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]; }
HYPRE_Int hypre_SMGResidualSetup( void *residual_vdata, hypre_StructMatrix *A, hypre_StructVector *x, hypre_StructVector *b, hypre_StructVector *r ) { HYPRE_Int ierr = 0; hypre_SMGResidualData *residual_data = residual_vdata; hypre_IndexRef base_index = (residual_data -> base_index); hypre_IndexRef base_stride = (residual_data -> base_stride); hypre_StructGrid *grid; hypre_StructStencil *stencil; hypre_BoxArray *base_points; hypre_ComputeInfo *compute_info; hypre_ComputePkg *compute_pkg; /*---------------------------------------------------------- * Set up base points and the compute package *----------------------------------------------------------*/ grid = hypre_StructMatrixGrid(A); stencil = hypre_StructMatrixStencil(A); base_points = hypre_BoxArrayDuplicate(hypre_StructGridBoxes(grid)); hypre_ProjectBoxArray(base_points, base_index, base_stride); hypre_CreateComputeInfo(grid, stencil, &compute_info); hypre_ComputeInfoProjectComp(compute_info, base_index, base_stride); hypre_ComputePkgCreate(compute_info, hypre_StructVectorDataSpace(x), 1, grid, &compute_pkg); /*---------------------------------------------------------- * Set up the residual data structure *----------------------------------------------------------*/ (residual_data -> A) = hypre_StructMatrixRef(A); (residual_data -> x) = hypre_StructVectorRef(x); (residual_data -> b) = hypre_StructVectorRef(b); (residual_data -> r) = hypre_StructVectorRef(r); (residual_data -> base_points) = base_points; (residual_data -> compute_pkg) = compute_pkg; /*----------------------------------------------------- * Compute flops *-----------------------------------------------------*/ (residual_data -> flops) = (hypre_StructMatrixGlobalSize(A) + hypre_StructVectorGlobalSize(x)) / (hypre_IndexX(base_stride) * hypre_IndexY(base_stride) * hypre_IndexZ(base_stride) ); return ierr; }
int hypre_StructMatvecSetup( void *matvec_vdata, hypre_StructMatrix *A, hypre_StructVector *x ) { int ierr = 0; hypre_StructMatvecData *matvec_data = matvec_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_Index unit_stride; hypre_ComputePkg *compute_pkg; /*---------------------------------------------------------- * Set up the compute package *----------------------------------------------------------*/ grid = hypre_StructMatrixGrid(A); stencil = hypre_StructMatrixStencil(A); hypre_CreateComputeInfo(grid, stencil, &send_boxes, &recv_boxes, &send_processes, &recv_processes, &indt_boxes, &dept_boxes); hypre_SetIndex(unit_stride, 1, 1, 1); hypre_ComputePkgCreate(send_boxes, recv_boxes, unit_stride, unit_stride, send_processes, recv_processes, indt_boxes, dept_boxes, unit_stride, grid, hypre_StructVectorDataSpace(x), 1, &compute_pkg); /*---------------------------------------------------------- * Set up the matvec data structure *----------------------------------------------------------*/ (matvec_data -> A) = hypre_StructMatrixRef(A); (matvec_data -> x) = hypre_StructVectorRef(x); (matvec_data -> compute_pkg) = compute_pkg; return ierr; }
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_SMGSetStructVectorConstantValues( hypre_StructVector *vector, double values, hypre_BoxArray *box_array, hypre_Index stride ) { int ierr = 0; hypre_Box *v_data_box; int vi; double *vp; hypre_Box *box; hypre_Index loop_size; hypre_IndexRef start; int loopi, loopj, loopk; int i; /*----------------------------------------------------------------------- * Set the vector coefficients *-----------------------------------------------------------------------*/ hypre_ForBoxI(i, box_array) { box = hypre_BoxArrayBox(box_array, i); start = hypre_BoxIMin(box); v_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(vector), i); vp = hypre_StructVectorBoxData(vector, i); hypre_BoxGetStrideSize(box, stride, loop_size); hypre_BoxLoop1Begin(loop_size, v_data_box, start, stride, vi); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,vi #include "hypre_box_smp_forloop.h" hypre_BoxLoop1For(loopi, loopj, loopk, vi) { vp[vi] = values; } hypre_BoxLoop1End(vi); }
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]; } }
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_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; }
