim, Number > ryujin::serialize_tensor ( const dealii::Tensor< rank, dim, Number > &  serial, const unsigned int  k  )

inline

Variant of above function for serial Tensors that simply returns the given tensor.

Definition at line 438 of file simd.h.

assign_serial_tensor() [1/2]

template<int rank, int dim, std::size_t width, typename Number >

DEAL_II_ALWAYS_INLINE void ryujin::assign_serial_tensor ( dealii::Tensor< rank, dim, dealii::VectorizedArray< Number, width > > &  result, const dealii::Tensor< rank, dim, Number > &  serial, const unsigned int  k  )

inline

Update the the k-th serial component of a Tensor of VectorizedArray

Definition at line 454 of file simd.h.

References ryujin::assign_serial_tensor().

Referenced by ryujin::assign_serial_tensor(), and ryujin::Postprocessor< ryujin::Euler::Description, dim, double >::compute().

assign_serial_tensor() [2/2]

template<int rank, int dim, typename Number >

DEAL_II_ALWAYS_INLINE void ryujin::assign_serial_tensor ( dealii::Tensor< rank, dim, Number > &  result, const dealii::Tensor< rank, dim, Number > &  serial, const unsigned int  k  )

inline

Variant of above function for serial Tensors that simply assigns the given tensor as is.

Definition at line 478 of file simd.h.

Variable Documentation

get_stride_size

template<typename T >

unsigned int ryujin::get_stride_size = 1

Return the stride size:

Definition at line 49 of file simd.h.