SStruct System and Object Interface

group SStruct System and Object Interface

A semi-structured-grid conceptual interface.

This interface represents a semi-structured-grid conceptual view of a linear system.

SStruct Grids

typedef struct hypre_SStructGrid_struct *HYPRE_SStructGrid

A grid object is constructed out of several structured “parts” and an optional unstructured “part”.

Each structured part has its own abstract index space.

typedef HYPRE_Int HYPRE_SStructVariable

An enumerated type that supports cell centered, node centered, face centered, and edge centered variables.

Face centered variables are split into x-face, y-face, and z-face variables, and edge centered variables are split into x-edge, y-edge, and z-edge variables. The edge centered variable types are only used in 3D. In 2D, edge centered variables are handled by the face centered types.

Variables are referenced relative to an abstract (cell centered) index in the following way:

cell centered variables are aligned with the index;
node centered variables are aligned with the cell corner at relative index (1/2, 1/2, 1/2);
x-face, y-face, and z-face centered variables are aligned with the faces at relative indexes (1/2, 0, 0), (0, 1/2, 0), and (0, 0, 1/2), respectively;
x-edge, y-edge, and z-edge centered variables are aligned with the edges at relative indexes (0, 1/2, 1/2), (1/2, 0, 1/2), and (1/2, 1/2, 0), respectively.

The supported identifiers are:

HYPRE_SSTRUCT_VARIABLE_CELL
HYPRE_SSTRUCT_VARIABLE_NODE
HYPRE_SSTRUCT_VARIABLE_XFACE
HYPRE_SSTRUCT_VARIABLE_YFACE
HYPRE_SSTRUCT_VARIABLE_ZFACE
HYPRE_SSTRUCT_VARIABLE_XEDGE
HYPRE_SSTRUCT_VARIABLE_YEDGE
HYPRE_SSTRUCT_VARIABLE_ZEDGE

NOTE: Although variables are referenced relative to a unique abstract cell-centered index, some variables are associated with multiple grid cells. For example, node centered variables in 3D are associated with 8 cells (away from boundaries). Although grid cells are distributed uniquely to different processes, variables may be owned by multiple processes because they may be associated with multiple cells.

HYPRE_Int HYPRE_SStructGridCreate(MPI_Comm comm, HYPRE_Int ndim, HYPRE_Int nparts, HYPRE_SStructGrid *grid): Create an ndim-dimensional grid object with nparts structured parts.

HYPRE_Int HYPRE_SStructGridDestroy(HYPRE_SStructGrid grid)

Destroy a grid object.

An object should be explicitly destroyed using this destructor when the user’s code no longer needs direct access to it. Once destroyed, the object must not be referenced again. Note that the object may not be deallocated at the completion of this call, since there may be internal package references to the object. The object will then be destroyed when all internal reference counts go to zero.

HYPRE_Int HYPRE_SStructGridSetExtents(HYPRE_SStructGrid grid, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper): Set the extents for a box on a structured part of the grid.

HYPRE_Int HYPRE_SStructGridSetVariables(HYPRE_SStructGrid grid, HYPRE_Int part, HYPRE_Int nvars, HYPRE_SStructVariable *vartypes): Describe the variables that live on a structured part of the grid.

HYPRE_Int HYPRE_SStructGridAddVariables(HYPRE_SStructGrid grid, HYPRE_Int part, HYPRE_Int *index, HYPRE_Int nvars, HYPRE_SStructVariable *vartypes)

Describe additional variables that live at a particular index.

These variables are appended to the array of variables set in HYPRE_SStructGridSetVariables, and are referenced as such.

NOTE: This routine is not yet supported.

HYPRE_Int HYPRE_SStructGridSetFEMOrdering(HYPRE_SStructGrid grid, HYPRE_Int part, HYPRE_Int *ordering)

Set the ordering of variables in a finite element problem.

This overrides the default ordering described below.

Array ordering is composed of blocks of size (1 + ndim). Each block indicates a specific variable in the element and the ordering of the blocks defines the ordering of the variables. A block contains a variable number followed by an offset direction relative to the element’s center. For example, a block containing (2, 1, -1, 0) means variable 2 on the edge located in the (1, -1, 0) direction from the center of the element. Note that here variable 2 must be of type ZEDGE for this to make sense. The ordering array must account for all variables in the element. This routine can only be called after HYPRE_SStructGridSetVariables.

The default ordering for element variables (var, i, j, k) varies fastest in index i, followed by j, then k, then var. For example, if var 0, var 1, and var 2 are declared to be XFACE, YFACE, and NODE variables, respectively, then the default ordering (in 2D) would first list the two XFACE variables, then the two YFACE variables, then the four NODE variables as follows:

(0,-1,0), (0,1,0), (1,0,-1), (1,0,1), (2,-1,-1), (2,1,-1), (2,-1,1), (2,1,1)

HYPRE_Int HYPRE_SStructGridSetNeighborPart(HYPRE_SStructGrid grid, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int nbor_part, HYPRE_Int *nbor_ilower, HYPRE_Int *nbor_iupper, HYPRE_Int *index_map, HYPRE_Int *index_dir)

Describe how regions just outside of a part relate to other parts.

This is done a box at a time.

Parts part and nbor_part must be different, except in the case where only cell-centered data is used.

Indexes should increase from ilower to iupper. It is not necessary that indexes increase from nbor_ilower to nbor_iupper.