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); } }
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_SMGResidualSetup( void *residual_vdata, hypre_StructMatrix *A, hypre_StructVector *x, hypre_StructVector *b, hypre_StructVector *r ) { int ierr = 0; hypre_SMGResidualData *residual_data = residual_vdata; hypre_IndexRef base_index = (residual_data -> base_index); hypre_IndexRef base_stride = (residual_data -> base_stride); hypre_Index unit_stride; 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_BoxArray *base_points; hypre_ComputePkg *compute_pkg; /*---------------------------------------------------------- * Set up base points and the compute package *----------------------------------------------------------*/ grid = hypre_StructMatrixGrid(A); stencil = hypre_StructMatrixStencil(A); hypre_SetIndex(unit_stride, 1, 1, 1); base_points = hypre_BoxArrayDuplicate(hypre_StructGridBoxes(grid)); hypre_ProjectBoxArray(base_points, base_index, base_stride); hypre_CreateComputeInfo(grid, stencil, &send_boxes, &recv_boxes, &send_processes, &recv_processes, &indt_boxes, &dept_boxes); hypre_ProjectBoxArrayArray(indt_boxes, base_index, base_stride); hypre_ProjectBoxArrayArray(dept_boxes, base_index, base_stride); hypre_ComputePkgCreate(send_boxes, recv_boxes, unit_stride, unit_stride, send_processes, recv_processes, indt_boxes, dept_boxes, base_stride, grid, hypre_StructVectorDataSpace(x), 1, &compute_pkg); /*---------------------------------------------------------- * Set up the residual data structure *----------------------------------------------------------*/ (residual_data -> A) = hypre_StructMatrixRef(A); (residual_data -> x) = hypre_StructVectorRef(x); (residual_data -> b) = hypre_StructVectorRef(b); (residual_data -> r) = hypre_StructVectorRef(r); (residual_data -> base_points) = base_points; (residual_data -> compute_pkg) = compute_pkg; /*----------------------------------------------------- * Compute flops *-----------------------------------------------------*/ (residual_data -> flops) = (hypre_StructMatrixGlobalSize(A) + hypre_StructVectorGlobalSize(x)) / (hypre_IndexX(base_stride) * hypre_IndexY(base_stride) * hypre_IndexZ(base_stride) ); return ierr; }
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_PointRelax( void *relax_vdata, hypre_StructMatrix *A, hypre_StructVector *b, hypre_StructVector *x ) { hypre_PointRelaxData *relax_data = (hypre_PointRelaxData *)relax_vdata; int max_iter = (relax_data -> max_iter); int zero_guess = (relax_data -> zero_guess); double weight = (relax_data -> weight); int num_pointsets = (relax_data -> num_pointsets); int *pointset_ranks = (relax_data -> pointset_ranks); hypre_Index *pointset_strides = (relax_data -> pointset_strides); hypre_StructVector *t = (relax_data -> t); int diag_rank = (relax_data -> diag_rank); hypre_ComputePkg **compute_pkgs = (relax_data -> compute_pkgs); 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 *b_data_box; hypre_Box *x_data_box; hypre_Box *t_data_box; int Ai; int bi; int xi; int ti; double *Ap; double *bp; double *xp; double *tp; hypre_IndexRef stride; hypre_IndexRef start; hypre_Index loop_size; hypre_StructStencil *stencil; hypre_Index *stencil_shape; int stencil_size; int iter, p, compute_i, i, j, si; int loopi, loopj, loopk; int pointset; int ierr = 0; /*---------------------------------------------------------- * Initialize some things and deal with special cases *----------------------------------------------------------*/ hypre_BeginTiming(relax_data -> time_index); hypre_StructMatrixDestroy(relax_data -> A); hypre_StructVectorDestroy(relax_data -> b); hypre_StructVectorDestroy(relax_data -> x); (relax_data -> A) = hypre_StructMatrixRef(A); (relax_data -> x) = hypre_StructVectorRef(x); (relax_data -> b) = hypre_StructVectorRef(b); (relax_data -> num_iterations) = 0; /* if max_iter is zero, return */ if (max_iter == 0) { /* if using a zero initial guess, return zero */ if (zero_guess) { hypre_StructVectorSetConstantValues(x, 0.0); } hypre_EndTiming(relax_data -> time_index); return ierr; } stencil = hypre_StructMatrixStencil(A); stencil_shape = hypre_StructStencilShape(stencil); stencil_size = hypre_StructStencilSize(stencil); /*---------------------------------------------------------- * Do