int
hypre_SMGRelaxSetupTempVec( void *relax_vdata,
hypre_StructMatrix *A,
hypre_StructVector *b,
hypre_StructVector *x )
{
hypre_SMGRelaxData *relax_data = relax_vdata;
hypre_StructVector *temp_vec = (relax_data -> temp_vec);
int ierr = 0;
/*----------------------------------------------------------
* Set up data
*----------------------------------------------------------*/
if ((relax_data -> temp_vec) == NULL)
{
temp_vec = hypre_StructVectorCreate(hypre_StructVectorComm(b),
hypre_StructVectorGrid(b));
hypre_StructVectorSetNumGhost(temp_vec, hypre_StructVectorNumGhost(b));
hypre_StructVectorInitialize(temp_vec);
hypre_StructVectorAssemble(temp_vec);
(relax_data -> temp_vec) = temp_vec;
}
(relax_data -> setup_temp_vec) = 0;
return ierr;
}
int
hypre_SMGRelaxSetupBaseBoxArray( void *relax_vdata,
hypre_StructMatrix *A,
hypre_StructVector *b,
hypre_StructVector *x )
{
hypre_SMGRelaxData *relax_data = relax_vdata;
hypre_StructGrid *grid;
hypre_BoxArray *boxes;
hypre_BoxArray *base_box_array;
int ierr = 0;
grid = hypre_StructVectorGrid(x);
boxes = hypre_StructGridBoxes(grid);
base_box_array = hypre_BoxArrayDuplicate(boxes);
hypre_ProjectBoxArray(base_box_array,
(relax_data -> base_index),
(relax_data -> base_stride));
(relax_data -> base_box_array) = base_box_array;
return ierr;
}
int
HYPRE_StructVectorSetBoxValues( HYPRE_StructVector vector,
int *ilower,
int *iupper,
double *values )
{
hypre_Index new_ilower;
hypre_Index new_iupper;
hypre_Box *new_value_box;
int d;
int ierr = 0;
hypre_ClearIndex(new_ilower);
hypre_ClearIndex(new_iupper);
for (d = 0; d < hypre_StructGridDim(hypre_StructVectorGrid(vector)); d++)
{
hypre_IndexD(new_ilower, d) = ilower[d];
hypre_IndexD(new_iupper, d) = iupper[d];
}
new_value_box = hypre_BoxCreate();
hypre_BoxSetExtents(new_value_box, new_ilower, new_iupper);
ierr = hypre_StructVectorSetBoxValues(vector, new_value_box, values, 0 );
hypre_BoxDestroy(new_value_box);
return ierr;
}
void *
hypre_StructKrylovCreateVector( void *vvector )
{
hypre_StructVector *vector = vvector;
hypre_StructVector *new_vector;
new_vector = hypre_StructVectorCreate( hypre_StructVectorComm(vector),
hypre_StructVectorGrid(vector) );
hypre_StructVectorInitialize(new_vector);
hypre_StructVectorAssemble(new_vector);
return ( (void *) new_vector );
}
int
hypre_StructCopy( hypre_StructVector *x,
hypre_StructVector *y )
{
int ierr = 0;
hypre_Box *x_data_box;
hypre_Box *y_data_box;
int xi;
int yi;
double *xp;
double *yp;
hypre_BoxArray *boxes;
hypre_Box *box;
hypre_Index loop_size;
hypre_IndexRef start;
hypre_Index unit_stride;
int i;
int loopi, loopj, loopk;
hypre_SetIndex(unit_stride, 1, 1, 1);
boxes = hypre_StructGridBoxes(hypre_StructVectorGrid(y));
hypre_ForBoxI(i, boxes)
{
box = hypre_BoxArrayBox(boxes, i);
start = hypre_BoxIMin(box);
x_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(x), i);
y_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(y), i);
xp = hypre_StructVectorBoxData(x, i);
yp = hypre_StructVectorBoxData(y, i);
hypre_BoxGetSize(box, loop_size);
hypre_BoxLoop2Begin(loop_size,
x_data_box, start, unit_stride, xi,
y_data_box, start, unit_stride, yi);
#define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,xi,yi
#include "hypre_box_smp_forloop.h"
hypre_BoxLoop2For(loopi, loopj, loopk, xi, yi)
{
yp[yi] = xp[xi];
}
hypre_BoxLoop2End(xi, yi);
}
int
hypre_SMGAxpy( double alpha,
hypre_StructVector *x,
hypre_StructVector *y,
hypre_Index base_index,
hypre_Index base_stride )
{
int ierr = 0;
hypre_Box *x_data_box;
hypre_Box *y_data_box;
int xi;
int yi;
double *xp;
double *yp;
hypre_BoxArray *boxes;
hypre_Box *box;
hypre_Index loop_size;
hypre_IndexRef start;
int i;
int loopi, loopj, loopk;
box = hypre_BoxCreate();
boxes = hypre_StructGridBoxes(hypre_StructVectorGrid(y));
hypre_ForBoxI(i, boxes)
{
hypre_CopyBox(hypre_BoxArrayBox(boxes, i), box);
hypre_ProjectBox(box, base_index, base_stride);
start = hypre_BoxIMin(box);
x_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(x), i);
y_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(y), i);
xp = hypre_StructVectorBoxData(x, i);
yp = hypre_StructVectorBoxData(y, i);
hypre_BoxGetStrideSize(box, base_stride, loop_size);
hypre_BoxLoop2Begin(loop_size,
x_data_box, start, base_stride, xi,
y_data_box, start, base_stride, yi);
#define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,xi,yi
#include "hypre_box_smp_forloop.h"
hypre_BoxLoop2For(loopi, loopj, loopk, xi, yi)
{
yp[yi] += alpha * xp[xi];
}
void *
hypre_StructKrylovCreateVector( void *vvector )
{
hypre_StructVector *vector = vvector;
hypre_StructVector *new_vector;
HYPRE_Int *num_ghost= hypre_StructVectorNumGhost(vector);
new_vector = hypre_StructVectorCreate( hypre_StructVectorComm(vector),
hypre_StructVectorGrid(vector) );
