int hypre_PrintBoxArrayData( FILE *file, hypre_BoxArray *box_array, hypre_BoxArray *data_space, int num_values, double *data ) { int ierr = 0; hypre_Box *box; hypre_Box *data_box; int data_box_volume; int datai; hypre_Index loop_size; hypre_IndexRef start; hypre_Index stride; int i, j; int loopi, loopj, loopk; /*---------------------------------------- * Print data *----------------------------------------*/ hypre_SetIndex(stride, 1, 1, 1); hypre_ForBoxI(i, box_array) { box = hypre_BoxArrayBox(box_array, i); data_box = hypre_BoxArrayBox(data_space, i); start = hypre_BoxIMin(box); data_box_volume = hypre_BoxVolume(data_box); hypre_BoxGetSize(box, loop_size); hypre_BoxLoop1Begin(loop_size, data_box, start, stride, datai); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,datai #include "hypre_box_smp_forloop.h" hypre_BoxLoop1For(loopi, loopj, loopk, datai) { for (j = 0; j < num_values; j++) { fprintf(file, "%d: (%d, %d, %d; %d) %e\n", i, hypre_IndexX(start) + loopi, hypre_IndexY(start) + loopj, hypre_IndexZ(start) + loopk, j, data[datai + j*data_box_volume]); } } hypre_BoxLoop1End(datai); data += num_values*data_box_volume; }
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); }
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); }
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); }
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_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); } }
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); } }
HYPRE_Int hypre_PFMGBuildCoarseOp5( hypre_StructMatrix *A, hypre_StructMatrix *P, hypre_StructMatrix *R, HYPRE_Int cdir, hypre_Index cindex, hypre_Index cstride, hypre_StructMatrix *RAP ) { hypre_Index index; hypre_Index index_temp; hypre_StructGrid *fgrid; hypre_BoxArray *fgrid_boxes; hypre_Box *fgrid_box; HYPRE_Int *fgrid_ids; hypre_StructGrid *cgrid; hypre_BoxArray *cgrid_boxes; hypre_Box *cgrid_box; HYPRE_Int *cgrid_ids; hypre_IndexRef cstart, bfstart, stridef; hypre_Index fstart, bcstart, stridec; hypre_Index loop_size; HYPRE_Int constant_coefficient; HYPRE_Int fi, ci, fbi; HYPRE_Int loopi, loopj, loopk; hypre_Box *A_dbox; hypre_Box *P_dbox; hypre_Box *RAP_dbox; hypre_BoxArray *bdy_boxes, *tmp_boxes; hypre_Box *bdy_box, *fcbox; double *pb, *pa; double *a_cc, *a_cw, *a_ce, *a_cb, *a_ca; double *rap_cc, *rap_cw, *rap_ce; double *rap_cb, *rap_ca; double west, east; double center_int, center_bdy; HYPRE_Int iA, iAm1, iAp1; HYPRE_Int iAc; HYPRE_Int iP, iPm1, iPp1; HYPRE_Int OffsetA; HYPRE_Int OffsetP; stridef = cstride; hypre_SetIndex(stridec, 1, 1, 1); fgrid = hypre_StructMatrixGrid(A); fgrid_boxes = hypre_StructGridBoxes(fgrid); fgrid_ids = hypre_StructGridIDs(fgrid); cgrid = hypre_StructMatrixGrid(RAP); cgrid_boxes = hypre_StructGridBoxes(cgrid); cgrid_ids = hypre_StructGridIDs(cgrid); constant_coefficient = hypre_StructMatrixConstantCoefficient(RAP); hypre_assert( hypre_StructMatrixConstantCoefficient(A) == constant_coefficient ); if ( constant_coefficient==0 ) { hypre_assert( hypre_StructMatrixConstantCoefficient(R) == 0 ); hypre_assert( hypre_StructMatrixConstantCoefficient(P) == 0 ); } else /* 1 or 2 */ { hypre_assert( hypre_StructMatrixConstantCoefficient(R) == 1 ); hypre_assert( hypre_StructMatrixConstantCoefficient(P) == 1 ); } fcbox = hypre_BoxCreate(); bdy_boxes = hypre_BoxArrayCreate(0); tmp_boxes = hypre_BoxArrayCreate(0); fi = 0; hypre_ForBoxI(ci, cgrid_boxes) { while (fgrid_ids[fi] != cgrid_ids[ci]) { fi++; } cgrid_box = hypre_BoxArrayBox(cgrid_boxes, ci); fgrid_box = hypre_BoxArrayBox(fgrid_boxes, fi); cstart = hypre_BoxIMin(cgrid_box); hypre_StructMapCoarseToFine(cstart, cindex, cstride, fstart); A_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(A), fi); P_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(P), fi); RAP_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(RAP), ci); /*----------------------------------------------------------------- * Extract pointers for interpolation operator: * pb is pointer for weight for f-point below c-point * pa is pointer for weight for f-point above c-point *-----------------------------------------------------------------*/ hypre_SetIndex(index_temp,0,-1,0); MapIndex(index_temp, cdir, index); pa = hypre_StructMatrixExtractPointerByIndex(P, fi, index); hypre_SetIndex(index_temp,0,1,0); MapIndex(index_temp, cdir, index); pb = hypre_StructMatrixExtractPointerByIndex(P, fi, index) - hypre_BoxOffsetDistance(P_dbox, index); /*----------------------------------------------------------------- * Extract pointers for 5-point fine grid operator: * * a_cc is pointer for center coefficient * a_cw is pointer for west coefficient * a_ce is pointer for east coefficient * a_cb is pointer for below coefficient * a_ca is pointer for above coefficient *-----------------------------------------------------------------*/ hypre_SetIndex(index_temp,0,0,0); MapIndex(index_temp, cdir, index); a_cc = hypre_StructMatrixExtractPointerByIndex(A, fi, index); hypre_SetIndex(index_temp,-1,0,0); MapIndex(index_temp, cdir, index); a_cw = hypre_StructMatrixExtractPointerByIndex(A, fi, index); hypre_SetIndex(index_temp,1,0,0); MapIndex(index_temp, cdir, index); a_ce = hypre_StructMatrixExtractPointerByIndex(A, fi, index); hypre_SetIndex(index_temp,0,-1,0); MapIndex(index_temp, cdir, index); a_cb = hypre_StructMatrixExtractPointerByIndex(A, fi, index); hypre_SetIndex(index_temp,0,1,0); MapIndex(index_temp, cdir, index); a_ca = hypre_StructMatrixExtractPointerByIndex(A, fi, index); /*----------------------------------------------------------------- * Extract pointers for coarse grid operator * rap_cc is pointer for center coefficient (etc.) *-----------------------------------------------------------------*/ hypre_SetIndex(index_temp,0,0,0); MapIndex(index_temp, cdir, index); rap_cc = hypre_StructMatrixExtractPointerByIndex(RAP, ci, index); hypre_SetIndex(index_temp,-1,0,0); MapIndex(index_temp, cdir, index); rap_cw = hypre_StructMatrixExtractPointerByIndex(RAP, ci, index); hypre_SetIndex(index_temp,1,0,0); MapIndex(index_temp, cdir, index); rap_ce = hypre_StructMatrixExtractPointerByIndex(RAP, ci, index); hypre_SetIndex(index_temp,0,-1,0); MapIndex(index_temp, cdir, index); rap_cb = hypre_StructMatrixExtractPointerByIndex(RAP, ci, index); hypre_SetIndex(index_temp,0,1,0); MapIndex(index_temp, cdir, index); rap_ca = hypre_StructMatrixExtractPointerByIndex(RAP, ci, index); /*----------------------------------------------------------------- * Define offsets for fine grid stencil and interpolation * * In the BoxLoop below I assume iA and iP refer to data associated * with the point which we are building the stencil for. The below * Offsets are used in refering to data associated with other points. *-----------------------------------------------------------------*/ hypre_SetIndex(index_temp,0,1,0); MapIndex(index_temp, cdir, index); OffsetP = hypre_BoxOffsetDistance(P_dbox,index); OffsetA = hypre_BoxOffsetDistance(A_dbox,index); /*-------------------------------------------------------------- * Loop for symmetric 5-point fine grid operator; produces a * symmetric 5-point coarse grid operator. *--------------------------------------------------------------*/ if ( constant_coefficient==0 ) { hypre_BoxGetSize(cgrid_box, loop_size); hypre_BoxLoop3Begin(loop_size, P_dbox, cstart, stridec, iP, A_dbox, fstart, stridef, iA, RAP_dbox, cstart, stridec, iAc); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,iP,iA,iAc,iAm1,iAp1,iPm1,iPp1,west,east #include "hypre_box_smp_forloop.h" hypre_BoxLoop3For(loopi, loopj, loopk, iP, iA, iAc) { iAm1 = iA - OffsetA; iAp1 = iA + OffsetA; iPm1 = iP - OffsetP; iPp1 = iP + OffsetP; rap_cb[iAc] = a_cb[iA] * pa[iPm1]; rap_ca[iAc] = a_ca[iA] * pb[iPp1]; west = a_cw[iA] + 0.5 * a_cw[iAm1] + 0.5 * a_cw[iAp1]; east = a_ce[iA] + 0.5 * a_ce[iAm1] + 0.5 * a_ce[iAp1]; /*----------------------------------------------------- * Prevent non-zero entries reaching off grid *-----------------------------------------------------*/ if(a_cw[iA] == 0.0) west = 0.0; if(a_ce[iA] == 0.0) east = 0.0; rap_cw[iAc] = west; rap_ce[iAc] = east; rap_cc[iAc] = a_cc[iA] + a_cw[iA] + a_ce[iA] + a_cb[iA] * pb[iP] + a_ca[iA] * pa[iP] - west - east; } hypre_BoxLoop3End(iP, iA, iAc); }
int hypre_StructMatvecCompute( void *matvec_vdata, double alpha, hypre_StructMatrix *A, hypre_StructVector *x, double beta, hypre_StructVector *y ) { int ierr = 0; hypre_StructMatvecData *matvec_data = matvec_vdata; hypre_ComputePkg *compute_pkg; hypre_CommHandle *comm_handle; hypre_BoxArrayArray *compute_box_aa; hypre_BoxArray *compute_box_a; hypre_Box *compute_box; hypre_Box *A_data_box; hypre_Box *x_data_box; hypre_Box *y_data_box; int Ai; int xi; int xoff0; int xoff1; int xoff2; int xoff3; int xoff4; int xoff5; int xoff6; int yi; double *Ap0; double *Ap1; double *Ap2; double *Ap3; double *Ap4; double *Ap5; double *Ap6; double *xp; double *yp; hypre_BoxArray *boxes; hypre_Box *box; hypre_Index loop_size; hypre_IndexRef start; hypre_IndexRef stride; hypre_StructStencil *stencil; hypre_Index *stencil_shape; int stencil_size; int depth; double temp; int compute_i, i, j, si; int loopi, loopj, loopk; /*----------------------------------------------------------------------- * Initialize some things *-----------------------------------------------------------------------*/ compute_pkg = (matvec_data -> compute_pkg); stride = hypre_ComputePkgStride(compute_pkg); /*----------------------------------------------------------------------- * Do (alpha == 0.0) computation *-----------------------------------------------------------------------*/ if (alpha == 0.0) { boxes = hypre_StructGridBoxes(hypre_StructMatrixGrid(A)); 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, 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] *= beta; } hypre_BoxLoop1End(yi); }