zero_guess iteration *----------------------------------------------------------*/ p = 0; iter = 0; if (zero_guess) { pointset = pointset_ranks[p]; compute_pkg = compute_pkgs[pointset]; stride = pointset_strides[pointset]; for (compute_i = 0; compute_i < 2; compute_i++) { switch(compute_i) { case 0: { compute_box_aa = hypre_ComputePkgIndtBoxes(compute_pkg); } break; case 1: { compute_box_aa = hypre_ComputePkgDeptBoxes(compute_pkg); } break; } hypre_ForBoxArrayI(i, compute_box_aa) { compute_box_a = hypre_BoxArrayArrayBoxArray(compute_box_aa, i); A_data_box = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(A), i); b_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(b), i); x_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(x), i); Ap = hypre_StructMatrixBoxData(A, i, diag_rank); bp = hypre_StructVectorBoxData(b, i); xp = hypre_StructVectorBoxData(x, i); hypre_ForBoxI(j, compute_box_a) { compute_box = hypre_BoxArrayBox(compute_box_a, j); start = hypre_BoxIMin(compute_box); hypre_BoxGetStrideSize(compute_box, stride, loop_size); hypre_BoxLoop3Begin(loop_size, A_data_box, start, stride, Ai, b_data_box, start, stride, bi, x_data_box, start, stride, xi); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,Ai,bi,xi #include "hypre_box_smp_forloop.h" hypre_BoxLoop3For(loopi, loopj, loopk, Ai, bi, xi) { xp[xi] = bp[bi] / Ap[Ai]; } hypre_BoxLoop3End(Ai, bi, xi); } } }
int hypre_SMGResidual( void *residual_vdata, hypre_StructMatrix *A, hypre_StructVector *x, hypre_StructVector *b, hypre_StructVector *r ) { int ierr = 0; hypre_SMGResidualData *residual_data = residual_vdata; hypre_IndexRef base_stride = (residual_data -> base_stride); hypre_BoxArray *base_points = (residual_data -> base_points); hypre_ComputePkg *compute_pkg = (residual_data -> 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 *b_data_box; hypre_Box *r_data_box; int Ai; int xi; int bi; int ri; double *Ap; double *xp; double *bp; double *rp; hypre_Index loop_size; hypre_IndexRef start; hypre_StructStencil *stencil; hypre_Index *stencil_shape; int stencil_size; int compute_i, i, j, si; int loopi, loopj, loopk; hypre_BeginTiming(residual_data -> time_index); /*----------------------------------------------------------------------- * Compute residual r = b - Ax *-----------------------------------------------------------------------*/ stencil = hypre_StructMatrixStencil(A); stencil_shape = hypre_StructStencilShape(stencil); stencil_size = hypre_StructStencilSize(stencil); for (compute_i = 0; compute_i < 2; compute_i++) { switch(compute_i) { case 0: { xp = hypre_StructVectorData(x); hypre_InitializeIndtComputations(compute_pkg, xp, &comm_handle); compute_box_aa = hypre_ComputePkgIndtBoxes(compute_pkg); /*---------------------------------------- * Copy b into r *----------------------------------------*/ compute_box_a = base_points; hypre_ForBoxI(i, compute_box_a) { compute_box = hypre_BoxArrayBox(compute_box_a, i); start = hypre_BoxIMin(compute_box); b_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(b), i); r_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(r), i); bp = hypre_StructVectorBoxData(b, i); rp = hypre_StructVectorBoxData(r, i); hypre_BoxGetStrideSize(compute_box, base_stride, loop_size); hypre_BoxLoop2Begin(loop_size, b_data_box, start, base_stride, bi, r_data_box, start, base_stride, ri); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,bi,ri #include "hypre_box_smp_forloop.h" hypre_BoxLoop2For(loopi, loopj, loopk, bi, ri) { rp[ri] = bp[bi]; } hypre_BoxLoop2End(bi, ri); } } break; case 1: { hypre_FinalizeIndtComputations(comm_handle); compute_box_aa = hypre_ComputePkgDeptBoxes(compute_pkg); } break; } /*-------------------------------------------------------------------- * Compute r -= A*x *--------------------------------------------------------------------*/ hypre_ForBoxArrayI(i, compute_box_aa) { compute_box_a = hypre_BoxArrayArrayBoxArray(compute_box_aa, i); A_data_box = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(A), i); x_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(x), i); r_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(r), i); rp = hypre_StructVectorBoxData(r, i); hypre_ForBoxI(j, compute_box_a) { compute_box = hypre_BoxArrayBox(compute_box_a, j); start = hypre_BoxIMin(compute_box); for (si = 0; si < stencil_size; si++) { Ap = hypre_StructMatrixBoxData(A, i, si); xp = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[si]); hypre_BoxGetStrideSize(compute_box, base_stride, loop_size); hypre_BoxLoop3Begin(loop_size, A_data_box, start, base_stride, Ai, x_data_box, start, base_stride, xi, r_data_box, start, base_stride, ri); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,Ai,xi,ri #include "hypre_box_smp_forloop.h" hypre_BoxLoop3For(loopi, loopj, loopk, Ai, xi, ri) { rp[ri] -= Ap[Ai] * xp[xi]; } hypre_BoxLoop3End(Ai, xi, ri); }
