Level 0: Miscellaneous additions

Collaboration diagram for Level 0: Miscellaneous additions:

Modules
	Timer class
	t.tic(); t.toc(); t.diff();
	Functions and functors
	Lowlevel helper functions and classes

Classes
class	dg::Message
	small class holding a stringstream More...
class	dg::Error
	class intended for the use in throw statements More...
struct	dg::Fail
	Indicate failure to converge. More...
struct	dg::MultiMatrix< MatrixType, ContainerType >
	Struct that applies given matrices one after the other. More...
class	dg::NoRoot1d
	Exception class, that stores boundaries for 1D root finding. More...

Functions
template<class ContainerType >
void	dg::transpose (unsigned nx, unsigned ny, const ContainerType &in, ContainerType &out)
	Transpose vector. More...
template<class ContainerType >
void	dg::extend_line (unsigned nx, unsigned ny, const ContainerType &in, ContainerType &out)
	Copy a line into rows of output vector. More...
template<class ContainerType >
void	dg::extend_column (unsigned nx, unsigned ny, const ContainerType &in, ContainerType &out)
	Copy a line into columns of output vector. More...
static void	dg::abort_program (int code=-1)
	Abort program (both MPI and non-MPI) More...
static void	dg::mpi_init (int argc, char *argv[])
	Convencience shortcut: Calls MPI_Init or MPI_Init_thread. More...
static void	dg::mpi_init2d (dg::bc bcx, dg::bc bcy, MPI_Comm &comm, std::istream &is=std::cin, bool verbose=true)
	Read in number of processses and create Cartesian MPI communicator. More...
static void	dg::mpi_init2d (unsigned &n, unsigned &Nx, unsigned &Ny, MPI_Comm comm, std::istream &is=std::cin, bool verbose=true)
	Read in number of processes and broadcast to process group. More...
static void	dg::mpi_init2d (dg::bc bcx, dg::bc bcy, unsigned &n, unsigned &Nx, unsigned &Ny, MPI_Comm &comm, std::istream &is=std::cin, bool verbose=true)
	Read in number of processses and grid size and create Cartesian MPI communicator. More...
static void	dg::mpi_init3d (dg::bc bcx, dg::bc bcy, dg::bc bcz, MPI_Comm &comm, std::istream &is=std::cin, bool verbose=true)
	Read in number of processses and create Cartesian MPI communicator. More...
static void	dg::mpi_init3d (unsigned &n, unsigned &Nx, unsigned &Ny, unsigned &Nz, MPI_Comm comm, std::istream &is=std::cin, bool verbose=true)
	Read in number of processes and broadcast to process group. More...
static void	dg::mpi_init3d (dg::bc bcx, dg::bc bcy, dg::bc bcz, unsigned &n, unsigned &Nx, unsigned &Ny, unsigned &Nz, MPI_Comm &comm, std::istream &is=std::cin, bool verbose=true)
	Read in number of processses and grid size and create Cartesian MPI communicator. More...
template<class real_type >
thrust::host_vector< real_type >	dg::forward_transform (const thrust::host_vector< real_type > &in, const aRealTopology2d< real_type > &g)
	Transform a vector from dg::xspace (nodal values) to dg::lspace (modal values) More...
template<class UnaryOp , class real_type >
dg::Operator< real_type >	dg::create::modal_filter (UnaryOp op, const DLT< real_type > &dlt)
	Create a modal filter block \( V D V^{-1}\). More...
template<class T >
T	dg::gcd (T a, T b)
	Greatest common divisor. More...
template<class T >
T	dg::lcm (T a, T b)
	Least common multiple. More...
bool	dg::is_same (double x, double y, double eps=1e-15)
bool	dg::is_same (float x, float y, float eps=1e-6)
bool	dg::is_divisable (double a, double b, double eps=1e-15)
bool	dg::is_divisable (float a, float b, float eps=1e-6)

Detailed Description

Function Documentation

abort_program()

static void dg::abort_program ( int  code = -1 )

inlinestatic

Abort program (both MPI and non-MPI)

#ifdef MPI_VERSION
    MPI_Abort(MPI_COMM_WORLD, code);
#endif //WITH_MPI
    exit( code);

Parameters

code	The abortion code that will be signalled to the caller (of the program)

extend_column()

template<class ContainerType >

void dg::extend_column	(	unsigned	nx,
		unsigned	ny,
		const ContainerType &	in,
		ContainerType &	out
	)