hypre_StructVectorSetNumGhost(new_vector, num_ghost);
hypre_StructVectorInitialize(new_vector);
hypre_StructVectorAssemble(new_vector);
return ( (void *) new_vector );
}
HYPRE_Int
hypre_StructVectorSetRandomValues( hypre_StructVector *vector,
HYPRE_Int seed )
{
hypre_Box *v_data_box;
HYPRE_Int vi;
double *vp;
hypre_BoxArray *boxes;
hypre_Box *box;
hypre_Index loop_size;
hypre_IndexRef start;
hypre_Index unit_stride;
HYPRE_Int i;
/*-----------------------------------------------------------------------
* Set the vector coefficients
*-----------------------------------------------------------------------*/
srand( seed );
hypre_SetIndex(unit_stride, 1, 1, 1);
boxes = hypre_StructGridBoxes(hypre_StructVectorGrid(vector));
hypre_ForBoxI(i, boxes)
{
box = hypre_BoxArrayBox(boxes, i);
start = hypre_BoxIMin(box);
v_data_box =
hypre_BoxArrayBox(hypre_StructVectorDataSpace(vector), i);
vp = hypre_StructVectorBoxData(vector, i);
hypre_BoxGetSize(box, loop_size);
hypre_BoxLoop1Begin(hypre_StructVectorDim(vector), loop_size,
v_data_box, start, unit_stride, vi);
#ifdef HYPRE_USING_OPENMP
#pragma omp parallel for private(HYPRE_BOX_PRIVATE,vi ) HYPRE_SMP_SCHEDULE
#endif
hypre_BoxLoop1For(vi)
{
vp[vi] = 2.0*rand()/RAND_MAX - 1.0;
}
hypre_BoxLoop1End(vi);
}
HYPRE_Int
hypre_SparseMSGInterpSetup( void *interp_vdata,
hypre_StructMatrix *P,
hypre_StructVector *xc,
hypre_StructVector *e,
hypre_Index cindex,
hypre_Index findex,
hypre_Index stride,
hypre_Index strideP )
{
hypre_SparseMSGInterpData *interp_data = interp_vdata;
hypre_StructGrid *grid;
hypre_StructStencil *stencil;
hypre_ComputeInfo *compute_info;
hypre_ComputePkg *compute_pkg;
HYPRE_Int ierr = 0;
/*----------------------------------------------------------
* Set up the compute package
*----------------------------------------------------------*/
grid = hypre_StructVectorGrid(e);
stencil = hypre_StructMatrixStencil(P);
hypre_CreateComputeInfo(grid, stencil, &compute_info);
hypre_ComputeInfoProjectSend(compute_info, cindex, stride);
hypre_ComputeInfoProjectRecv(compute_info, cindex, stride);
hypre_ComputeInfoProjectComp(compute_info, findex, stride);
hypre_ComputePkgCreate(compute_info, hypre_StructVectorDataSpace(e), 1,
grid, &compute_pkg);
/*----------------------------------------------------------
* Set up the interp data structure
*----------------------------------------------------------*/
(interp_data -> P) = hypre_StructMatrixRef(P);
(interp_data -> compute_pkg) = compute_pkg;
hypre_CopyIndex(cindex, (interp_data -> cindex));
hypre_CopyIndex(findex, (interp_data -> findex));
hypre_CopyIndex(stride, (interp_data -> stride));
hypre_CopyIndex(strideP, (interp_data -> strideP));
return ierr;
}
int
hypre_StructVectorInitializeShell( hypre_StructVector *vector )
{
int ierr = 0;
hypre_StructGrid *grid;
int *num_ghost;
hypre_BoxArray *data_space;
hypre_BoxArray *boxes;
hypre_Box *box;
hypre_Box *data_box;
int *data_indices;
int data_size;
int i, d;
/*-----------------------------------------------------------------------
* Set up data_space
*-----------------------------------------------------------------------*/
grid = hypre_StructVectorGrid(vector);
if (hypre_StructVectorDataSpace(vector) == NULL)
{
num_ghost = hypre_StructVectorNumGhost(vector);
boxes = hypre_StructGridBoxes(grid);
data_space = hypre_BoxArrayCreate(hypre_BoxArraySize(boxes));
hypre_ForBoxI(i, boxes)
{
box = hypre_BoxArrayBox(boxes, i);
data_box = hypre_BoxArrayBox(data_space, i);
hypre_CopyBox(box, data_box);
for (d = 0; d < 3; d++)
{
hypre_BoxIMinD(data_box, d) -= num_ghost[2*d];
hypre_BoxIMaxD(data_box, d) += num_ghost[2*d + 1];
}
}
HYPRE_Int
HYPRE_StructVectorSetValues( HYPRE_StructVector vector,
HYPRE_Int *grid_index,
double values )
{
hypre_Index new_grid_index;
HYPRE_Int d;
hypre_ClearIndex(new_grid_index);
for (d = 0; d < hypre_StructGridDim(hypre_StructVectorGrid(vector)); d++)
{
hypre_IndexD(new_grid_index, d) = grid_index[d];
}
hypre_StructVectorSetValues(vector, new_grid_index, &values, 0, -1, 0);
return hypre_error_flag;
}
int
hypre_StructScale( double alpha,
hypre_StructVector *y )
{
int ierr = 0;
hypre_Box *y_data_box;
int yi;
double *yp;
hypre_BoxArray *boxes;
hypre_Box *box;
hypre_Index loop_size;
hypre_IndexRef start;
hypre_Index unit_stride;
int i;
int loopi, loopj, loopk;
hypre_SetIndex(unit_stride, 1, 1, 1);
boxes = hypre_StructGridBoxes(hypre_StructVectorGrid(y));
hypre_ForBoxI(i, boxes)
{
box = hypre_BoxArrayBox(boxes, i);
start = hypre_BoxIMin(box);
y_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(y), i);
yp = hypre_StructVectorBoxData(y, i);
hypre_BoxGetSize(box, loop_size);
hypre_BoxLoop1Begin(loop_size,
y_data_box, start, unit_stride, yi);
#define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,yi