int hypre_SMGResidual( void *residual_vdata, hypre_StructMatrix *A, hypre_StructVector *x, hypre_StructVector *b, hypre_StructVector *r ) { int ierr = 0; hypre_SMGResidualData *residual_data = (hypre_SMGResidualData *)residual_vdata; hypre_IndexRef base_stride = (residual_data -> base_stride); hypre_BoxArray *base_points = (residual_data -> base_points); hypre_ComputePkg *compute_pkg = (residual_data -> 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 *b_data_box; hypre_Box *r_data_box; int Ai; int xi; int bi; int ri; double *Ap; double *xp; double *bp; double *rp; hypre_Index loop_size; hypre_IndexRef start; hypre_StructStencil *stencil; hypre_Index *stencil_shape; int stencil_size; int compute_i, i, j, si; int loopi, loopj, loopk; hypre_BeginTiming(residual_data -> time_index); /*----------------------------------------------------------------------- * Compute residual r = b - Ax *-----------------------------------------------------------------------*/ stencil = hypre_StructMatrixStencil(A); stencil_shape = hypre_StructStencilShape(stencil); stencil_size = hypre_StructStencilSize(stencil); for (compute_i = 0; compute_i < 2; compute_i++) { switch(compute_i) { case 0: { xp = hypre_StructVectorData(x); hypre_InitializeIndtComputations(compute_pkg, xp, &comm_handle); compute_box_aa = hypre_ComputePkgIndtBoxes(compute_pkg); /*---------------------------------------- * Copy b into r *----------------------------------------*/ compute_box_a = base_points; hypre_ForBoxI(i, compute_box_a) { compute_box = hypre_BoxArrayBox(compute_box_a, i); start = hypre_BoxIMin(compute_box); b_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(b), i); r_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(r), i); bp = hypre_StructVectorBoxData(b, i); rp = hypre_StructVectorBoxData(r, i); hypre_BoxGetStrideSize(compute_box, base_stride, loop_size); hypre_BoxLoop2Begin(loop_size, b_data_box, start, base_stride, bi, r_data_box, start, base_stride, ri); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,bi,ri #include "hypre_box_smp_forloop.h" hypre_BoxLoop2For(loopi, loopj, loopk, bi, ri) { rp[ri] = bp[bi]; } hypre_BoxLoop2End(bi, ri); } } break; case 1: { hypre_FinalizeIndtComputations(comm_handle); compute_box_aa = hypre_ComputePkgDeptBoxes(compute_pkg); } break; } /*-------------------------------------------------------------------- * Compute r -= A*x *--------------------------------------------------------------------*/ hypre_ForBoxArrayI(i, compute_box_aa) { compute_box_a = hypre_BoxArrayArrayBoxArray(compute_box_aa, i); A_data_box = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(A), i); x_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(x), i); r_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(r), i); rp = hypre_StructVectorBoxData(r, i); hypre_ForBoxI(j, compute_box_a) { compute_box = hypre_BoxArrayBox(compute_box_a, j); start = hypre_BoxIMin(compute_box); for (si = 0; si < stencil_size; si++) { Ap = hypre_StructMatrixBoxData(A, i, si); xp = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[si]); hypre_BoxGetStrideSize(compute_box, base_stride, loop_size); hypre_BoxLoop3Begin(loop_size, A_data_box, start, base_stride, Ai, x_data_box, start, base_stride, xi, r_data_box, start, base_stride, ri); #if 0 /* The following portion is preprocessed to be handled by ROSE outliner */ #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,Ai,xi,ri #include "hypre_box_smp_forloop.h" hypre_BoxLoop3For(loopi, loopj, loopk, Ai, xi, ri) { rp[ri] -= Ap[Ai] * xp[xi]; } hypre_BoxLoop3End(Ai, xi, ri); #else for (hypre__block = 0; hypre__block < hypre__num_blocks; hypre__block++) { loopi = 0; loopj = 0; loopk = 0; hypre__nx = hypre__mx; hypre__ny = hypre__my; hypre__nz = hypre__mz; if (hypre__num_blocks > 1) { if (hypre__dir == 0) { loopi = hypre__block * hypre__div + (((hypre__mod) < (hypre__block)) ? (hypre__mod) : (hypre__block)); hypre__nx = hypre__div + ((hypre__mod > hypre__block) ? 