int hypre_SMG2BuildRAPSym( hypre_StructMatrix *A, hypre_StructMatrix *PT, hypre_StructMatrix *R, hypre_StructMatrix *RAP, hypre_Index cindex, hypre_Index cstride ) { hypre_Index index; hypre_StructStencil *fine_stencil; int fine_stencil_size; hypre_StructGrid *fgrid; int *fgrid_ids; hypre_StructGrid *cgrid; hypre_BoxArray *cgrid_boxes; int *cgrid_ids; hypre_Box *cgrid_box; hypre_IndexRef cstart; hypre_Index stridec; hypre_Index fstart; hypre_IndexRef stridef; hypre_Index loop_size; int fi, ci; int loopi, loopj, loopk; hypre_Box *A_dbox; hypre_Box *PT_dbox; hypre_Box *R_dbox; hypre_Box *RAP_dbox; double *pa, *pb; double *ra, *rb; double *a_cc, *a_cw, *a_ce, *a_cs, *a_cn; double *a_csw, *a_cse, *a_cnw; double *rap_cc, *rap_cw, *rap_cs; double *rap_csw, *rap_cse; int iA, iAm1, iAp1; int iAc; int iP, iP1; int iR; int yOffsetA; int xOffsetP; int yOffsetP; int ierr = 0; fine_stencil = hypre_StructMatrixStencil(A); fine_stencil_size = hypre_StructStencilSize(fine_stencil); stridef = cstride; hypre_SetIndex(stridec, 1, 1, 1); fgrid = hypre_StructMatrixGrid(A); fgrid_ids = hypre_StructGridIDs(fgrid); cgrid = hypre_StructMatrixGrid(RAP); 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++; } cgrid_box = hypre_BoxArrayBox(cgrid_boxes, ci); cstart = hypre_BoxIMin(cgrid_box); hypre_StructMapCoarseToFine(cstart, cindex, cstride, fstart); A_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(A), fi); PT_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(PT), fi); R_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(R), fi); RAP_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(RAP), ci); /*----------------------------------------------------------------- * Extract pointers for interpolation operator: * pa is pointer for weight for f-point above c-point * pb is pointer for weight for f-point below c-point *-----------------------------------------------------------------*/ hypre_SetIndex(index,0,1,0); pa = hypre_StructMatrixExtractPointerByIndex(PT, fi, index); hypre_SetIndex(index,0,-1,0); pb = hypre_StructMatrixExtractPointerByIndex(PT, fi, index); /*----------------------------------------------------------------- * Extract pointers for restriction operator: * ra is pointer for weight for f-point above c-point * rb is pointer for weight for f-point below c-point *-----------------------------------------------------------------*/ hypre_SetIndex(index,0,1,0); ra = hypre_StructMatrixExtractPointerByIndex(R, fi, index); hypre_SetIndex(index,0,-1,0); rb = hypre_StructMatrixExtractPointerByIndex(R, fi, index); /*----------------------------------------------------------------- * Extract pointers for 5-point fine grid operator: * * a_cc is pointer for center coefficient * a_cw is pointer for west coefficient * a_ce is pointer for east coefficient * a_cs is pointer for south coefficient * a_cn is pointer for north coefficient *-----------------------------------------------------------------*/ hypre_SetIndex(index,0,0,0); a_cc = hypre_StructMatrixExtractPointerByIndex(A, fi, index); hypre_SetIndex(index,-1,0,0); a_cw = hypre_StructMatrixExtractPointerByIndex(A, fi, index); hypre_SetIndex(index,1,0,0); a_ce = hypre_StructMatrixExtractPointerByIndex(A, fi, index); hypre_SetIndex(index,0,-1,0); a_cs = hypre_StructMatrixExtractPointerByIndex(A, fi, index); hypre_SetIndex(index,0,1,0); a_cn = hypre_StructMatrixExtractPointerByIndex(A, fi, index); /*----------------------------------------------------------------- * Extract additional pointers for 9-point fine grid operator: * * a_csw is pointer for southwest coefficient * a_cse is pointer for southeast coefficient * a_cnw is pointer for northwest coefficient * a_cne is pointer for northeast coefficient *-----------------------------------------------------------------*/ if(fine_stencil_size > 5) { hypre_SetIndex(index,-1,-1,0); a_csw = hypre_StructMatrixExtractPointerByIndex(A, fi, index); hypre_SetIndex(index,1,-1,0); a_cse = hypre_StructMatrixExtractPointerByIndex(A, fi, index); hypre_SetIndex(index,-1,1,0); a_cnw = hypre_StructMatrixExtractPointerByIndex(A, fi, index); } /*----------------------------------------------------------------- * Extract pointers for coarse grid operator - always 9-point: * * We build only the lower triangular part (plus diagonal). * * rap_cc is pointer for center coefficient (etc.) *-----------------------------------------------------------------*/ hypre_SetIndex(index,0,0,0); rap_cc = hypre_StructMatrixExtractPointerByIndex(RAP, ci, index); hypre_SetIndex(index,-1,0,0); rap_cw = hypre_StructMatrixExtractPointerByIndex(RAP, ci, index); hypre_SetIndex(index,0,-1,0); rap_cs = hypre_StructMatrixExtractPointerByIndex(RAP, ci, index); hypre_SetIndex(index,-1,-1,0); rap_csw = hypre_StructMatrixExtractPointerByIndex(RAP, ci, index); hypre_SetIndex(index,1,-1,0); rap_cse = hypre_StructMatrixExtractPointerByIndex(RAP, ci, index); /*----------------------------------------------------------------- * Define offsets for fine grid stencil and interpolation * * In the BoxLoop below I assume iA and iP refer to data associated * with the point which we are building the stencil for. The below * Offsets are used in refering to data associated with other points. *-----------------------------------------------------------------*/ hypre_SetIndex(index,0,1,0); yOffsetA = hypre_BoxOffsetDistance(A_dbox,index); yOffsetP = hypre_BoxOffsetDistance(PT_dbox,index); hypre_SetIndex(index,1,0,0); xOffsetP = hypre_BoxOffsetDistance(PT_dbox,index); /*----------------------------------------------------------------- * Switch statement to direct control to