#include "hypre_box_smp_forloop.h"
hypre_BoxLoop1For(loopi, loopj, loopk, yi)
{
yp[yi] *= alpha;
}
hypre_BoxLoop1End(yi);
}
void *
hypre_StructKrylovCreateVectorArray(int n, void *vvector )
{
hypre_StructVector *vector = vvector;
hypre_StructVector **new_vector;
int i;
new_vector = hypre_CTAlloc(hypre_StructVector*,n);
for (i=0; i < n; i++)
{
HYPRE_StructVectorCreate(hypre_StructVectorComm(vector),
hypre_StructVectorGrid(vector),
(HYPRE_StructVector *) &new_vector[i] );
HYPRE_StructVectorInitialize((HYPRE_StructVector) new_vector[i]);
HYPRE_StructVectorAssemble((HYPRE_StructVector) new_vector[i]);
}
return ( (void *) new_vector );
}
int
HYPRE_StructVectorSetValues( HYPRE_StructVector vector,
int *grid_index,
double values )
{
hypre_Index new_grid_index;
int d;
int ierr = 0;
hypre_ClearIndex(new_grid_index);
for (d = 0; d < hypre_StructGridDim(hypre_StructVectorGrid(vector)); d++)
{
hypre_IndexD(new_grid_index, d) = grid_index[d];
}
ierr = hypre_StructVectorSetValues(vector, new_grid_index, values, 0);
return ierr;
}
HYPRE_Int
hypre_StructScale( double alpha,
hypre_StructVector *y )
{
hypre_Box *y_data_box;
HYPRE_Int yi;
double *yp;
hypre_BoxArray *boxes;
hypre_Box *box;
hypre_Index loop_size;
hypre_IndexRef start;
hypre_Index unit_stride;
HYPRE_Int i;
hypre_SetIndex(unit_stride, 1, 1, 1);
boxes = hypre_StructGridBoxes(hypre_StructVectorGrid(y));
hypre_ForBoxI(i, boxes)
{
box = hypre_BoxArrayBox(boxes, i);
start = hypre_BoxIMin(box);
y_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(y), i);
yp = hypre_StructVectorBoxData(y, i);
hypre_BoxGetSize(box, loop_size);
hypre_BoxLoop1Begin(hypre_StructVectorDim(y), loop_size,
y_data_box, start, unit_stride, yi);
#ifdef HYPRE_USING_OPENMP
#pragma omp parallel for private(HYPRE_BOX_PRIVATE,yi) HYPRE_SMP_SCHEDULE
#endif
hypre_BoxLoop1For(yi)
{
yp[yi] *= alpha;
}
hypre_BoxLoop1End(yi);
}
HYPRE_Int
hypre_StructVectorDestroy( hypre_StructVector *vector )
{
if (vector)
{
hypre_StructVectorRefCount(vector) --;
if (hypre_StructVectorRefCount(vector) == 0)
{
if (hypre_StructVectorDataAlloced(vector))
{
hypre_SharedTFree(hypre_StructVectorData(vector));
}
hypre_TFree(hypre_StructVectorDataIndices(vector));
hypre_BoxArrayDestroy(hypre_StructVectorDataSpace(vector));
hypre_StructGridDestroy(hypre_StructVectorGrid(vector));
hypre_TFree(vector);
}
}
return hypre_error_flag;
}
hypre_StructVector *
hypre_StructVectorCreate( MPI_Comm comm,
hypre_StructGrid *grid )
{
hypre_StructVector *vector;
int i;
vector = hypre_CTAlloc(hypre_StructVector, 1);
hypre_StructVectorComm(vector) = comm;
hypre_StructGridRef(grid, &hypre_StructVectorGrid(vector));
hypre_StructVectorDataAlloced(vector) = 1;
hypre_StructVectorRefCount(vector) = 1;
/* set defaults */
for (i = 0; i < 6; i++)
hypre_StructVectorNumGhost(vector)[i] = 1;
return vector;
}
HYPRE_Int
hypre_SparseMSGInterp( void *interp_vdata,
hypre_StructMatrix *P,
hypre_StructVector *xc,
hypre_StructVector *e )
{
HYPRE_Int ierr = 0;
hypre_SparseMSGInterpData *interp_data = interp_vdata;
hypre_ComputePkg *compute_pkg;
hypre_IndexRef cindex;
hypre_IndexRef findex;
hypre_IndexRef stride;
hypre_IndexRef strideP;
hypre_StructGrid *fgrid;
HYPRE_Int *fgrid_ids;
hypre_StructGrid *cgrid;
hypre_BoxArray *cgrid_boxes;
HYPRE_Int *cgrid_ids;
hypre_CommHandle *comm_handle;
hypre_BoxArrayArray *compute_box_aa;
hypre_BoxArray *compute_box_a;
hypre_Box *compute_box;
hypre_Box *P_dbox;
hypre_Box *xc_dbox;
hypre_Box *e_dbox;
HYPRE_Int Pi;
HYPRE_Int xci;
HYPRE_Int ei;
double *Pp0, *Pp1;
double *xcp;
double *ep, *ep0, *ep1;
hypre_Index loop_size;
hypre_Index start;
hypre_Index startc;
hypre_Index startP;
hypre_Index stridec;
hypre_StructStencil *stencil;
hypre_Index *stencil_shape;
HYPRE_Int compute_i, fi, ci, j;
HYPRE_Int loopi, loopj, loopk;
/*-----------------------------------------------------------------------
* Initialize some things
*-----------------------------------------------------------------------*/
hypre_BeginTiming(interp_data -> time_index);
compute_pkg = (interp_data -> compute_pkg);
cindex = (interp_data -> cindex);
findex = (interp_data -> findex);
stride = (interp_data -> stride);
strideP = (interp_data -> strideP);
stencil = hypre_StructMatrixStencil(P);
stencil_shape = hypre_StructStencilShape(stencil);
hypre_SetIndex(stridec, 1, 1, 1);
/*-----------------------------------------------------------------------
* Compute e at coarse points (injection)
*-----------------------------------------------------------------------*/
fgrid = hypre_StructVectorGrid(e);
fgrid_ids = hypre_StructGridIDs(fgrid);
cgrid = hypre_StructVectorGrid(xc);
cgrid_boxes = hypre_StructGridBoxes(cgrid);
cgrid_ids = hypre_StructGridIDs(cgrid);
fi = 0;
hypre_ForBoxI(ci, cgrid_boxes)
{
while (fgrid_ids[fi] != cgrid_ids[ci])
{
fi++;