Copy a line into columns of output vector.

out[i*nx+j] = in[i]

Template Parameters

ContainerType

Any class for which a specialization of TensorTraits exists and which fulfills the requirements of the there defined data and execution policies derived from AnyVectorTag and AnyPolicyTag. Among others dg::HVec (serial), dg::DVec (cuda / omp), dg::MHVec (mpi + serial) or dg::MDVec (mpi + cuda / omp) std::vector<dg::DVec> (vector of shared device vectors), std::array<double, 4> (array of 4 doubles) or std::map < std::string, dg::DVec> ( a map of named vectors) double (scalar) and other primitive types ... If there are several ContainerTypes in the argument list, then TensorTraits must exist for all of them

See also

See The dg dispatch system for a detailed explanation of our type dispatch system

Parameters

nx	number of columns (size of contiguous chunks) in output vector
ny	size of input vector, number of rows in output vector
in	input (size ny)
out	output (size nx*ny may not alias in)

extend_line()

template<class ContainerType >

void dg::extend_line	(	unsigned	nx,
		unsigned	ny,
		const ContainerType &	in,
		ContainerType &	out
	)

Copy a line into rows of output vector.

Just append the line of size nx until the output vector size of nx*ny is reached: out[i*nx+j] = in[j]

Template Parameters

ContainerType

Any class for which a specialization of TensorTraits exists and which fulfills the requirements of the there defined data and execution policies derived from AnyVectorTag and AnyPolicyTag. Among others dg::HVec (serial), dg::DVec (cuda / omp), dg::MHVec (mpi + serial) or dg::MDVec (mpi + cuda / omp) std::vector<dg::DVec> (vector of shared device vectors), std::array<double, 4> (array of 4 doubles) or std::map < std::string, dg::DVec> ( a map of named vectors) double (scalar) and other primitive types ... If there are several ContainerTypes in the argument list, then TensorTraits must exist for all of them

See also

See The dg dispatch system for a detailed explanation of our type dispatch system

Parameters

nx	size of the input vector/ number of columns (size of contiguous chunks) in output vector
ny	number of rows in output vector
in	input (size nx)
out	output (size nx*ny; may not alias in)

forward_transform()

template<class real_type >

thrust::host_vector< real_type > dg::forward_transform	(	const thrust::host_vector< real_type > &	in,
		const aRealTopology2d< real_type > &	g
	)

Transform a vector from dg::xspace (nodal values) to dg::lspace (modal values)

Parameters

in	input
g	grid

Returns

the vector in LSPACE

See also

fast_transform

gcd()

template<class T >

T dg::gcd	(	T	a,
		T	b
	)

Greatest common divisor.

Parameters

a	First number
b	Second number

Returns

greatest common divisor

is_divisable() [1/2]

bool dg::is_divisable ( double  a, double  b, double  eps = 1e-15  )

inline

Checks if two number are integer divisable \(a/b \in \mathbb{Z}\) within accuracy

Attention

Does not check for equal sign!

is_divisable() [2/2]

bool dg::is_divisable ( float  a, float  b, float  eps = 1e-6  )

inline

Checks if two number are integer divisable \(a/b \in \mathbb{Z}\) within accuracy

Attention

Does not check for equal sign!

is_same() [1/2]

bool dg::is_same ( double  x, double  y, double  eps = 1e-15  )

inline

Checks if two number are same within accuracy

is_same() [2/2]

bool dg::is_same ( float  x, float  y, float  eps = 1e-6  )

inline

Checks if two number are same within accuracy

lcm()

template<class T >

T dg::lcm	(	T	a,
		T	b
	)

Least common multiple.

Parameters

a	Fist number
b	Second number

Returns

Least common multiple

modal_filter()

template<class UnaryOp , class real_type >

dg::Operator< real_type > dg::create::modal_filter	(	UnaryOp	op,
		const DLT< real_type > &	dlt
	)

Create a modal filter block \( V D V^{-1}\).

where \( V\) is the Vandermonde matrix (the backward transformation matrix) and \( D \) is a diagonal matrix with \( D_{ii} = \sigma(i)\)

See also

A discussion of the effects of the modal filter on advection schemes can be found here https://mwiesenberger.github.io/advection

Note

basically the result is that it is usually not advantageous to use a modal filter