apropriate BoxLoop depending * on stencil size. Default is full 9-point. *-----------------------------------------------------------------*/ switch (fine_stencil_size) { /*-------------------------------------------------------------- * Loop for symmetric 5-point fine grid operator; produces a * symmetric 9-point coarse grid operator. We calculate only the * lower triangular stencil entries: (southwest, south, southeast, * west, and center). *--------------------------------------------------------------*/ case 5: hypre_BoxGetSize(cgrid_box, loop_size); hypre_BoxLoop4Begin(loop_size, PT_dbox, cstart, stridec, iP, R_dbox, cstart, stridec, iR, A_dbox, fstart, stridef, iA, RAP_dbox, cstart, stridec, iAc); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,iP,iR,iA,iAc,iAm1,iAp1,iP1 #include "hypre_box_smp_forloop.h" hypre_BoxLoop4For(loopi, loopj, loopk, iP, iR, iA, iAc) { iAm1 = iA - yOffsetA; iAp1 = iA + yOffsetA; iP1 = iP - yOffsetP - xOffsetP; rap_csw[iAc] = rb[iR] * a_cw[iAm1] * pa[iP1]; iP1 = iP - yOffsetP; rap_cs[iAc] = rb[iR] * a_cc[iAm1] * pa[iP1] + rb[iR] * a_cs[iAm1] + a_cs[iA] * pa[iP1]; iP1 = iP - yOffsetP + xOffsetP; rap_cse[iAc] = rb[iR] * a_ce[iAm1] * pa[iP1]; iP1 = iP - xOffsetP; rap_cw[iAc] = a_cw[iA] + rb[iR] * a_cw[iAm1] * pb[iP1] + ra[iR] * a_cw[iAp1] * pa[iP1]; rap_cc[iAc] = a_cc[iA] + rb[iR] * a_cc[iAm1] * pb[iP] + ra[iR] * a_cc[iAp1] * pa[iP] + rb[iR] * a_cn[iAm1] + ra[iR] * a_cs[iAp1] + a_cs[iA] * pb[iP] + a_cn[iA] * pa[iP]; } hypre_BoxLoop4End(iP, iR, iA, iAc); break; /*-------------------------------------------------------------- * Loop for symmetric 9-point fine grid operator; produces a * symmetric 9-point coarse grid operator. We calculate only the * lower triangular stencil entries: (southwest, south, southeast, * west, and center). *--------------------------------------------------------------*/ default: hypre_BoxGetSize(cgrid_box, loop_size); hypre_BoxLoop4Begin(loop_size, PT_dbox, cstart, stridec, iP, R_dbox, cstart, stridec, iR, A_dbox, fstart, stridef, iA, RAP_dbox, cstart, stridec, iAc); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,iP,iR,iA,iAc,iAm1,iAp1,iP1 #include "hypre_box_smp_forloop.h" hypre_BoxLoop4For(loopi, loopj, loopk, iP, iR, iA, iAc) { iAm1 = iA - yOffsetA; iAp1 = iA + yOffsetA; iP1 = iP - yOffsetP - xOffsetP; rap_csw[iAc] = rb[iR] * a_cw[iAm1] * pa[iP1] + rb[iR] * a_csw[iAm1] + a_csw[iA] * pa[iP1]; iP1 = iP - yOffsetP; rap_cs[iAc] = rb[iR] * a_cc[iAm1] * pa[iP1] + rb[iR] * a_cs[iAm1] + a_cs[iA] * pa[iP1]; iP1 = iP - yOffsetP + xOffsetP; rap_cse[iAc] = rb[iR] * a_ce[iAm1] * pa[iP1] + rb[iR] * a_cse[iAm1] + a_cse[iA] * pa[iP1]; iP1 = iP - xOffsetP; rap_cw[iAc] = a_cw[iA] + rb[iR] * a_cw[iAm1] * pb[iP1] + ra[iR] * a_cw[iAp1] * pa[iP1] + rb[iR] * a_cnw[iAm1] + ra[iR] * a_csw[iAp1] + a_csw[iA] * pb[iP1] + a_cnw[iA] * pa[iP1]; rap_cc[iAc] = a_cc[iA] + rb[iR] * a_cc[iAm1] * pb[iP] + ra[iR] * a_cc[iAp1] * pa[iP] + rb[iR] * a_cn[iAm1] + ra[iR] * a_cs[iAp1] + a_cs[iA] * pb[iP] + a_cn[iA] * pa[iP]; } hypre_BoxLoop4End(iP, iR, iA, iAc); break; } /* end switch statement */
/*-------------------------------------------------------------------------- * hypre_FacZeroCFSten: Zeroes the coarse stencil coefficients that reach * into an underlying coarsened refinement box. * Algo: For each cbox * { * 1) refine cbox and expand by one in each direction * 2) boxman_intersect with the fboxman * 3) loop over intersection boxes to see if stencil * reaches over. * } *--------------------------------------------------------------------------*/ HYPRE_Int hypre_FacZeroCFSten( hypre_SStructPMatrix *Af, hypre_SStructPMatrix *Ac, hypre_SStructGrid *grid, HYPRE_Int fine_part, hypre_Index rfactors ) { hypre_BoxManager *fboxman; hypre_BoxManEntry **boxman_entries; HYPRE_Int nboxman_entries; hypre_SStructPGrid *p_cgrid; hypre_Box fgrid_box; hypre_StructGrid *cgrid; hypre_BoxArray *cgrid_boxes; hypre_Box *cgrid_box; hypre_Box scaled_box; hypre_Box *shift_ibox; hypre_StructMatrix *smatrix; hypre_StructStencil *stencils; HYPRE_Int stencil_size; hypre_Index refine_factors, upper_shift; hypre_Index stride; hypre_Index stencil_shape; hypre_Index zero_index, ilower, iupper; HYPRE_Int nvars, var1, var2; HYPRE_Int ndim; hypre_Box *ac_dbox; HYPRE_Real *ac_ptr; hypre_Index loop_size; HYPRE_Int iac; HYPRE_Int ci, i, j; HYPRE_Int abs_shape; HYPRE_Int ierr = 0; p_cgrid = hypre_SStructPMatrixPGrid(Ac); nvars = hypre_SStructPMatrixNVars(Ac); ndim = hypre_SStructPGridNDim(p_cgrid); hypre_BoxInit(&fgrid_box, ndim); hypre_BoxInit(&scaled_box, ndim); hypre_ClearIndex(zero_index); hypre_ClearIndex(stride); hypre_ClearIndex(upper_shift); for (i= 0; i< ndim; i++) { stride[i]= 1; upper_shift[i]= rfactors[i]-1; } hypre_CopyIndex(rfactors, refine_factors); if (ndim < 3) { for (i= ndim; i< 3; i++) { refine_factors[i]= 1; } } for (var1= 0; var1< nvars; var1++) { cgrid= hypre_SStructPGridSGrid(hypre_SStructPMatrixPGrid(Ac), var1); cgrid_boxes= hypre_StructGridBoxes(cgrid); fboxman= hypre_SStructGridBoxManager(grid, fine_part, var1); /*------------------------------------------------------------------ * For each parent coarse box find all fboxes