}
compute_box = hypre_BoxArrayBox(cgrid_boxes, ci);
hypre_CopyIndex(hypre_BoxIMin(compute_box), startc);
hypre_StructMapCoarseToFine(startc, cindex, stride, start);
e_dbox = hypre_BoxArrayBox(hypre_StructVectorDataSpace(e), fi);
xc_dbox = hypre_BoxArrayBox(hypre_StructVectorDataSpace(xc), ci);
ep = hypre_StructVectorBoxData(e, fi);
xcp = hypre_StructVectorBoxData(xc, ci);
hypre_BoxGetSize(compute_box, loop_size);
hypre_BoxLoop2Begin(loop_size,
e_dbox, start, stride, ei,
xc_dbox, startc, stridec, xci);
#define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,ei,xci
#include "hypre_box_smp_forloop.h"
hypre_BoxLoop2For(loopi, loopj, loopk, ei, xci)
{
ep[ei] = xcp[xci];
}
hypre_BoxLoop2End(ei, xci);
}
/*-----------------------------------------------------------------------
* Compute e at fine points
*-----------------------------------------------------------------------*/
for (compute_i = 0; compute_i < 2; compute_i++)
{
switch(compute_i)
{
case 0:
{
ep = hypre_StructVectorData(e);
hypre_InitializeIndtComputations(compute_pkg, ep, &comm_handle);
compute_box_aa = hypre_ComputePkgIndtBoxes(compute_pkg);
}
break;
case 1:
{
hypre_FinalizeIndtComputations(comm_handle);
compute_box_aa = hypre_ComputePkgDeptBoxes(compute_pkg);
}
break;
}
hypre_ForBoxArrayI(fi, compute_box_aa)
{
compute_box_a = hypre_BoxArrayArrayBoxArray(compute_box_aa, fi);
P_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(P), fi);
e_dbox = hypre_BoxArrayBox(hypre_StructVectorDataSpace(e), fi);
Pp0 = hypre_StructMatrixBoxData(P, fi, 0);
Pp1 = hypre_StructMatrixBoxData(P, fi, 1);
ep = hypre_StructVectorBoxData(e, fi);
ep0 = ep + hypre_BoxOffsetDistance(e_dbox, stencil_shape[0]);
ep1 = ep + hypre_BoxOffsetDistance(e_dbox, stencil_shape[1]);
hypre_ForBoxI(j, compute_box_a)
{
compute_box = hypre_BoxArrayBox(compute_box_a, j);
hypre_CopyIndex(hypre_BoxIMin(compute_box), start);
hypre_StructMapFineToCoarse(start, findex, stride, startc);
hypre_StructMapCoarseToFine(startc, cindex, strideP, startP);
hypre_BoxGetStrideSize(compute_box, stride, loop_size);
hypre_BoxLoop2Begin(loop_size,
P_dbox, startP, strideP, Pi,
e_dbox, start, stride, ei);
#define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,Pi,ei
#include "hypre_box_smp_forloop.h"
hypre_BoxLoop2For(loopi, loopj, loopk, Pi, ei)
{
ep[ei] = (Pp0[Pi] * ep0[ei] +
Pp1[Pi] * ep1[ei]);
}
hypre_BoxLoop2End(Pi, ei);
}
}
int
hypre_PointRelaxSetup( void *relax_vdata,
hypre_StructMatrix *A,
hypre_StructVector *b,
hypre_StructVector *x )
{
hypre_PointRelaxData *relax_data = (hypre_PointRelaxData *)relax_vdata;
int num_pointsets = (relax_data -> num_pointsets);
int *pointset_sizes = (relax_data -> pointset_sizes);
hypre_Index *pointset_strides = (relax_data -> pointset_strides);
hypre_Index **pointset_indices = (relax_data -> pointset_indices);
hypre_StructVector *t;
int diag_rank;
hypre_ComputePkg **compute_pkgs;
hypre_Index unit_stride;
hypre_Index diag_index;
hypre_IndexRef stride;
hypre_IndexRef index;
hypre_StructGrid *grid;
hypre_StructStencil *stencil;
hypre_BoxArrayArray *send_boxes;
hypre_BoxArrayArray *recv_boxes;
int **send_processes;
int **recv_processes;
hypre_BoxArrayArray *indt_boxes;
hypre_BoxArrayArray *dept_boxes;
hypre_BoxArrayArray *orig_indt_boxes;
hypre_BoxArrayArray *orig_dept_boxes;
hypre_BoxArrayArray *box_aa;
hypre_BoxArray *box_a;
hypre_Box *box;
int box_aa_size;
int box_a_size;
hypre_BoxArrayArray *new_box_aa;
hypre_BoxArray *new_box_a;
hypre_Box *new_box;
double scale;
int frac;
int i, j, k, p, m, compute_i;
int ierr = 0;
/*----------------------------------------------------------
* Set up the temp vector
*----------------------------------------------------------*/
if ((relax_data -> t) == NULL)
{
t = hypre_StructVectorCreate(hypre_StructVectorComm(b),
hypre_StructVectorGrid(b));
hypre_StructVectorSetNumGhost(t, hypre_StructVectorNumGhost(b));
hypre_StructVectorInitialize(t);
hypre_StructVectorAssemble(t);
(relax_data -> t) = t;
}
/*----------------------------------------------------------
* Find the matrix diagonal
*----------------------------------------------------------*/
grid = hypre_StructMatrixGrid(A);
stencil = hypre_StructMatrixStencil(A);
hypre_SetIndex(diag_index, 0, 0, 0);
diag_rank = hypre_StructStencilElementRank(stencil, diag_index);
/*----------------------------------------------------------
* Set up the compute packages
*----------------------------------------------------------*/
hypre_SetIndex(unit_stride, 1, 1, 1);
compute_pkgs = hypre_CTAlloc(hypre_ComputePkg *, num_pointsets);
for (p = 0; p < num_pointsets; p++)
{
hypre_CreateComputeInfo(grid, stencil,
&send_boxes, &recv_boxes,
&send_processes, &recv_processes,
&orig_indt_boxes, &orig_dept_boxes);
stride = pointset_strides[p];