1 : 0); } else if (hypre__dir == 1) { loopj = hypre__block * hypre__div + (((hypre__mod) < (hypre__block)) ? (hypre__mod) : (hypre__block)); hypre__ny = hypre__div + ((hypre__mod > hypre__block) ? 1 : 0); } else if (hypre__dir == 2) { loopk = hypre__block * hypre__div + (((hypre__mod) < (hypre__block)) ? (hypre__mod) : (hypre__block)); hypre__nz = hypre__div + ((hypre__mod > hypre__block) ? 1 : 0); } }; Ai = hypre__i1start + loopi * hypre__sx1 + loopj * hypre__sy1 + loopk * hypre__sz1; xi = hypre__i2start + loopi * hypre__sx2 + loopj * hypre__sy2 + loopk * hypre__sz2; ri = hypre__i3start + loopi * hypre__sx3 + loopj * hypre__sy3 + loopk * hypre__sz3; //begin of the loop #if 0 // for (loopk = 0; loopk < hypre__nz; loopk++) { for (loopj = 0; loopj < hypre__ny; loopj++) { for (loopi = 0; loopi < hypre__nx; loopi++) { { rp[ri] -= Ap[Ai] * xp[xi]; } Ai += hypre__sx1; xi += hypre__sx2; ri += hypre__sx3; } Ai += hypre__sy1 - hypre__nx * hypre__sx1; xi += hypre__sy2 - hypre__nx * hypre__sx2; ri += hypre__sy3 - hypre__nx * hypre__sx3; } Ai += hypre__sz1 - hypre__ny * hypre__sy1; xi += hypre__sz2 - hypre__ny * hypre__sy2; ri += hypre__sz3 - hypre__ny * hypre__sy3; } // end of the loop #else #if BLCR_CHECKPOINTING // Only checkpoint it at the first occurrance. if (g_checkpoint_flag == 0) { int err; cr_checkpoint_args_t cr_args; cr_checkpoint_handle_t cr_handle; cr_initialize_checkpoint_args_t(&cr_args); cr_args.cr_scope = CR_SCOPE_PROC;// a process cr_args.cr_target = 0; //self cr_args.cr_signal = SIGKILL; // kill after checkpointing cr_args.cr_fd = open("dump.yy", O_WRONLY|O_CREAT|O_LARGEFILE, 0400); if (cr_args.cr_fd < 0) { printf("Error: cannot open file for checkpoiting context\n"); abort(); } g_checkpoint_flag ++; printf("Checkpoiting: starting here ..\n"); err = cr_request_checkpoint(&cr_args, &cr_handle); if (err < 0) { printf("cannot request checkpoining! err=%d\n",err); abort(); } // block until the request is served cr_enter_cs(cr); cr_leave_cs(cr); printf("Checkpoiting: restarting here ..\n"); } #endif OUT__1__6755__(&Ai,&xi,&ri,&Ap,&xp,&rp,&loopi,&loopj,&loopk,&hypre__sx1,&hypre__sy1,&hypre__sz1,&hypre__sx2,&hypre__sy2,&hypre__sz2,&hypre__sx3,&hypre__sy3,&hypre__sz3,&hypre__nx,&hypre__ny,&hypre__nz); #endif } };
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; }
HYPRE_Int hypre_SMGResidual( void *residual_vdata, hypre_StructMatrix *A, hypre_StructVector *x, hypre_StructVector *b, hypre_StructVector *r ) { HYPRE_Int ierr; hypre_SMGResidualData *residual_data = residual_vdata; hypre_IndexRef base_stride = (residual_data -> base_stride); hypre_BoxArray *base_points = (residual_data -> base_points); hypre_ComputePkg *compute_pkg = (residual_data -> 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 *b_data_box; hypre_Box *r_data_box; HYPRE_Int Ai; HYPRE_Int xi; HYPRE_Int bi; HYPRE_Int ri; double *Ap0; double *xp0; double *bp; double *rp; hypre_Index loop_size; hypre_IndexRef start; hypre_StructStencil *stencil; hypre_Index *stencil_shape; HYPRE_Int stencil_size; HYPRE_Int compute_i, i, j, si; HYPRE_Int loopi, loopj, loopk; double *Ap1, *Ap2; double *Ap3, *Ap4; double *Ap5, *Ap6; double *Ap7, *Ap8, *Ap9; double *Ap10, *Ap11, *Ap12, *Ap13, *Ap14; double *Ap15, *Ap16, *Ap17, *Ap18; double *Ap19, *Ap20, *Ap21, *Ap22, *Ap23, *Ap24, *Ap25, *Ap26; double *xp1, *xp2; double *xp3, *xp4; double *xp5, *xp6; double *xp7, *xp8, *xp9; double *xp10, *xp11, *xp12, *xp13, *xp14; double *xp15, *xp16, *xp17, *xp18; double *xp19, *xp20, *xp21, *xp22, *xp23, *xp24, *xp25, *xp26; hypre_BeginTiming(residual_data -> time_index); /*----------------------------------------------------------------------- * Compute residual r = b - Ax *-----------------------------------------------------------------------*/ stencil = hypre_StructMatrixStencil(A); stencil_shape = hypre_StructStencilShape(stencil); stencil_size = hypre_StructStencilSize(stencil); for (compute_i = 