The index_map describes the mapping of indexes 0, 1, and 2 on part part to the corresponding indexes on part nbor_part. For example, triple (1, 2, 0) means that indexes 0, 1, and 2 on part part map to indexes 1, 2, and 0 on part nbor_part, respectively.

The index_dir describes the direction of the mapping in index_map. For example, triple (1, 1, -1) means that for indexes 0 and 1, increasing values map to increasing values on nbor_part, while for index 2, decreasing values map to increasing values.

NOTE: All parts related to each other via this routine must have an identical list of variables and variable types. For example, if part 0 has only two variables on it, a cell centered variable and a node centered variable, and we declare part 1 to be a neighbor of part 0, then part 1 must also have only two variables on it, and they must be of type cell and node. In addition, variables associated with FACEs or EDGEs must be grouped together and listed in X, Y, Z order. This is to enable the code to correctly associate variables on one part with variables on its neighbor part when a coordinate transformation is specified. For example, an XFACE variable on one part may correspond to a YFACE variable on a neighbor part under a particular tranformation, and the code determines this association by assuming that the variable lists are as noted here.

HYPRE_Int HYPRE_SStructGridSetSharedPart(HYPRE_SStructGrid grid, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int *offset, HYPRE_Int shared_part, HYPRE_Int *shared_ilower, HYPRE_Int *shared_iupper, HYPRE_Int *shared_offset, HYPRE_Int *index_map, HYPRE_Int *index_dir)

Describe how regions inside a part are shared with regions in other parts.

Parts part and shared_part must be different.

Indexes should increase from ilower to iupper. It is not necessary that indexes increase from shared_ilower to shared_iupper. This is to maintain consistency with the SetNeighborPart function, which is also able to describe shared regions but in a more limited fashion.

The offset is a triple (in 3D) used to indicate the dimensionality of the shared set of data and its position with respect to the box extents ilower and iupper on part part. The dimensionality is given by the number of 0’s in the triple, and the position is given by plus or minus 1’s. For example: (0, 0, 0) indicates sharing of all data in the given box; (1, 0, 0) indicates sharing of data on the faces in the (1, 0, 0) direction; (1, 0, -1) indicates sharing of data on the edges in the (1, 0, -1) direction; and (1, -1, 1) indicates sharing of data on the nodes in the (1, -1, 1) direction. To ensure the dimensionality, it is required that for every nonzero entry, the corresponding extents of the box are the same. For example, if offset is (0, 1, 0), then (2, 1, 3) and (10, 1, 15) are valid box extents, whereas (2, 1, 3) and (10, 7, 15) are invalid (because 1 and 7 are not the same).

The shared_offset is used in the same way as offset, but with respect to the box extents shared_ilower and shared_iupper on part shared_part.

The index_map describes the mapping of indexes 0, 1, and 2 on part part to the corresponding indexes on part shared_part. For example, triple (1, 2, 0) means that indexes 0, 1, and 2 on part part map to indexes 1, 2, and 0 on part shared_part, respectively.

The index_dir describes the direction of the mapping in index_map. For example, triple (1, 1, -1) means that for indexes 0 and 1, increasing values map to increasing values on shared_part, while for index 2, decreasing values map to increasing values.

NOTE: All parts related to each other via this routine must have an identical list of variables and variable types. For example, if part 0 has only two variables on it, a cell centered variable and a node centered variable, and we declare part 1 to have shared regions with part 0, then part 1 must also have only two variables on it, and they must be of type cell and node. In addition, variables associated with FACEs or EDGEs must be grouped together and listed in X, Y, Z order. This is to enable the code to correctly associate variables on one part with variables on a shared part when a coordinate transformation is specified. For example, an XFACE variable on one part may correspond to a YFACE variable on a shared part under a particular tranformation, and the code determines this association by assuming that the variable lists are as noted here.

HYPRE_Int HYPRE_SStructGridSetAMRPart(HYPRE_SStructGrid grid, HYPRE_Int coarse_part, HYPRE_Int fine_part, HYPRE_Int *coarse_index, HYPRE_Int *fine_index, HYPRE_Int *rfactors)

AMRNEW.

Declare a part to be a refinement of another part in an AMR hierarchy.

The index space of fine_part is defined to be a refinement of the index space of coarse_part by a refinement factor in each dimension given by rfactors. The two index spaces are aligned based on coarse_index and fine_index, which specifies the fine index of the lower left cell of the given coarse index. This induces notions of real and slave variables and interpolation between them, which impacts how vector and matrix values are set.

By default, real variables are defined as follows, and the remaining variables are slave variables (this may be changed by the user through the routine HYPRE_SStructGridSetAMRRefSlaves:

fine variables on the interior of overlapping coarse-fine regions;
coarse variables on the boundary of overlapping coarse-fine regions;
all variables in non-overlapping regions.

By default, interpolation (and restriction) is defined to be the natural finite element interpolation corresponding to each variable type, but this may be changed by the user via the other SStructGridSetAMR routines.

There are two basic steps for changing interpolation, both optional. The first is through the SStructGridSetAMRRef routines using a reference overlapping coarse-fine patch. This reference patch is applied throughout the entire part to define the global interpolation operator. Interpolation may then be changed at individual locations in the grid through the HYPRE_SStructGridSetAMRInterp routine.

The reference coarse-fine patch consists of a single coarse cell and its refinement, where the coarse and fine reference patches are assumed to have a lower left index of zero. Coarse and fine variables are referenced in the patch by their associated cell indexes (in the same way that variables are referenced on the grid). For example, in 2D, for a refinement factor of two in both directions and a nodal variable type, the lower left fine variable would be referenced by the index (-1,-1) and the upper right variable by index (1,1). Similarly, the lower left coarse variable would be referenced by index (-1,-1) and the upper right with index (0,0).