for (compute_i = 0; compute_i < 2; compute_i++)
{
switch(compute_i)
{
case 0:
box_aa = orig_indt_boxes;
break;
case 1:
box_aa = orig_dept_boxes;
break;
}
box_aa_size = hypre_BoxArrayArraySize(box_aa);
new_box_aa = hypre_BoxArrayArrayCreate(box_aa_size);
for (i = 0; i < box_aa_size; i++)
{
box_a = hypre_BoxArrayArrayBoxArray(box_aa, i);
box_a_size = hypre_BoxArraySize(box_a);
new_box_a = hypre_BoxArrayArrayBoxArray(new_box_aa, i);
hypre_BoxArraySetSize(new_box_a, box_a_size * pointset_sizes[p]);
k = 0;
for (m = 0; m < pointset_sizes[p]; m++)
{
index = pointset_indices[p][m];
for (j = 0; j < box_a_size; j++)
{
box = hypre_BoxArrayBox(box_a, j);
new_box = hypre_BoxArrayBox(new_box_a, k);
hypre_CopyBox(box, new_box);
hypre_ProjectBox(new_box, index, stride);
k++;
}
}
}
switch(compute_i)
{
case 0:
indt_boxes = new_box_aa;
break;
case 1:
dept_boxes = new_box_aa;
break;
}
}
hypre_ComputePkgCreate(send_boxes, recv_boxes,
unit_stride, unit_stride,
send_processes, recv_processes,
indt_boxes, dept_boxes,
stride, grid,
hypre_StructVectorDataSpace(x), 1,
&compute_pkgs[p]);
hypre_BoxArrayArrayDestroy(orig_indt_boxes);
hypre_BoxArrayArrayDestroy(orig_dept_boxes);
}
/*----------------------------------------------------------
* Set up the relax data structure
*----------------------------------------------------------*/
(relax_data -> A) = hypre_StructMatrixRef(A);
(relax_data -> x) = hypre_StructVectorRef(x);
(relax_data -> b) = hypre_StructVectorRef(b);
(relax_data -> diag_rank) = diag_rank;
(relax_data -> compute_pkgs) = compute_pkgs;
/*-----------------------------------------------------
* Compute flops
*-----------------------------------------------------*/
scale = 0.0;
for (p = 0; p < num_pointsets; p++)
{
stride = pointset_strides[p];
frac = hypre_IndexX(stride);
frac *= hypre_IndexY(stride);
frac *= hypre_IndexZ(stride);
scale += (pointset_sizes[p] / frac);
}
(relax_data -> flops) = scale * (hypre_StructMatrixGlobalSize(A) +
hypre_StructVectorGlobalSize(x));
return ierr;
}
int
hypre_SemiRestrictSetup( void *restrict_vdata,
hypre_StructMatrix *R,
int R_stored_as_transpose,
hypre_StructVector *r,
hypre_StructVector *rc,
hypre_Index cindex,
hypre_Index findex,
hypre_Index stride )
{
hypre_SemiRestrictData *restrict_data = (hypre_SemiRestrictData *)restrict_vdata;
hypre_StructGrid *grid;
hypre_StructStencil *stencil;
hypre_BoxArrayArray *send_boxes;
hypre_BoxArrayArray *recv_boxes;
int **send_processes;
int **recv_processes;
hypre_BoxArrayArray *indt_boxes;
hypre_BoxArrayArray *dept_boxes;
hypre_ComputePkg *compute_pkg;
int ierr = 0;
/*----------------------------------------------------------
* Set up the compute package
*----------------------------------------------------------*/
grid = hypre_StructVectorGrid(r);
stencil = hypre_StructMatrixStencil(R);
hypre_CreateComputeInfo(grid, stencil,
&send_boxes, &recv_boxes,
&send_processes, &recv_processes,
&indt_boxes, &dept_boxes);
hypre_ProjectBoxArrayArray(send_boxes, findex, stride);
hypre_ProjectBoxArrayArray(recv_boxes, findex, stride);
hypre_ProjectBoxArrayArray(indt_boxes, cindex, stride);
hypre_ProjectBoxArrayArray(dept_boxes, cindex, stride);
hypre_ComputePkgCreate(send_boxes, recv_boxes,
stride, stride,
send_processes, recv_processes,
indt_boxes, dept_boxes,
stride, grid,
hypre_StructVectorDataSpace(r), 1,
&compute_pkg);
/*----------------------------------------------------------
* Set up the restrict data structure
*----------------------------------------------------------*/
(restrict_data -> R) = hypre_StructMatrixRef(R);
(restrict_data -> R_stored_as_transpose) = R_stored_as_transpose;
(restrict_data -> compute_pkg) = compute_pkg;
hypre_CopyIndex(cindex ,(restrict_data -> cindex));
hypre_CopyIndex(stride ,(restrict_data -> stride));
return ierr;
}
int
hypre_SemiRestrict( void *restrict_vdata,
hypre_StructMatrix *R,
hypre_StructVector *r,
hypre_StructVector *rc )
{
int ierr = 0;
hypre_SemiRestrictData *restrict_data = (hypre_SemiRestrictData *)restrict_vdata;
int R_stored_as_transpose;
hypre_ComputePkg *compute_pkg;
hypre_IndexRef cindex;
hypre_IndexRef stride;
hypre_StructGrid *fgrid;
int *fgrid_ids;
hypre_StructGrid *cgrid;
hypre_BoxArray *cgrid_boxes;
int *cgrid_ids;
hypre_CommHandle *comm_handle;
hypre_BoxArrayArray *compute_box_aa;
hypre_BoxArray *compute_box_a;
hypre_Box *compute_box;
hypre_Box *R_dbox;
hypre_Box *r_dbox;
hypre_Box *rc_dbox;
int Ri;
int ri;
int rci;
double *Rp0, *Rp1;
double *rp, *rp0, *rp1;
double *rcp;
hypre_Index loop_size;
hypre_IndexRef start;
hypre_Index startc;
hypre_Index stridec;
hypre_StructStencil *stencil;
hypre_Index *stencil_shape;
int compute_i, fi, ci, j;
int loopi, loopj, loopk;
/*-----------------------------------------------------------------------
* Initialize some things.