0; compute_i < 2; compute_i++) { switch(compute_i) { case 0: { xp0 = hypre_StructVectorData(x); hypre_InitializeIndtComputations(compute_pkg, xp0, &comm_handle); compute_box_aa = hypre_ComputePkgIndtBoxes(compute_pkg); /*---------------------------------------- * Copy b into r *----------------------------------------*/ compute_box_a = base_points; hypre_ForBoxI(i, compute_box_a) { compute_box = hypre_BoxArrayBox(compute_box_a, i); start = hypre_BoxIMin(compute_box); b_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(b), i); r_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(r), i); bp = hypre_StructVectorBoxData(b, i); rp = hypre_StructVectorBoxData(r, i); hypre_BoxGetStrideSize(compute_box, base_stride, loop_size); hypre_BoxLoop2Begin(loop_size, b_data_box, start, base_stride, bi, r_data_box, start, base_stride, ri); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,bi,ri #include "hypre_box_smp_forloop.h" hypre_BoxLoop2For(loopi, loopj, loopk, bi, ri) { rp[ri] = bp[bi]; } hypre_BoxLoop2End(bi, ri); } } break; case 1: { hypre_FinalizeIndtComputations(comm_handle); compute_box_aa = hypre_ComputePkgDeptBoxes(compute_pkg); } break; } /*-------------------------------------------------------------------- * Compute r -= A*x *--------------------------------------------------------------------*/ hypre_ForBoxArrayI(i, compute_box_aa) { compute_box_a = hypre_BoxArrayArrayBoxArray(compute_box_aa, i); A_data_box = hypre_BoxArrayBox(hypre_StructMatrixDataSpace(A), i); x_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(x), i); r_data_box = hypre_BoxArrayBox(hypre_StructVectorDataSpace(r), i); rp = hypre_StructVectorBoxData(r, i); /*-------------------------------------------------------------- * Switch statement to direct control (based on stencil size) to * code to get pointers and offsets fo A and x. *--------------------------------------------------------------*/ switch (stencil_size) { case 1: Ap0 = hypre_StructMatrixBoxData(A, i, 0); xp0 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[0]); break; case 3: Ap0 = hypre_StructMatrixBoxData(A, i, 0); Ap1 = hypre_StructMatrixBoxData(A, i, 1); Ap2 = hypre_StructMatrixBoxData(A, i, 2); xp0 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[0]); xp1 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[1]); xp2 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[2]); break; case 5: Ap0 = hypre_StructMatrixBoxData(A, i, 0); Ap1 = hypre_StructMatrixBoxData(A, i, 1); Ap2 = hypre_StructMatrixBoxData(A, i, 2); Ap3 = hypre_StructMatrixBoxData(A, i, 3); Ap4 = hypre_StructMatrixBoxData(A, i, 4); xp0 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[0]); xp1 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[1]); xp2 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[2]); xp3 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[3]); xp4 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[4]); break; case 7: Ap0 = hypre_StructMatrixBoxData(A, i, 0); Ap1 = hypre_StructMatrixBoxData(A, i, 1); Ap2 = hypre_StructMatrixBoxData(A, i, 2); Ap3 = hypre_StructMatrixBoxData(A, i, 3); Ap4 = hypre_StructMatrixBoxData(A, i, 4); Ap5 = hypre_StructMatrixBoxData(A, i, 5); Ap6 = hypre_StructMatrixBoxData(A, i, 6); xp0 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[0]); xp1 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[1]); xp2 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[2]); xp3 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[3]); xp4 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[4]); xp5 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[5]); xp6 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[6]); break; case 9: Ap0 = hypre_StructMatrixBoxData(A, i, 0); Ap1 = hypre_StructMatrixBoxData(A, i, 1); Ap2 = hypre_StructMatrixBoxData(A, i, 2); Ap3 = hypre_StructMatrixBoxData(A, i, 3); Ap4 = hypre_StructMatrixBoxData(A, i, 4); Ap5 = hypre_StructMatrixBoxData(A, i, 5); Ap6 = hypre_StructMatrixBoxData(A, i, 6); Ap7 = hypre_StructMatrixBoxData(A, i, 7); Ap8 = hypre_StructMatrixBoxData(A, i, 8); xp0 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[0]); xp1 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[1]); xp2 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[2]); xp3 