HYPRE_Int HYPRE_SStructGridSetAMRRefSlaves(HYPRE_SStructGrid grid, HYPRE_Int coarse_part, HYPRE_Int var, HYPRE_Int nslaves, HYPRE_Int *slaves)

AMRNEW.

Define the slave variables in the reference coarse-fine patch. The argument slaves is an array of blocks of size ndim containing the associated cell indexes for the slave variables.

See HYPRE_SStructGridSetAMRPart for details on the coarse-fine patch. This routine must be called after HYPRE_SStructGridSetAMRPart and before any other SStructGridSetAMR routines.

HYPRE_Int HYPRE_SStructGridSetAMRRefInterp(HYPRE_SStructGrid grid, HYPRE_Int coarse_part, HYPRE_Int scf, HYPRE_Int svar, HYPRE_Int *sindex, HYPRE_Int nvalues, HYPRE_Int *vars, HYPRE_Int *cf, HYPRE_Int *indexes, HYPRE_Complex *values)

AMRNEW.

Set interpolation on the reference coarse-fine patch. Interpolation maps real variables to all variables (real and slave). Real variables are mapped to real variables identically. Slave coarse variables are not interpolated at all. Users may only change interpolation from real variables (coarse and fine) to slave fine variables scf=1 (see special cell-centered case below). The argument sindex is a fine reference patch index for slave variable svar. The array indexes contains both coarse and fine reference patch indexes for variables vars as specified in cf by a 0 (coarse) or a 1 (fine).

Cell-centered variables are treated as a special case, where coupling occurs through fictitious slave face variables. Although these variables are associated with faces, they should actually be thought of as being centered either at cells just outside of the patch for fine faces (scf=1) or at the patch center for coarse faces (scf=0). These fictitious variables are also referenced differently from other variables, where, for convenience, sindex always specifies cells just outside of the patch (in the coarse case in particular, this approach enables a distinction between the multiple coarse faces and interpolation formulas). Interpolation may be from any neighboring real variables, inside or outside of the patch.

Idea for the cell-centered case: To reduce the need to fix up interpolation at individual grid locations, consider adding a rule to replace interpolation from nonexistent coarse variables with interpolation from existing underlying fine variables. For example, in those cases, take the average of the closest fine variables (in the odd refinement case, this is just injection from the underlying fine variable).

See HYPRE_SStructGridSetAMRPart for details on the coarse-fine patch. Note: Currently, each call must set an entire row of interpolation.

HYPRE_Int HYPRE_SStructGridSetAMRRefRestrictT(HYPRE_SStructGrid grid, HYPRE_Int coarse_part, HYPRE_Int scf, HYPRE_Int svar, HYPRE_Int *sindex, HYPRE_Int nvalues, HYPRE_Int *vars, HYPRE_Int *cf, HYPRE_Int *indexes, HYPRE_Complex *values)

AMRNEW.

Set (the transpose of) restriction on the reference coarse-fine patch. By default, restriction is the transpose of interpolation. Usually, this routine should only be called to set up a nonsymmetric operator. Usage is the same as for setting up interpolation.

HYPRE_Int HYPRE_SStructGridSetAMRInterp(HYPRE_SStructGrid grid, HYPRE_Int coarse_part, HYPRE_Int *coarse_index, HYPRE_Int scf, HYPRE_Int svar, HYPRE_Int *sindex, HYPRE_Int nvalues, HYPRE_Int *vars, HYPRE_Int *cf, HYPRE_Int *indexes, HYPRE_Complex *values)

AMRNEW.

Set interpolation at a specific grid index on a refined part. Usage is the same as for setting interpolation on the reference coarse-fine patch, except that an additional argument coarse_index is given.

HYPRE_Int HYPRE_SStructGridSetAMRRestrictT(HYPRE_SStructGrid grid, HYPRE_Int coarse_part, HYPRE_Int *coarse_index, HYPRE_Int scf, HYPRE_Int svar, HYPRE_Int *sindex, HYPRE_Int nvalues, HYPRE_Int *vars, HYPRE_Int *cf, HYPRE_Int *indexes, HYPRE_Complex *values)

AMRNEW.

Set restriction at a specific grid index on a refined part. Usage is the same as for setting restriction on the reference coarse-fine patch, except that an additional argument coarse_index is given.

HYPRE_Int HYPRE_SStructGridAddUnstructuredPart(HYPRE_SStructGrid grid, HYPRE_Int ilower, HYPRE_Int iupper)

Add an unstructured part to the grid.

The variables in the unstructured part of the grid are referenced by a global rank between 0 and the total number of unstructured variables minus one. Each process owns some unique consecutive range of variables, defined by ilower and iupper.

NOTE: This is just a placeholder. This part of the interface is not finished.

HYPRE_Int HYPRE_SStructGridAssemble(HYPRE_SStructGrid grid): Finalize the construction of the grid before using.

HYPRE_Int HYPRE_SStructGridSetObjectType(HYPRE_SStructGrid grid, HYPRE_Int type)

AMRNEW.

Set the storage type of associated matrix and vector objects. This object type is inherited by SStructGraph, SStructMatrix, and SStructVector at their creation, but can also be changed through the corresponding SetObjectType routines. Setting an object type with this routine helps to minimize the number of overall calls to SetObjectType, but is particularly useful when grid-based operators are needed, such as a discrete gradient (HYPRE_SStructGridGetGradient).