Template Parameters

UnaryOp

Model of Unary Function real_type sigma(unsigned) The input will be the modal number i where \( i=0,...,n-1\) and n is the number of polynomial coefficients in use. The output is the filter strength for the given mode number

Parameters

op	the unary function
dlt	provide the forward, backward transformation and number of coefficients n

Returns

The product \( V D V^{-1}\)

Note

The idea is to use the result in connection with dg::create::fast_transform() to create a matrix that applies the filter to vectors. For example to create a modal filter that acts in two dimensions:

// create filter:
auto filter = dg::create::fast_transform(
     dg::create::modal_filter( op, grid.dltx()),
     dg::create::modal_filter( op, grid.dlty()), grid);
//apply filter:
dg::blas2::symv( filter, x, y);

mpi_init()

static void dg::mpi_init ( int  argc, char *  argv[]  )

inlinestatic

Convencience shortcut: Calls MPI_Init or MPI_Init_thread.

Shortcut for

#ifdef _OPENMP
    int provided, error;
    error = MPI_Init_thread(&argc, &argv, MPI_THREAD_FUNNELED, &provided);
    assert( error == MPI_SUCCESS && "Threaded MPI lib required!\n");
#else
    MPI_Init(&argc, &argv);
#endif

Parameters

argc	command line argument number
argv	command line arguments

mpi_init2d() [1/3]

static void dg::mpi_init2d ( dg::bc  bcx, dg::bc  bcy, MPI_Comm &  comm, std::istream &  is = std::cin, bool  verbose = true  )

inlinestatic

Read in number of processses and create Cartesian MPI communicator.

Also sets the GPU a process should use via rank% num_devices_per_node if THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_CUDA

Parameters

bcx	if bcx==dg::PER then the communicator is periodic in x
bcy	if bcy==dg::PER then the communicator is periodic in y
comm	(write only) MPI_COMM_WORLD as a 2d Cartesian MPI communicator
is	Input stream rank 0 reads parameters (npx, npy)
verbose	If true, rank 0 prints queries and information on std::cout

Attention

Before creating a second Cartesian communicator consider freeing existing ones with MPI_Comm_free. (Using mpi_init2d and mpi_init3d in the same program has sometimes led to Segmentation faults in the past)

mpi_init2d() [2/3]

static void dg::mpi_init2d ( dg::bc  bcx, dg::bc  bcy, unsigned &  n, unsigned &  Nx, unsigned &  Ny, MPI_Comm &  comm, std::istream &  is = std::cin, bool  verbose = true  )

inlinestatic

Read in number of processses and grid size and create Cartesian MPI communicator.

Also sets the GPU a process should use via rank% num_devices_per_node if THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_CUDA

Parameters

bcx	if bcx==dg::PER then the communicator is periodic in x
bcy	if bcy==dg::PER then the communicator is periodic in y
n	rank 0 reads in from is and broadcasts to all processes in MPI_COMM_WORLD
Nx	rank 0 reads in from is and broadcasts to all processes in MPI_COMM_WORLD
Ny	rank 0 reads in from is and broadcasts to all processes in MPI_COMM_WORLD
comm	(write only) MPI_COMM_WORLD as a 2d Cartesian MPI communicator
is	Input stream rank 0 reads parameters (npx, npy, n, Nx, Ny)
verbose	If true, rank 0 prints queries and information on std::cout

Attention

Before creating a second Cartesian communicator consider freeing existing ones with MPI_Comm_free. (Using mpi_init2d and mpi_init3d in the same program has sometimes led to Segmentation faults in the past)

mpi_init2d() [3/3]

static void dg::mpi_init2d ( unsigned &  n, unsigned &  Nx, unsigned &  Ny, MPI_Comm  comm, std::istream &  is = std::cin, bool  verbose = true  )

inlinestatic

Read in number of processes and broadcast to process group.

Parameters

n	rank 0 reads in from is and broadcasts to all processes in comm
Nx	rank 0 reads in from is and broadcasts to all processes in comm
Ny	rank 0 reads in from is and broadcasts to all processes in comm
comm	(read only) a 2d Cartesian MPI communicator
is	Input stream rank 0 reads parameters (n, Nx, Ny)
verbose	If true, rank 0 prints queries and information on std::cout

mpi_init3d() [1/3]

static void dg::mpi_init3d ( dg::bc  bcx, dg::bc  bcy, dg::bc  bcz, MPI_Comm &  comm, std::istream &  is = std::cin, bool  verbose = true  )

inlinestatic

Read in number of processses and create Cartesian MPI communicator.