that may be connected * through a stencil entry- refine this box, expand it by one * in each direction, and boxman_intersect with fboxman *------------------------------------------------------------------*/ hypre_ForBoxI(ci, cgrid_boxes) { cgrid_box= hypre_BoxArrayBox(cgrid_boxes, ci); hypre_StructMapCoarseToFine(hypre_BoxIMin(cgrid_box), zero_index, refine_factors, hypre_BoxIMin(&scaled_box)); hypre_StructMapCoarseToFine(hypre_BoxIMax(cgrid_box), upper_shift, refine_factors, hypre_BoxIMax(&scaled_box)); hypre_SubtractIndexes(hypre_BoxIMin(&scaled_box), stride, 3, hypre_BoxIMin(&scaled_box)); hypre_AddIndexes(hypre_BoxIMax(&scaled_box), stride, 3, hypre_BoxIMax(&scaled_box)); hypre_BoxManIntersect(fboxman, hypre_BoxIMin(&scaled_box), hypre_BoxIMax(&scaled_box), &boxman_entries, &nboxman_entries); for (var2= 0; var2< nvars; var2++) { stencils= hypre_SStructPMatrixSStencil(Ac, var1, var2); if (stencils != NULL) { stencil_size= hypre_StructStencilSize(stencils); smatrix = hypre_SStructPMatrixSMatrix(Ac, var1, var2); ac_dbox = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(smatrix), ci); /*--------------------------------------------------------- * Find the stencil coefficients that must be zeroed off. * Loop over all possible boxes. *---------------------------------------------------------*/ for (i= 0; i< stencil_size; i++) { hypre_CopyIndex(hypre_StructStencilElement(stencils, i), stencil_shape); AbsStencilShape(stencil_shape, abs_shape); if (abs_shape) /* non-centre stencils are zeroed */ { /* look for connecting fboxes that must be zeroed. */ for (j= 0; j< nboxman_entries; j++) { hypre_BoxManEntryGetExtents(boxman_entries[j], ilower, iupper); hypre_BoxSetExtents(&fgrid_box, ilower, iupper); shift_ibox= hypre_CF_StenBox(&fgrid_box, cgrid_box, stencil_shape, refine_factors, ndim); if ( hypre_BoxVolume(shift_ibox) ) { ac_ptr= hypre_StructMatrixExtractPointerByIndex(smatrix, ci, stencil_shape); hypre_BoxGetSize(shift_ibox, loop_size); hypre_BoxLoop1Begin(ndim, loop_size, ac_dbox, hypre_BoxIMin(shift_ibox), stride, iac); #ifdef HYPRE_USING_OPENMP #pragma omp parallel for private(HYPRE_BOX_PRIVATE,iac) HYPRE_SMP_SCHEDULE #endif hypre_BoxLoop1For(iac) { ac_ptr[iac] = 0.0; } hypre_BoxLoop1End(iac); } /* if ( hypre_BoxVolume(shift_ibox) ) */ hypre_BoxDestroy(shift_ibox); } /* for (j= 0; j< nboxman_entries; j++) */ } /* if (abs_shape) */ } /* for (i= 0; i< stencil_size; i++) */ } /* if (stencils != NULL) */ } /* for (var2= 0; var2< nvars; var2++) */ hypre_TFree(boxman_entries); } /* hypre_ForBoxI ci */
HYPRE_Int hypre_PrintBoxArrayData( FILE *file, hypre_BoxArray *box_array, hypre_BoxArray *data_space, HYPRE_Int num_values, HYPRE_Int dim, HYPRE_Complex *data ) { hypre_Box *box; hypre_Box *data_box; HYPRE_Int data_box_volume; HYPRE_Int datai; hypre_Index loop_size; hypre_IndexRef start; hypre_Index stride; hypre_Index index; HYPRE_Int i, j, d; HYPRE_Complex value; /*---------------------------------------- * Print data *----------------------------------------*/ hypre_SetIndex(stride, 1); hypre_ForBoxI(i, box_array) { box = hypre_BoxArrayBox(box_array, i); data_box = hypre_BoxArrayBox(data_space, i); start = hypre_BoxIMin(box); data_box_volume = hypre_BoxVolume(data_box); hypre_BoxGetSize(box, loop_size); hypre_BoxLoop1Begin(dim, loop_size, data_box, start, stride, datai); hypre_BoxLoop1For(datai) { /* Print lines of the form: "%d: (%d, %d, %d; %d) %.14e\n" */ hypre_BoxLoopGetIndex(index); for (j = 0; j < num_values; j++) { hypre_fprintf(file, "%d: (%d", i, hypre_IndexD(start, 0) + hypre_IndexD(index, 0)); for (d = 1; d < dim; d++) { hypre_fprintf(file, ", %d", hypre_IndexD(start, d) + hypre_IndexD(index, d)); } value = data[datai + j*data_box_volume]; #ifdef HYPRE_COMPLEX hypre_fprintf(file, "; %d) %.14e , %.14e\n", j, hypre_creal(value), hypre_cimag(value)); #else hypre_fprintf(file, "; %d) %.14e\n", j, value); #endif } } hypre_BoxLoop1End(datai); data += num_values*data_box_volume; }
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]; }
HYPRE_Int HYPRE_StructDiagScale( HYPRE_StructSolver solver, HYPRE_StructMatrix HA, HYPRE_StructVector Hy, HYPRE_StructVector Hx ) { hypre_StructMatrix *A = (hypre_StructMatrix *) HA; hypre_StructVector *y = (hypre_StructVector *) Hy; hypre_StructVector *x = (hypre_StructVector *) Hx; hypre_BoxArray *boxes; hypre_Box *box; hypre_Box *A_data_box; hypre_Box *y_data_box; hypre_Box *x_data_box; double *Ap; double *yp; double *xp; HYPRE_Int Ai; HYPRE_Int yi; HYPRE_Int xi; hypre_Index index; hypre_IndexRef start; hypre_Index stride; hypre_Index loop_size; HYPRE_Int i; /* x = D^{-1} y */ hypre_SetIndex(stride, 1, 1, 1); boxes = hypre_StructGridBoxes(hypre_StructMatrixGrid(A)); hypre_ForBoxI(i, boxes) { box = hypre_BoxArrayBox(boxes, i); A_data_box = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(A), i); x_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(x), i); y_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(y), i); hypre_SetIndex(index, 0, 0, 0); Ap = hypre_StructMatrixExtractPointerByIndex(A, i, index); xp = hypre_StructVectorBoxData(x, i); yp = hypre_StructVectorBoxData(y, i); start = hypre_BoxIMin(box); hypre_BoxGetSize(box, loop_size); hypre_BoxLoop3Begin(hypre_StructVectorDim(Hx), loop_size, A_data_box, start, stride, Ai, x_data_box, start, stride, xi, y_data_box, start, stride, yi); #ifdef HYPRE_USING_OPENMP #pragma omp parallel for private(HYPRE_BOX_PRIVATE,yi,xi,Ai) HYPRE_SMP_SCHEDULE #endif hypre_BoxLoop3For(Ai, xi, yi) { xp[xi] = yp[yi] / Ap[Ai]; } hypre_BoxLoop3End(Ai, xi, yi); }