*-----------------------------------------------------------------------*/
hypre_BeginTiming(restrict_data -> time_index);
R_stored_as_transpose = (restrict_data -> R_stored_as_transpose);
compute_pkg = (restrict_data -> compute_pkg);
cindex = (restrict_data -> cindex);
stride = (restrict_data -> stride);
stencil = hypre_StructMatrixStencil(R);
stencil_shape = hypre_StructStencilShape(stencil);
hypre_SetIndex(stridec, 1, 1, 1);
/*--------------------------------------------------------------------
* Restrict the residual.
*--------------------------------------------------------------------*/
fgrid = hypre_StructVectorGrid(r);
fgrid_ids = hypre_StructGridIDs(fgrid);
cgrid = hypre_StructVectorGrid(rc);
cgrid_boxes = hypre_StructGridBoxes(cgrid);
cgrid_ids = hypre_StructGridIDs(cgrid);
for (compute_i = 0; compute_i < 2; compute_i++)
{
switch(compute_i)
{
case 0:
{
rp = hypre_StructVectorData(r);
hypre_InitializeIndtComputations(compute_pkg, rp, &comm_handle);
compute_box_aa = hypre_ComputePkgIndtBoxes(compute_pkg);
}
break;
case 1:
{
hypre_FinalizeIndtComputations(comm_handle);
compute_box_aa = hypre_ComputePkgDeptBoxes(compute_pkg);
}
break;
}
fi = 0;
hypre_ForBoxArrayI(ci, cgrid_boxes)
{
while (fgrid_ids[fi] != cgrid_ids[ci])
{
fi++;
}
compute_box_a = hypre_BoxArrayArrayBoxArray(compute_box_aa, fi);
R_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(R), fi);
r_dbox = hypre_BoxArrayBox(hypre_StructVectorDataSpace(r), fi);
rc_dbox = hypre_BoxArrayBox(hypre_StructVectorDataSpace(rc), ci);
if (R_stored_as_transpose)
{
Rp0 = hypre_StructMatrixBoxData(R, fi, 1) -
hypre_BoxOffsetDistance(R_dbox, stencil_shape[1]);
Rp1 = hypre_StructMatrixBoxData(R, fi, 0);
}
else
{
Rp0 = hypre_StructMatrixBoxData(R, fi, 0);
Rp1 = hypre_StructMatrixBoxData(R, fi, 1);
}
rp = hypre_StructVectorBoxData(r, fi);
rp0 = rp + hypre_BoxOffsetDistance(r_dbox, stencil_shape[0]);
rp1 = rp + hypre_BoxOffsetDistance(r_dbox, stencil_shape[1]);
rcp = hypre_StructVectorBoxData(rc, ci);
hypre_ForBoxI(j, compute_box_a)
{
compute_box = hypre_BoxArrayBox(compute_box_a, j);
start = hypre_BoxIMin(compute_box);
hypre_StructMapFineToCoarse(start, cindex, stride, startc);
hypre_BoxGetStrideSize(compute_box, stride, loop_size);
hypre_BoxLoop3Begin(loop_size,
R_dbox, startc, stridec, Ri,
r_dbox, start, stride, ri,
rc_dbox, startc, stridec, rci);
#define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,Ri,ri,rci
#include "hypre_box_smp_forloop.h"
hypre_BoxLoop3For(loopi, loopj, loopk, Ri, ri, rci)
{
rcp[rci] = rp[ri] + (Rp0[Ri] * rp0[ri] +
Rp1[Ri] * rp1[ri]);
}
hypre_BoxLoop3End(Ri, ri, rci);
}
}
int
hypre_SemiInterpSetup( void *interp_vdata,
hypre_StructMatrix *P,
int P_stored_as_transpose,
hypre_StructVector *xc,
hypre_StructVector *e,
hypre_Index cindex,
hypre_Index findex,
hypre_Index stride )
{
hypre_SemiInterpData *interp_data = interp_vdata;
hypre_StructGrid *grid;
hypre_StructStencil *stencil;
hypre_BoxArrayArray *send_boxes;
hypre_BoxArrayArray *recv_boxes;
int **send_processes;
int **recv_processes;
hypre_BoxArrayArray *indt_boxes;
hypre_BoxArrayArray *dept_boxes;
hypre_ComputePkg *compute_pkg;
int ierr = 0;
/*----------------------------------------------------------
* Set up the compute package
*----------------------------------------------------------*/
grid = hypre_StructVectorGrid(e);
stencil = hypre_StructMatrixStencil(P);
hypre_CreateComputeInfo(grid, stencil,
&send_boxes, &recv_boxes,
&send_processes, &recv_processes,
&indt_boxes, &dept_boxes);
hypre_ProjectBoxArrayArray(send_boxes, cindex, stride);
hypre_ProjectBoxArrayArray(recv_boxes, cindex, stride);
hypre_ProjectBoxArrayArray(indt_boxes, findex, stride);
hypre_ProjectBoxArrayArray(dept_boxes, findex, stride);
hypre_ComputePkgCreate(send_boxes, recv_boxes,
stride, stride,
send_processes, recv_processes,
indt_boxes, dept_boxes,
stride, grid,
hypre_StructVectorDataSpace(e), 1,
&compute_pkg);
/*----------------------------------------------------------
* Set up the interp data structure
*----------------------------------------------------------*/
(interp_data -> P) = hypre_StructMatrixRef(P);