Currently, type can be HYPRE_SSTRUCT, HYPRE_STRUCT, or HYPRE_PARCSR. The default is HYPRE_SSTRUCT.

HYPRE_Int HYPRE_SStructGridGetGradient(HYPRE_SStructGrid grid, void **gradient)

AMRNEW.

Get a reference to a discrete gradient matrix object.

RDF: Not sure yet if this routine really needs to exist.

HYPRE_Int HYPRE_SStructGridSetPeriodic(HYPRE_SStructGrid grid, HYPRE_Int part, HYPRE_Int *periodic)

Set the periodicity on a particular part.

The argument periodic is an ndim-dimensional integer array that contains the periodicity for each dimension. A zero value for a dimension means non-periodic, while a nonzero value means periodic and contains the actual period. For example, periodicity in the first and third dimensions for a 10x11x12 part is indicated by the array [10,0,12].

NOTE: Some of the solvers in hypre have power-of-two restrictions on the size of the periodic dimensions.

HYPRE_Int HYPRE_SStructGridSetNumGhost(HYPRE_SStructGrid grid, HYPRE_Int *num_ghost): Setting ghost in the sgrids.

HYPRE_Int HYPRE_SStructGridGetVariableBox(HYPRE_SStructGrid grid, HYPRE_Int part, HYPRE_Int var, HYPRE_Int *cell_ilower, HYPRE_Int *cell_iupper, HYPRE_Int *var_ilower, HYPRE_Int *var_iupper): Helper function for returning the box dimensions of a (part,var)-structured grid.

HYPRE_Int HYPRE_SStructGridProjectBox(HYPRE_SStructGrid grid, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int *origin, HYPRE_Int *stride)

Project the box described by ilower and iupper onto the strided index space that contains the index origin and has stride stride.

This routine is useful when dealing with rectangular matrices.

HYPRE_Int HYPRE_SStructGridCoarsen(HYPRE_SStructGrid fgrid, HYPRE_Index *strides, HYPRE_SStructGrid *cgrid): Coarsen grid by factor strides to create cgrid.

HYPRE_Int HYPRE_SStructGridPrintGLVis(HYPRE_SStructGrid grid, const char *meshprefix, HYPRE_Real *trans, HYPRE_Real *origin)

HYPRE_SSTRUCT_VARIABLE_UNDEFINED

HYPRE_SSTRUCT_VARIABLE_CELL

HYPRE_SSTRUCT_VARIABLE_NODE

HYPRE_SSTRUCT_VARIABLE_XFACE

HYPRE_SSTRUCT_VARIABLE_YFACE

HYPRE_SSTRUCT_VARIABLE_ZFACE

HYPRE_SSTRUCT_VARIABLE_XEDGE

HYPRE_SSTRUCT_VARIABLE_YEDGE

HYPRE_SSTRUCT_VARIABLE_ZEDGE

SStruct Stencils

typedef struct hypre_SStructStencil_struct *HYPRE_SStructStencil: The stencil object.

HYPRE_Int HYPRE_SStructStencilCreate(HYPRE_Int ndim, HYPRE_Int size, HYPRE_SStructStencil *stencil): Create a stencil object for the specified number of spatial dimensions and stencil entries.

HYPRE_Int HYPRE_SStructStencilDestroy(HYPRE_SStructStencil stencil): Destroy a stencil object.

HYPRE_Int HYPRE_SStructStencilSetEntry(HYPRE_SStructStencil stencil, HYPRE_Int entry, HYPRE_Int *offset, HYPRE_Int var): Set a stencil entry.

HYPRE_Int HYPRE_SStructStencilPrint(FILE *file, HYPRE_SStructStencil stencil)

HYPRE_Int HYPRE_SStructStencilRead(FILE *file, HYPRE_SStructStencil *stencil_ptr)

SStruct Graphs

typedef struct hypre_SStructGraph_struct *HYPRE_SStructGraph: The graph object is used to describe the nonzero structure of a matrix.

HYPRE_Int HYPRE_SStructGraphCreate(MPI_Comm comm, HYPRE_SStructGrid grid, HYPRE_SStructGraph *graph): Create a graph object.

HYPRE_Int HYPRE_SStructGraphDestroy(HYPRE_SStructGraph graph): Destroy a graph object.

HYPRE_Int HYPRE_SStructGraphSetDomainGrid(HYPRE_SStructGraph graph, HYPRE_SStructGrid domain_grid): Set the domain grid.

HYPRE_Int HYPRE_SStructGraphSetStencil(HYPRE_SStructGraph graph, HYPRE_Int part, HYPRE_Int var, HYPRE_SStructStencil stencil): Set the stencil for a variable on a structured part of the grid.

HYPRE_Int HYPRE_SStructGraphSetFEM(HYPRE_SStructGraph graph, HYPRE_Int part): Indicate that an FEM approach will be used to set matrix values on this part.

HYPRE_Int HYPRE_SStructGraphSetFEMSparsity(HYPRE_SStructGraph graph, HYPRE_Int part, HYPRE_Int nsparse, HYPRE_Int *sparsity)

Set the finite element stiffness matrix sparsity.

This overrides the default full sparsity pattern described below.

Array sparsity contains nsparse row/column tuples (I,J) that indicate the nonzeroes of the local stiffness matrix. The layout of the values passed into the routine HYPRE_SStructMatrixAddFEMValues is determined here.