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[3]); xp4 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[4]); xp5 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[5]); xp6 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[6]); xp7 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[7]); xp8 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[8]); break; case 15: Ap0 = hypre_StructMatrixBoxData(A, i, 0); Ap1 = hypre_StructMatrixBoxData(A, i, 1); Ap2 = hypre_StructMatrixBoxData(A, i, 2); Ap3 = hypre_StructMatrixBoxData(A, i, 3); Ap4 = hypre_StructMatrixBoxData(A, i, 4); Ap5 = hypre_StructMatrixBoxData(A, i, 5); Ap6 = hypre_StructMatrixBoxData(A, i, 6); Ap7 = hypre_StructMatrixBoxData(A, i, 7); Ap8 = hypre_StructMatrixBoxData(A, i, 8); Ap9 = hypre_StructMatrixBoxData(A, i, 9); Ap10 = hypre_StructMatrixBoxData(A, i, 10); Ap11 = hypre_StructMatrixBoxData(A, i, 11); Ap12 = hypre_StructMatrixBoxData(A, i, 12); Ap13 = hypre_StructMatrixBoxData(A, i, 13); Ap14 = hypre_StructMatrixBoxData(A, i, 14); xp0 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[0]); xp1 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[1]); xp2 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[2]); xp3 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[3]); xp4 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[4]); xp5 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[5]); xp6 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[6]); xp7 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[7]); xp8 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[8]); xp9 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[9]); xp10 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[10]); xp11 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[11]); xp12 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[12]); xp13 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[13]); xp14 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[14]); break; case 19: Ap0 = hypre_StructMatrixBoxData(A, i, 0); Ap1 = hypre_StructMatrixBoxData(A, i, 1); Ap2 = hypre_StructMatrixBoxData(A, i, 2); Ap3 = hypre_StructMatrixBoxData(A, i, 3); Ap4 = hypre_StructMatrixBoxData(A, i, 4); Ap5 = hypre_StructMatrixBoxData(A, i, 5); Ap6 = hypre_StructMatrixBoxData(A, i, 6); Ap7 = hypre_StructMatrixBoxData(A, i, 7); Ap8 = hypre_StructMatrixBoxData(A, i, 8); Ap9 = hypre_StructMatrixBoxData(A, i, 9); Ap10 = hypre_StructMatrixBoxData(A, i, 10); Ap11 = hypre_StructMatrixBoxData(A, i, 11); Ap12 = hypre_StructMatrixBoxData(A, i, 12); Ap13 = hypre_StructMatrixBoxData(A, i, 13); Ap14 = hypre_StructMatrixBoxData(A, i, 14); Ap15 = hypre_StructMatrixBoxData(A, i, 15); Ap16 = hypre_StructMatrixBoxData(A, i, 16); Ap17 = hypre_StructMatrixBoxData(A, i, 17); Ap18 = hypre_StructMatrixBoxData(A, i, 18); xp0 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[0]); xp1 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[1]); xp2 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[2]); xp3 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[3]); xp4 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[4]); xp5 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[5]); xp6 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[6]); xp7 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[7]); xp8 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[8]); xp9 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[9]); xp10 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[10]); xp11 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[11]); xp12 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[12]); xp13 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[13]); xp14 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[14]); xp15 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[15]); xp16 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[16]); xp17 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[17]); xp18 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[18]); break; case 27: Ap0 = hypre_StructMatrixBoxData(A, i, 0); Ap1 = hypre_StructMatrixBoxData(A, i, 1); Ap2 = hypre_StructMatrixBoxData(A, i, 2); Ap3 = hypre_StructMatrixBoxData(A, i, 3); Ap4 = hypre_StructMatrixBoxData(A, i, 