(interp_data -> P_stored_as_transpose) = P_stored_as_transpose;
(interp_data -> compute_pkg) = compute_pkg;
hypre_CopyIndex(cindex, (interp_data -> cindex));
hypre_CopyIndex(findex, (interp_data -> findex));
hypre_CopyIndex(stride, (interp_data -> stride));
return ierr;
}
HYPRE_Int
hypre_SemiInterp( void *interp_vdata,
hypre_StructMatrix *P,
hypre_StructVector *xc,
hypre_StructVector *e )
{
hypre_SemiInterpData *interp_data = interp_vdata;
HYPRE_Int P_stored_as_transpose;
hypre_ComputePkg *compute_pkg;
hypre_IndexRef cindex;
hypre_IndexRef findex;
hypre_IndexRef stride;
hypre_StructGrid *fgrid;
HYPRE_Int *fgrid_ids;
hypre_StructGrid *cgrid;
hypre_BoxArray *cgrid_boxes;
HYPRE_Int *cgrid_ids;
hypre_CommHandle *comm_handle;
hypre_BoxArrayArray *compute_box_aa;
hypre_BoxArray *compute_box_a;
hypre_Box *compute_box;
hypre_Box *P_dbox;
hypre_Box *xc_dbox;
hypre_Box *e_dbox;
HYPRE_Int Pi;
HYPRE_Int xci;
HYPRE_Int ei;
HYPRE_Int constant_coefficient;
HYPRE_Real *Pp0, *Pp1;
HYPRE_Real *xcp;
HYPRE_Real *ep, *ep0, *ep1;
hypre_Index loop_size;
hypre_Index start;
hypre_Index startc;
hypre_Index stridec;
hypre_StructStencil *stencil;
hypre_Index *stencil_shape;
HYPRE_Int compute_i, fi, ci, j;
/*-----------------------------------------------------------------------
* Initialize some things
*-----------------------------------------------------------------------*/
hypre_BeginTiming(interp_data -> time_index);
P_stored_as_transpose = (interp_data -> P_stored_as_transpose);
compute_pkg = (interp_data -> compute_pkg);
cindex = (interp_data -> cindex);
findex = (interp_data -> findex);
stride = (interp_data -> stride);
stencil = hypre_StructMatrixStencil(P);
stencil_shape = hypre_StructStencilShape(stencil);
constant_coefficient = hypre_StructMatrixConstantCoefficient(P);
hypre_assert( constant_coefficient==0 || constant_coefficient==1 );
/* ... constant_coefficient==2 for P shouldn't happen, see
hypre_PFMGCreateInterpOp in pfmg_setup_interp.c */
if (constant_coefficient) hypre_StructVectorClearBoundGhostValues(e, 0);
hypre_SetIndex3(stridec, 1, 1, 1);
/*-----------------------------------------------------------------------
* Compute e at coarse points (injection)
*-----------------------------------------------------------------------*/
fgrid = hypre_StructVectorGrid(e);
fgrid_ids = hypre_StructGridIDs(fgrid);
cgrid = hypre_StructVectorGrid(xc);
cgrid_boxes = hypre_StructGridBoxes(cgrid);
cgrid_ids = hypre_StructGridIDs(cgrid);
fi = 0;
hypre_ForBoxI(ci, cgrid_boxes)
{
while (fgrid_ids[fi] != cgrid_ids[ci])
{
fi++;
}
compute_box = hypre_BoxArrayBox(cgrid_boxes, ci);
hypre_CopyIndex(hypre_BoxIMin(compute_box), startc);
hypre_StructMapCoarseToFine(startc, cindex, stride, start);
e_dbox = hypre_BoxArrayBox(hypre_StructVectorDataSpace(e), fi);
xc_dbox = hypre_BoxArrayBox(hypre_StructVectorDataSpace(xc), ci);
ep = hypre_StructVectorBoxData(e, fi);
xcp = hypre_StructVectorBoxData(xc, ci);
hypre_BoxGetSize(compute_box, loop_size);
hypre_BoxLoop2Begin(hypre_StructMatrixNDim(P), loop_size,
e_dbox, start, stride, ei,
xc_dbox, startc, stridec, xci);
#ifdef HYPRE_USING_OPENMP
#pragma omp parallel for private(HYPRE_BOX_PRIVATE,ei,xci) HYPRE_SMP_SCHEDULE
#endif
hypre_BoxLoop2For(ei, xci)
{
ep[ei] = xcp[xci];
}
hypre_BoxLoop2End(ei, xci);
}
/*-----------------------------------------------------------------------
* Compute e at fine points
*-----------------------------------------------------------------------*/
for (compute_i = 0; compute_i < 2; compute_i++)
{
switch(compute_i)
{
case 0:
{
ep = hypre_StructVectorData(e);
hypre_InitializeIndtComputations(compute_pkg, ep, &comm_handle);
compute_box_aa = hypre_ComputePkgIndtBoxes(compute_pkg);
}
break;
case 1:
{
hypre_FinalizeIndtComputations(comm_handle);
compute_box_aa = hypre_ComputePkgDeptBoxes(compute_pkg);
}
break;
}
hypre_ForBoxArrayI(fi, compute_box_aa)
{
compute_box_a = hypre_BoxArrayArrayBoxArray(compute_box_aa, fi);
P_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(P), fi);
e_dbox = hypre_BoxArrayBox(hypre_StructVectorDataSpace(e), fi);
if (P_stored_as_transpose)