The default sparsity is full (each variable is coupled to all others), and the values passed into the routine HYPRE_SStructMatrixAddFEMValues are assumed to be by rows (that is, column indices vary fastest).

HYPRE_Int HYPRE_SStructGraphAddEntries(HYPRE_SStructGraph graph, HYPRE_Int part, HYPRE_Int *index, HYPRE_Int var, HYPRE_Int to_part, HYPRE_Int *to_index, HYPRE_Int to_var)

Add a non-stencil graph entry at a particular index.

This graph entry is appended to the existing graph entries, and is referenced as such.

NOTE: Users are required to set graph entries on all processes that own the associated variables. This means that some data will be multiply defined.

HYPRE_Int HYPRE_SStructGraphAssemble(HYPRE_SStructGraph graph): Finalize the construction of the graph before using.

HYPRE_Int HYPRE_SStructGraphSetObjectType(HYPRE_SStructGraph graph, HYPRE_Int type)

Set the storage type of the associated matrix object.

It is used before AddEntries and Assemble to compute the right ranks in the graph.

NOTE: This routine is only necessary for implementation reasons, and will eventually be removed.

See also

HYPRE_SStructMatrixSetObjectType

HYPRE_Int HYPRE_SStructGraphPrint(FILE *file, HYPRE_SStructGraph graph)

HYPRE_Int HYPRE_SStructGraphRead(FILE *file, HYPRE_SStructGrid grid, HYPRE_SStructStencil **stencils, HYPRE_SStructGraph *graph_ptr)

SStruct Matrices

typedef struct hypre_SStructMatrix_struct *HYPRE_SStructMatrix: The matrix object.

HYPRE_Int HYPRE_SStructMatrixCreate(MPI_Comm comm, HYPRE_SStructGraph graph, HYPRE_SStructMatrix *matrix)

Create a matrix object.

Matrices may have different range and domain grids, that is, they need not be square. By default, the range and domain grids are the same as the argument grid in HYPRE_SStructMatrixCreate. Both grids live on a common fine index space and should have the same number of boxes. The actual range is a coarsening of the range grid with coarsening factor ran_stride specified in HYPRE_SStructMatrixSetRangeStride. Similarly, the actual domain is a coarsening of the domain grid with factor dom_stride specified in HYPRE_SStructMatrixSetDomainStride. Currently, either ran_stride or dom_stride or both must be all ones (i.e., no coarsening).

HYPRE_Int HYPRE_SStructMatrixDestroy(HYPRE_SStructMatrix matrix): Destroy a matrix object.

HYPRE_Int HYPRE_SStructMatrixInitialize(HYPRE_SStructMatrix matrix): Prepare a matrix object for setting coefficient values.

HYPRE_Int HYPRE_SStructMatrixSetDomainStride(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *dom_stride)

(Optional) Set matrix domain stride part by part.

For more information, see HYPRE_SStructMatrixCreate.

HYPRE_Int HYPRE_SStructMatrixSetRangeStride(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *ran_stride)

(Optional) Set matrix range stride part by part.

For more information, see HYPRE_SStructMatrixCreate.

HYPRE_Int HYPRE_SStructMatrixSetConstantEntries(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int var, HYPRE_Int to_var, HYPRE_Int nentries, HYPRE_Int *centries)

(Optional) Set matrix coefficients which are constant over the grid.

This considers a block matrix associated with the tuple (part, var, to_var). The centries array is of length nentries.

HYPRE_Int HYPRE_SStructMatrixSetEarlyAssemble(HYPRE_SStructMatrix matrix, HYPRE_Int early_assemble)

(Optional, GPU only) Sets if the matrix assemble routine does reductions of the IJ part before calling HYPRE_SStructMatrixAssemble.

See also the comments of HYPRE_IJMatrixSetEarlyAssemble. This early assemble feature may save the peak memory usage but requires extra work.

HYPRE_Int HYPRE_SStructMatrixSetValues(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *index, HYPRE_Int var, HYPRE_Int nentries, HYPRE_Int *entries, HYPRE_Complex *values)

Set matrix coefficients index by index.

The values array is of length nentries.

NOTE: For better efficiency, use HYPRE_SStructMatrixSetBoxValues to set coefficients a box at a time.

NOTE: Users are required to set values on all processes that own the associated variables. This means that some data will be multiply defined.

NOTE: The entries in this routine must all be of the same type: either stencil or non-stencil, but not both. Also, if they are stencil entries, they must all represent couplings to the same variable type (there are no such restrictions for non-stencil entries).

HYPRE_Int HYPRE_SStructMatrixAddToValues(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *index, HYPRE_Int var, HYPRE_Int nentries, HYPRE_Int *entries, HYPRE_Complex *values)

Add to matrix coefficients index by index.

The values array is of length nentries.

NOTE: For better efficiency, use HYPRE_SStructMatrixAddToBoxValues to set coefficients a box at a time.

NOTE: Users are required to set values on all processes that own the associated variables. This means that some data will be multiply defined.

NOTE: The entries in this routine must all be of the same type: either stencil or non-stencil, but not both. Also, if they are stencil entries, they must all represent couplings to the same variable type.

HYPRE_Int HYPRE_SStructMatrixAddFEMValues(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *index, HYPRE_Complex *values)

Add finite element stiffness matrix coefficients index by index.

The layout of the data in values is determined by the routines HYPRE_SStructGridSetFEMOrdering and HYPRE_SStructGraphSetFEMSparsity.

NOTE: For better efficiency, use HYPRE_SStructMatrixAddFEMBoxValues to set coefficients a box at a time.