4); Ap5 = hypre_StructMatrixBoxData(A, i, 5); Ap6 = hypre_StructMatrixBoxData(A, i, 6); Ap7 = hypre_StructMatrixBoxData(A, i, 7); Ap8 = hypre_StructMatrixBoxData(A, i, 8); Ap9 = hypre_StructMatrixBoxData(A, i, 9); Ap10 = hypre_StructMatrixBoxData(A, i, 10); Ap11 = hypre_StructMatrixBoxData(A, i, 11); Ap12 = hypre_StructMatrixBoxData(A, i, 12); Ap13 = hypre_StructMatrixBoxData(A, i, 13); Ap14 = hypre_StructMatrixBoxData(A, i, 14); Ap15 = hypre_StructMatrixBoxData(A, i, 15); Ap16 = hypre_StructMatrixBoxData(A, i, 16); Ap17 = hypre_StructMatrixBoxData(A, i, 17); Ap18 = hypre_StructMatrixBoxData(A, i, 18); Ap19 = hypre_StructMatrixBoxData(A, i, 19); Ap20 = hypre_StructMatrixBoxData(A, i, 20); Ap21 = hypre_StructMatrixBoxData(A, i, 21); Ap22 = hypre_StructMatrixBoxData(A, i, 22); Ap23 = hypre_StructMatrixBoxData(A, i, 23); Ap24 = hypre_StructMatrixBoxData(A, i, 24); Ap25 = hypre_StructMatrixBoxData(A, i, 25); Ap26 = hypre_StructMatrixBoxData(A, i, 26); xp0 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[0]); xp1 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[1]); xp2 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[2]); xp3 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[3]); xp4 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[4]); xp5 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[5]); xp6 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[6]); xp7 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[7]); xp8 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[8]); xp9 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[9]); xp10 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[10]); xp11 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[11]); xp12 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[12]); xp13 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[13]); xp14 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[14]); xp15 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[15]); xp16 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[16]); xp17 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[17]); xp18 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[18]); xp19 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[19]); xp20 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[20]); xp21 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[21]); xp22 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[22]); xp23 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[23]); xp24 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[24]); xp25 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[25]); xp26 = hypre_StructVectorBoxData(x, i) + hypre_BoxOffsetDistance(x_data_box, stencil_shape[26]); break; default: ; } hypre_ForBoxI(j, compute_box_a) { compute_box = hypre_BoxArrayBox(compute_box_a, j); start = hypre_BoxIMin(compute_box); /*------------------------------------------------------ * Switch statement to direct control to appropriate * box loop depending on stencil size *------------------------------------------------------*/ switch (stencil_size) { case 1: hypre_BoxGetStrideSize(compute_box, base_stride, loop_size); hypre_BoxLoop3Begin(loop_size, A_data_box, start, base_stride, Ai, x_data_box, start, base_stride, xi, r_data_box, start, base_stride, ri); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,Ai,xi,ri #include "hypre_box_smp_forloop.h" hypre_BoxLoop3For(loopi, loopj, loopk, Ai, xi, ri) { rp[ri] = rp[ri] - Ap0[Ai] * xp0[xi]; } hypre_BoxLoop3End(Ai, xi, ri); break; case 3: hypre_BoxGetStrideSize(compute_box, base_stride, loop_size); hypre_BoxLoop3Begin(loop_size, A_data_box, start, base_stride, Ai, x_data_box, start, base_stride, xi, r_data_box, start, base_stride, ri); #define HYPRE_BOX_SMP_PRIVATE loopk,loopi,loopj,Ai,xi,ri #include "hypre_box_smp_forloop.h" hypre_BoxLoop3For(loopi, loopj, loopk, Ai, xi, ri) { rp[ri] = rp[ri] - Ap0[Ai] * xp0[xi] - Ap1[Ai] * xp1[xi] - Ap2[Ai] * xp2[xi]; }
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_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); }
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_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); } }
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; }