{
if ( constant_coefficient )
{
Pp0 = hypre_StructMatrixBoxData(P, fi, 1);
Pp1 = hypre_StructMatrixBoxData(P, fi, 0) -
hypre_CCBoxOffsetDistance(P_dbox, stencil_shape[0]);
}
else
{
Pp0 = hypre_StructMatrixBoxData(P, fi, 1);
Pp1 = hypre_StructMatrixBoxData(P, fi, 0) -
hypre_BoxOffsetDistance(P_dbox, stencil_shape[0]);
}
}
else
{
Pp0 = hypre_StructMatrixBoxData(P, fi, 0);
Pp1 = hypre_StructMatrixBoxData(P, fi, 1);
}
ep = hypre_StructVectorBoxData(e, fi);
ep0 = ep + hypre_BoxOffsetDistance(e_dbox, stencil_shape[0]);
ep1 = ep + hypre_BoxOffsetDistance(e_dbox, stencil_shape[1]);
hypre_ForBoxI(j, compute_box_a)
{
compute_box = hypre_BoxArrayBox(compute_box_a, j);
hypre_CopyIndex(hypre_BoxIMin(compute_box), start);
hypre_StructMapFineToCoarse(start, findex, stride, startc);
hypre_BoxGetStrideSize(compute_box, stride, loop_size);
if ( constant_coefficient )
{
Pi = hypre_CCBoxIndexRank( P_dbox, startc );
hypre_BoxLoop1Begin(hypre_StructMatrixNDim(P), loop_size,
e_dbox, start, stride, ei);
#ifdef HYPRE_USING_OPENMP
#pragma omp parallel for private(HYPRE_BOX_PRIVATE,ei) HYPRE_SMP_SCHEDULE
#endif
hypre_BoxLoop1For(ei)
{
ep[ei] = (Pp0[Pi] * ep0[ei] +
Pp1[Pi] * ep1[ei]);
}
hypre_BoxLoop1End(ei);
}
else
{
hypre_BoxLoop2Begin(hypre_StructMatrixNDim(P), loop_size,
P_dbox, startc, stridec, Pi,
e_dbox, start, stride, ei);
#ifdef HYPRE_USING_OPENMP
#pragma omp parallel for private(HYPRE_BOX_PRIVATE,Pi,ei) HYPRE_SMP_SCHEDULE
#endif
hypre_BoxLoop2For(Pi, ei)
{
ep[ei] = (Pp0[Pi] * ep0[ei] +
Pp1[Pi] * ep1[ei]);
}
hypre_BoxLoop2End(Pi, ei);
}
}
double
hypre_StructOverlapInnerProd( hypre_StructVector *x,
hypre_StructVector *y )
{
double final_innerprod_result;
double local_result, overlap_result;
double process_result;
hypre_Box *x_data_box;
hypre_Box *y_data_box;
hypre_BoxArray *overlap_boxes;
HYPRE_Int xi;
HYPRE_Int yi;
double *xp;
double *yp;
hypre_BoxArray *boxes;
hypre_Box *boxi, *boxj, intersect_box;
hypre_StructGrid *grid= hypre_StructVectorGrid(y);
hypre_BoxManager *boxman = hypre_StructGridBoxMan(grid);
hypre_BoxArray *neighbor_boxes;
HYPRE_Int *neighbors_procs= NULL;
hypre_BoxArray *selected_nboxes;
hypre_BoxArray *tmp_box_array, *tmp2_box_array;
hypre_Index loop_size;
hypre_IndexRef start;
hypre_Index unit_stride;
HYPRE_Int i, j;
HYPRE_Int myid;
HYPRE_Int boxarray_size;
HYPRE_Int loopi, loopj, loopk;
#ifdef HYPRE_USE_PTHREADS
HYPRE_Int threadid = hypre_GetThreadID();
#endif
local_result = 0.0;
process_result = 0.0;
hypre_SetIndex(unit_stride, 1, 1, 1);
hypre_MPI_Comm_rank(hypre_StructVectorComm(y), &myid);
/*-----------------------------------------------------------------------
* Determine the overlapped boxes on this local processor.
*-----------------------------------------------------------------------*/
boxes = hypre_StructGridBoxes(hypre_StructVectorGrid(y));
boxarray_size= hypre_BoxArraySize(boxes);
/*-----------------------------------------------------------------------
* To compute the inner product over this local processor, given a box,
* the inner product between x & y is computed over the whole box and
* over any overlapping between this box and overlap_boxes. The latter
* result is subtracted from the former. Overlapping between more than
* two boxes are handled.
*-----------------------------------------------------------------------*/
hypre_ForBoxI(i, boxes)
{
boxi = hypre_BoxArrayBox(boxes, i);
start = hypre_BoxIMin(boxi);
x_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(x), i);
y_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(y), i);
xp = hypre_StructVectorBoxData(x, i);
yp = hypre_StructVectorBoxData(y, i);
hypre_BoxGetSize(boxi, loop_size);
#ifdef HYPRE_USE_PTHREADS
local_result_ref[threadid] = &local_result;
#endif
hypre_BoxLoop2Begin(loop_size,
x_data_box, start, unit_stride, xi,
y_data_box, start, unit_stride, yi);
#define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,xi,yi
#define HYPRE_SMP_REDUCTION_OP +
#define HYPRE_SMP_REDUCTION_VARS local_result
#include "hypre_box_smp_forloop.h"
hypre_BoxLoop2For(loopi, loopj, loopk, xi, yi)
{
local_result += xp[xi] * yp[yi];
}