HYPRE_Int HYPRE_SStructMatrixGetValues(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *index, HYPRE_Int var, HYPRE_Int nentries, HYPRE_Int *entries, HYPRE_Complex *values)

Get matrix coefficients index by index.

The values array is of length nentries.

NOTE: For better efficiency, use HYPRE_SStructMatrixGetBoxValues to get coefficients a box at a time.

NOTE: Users may get values on any process that owns the associated variables.

NOTE: The entries in this routine must all be of the same type: either stencil or non-stencil, but not both. Also, if they are stencil entries, they must all represent couplings to the same variable type (there are no such restrictions for non-stencil entries).

HYPRE_Int HYPRE_SStructMatrixGetFEMValues(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *index, HYPRE_Complex *values)

Get finite element stiffness matrix coefficients index by index.

The layout of the data in values is determined by the routines HYPRE_SStructGridSetFEMOrdering and HYPRE_SStructGraphSetFEMSparsity.

HYPRE_Int HYPRE_SStructMatrixSetBoxValues(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Int nentries, HYPRE_Int *entries, HYPRE_Complex *values)

Set matrix coefficients a box at a time.

The data in values is ordered as follows:

m = 0;
for (k = ilower[2]; k <= iupper[2]; k++)
   for (j = ilower[1]; j <= iupper[1]; j++)
      for (i = ilower[0]; i <= iupper[0]; i++)
         for (entry = 0; entry < nentries; entry++)
         {
            values[m] = ...;
            m++;
         }

NOTE: Users are required to set values on all processes that own the associated variables. This means that some data will be multiply defined.

NOTE: The entries in this routine must all be of the same type: either stencil or non-stencil, but not both. Also, if they are stencil entries, they must all represent couplings to the same variable type (there are no such restrictions for non-stencil entries).

HYPRE_Int HYPRE_SStructMatrixAddToBoxValues(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Int nentries, HYPRE_Int *entries, HYPRE_Complex *values)

Add to matrix coefficients a box at a time.

The data in values is ordered as in HYPRE_SStructMatrixSetBoxValues.

NOTE: Users are required to set values on all processes that own the associated variables. This means that some data will be multiply defined.

NOTE: The entries in this routine must all be of stencil type. Also, they must all represent couplings to the same variable type.

HYPRE_Int HYPRE_SStructMatrixSetBoxValues2(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Int nentries, HYPRE_Int *entries, HYPRE_Int *vilower, HYPRE_Int *viupper, HYPRE_Complex *values)

Set matrix coefficients a box at a time.

The values array is logically box shaped with value-box extents vilower and viupper that must contain the set-box extents ilower and iupper . The data in the values array is ordered as in HYPRE_SStructMatrixSetBoxValues, but based on the value-box extents.

HYPRE_Int HYPRE_SStructMatrixAddToBoxValues2(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Int nentries, HYPRE_Int *entries, HYPRE_Int *vilower, HYPRE_Int *viupper, HYPRE_Complex *values)

Add to matrix coefficients a box at a time.

The data in values is ordered as in HYPRE_SStructMatrixSetBoxValues2.

HYPRE_Int HYPRE_SStructMatrixAddFEMBoxValues(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Complex *values)

Add finite element stiffness matrix coefficients a box at a time.

The data in values is organized as an array of element matrices ordered as in HYPRE_SStructMatrixSetBoxValues. The layout of the data entries of each element matrix is determined by the routines HYPRE_SStructGridSetFEMOrdering and HYPRE_SStructGraphSetFEMSparsity.

HYPRE_Int HYPRE_SStructMatrixAssemble(HYPRE_SStructMatrix matrix): Finalize the construction of the matrix before using.

HYPRE_Int HYPRE_SStructMatrixGetBoxValues(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Int nentries, HYPRE_Int *entries, HYPRE_Complex *values)

Get matrix coefficients a box at a time.

The data in values is ordered as in HYPRE_SStructMatrixSetBoxValues.

NOTE: Users may get values on any process that owns the associated variables.

NOTE: The entries in this routine must all be of stencil type. Also, they must all represent couplings to the same variable type.

HYPRE_Int HYPRE_SStructMatrixGetBoxValues2(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Int nentries, HYPRE_Int *entries, HYPRE_Int *vilower, HYPRE_Int *viupper, HYPRE_Complex *values)

Get matrix coefficients a box at a time.

The data in values is ordered as in HYPRE_SStructMatrixSetBoxValues2.

HYPRE_Int HYPRE_SStructMatrixGetFEMBoxValues(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Complex *values): Does this even make sense to implement?

HYPRE_Int HYPRE_SStructMatrixSetSymmetric(HYPRE_SStructMatrix matrix, HYPRE_Int part, HYPRE_Int var, HYPRE_Int to_var, HYPRE_Int symmetric)

Define symmetry properties for the stencil entries in the matrix.

The boolean argument symmetric is applied to stencil entries on part part that couple variable var to variable to_var. A value of -1 may be used for part, var, or to_var to specify “all”. For example, if part and to_var are set to -1, then the boolean is applied to stencil entries on all parts that couple variable var to all other variables.

By default, matrices are assumed to be nonsymmetric. Significant storage savings can be made if the matrix is symmetric.

HYPRE_Int HYPRE_SStructMatrixSetNSSymmetric(HYPRE_SStructMatrix matrix, HYPRE_Int symmetric): Define symmetry properties for all non-stencil matrix entries.