int HYPRE_StructDiagScale( HYPRE_StructSolver solver, HYPRE_StructMatrix HA, HYPRE_StructVector Hy, HYPRE_StructVector Hx ) { hypre_StructMatrix *A = (hypre_StructMatrix *) HA; hypre_StructVector *y = (hypre_StructVector *) Hy; hypre_StructVector *x = (hypre_StructVector *) Hx; hypre_BoxArray *boxes; hypre_Box *box; hypre_Box *A_data_box; hypre_Box *y_data_box; hypre_Box *x_data_box; double *Ap; double *yp; double *xp; int Ai; int yi; int xi; hypre_Index index; hypre_IndexRef start; hypre_Index stride; hypre_Index loop_size; int i; int loopi, loopj, loopk; int ierr = 0; /* x = D^{-1} y */ hypre_SetIndex(stride, 1, 1, 1); boxes = hypre_StructGridBoxes(hypre_StructMatrixGrid(A)); hypre_ForBoxI(i, boxes) { box = hypre_BoxArrayBox(boxes, i); A_data_box = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(A), i); x_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(x), i); y_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(y), i); hypre_SetIndex(index, 0, 0, 0); Ap = hypre_StructMatrixExtractPointerByIndex(A, i, index); xp = hypre_StructVectorBoxData(x, i); yp = hypre_StructVectorBoxData(y, i); start = hypre_BoxIMin(box); hypre_BoxGetSize(box, loop_size); hypre_BoxLoop3Begin(loop_size, A_data_box, start, stride, Ai, x_data_box, start, stride, xi, y_data_box, start, stride, yi); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,yi,xi,Ai #include "hypre_box_smp_forloop.h" hypre_BoxLoop3For(loopi, loopj, loopk, Ai, xi, yi) { xp[xi] = yp[yi] / Ap[Ai]; } hypre_BoxLoop3End(Ai, xi, yi); }
HYPRE_Int hypre_CreateCommInfoFromStencil( hypre_StructGrid *grid, hypre_StructStencil *stencil, hypre_CommInfo **comm_info_ptr ) { HYPRE_Int ndim = hypre_StructGridNDim(grid); HYPRE_Int i,j,k, d, m, s, si; hypre_BoxArrayArray *send_boxes; hypre_BoxArrayArray *recv_boxes; HYPRE_Int **send_procs; HYPRE_Int **recv_procs; HYPRE_Int **send_rboxnums; HYPRE_Int **recv_rboxnums; hypre_BoxArrayArray *send_rboxes; hypre_BoxArrayArray *recv_rboxes; hypre_BoxArray *local_boxes; HYPRE_Int num_boxes; hypre_BoxManager *boxman; hypre_Index *stencil_shape; hypre_IndexRef stencil_offset; hypre_IndexRef pshift; hypre_Box *box; hypre_Box *hood_box; hypre_Box *grow_box; hypre_Box *extend_box; hypre_Box *int_box; hypre_Box *periodic_box; hypre_Box *stencil_box, *sbox; /* extents of the stencil grid */ HYPRE_Int *stencil_grid; HYPRE_Int grow[HYPRE_MAXDIM][2]; hypre_BoxManEntry **entries; hypre_BoxManEntry *entry; HYPRE_Int num_entries; hypre_BoxArray *neighbor_boxes = NULL; HYPRE_Int *neighbor_procs = NULL; HYPRE_Int *neighbor_ids = NULL; HYPRE_Int *neighbor_shifts = NULL; HYPRE_Int neighbor_count; HYPRE_Int neighbor_alloc; hypre_Index ilower, iupper; hypre_BoxArray *send_box_array; hypre_BoxArray *recv_box_array; hypre_BoxArray *send_rbox_array; hypre_BoxArray *recv_rbox_array; hypre_Box **cboxes; hypre_Box *cboxes_mem; HYPRE_Int *cboxes_neighbor_location; HYPRE_Int num_cboxes, cbox_alloc; hypre_Index istart, istop, sgindex; hypre_IndexRef start; hypre_Index loop_size, stride; HYPRE_Int num_periods, loc, box_id, id, proc_id; HYPRE_Int myid; MPI_Comm comm; /*------------------------------------------------------ * Initializations *------------------------------------------------------*/ hypre_SetIndex(ilower, 0); hypre_SetIndex(iupper, 0); hypre_SetIndex(istart, 0); hypre_SetIndex(istop, 0); hypre_SetIndex(sgindex, 0); local_boxes = hypre_StructGridBoxes(grid); num_boxes = hypre_BoxArraySize(local_boxes); num_periods = hypre_StructGridNumPeriods(grid); boxman = hypre_StructGridBoxMan(grid); comm = hypre_StructGridComm(grid); hypre_MPI_Comm_rank(comm, &myid); stencil_box = hypre_BoxCreate(ndim); hypre_SetIndex(hypre_BoxIMin(stencil_box), 0); hypre_SetIndex(hypre_BoxIMax(stencil_box), 2); /* Set initial values to zero */ stencil_grid = hypre_CTAlloc(HYPRE_Int, hypre_BoxVolume(stencil_box)); sbox = hypre_BoxCreate(ndim); hypre_SetIndex(stride, 1); /*------------------------------------------------------ * Compute the "grow" information from the stencil *------------------------------------------------------*/ stencil_shape = hypre_StructStencilShape(stencil); for (d = 0; d < ndim; d++) { grow[d][0] = 0; grow[d][1] = 0; } for (s = 0; s < hypre_StructStencilSize(stencil); s++) { stencil_offset = stencil_shape[s]; for (d = 0; d < ndim; d++) { m = stencil_offset[d]; istart[d] = 1; istop[d] = 1; if (m < 0) { istart[d] = 0; grow[d][0] = hypre_max(grow[d][0], -m); } else if (m > 0) { istop[d] = 2; grow[d][1] = hypre_max(grow[d][1], m); } } /* update stencil grid from the grow_stencil */ hypre_BoxSetExtents(sbox, istart, istop); start = hypre_BoxIMin(sbox); hypre_BoxGetSize(sbox, loop_size); hypre_BoxLoop1Begin(ndim, loop_size, stencil_box, start, stride, si); hypre_BoxLoopSetOneBlock(); hypre_BoxLoop1For(si) { stencil_grid[si] = 