Also sets the GPU a process should use via rank% num_devices_per_node if THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_CUDA

Parameters

bcx	if bcx==dg::PER then the communicator is periodic in x
bcy	if bcy==dg::PER then the communicator is periodic in y
bcz	if bcz==dg::PER then the communicator is periodic in z
comm	(write only) MPI_COMM_WORLD as a 3d Cartesian MPI communicator
is	Input stream rank 0 reads parameters (npx, npy, npz)
verbose	If true, rank 0 prints queries and information on std::cout

Attention

Before creating a second Cartesian communicator consider freeing existing ones with MPI_Comm_free. (Using mpi_init2d and mpi_init3d in the same program has sometimes led to Segmentation faults in the past)

mpi_init3d() [2/3]

static void dg::mpi_init3d ( dg::bc  bcx, dg::bc  bcy, dg::bc  bcz, unsigned &  n, unsigned &  Nx, unsigned &  Ny, unsigned &  Nz, MPI_Comm &  comm, std::istream &  is = std::cin, bool  verbose = true  )

inlinestatic

Read in number of processses and grid size and create Cartesian MPI communicator.

Also sets the GPU via rank% num_devices_per_node a process should use if THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_CUDA

Parameters

bcx	if bcx==dg::PER then the communicator is periodic in x
bcy	if bcy==dg::PER then the communicator is periodic in y
bcz	if bcz==dg::PER then the communicator is periodic in z
n	rank 0 reads in from is and broadcasts to all processes in MPI_COMM_WORLD
Nx	rank 0 reads in from is and broadcasts to all processes in MPI_COMM_WORLD
Ny	rank 0 reads in from is and broadcasts to all processes in MPI_COMM_WORLD
Nz	rank 0 reads in from is and broadcasts to all processes in MPI_COMM_WORLD
comm	(write only) MPI_COMM_WORLD as a 3d Cartesian MPI communicator
is	Input stream rank 0 reads parameters (npx, npy, npz, n, Nx, Ny, Nz)
verbose	If true, rank 0 prints queries and information on std::cout

Attention

Before creating a second Cartesian communicator consider freeing existing ones with MPI_Comm_free. (Using mpi_init2d and mpi_init3d in the same program has sometimes led to Segmentation faults in the past)

mpi_init3d() [3/3]

static void dg::mpi_init3d ( unsigned &  n, unsigned &  Nx, unsigned &  Ny, unsigned &  Nz, MPI_Comm  comm, std::istream &  is = std::cin, bool  verbose = true  )

inlinestatic

Read in number of processes and broadcast to process group.

Parameters

n	rank 0 reads in from is and broadcasts to all processes in comm
Nx	rank 0 reads in from is and broadcasts to all processes in comm
Ny	rank 0 reads in from is and broadcasts to all processes in comm
Nz	rank 0 reads in from is and broadcasts to all processes in comm
comm	(read only) a 3d Cartesian MPI communicator
is	Input stream rank 0 reads parameters (n, Nx, Ny, Nz )
verbose	If true, rank 0 prints queries and information on std::cout

transpose()

template<class ContainerType >

void dg::transpose	(	unsigned	nx,
		unsigned	ny,
		const ContainerType &	in,
		ContainerType &	out
	)

Transpose vector.

The equivalent of out[j*ny+i] = in[i*nx+j];

Template Parameters

ContainerType

Any class for which a specialization of TensorTraits exists and which fulfills the requirements of the there defined data and execution policies derived from AnyVectorTag and AnyPolicyTag. Among others dg::HVec (serial), dg::DVec (cuda / omp), dg::MHVec (mpi + serial) or dg::MDVec (mpi + cuda / omp) std::vector<dg::DVec> (vector of shared device vectors), std::array<double, 4> (array of 4 doubles) or std::map < std::string, dg::DVec> ( a map of named vectors) double (scalar) and other primitive types ... If there are several ContainerTypes in the argument list, then TensorTraits must exist for all of them

See also

See The dg dispatch system for a detailed explanation of our type dispatch system

Parameters

nx	number of columns in input vector (size of contiguous chunks) /rows in output vector
ny	number of rows in input vector /columns in output vector
in	input
out	output (may not alias in)