HYPRE_Int HYPRE_SStructMatrixSetObjectType(HYPRE_SStructMatrix matrix, HYPRE_Int type)

Set the storage type of the matrix object to be constructed.

Currently, type can be either HYPRE_SSTRUCT (the default), HYPRE_STRUCT, or HYPRE_PARCSR.

See also

HYPRE_SStructMatrixGetObject

HYPRE_Int HYPRE_SStructMatrixGetObject(HYPRE_SStructMatrix matrix, void **object)

Get a reference to the constructed matrix object.

See also

HYPRE_SStructMatrixSetObjectType

HYPRE_Int HYPRE_SStructMatrixGetGrid(HYPRE_SStructMatrix matrix, HYPRE_SStructGrid *grid): Returns the grid object of a HYPRE_SStructMatrix.

HYPRE_Int HYPRE_SStructMatrixPrint(const char *filename, HYPRE_SStructMatrix matrix, HYPRE_Int all)

Print the matrix to file.

This is mainly for debugging purposes.

HYPRE_Int HYPRE_SStructMatrixToIJMatrix(HYPRE_SStructMatrix matrix, HYPRE_Int fill_diagonal, HYPRE_IJMatrix *ijmatrix)

Converts a SStructMatrix to an IJMatrix.

This will sum the structured and unstructured components of the input SStructMatrix and construct the resulting sum as an IJMatrix. When the flag fill_diagonal is turned on, the diagonal coefficient of ghost rows is set to 1.

HYPRE_Int HYPRE_SStructMatrixRead(MPI_Comm comm, const char *filename, HYPRE_SStructMatrix *matrix_ptr)

Read the matrix from file.

This is mainly for debugging purposes.

SStruct Vectors

typedef struct hypre_SStructVector_struct *HYPRE_SStructVector: The vector object.

HYPRE_Int HYPRE_SStructVectorCreate(MPI_Comm comm, HYPRE_SStructGrid grid, HYPRE_SStructVector *vector): Create a vector object.

HYPRE_Int HYPRE_SStructVectorDestroy(HYPRE_SStructVector vector): Destroy a vector object.

HYPRE_Int HYPRE_SStructVectorInitialize(HYPRE_SStructVector vector): Prepare a vector object for setting coefficient values.

HYPRE_Int HYPRE_SStructVectorSetValues(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *index, HYPRE_Int var, HYPRE_Complex *value)

Set vector coefficients index by index.

NOTE: For better efficiency, use HYPRE_SStructVectorSetBoxValues to set coefficients a box at a time.

NOTE: Users are required to set values on all processes that own the associated variables. This means that some data will be multiply defined.

HYPRE_Int HYPRE_SStructVectorSetConstantValues(HYPRE_SStructVector vector, HYPRE_Complex value): Set vector coefficients to a constant value over the grid.

HYPRE_Int HYPRE_SStructVectorSetRandomValues(HYPRE_SStructVector vector, HYPRE_Int seed)

Set vector coefficients to random values between -1.0 and 1.0 over the grid.

The parameter seed controls the generation of random numbers.

HYPRE_Int HYPRE_SStructVectorAddToValues(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *index, HYPRE_Int var, HYPRE_Complex *value)

Add to vector coefficients index by index.

NOTE: For better efficiency, use HYPRE_SStructVectorAddToBoxValues to set coefficients a box at a time.

NOTE: Users are required to set values on all processes that own the associated variables. This means that some data will be multiply defined.

HYPRE_Int HYPRE_SStructVectorAddFEMValues(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *index, HYPRE_Complex *values)

Add finite element vector coefficients index by index.

The layout of the data in values is determined by the routine HYPRE_SStructGridSetFEMOrdering.

NOTE: For better efficiency, use HYPRE_SStructVectorAddFEMBoxValues to set coefficients a box at a time.

HYPRE_Int HYPRE_SStructVectorGetValues(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *index, HYPRE_Int var, HYPRE_Complex *value)

Get vector coefficients index by index.

Users must first call the routine HYPRE_SStructVectorGather to ensure that data owned by multiple processes is correct.

NOTE: For better efficiency, use HYPRE_SStructVectorGetBoxValues to get coefficients a box at a time.

NOTE: Users may only get values on processes that own the associated variables.

HYPRE_Int HYPRE_SStructVectorGetFEMValues(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *index, HYPRE_Complex *values)

Get finite element vector coefficients index by index.

The layout of the data in values is determined by the routine HYPRE_SStructGridSetFEMOrdering. Users must first call the routine HYPRE_SStructVectorGather to ensure that data owned by multiple processes is correct.

HYPRE_Int HYPRE_SStructVectorSetBoxValues(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Complex *values)

Set vector coefficients a box at a time.

The data in values is ordered as follows:

m = 0;
for (k = ilower[2]; k <= iupper[2]; k++)
   for (j = ilower[1]; j <= iupper[1]; j++)
      for (i = ilower[0]; i <= iupper[0]; i++)
      {
         values[m] = ...;
         m++;
      }

NOTE: Users are required to set values on all processes that own the associated variables. This means that some data will be multiply defined.

HYPRE_Int HYPRE_SStructVectorAddToBoxValues(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Complex *values)

Add to vector coefficients a box at a time.

The data in values is ordered as in HYPRE_SStructVectorSetBoxValues.

NOTE: Users are required to set values on all processes that own the associated variables. This means that some data will be multiply defined.

HYPRE_Int HYPRE_SStructVectorSetBoxValues2(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Int *vilower, HYPRE_Int *viupper, HYPRE_Complex *values)

Set vector coefficients a box at a time.