1; } hypre_BoxLoop1End(si); } /*------------------------------------------------------ * Compute send/recv boxes and procs for each local box *------------------------------------------------------*/ /* initialize: for each local box, we create an array of send/recv info */ send_boxes = hypre_BoxArrayArrayCreate(num_boxes, ndim); recv_boxes = hypre_BoxArrayArrayCreate(num_boxes, ndim); send_procs = hypre_CTAlloc(HYPRE_Int *, num_boxes); recv_procs = hypre_CTAlloc(HYPRE_Int *, num_boxes); /* Remote boxnums and boxes describe data on the opposing processor, so some shifting of boxes is needed below for periodic neighbor boxes. Remote box info is also needed for receives to allow for reverse communication. */ send_rboxnums = hypre_CTAlloc(HYPRE_Int *, num_boxes); send_rboxes = hypre_BoxArrayArrayCreate(num_boxes, ndim); recv_rboxnums = hypre_CTAlloc(HYPRE_Int *, num_boxes); recv_rboxes = hypre_BoxArrayArrayCreate(num_boxes, ndim); grow_box = hypre_BoxCreate(hypre_StructGridNDim(grid)); extend_box = hypre_BoxCreate(hypre_StructGridNDim(grid)); int_box = hypre_BoxCreate(hypre_StructGridNDim(grid)); periodic_box = hypre_BoxCreate(hypre_StructGridNDim(grid)); /* storage we will use and keep track of the neighbors */ neighbor_alloc = 30; /* initial guess at max size */ neighbor_boxes = hypre_BoxArrayCreate(neighbor_alloc, ndim); neighbor_procs = hypre_CTAlloc(HYPRE_Int, neighbor_alloc); neighbor_ids = hypre_CTAlloc(HYPRE_Int, neighbor_alloc); neighbor_shifts = hypre_CTAlloc(HYPRE_Int, neighbor_alloc); /* storage we will use to collect all of the intersected boxes (the send and recv regions for box i (this may not be enough in the case of periodic boxes, so we will have to check) */ cbox_alloc = hypre_BoxManNEntries(boxman); cboxes_neighbor_location = hypre_CTAlloc(HYPRE_Int, cbox_alloc); cboxes = hypre_CTAlloc(hypre_Box *, cbox_alloc); cboxes_mem = hypre_CTAlloc(hypre_Box, cbox_alloc); /******* loop through each local box **************/ for (i = 0; i < num_boxes; i++) { /* get the box */ box = hypre_BoxArrayBox(local_boxes, i); box_id = i; /* grow box local i according to the stencil*/ hypre_CopyBox(box, grow_box); for (d = 0; d < ndim; d++) { hypre_BoxIMinD(grow_box, d) -= grow[d][0]; hypre_BoxIMaxD(grow_box, d) += grow[d][1]; } /* extend_box - to find the list of potential neighbors, we need to grow the local box a bit differently in case, for example, the stencil grows in one dimension [0] and not the other [1] */ hypre_CopyBox(box, extend_box); for (d = 0; d < ndim; d++) { hypre_BoxIMinD(extend_box, d) -= hypre_max(grow[d][0],grow[d][1]); hypre_BoxIMaxD(extend_box, d) += hypre_max(grow[d][0],grow[d][1]); } /*------------------------------------------------ * Determine the neighbors of box i *------------------------------------------------*/ /* Do this by intersecting the extend box with the BoxManager. We must also check for periodic neighbors. */ neighbor_count = 0; hypre_BoxArraySetSize(neighbor_boxes, 0); /* shift the box by each period (k=0 is original box) */ for (k = 0; k < num_periods; k++) { hypre_CopyBox(extend_box, periodic_box); pshift = hypre_StructGridPShift(grid, k); hypre_BoxShiftPos(periodic_box, pshift); /* get the intersections */ hypre_BoxManIntersect(boxman, hypre_BoxIMin(periodic_box) , hypre_BoxIMax(periodic_box) , &entries , &num_entries); /* note: do we need to remove the intersection with our original box? no if periodic, yes if non-periodic (k=0) */ /* unpack entries (first check storage) */ if (neighbor_count + num_entries > neighbor_alloc) { neighbor_alloc = neighbor_count + num_entries + 5; neighbor_procs = hypre_TReAlloc(neighbor_procs, HYPRE_Int, neighbor_alloc); neighbor_ids = hypre_TReAlloc(neighbor_ids, HYPRE_Int, neighbor_alloc); neighbor_shifts = hypre_TReAlloc(neighbor_shifts, HYPRE_Int, neighbor_alloc); } /* check storage for the array */ hypre_BoxArraySetSize(neighbor_boxes, neighbor_count + num_entries); /* now unpack */ for (j = 0; j < num_entries; j++) { entry = entries[j]; proc_id = hypre_BoxManEntryProc(entry); id = hypre_BoxManEntryId(entry); /* don't keep box i in the non-periodic case*/ if (!k) { if((myid == proc_id) && (box_id == id)) { continue; } } hypre_BoxManEntryGetExtents(entry, ilower, iupper); hypre_BoxSetExtents(hypre_BoxArrayBox(neighbor_boxes, neighbor_count), ilower, iupper); /* shift the periodic boxes (needs to be the opposite of above) */ if (k) { hypre_BoxShiftNeg( hypre_BoxArrayBox(neighbor_boxes, neighbor_count), pshift); } neighbor_procs[neighbor_count] = proc_id; neighbor_ids[neighbor_count] = id; neighbor_shifts[neighbor_count] = k; neighbor_count++; } hypre_BoxArraySetSize(neighbor_boxes, neighbor_count); hypre_TFree(entries); } /* end of loop through periods k */ /* Now we have a list of all of the neighbors for box i! */ /* note: we don't want/need to remove duplicates - they should have different intersections (TO DO: put more thought into if there are ever any exceptions to this? - the intersection routine already eliminates duplicates - so what i mean is eliminating duplicates from multiple intersection calls in periodic case) */ /*------------------------------------------------ * Compute recv_box_array for box i *------------------------------------------------*/ /* check size of storage for cboxes */ /* let's make sure that we have enough storage in case each neighbor produces a send/recv region */ if (neighbor_count > cbox_alloc) { cbox_alloc = neighbor_count; cboxes_neighbor_location = hypre_TReAlloc(cboxes_neighbor_location, HYPRE_Int, cbox_alloc); cboxes = hypre_TReAlloc(cboxes, hypre_Box *, cbox_alloc); cboxes_mem = hypre_TReAlloc(cboxes_mem, hypre_Box, cbox_alloc); } /* Loop through each neighbor box. If the neighbor box intersects the grown box i (grown according to our stencil), then the intersection is a recv region. If the neighbor box was shifted to handle periodicity, we need to (positive) shift it back. */ num_cboxes = 0; for (k = 0; k < neighbor_count; k++) { hood_box = hypre_BoxArrayBox(neighbor_boxes, k); /* check the stencil grid to see if it makes sense to intersect */ for (d = 0; d < ndim; d++) { sgindex[d] = 1; s = hypre_BoxIMinD(hood_box, d) - hypre_BoxIMaxD(box, d); if (s > 0) { sgindex[d] = 2; } s = hypre_BoxIMinD(box, d) - hypre_BoxIMaxD(hood_box, d); if (s > 0) { sgindex[d] = 0; } } /* it makes sense only if we have at least one non-zero entry */ si = hypre_BoxIndexRank(stencil_box, sgindex); if (stencil_grid[si]) { /* intersect - result is int_box */ hypre_IntersectBoxes(grow_box, hood_box, int_box); /* if we have a positive volume box, this is a recv region */ if (hypre_BoxVolume(int_box)) { /* keep track of which neighbor: k... */ cboxes_neighbor_location[num_cboxes] = k; cboxes[num_cboxes] = &cboxes_mem[num_cboxes]; /* keep the intersected box */ hypre_CopyBox(int_box, cboxes[num_cboxes]); num_cboxes++; } } } /* end of loop through each neighbor */ /* create recv_box_array and recv_procs for box i */ recv_box_array = hypre_BoxArrayArrayBoxArray(recv_boxes, i); hypre_BoxArraySetSize(recv_box_array, num_cboxes); recv_procs[i] = hypre_CTAlloc(HYPRE_Int, num_cboxes); recv_rboxnums[i] = hypre_CTAlloc(HYPRE_Int, num_cboxes); recv_rbox_array = hypre_BoxArrayArrayBoxArray(recv_rboxes, i); hypre_BoxArraySetSize(recv_rbox_array, num_cboxes); for (m = 0; m < num_cboxes; m++) { loc = cboxes_neighbor_location[m]; recv_procs[i][m] = neighbor_procs[loc]; recv_rboxnums[i][m] = neighbor_ids[loc]; hypre_CopyBox(cboxes[m], hypre_BoxArrayBox(recv_box_array, m)); /* if periodic, positive shift before copying to the rbox_array */ if (neighbor_shifts[loc]) /* periodic if shift != 0 */ { pshift = hypre_StructGridPShift(grid, neighbor_shifts[loc]); hypre_BoxShiftPos(cboxes[m], pshift); } hypre_CopyBox(cboxes[m], hypre_BoxArrayBox(recv_rbox_array, m)); cboxes[m] = NULL; } /*------------------------------------------------ * Compute send_box_array for box i *------------------------------------------------*/ /* Loop through each neighbor box. If the grown neighbor box intersects box i, then the intersection is a send region. If the neighbor box was shifted to handle periodicity, we need to (positive) shift it back. */ num_cboxes = 0; for (k = 0; k < neighbor_count; k++) { hood_box = hypre_BoxArrayBox(neighbor_boxes, k); /* check the stencil grid to see if it makes sense to intersect */ for (d = 0; d < ndim; d++) { sgindex[d] = 1; s = hypre_BoxIMinD(box, d) - hypre_BoxIMaxD(hood_box, d); if (s > 0) { sgindex[d] = 2; } s = hypre_BoxIMinD(hood_box, d) - hypre_BoxIMaxD(box, d); if (s > 0) { sgindex[d] = 0; } } /* it makes sense only if we have at least one non-zero entry */ si = hypre_BoxIndexRank(stencil_box, sgindex); if (stencil_grid[si]) { /* grow the neighbor box and intersect */ hypre_CopyBox(hood_box, grow_box); for (d = 0; d < ndim; d++) { hypre_BoxIMinD(grow_box, d) -= grow[d][0]; hypre_BoxIMaxD(grow_box, d) += grow[d][1]; } hypre_IntersectBoxes(box, grow_box, int_box); /* if we have a positive volume box, this is a send region */ if (hypre_BoxVolume(int_box)) { /* keep track of which neighbor: k... */ cboxes_neighbor_location[num_cboxes] = k; cboxes[num_cboxes] = &cboxes_mem[num_cboxes]; /* keep the intersected box */ hypre_CopyBox(int_box, cboxes[num_cboxes]); num_cboxes++; } } }/* end of loop through neighbors */ /* create send_box_array and send_procs for box i */ send_box_array = hypre_BoxArrayArrayBoxArray(send_boxes, i); hypre_BoxArraySetSize(send_box_array, num_cboxes); send_procs[i] = hypre_CTAlloc(HYPRE_Int, num_cboxes); send_rboxnums[i] = hypre_CTAlloc(HYPRE_Int, num_cboxes); send_rbox_array = hypre_BoxArrayArrayBoxArray(send_rboxes, i); hypre_BoxArraySetSize(send_rbox_array, num_cboxes); for (m = 0; m < num_cboxes; m++) { loc = cboxes_neighbor_location[m]; send_procs[i][m] = neighbor_procs[loc]; send_rboxnums[i][m] = neighbor_ids[loc]; hypre_CopyBox(cboxes[m], hypre_BoxArrayBox(send_box_array, m)); /* if periodic, positive shift before copying to the rbox_array */ if (neighbor_shifts[loc]) /* periodic if shift != 0 */ { pshift = hypre_StructGridPShift(grid, neighbor_shifts[loc]); hypre_BoxShiftPos(cboxes[m], pshift); } hypre_CopyBox(cboxes[m], hypre_BoxArrayBox(send_rbox_array, m)); cboxes[m] = NULL; } } /* end of loop through each local box */