The values array is logically box shaped with value-box extents vilower and viupper that must contain the set-box extents ilower and iupper . The data in the values array is ordered as in HYPRE_SStructVectorSetBoxValues, but based on the value-box extents.

HYPRE_Int HYPRE_SStructVectorAddToBoxValues2(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Int *vilower, HYPRE_Int *viupper, HYPRE_Complex *values)

Add to vector coefficients a box at a time.

The data in values is ordered as in HYPRE_SStructVectorSetBoxValues2.

HYPRE_Int HYPRE_SStructVectorAddFEMBoxValues(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Complex *values)

Add finite element vector coefficients a box at a time.

The data in values is organized as an array of element vectors ordered as in HYPRE_SStructVectorSetBoxValues. The layout of the data entries of each element vector is determined by the routine HYPRE_SStructGridSetFEMOrdering.

HYPRE_Int HYPRE_SStructVectorAssemble(HYPRE_SStructVector vector): Finalize the construction of the vector before using.

HYPRE_Int HYPRE_SStructVectorGetBoxValues(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Complex *values)

Get vector coefficients a box at a time.

The data in values is ordered as in HYPRE_SStructVectorSetBoxValues. Users must first call the routine HYPRE_SStructVectorGather to ensure that data owned by multiple processes is correct.

NOTE: Users may only get values on processes that own the associated variables.

HYPRE_Int HYPRE_SStructVectorGetBoxValues2(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Int var, HYPRE_Int *vilower, HYPRE_Int *viupper, HYPRE_Complex *values)

Get vector coefficients a box at a time.

The data in values is ordered as in HYPRE_SStructVectorSetBoxValues2.

HYPRE_Int HYPRE_SStructVectorGetFEMBoxValues(HYPRE_SStructVector vector, HYPRE_Int part, HYPRE_Int *ilower, HYPRE_Int *iupper, HYPRE_Complex *values): Does this even make sense to implement?

HYPRE_Int HYPRE_SStructVectorGather(HYPRE_SStructVector vector)

Gather vector data so that efficient GetValues can be done.

This routine must be called prior to calling GetValues to ensure that correct and consistent values are returned, especially for non cell-centered data that is shared between more than one processor.

HYPRE_Int HYPRE_SStructVectorSetObjectType(HYPRE_SStructVector vector, HYPRE_Int type)

Set the storage type of the vector object to be constructed.

Currently, type can be either HYPRE_SSTRUCT (the default), HYPRE_STRUCT, or HYPRE_PARCSR.

See also

HYPRE_SStructVectorGetObject

HYPRE_Int HYPRE_SStructVectorGetObject(HYPRE_SStructVector vector, void **object)

Get a reference to the constructed vector object.

See also

HYPRE_SStructVectorSetObjectType

HYPRE_Int HYPRE_SStructVectorPrint(const char *filename, HYPRE_SStructVector vector, HYPRE_Int all)

Print the vector to file.

This is mainly for debugging purposes.

HYPRE_Int HYPRE_SStructVectorRead(MPI_Comm comm, const char *filename, HYPRE_SStructVector *vector_ptr)

Read the vector from file.

This is mainly for debugging purposes.

HYPRE_Int HYPRE_SStructVectorPrintGLVis(HYPRE_SStructVector vector, const char *fileprefix)

SStruct Other

HYPRE_Int HYPRE_SStructGetAMRObjects(HYPRE_SStructMatrix matrix, HYPRE_SStructVector rhs, void **matrix_object, void **rhs_object)

AMRNEW.

Get a reference to the constructed matrix and right-hand-side (rhs) objects for an AMR system. This routine is similar to the routines HYPRE_SStructMatrixGetObject and HYPRE_SStructVectorGetObject, but ensures that trivial equations such as Dirichlet conditions are also satisfied exactly in the AMR system.

Basic Matrix/vector routines

HYPRE_Int HYPRE_SStructVectorCopy(HYPRE_SStructVector x, HYPRE_SStructVector y): Copy vector x into y ( \(y \leftarrow x\)).

HYPRE_Int HYPRE_SStructVectorScale(HYPRE_Complex alpha, HYPRE_SStructVector y): Scale a vector by alpha ( \(y \leftarrow \alpha y\)).

HYPRE_Int HYPRE_SStructVectorAxpy(HYPRE_Complex alpha, HYPRE_SStructVector x, HYPRE_SStructVector y): Compute \(y = y + \alpha x\).

HYPRE_Int HYPRE_SStructVectorInnerProd(HYPRE_SStructVector x, HYPRE_SStructVector y, HYPRE_Real *result): Compute result, the inner product of vectors x and y.

HYPRE_Int HYPRE_SStructMatrixMatvec(HYPRE_Complex alpha, HYPRE_SStructMatrix A, HYPRE_SStructVector x, HYPRE_Complex beta, HYPRE_SStructVector y)

Matvec operator.

This operation is \(y = \alpha A x + \beta y\) . Note that you can do a simple matrix-vector multiply by setting \(\alpha=1\) and \(\beta=0\).

HYPRE_Int HYPRE_SStructMatrixMatmat(HYPRE_SStructMatrix A, HYPRE_SStructMatrix B, HYPRE_SStructMatrix *C): Matrix-matrix multiply.

HYPRE_Int HYPRE_SStructMatrixScale(HYPRE_SStructMatrix matrix, HYPRE_Complex scalar): Scale a matrix by scalar ( \(A \leftarrow \text{scalar} \cdot A\)).