HYPRE_Real
hypre_StructKrylovInnerProd( void *x,
void *y )
{
return ( hypre_StructInnerProd( (hypre_StructVector *) x,
(hypre_StructVector *) y ) );
}
HYPRE_Int
hypre_SStructPInnerProd( hypre_SStructPVector *px,
hypre_SStructPVector *py,
double *presult_ptr )
{
HYPRE_Int nvars = hypre_SStructPVectorNVars(px);
double presult;
double sresult;
HYPRE_Int var;
presult = 0.0;
for (var = 0; var < nvars; var++)
{
sresult = hypre_StructInnerProd(hypre_SStructPVectorSVector(px, var),
hypre_SStructPVectorSVector(py, var));
presult += sresult;
}
*presult_ptr = presult;
return hypre_error_flag;
}
hypre_int
main( hypre_int argc,
char *argv[] )
{
HYPRE_Int arg_index;
HYPRE_Int print_usage;
HYPRE_Int nx, ny, nz;
HYPRE_Int P, Q, R;
HYPRE_Int bx, by, bz;
HYPRE_StructGrid from_grid, to_grid;
HYPRE_StructVector from_vector, to_vector, check_vector;
HYPRE_CommPkg comm_pkg;
HYPRE_Int time_index;
HYPRE_Int num_procs, myid;
HYPRE_Int p, q, r;
HYPRE_Int dim;
HYPRE_Int nblocks ;
HYPRE_Int **ilower, **iupper, **iupper2;
HYPRE_Int istart[3];
HYPRE_Int i, ix, iy, iz, ib;
HYPRE_Int print_system = 0;
HYPRE_Real check;
/*-----------------------------------------------------------
* Initialize some stuff
*-----------------------------------------------------------*/
/* Initialize MPI */
hypre_MPI_Init(&argc, &argv);
hypre_MPI_Comm_size(hypre_MPI_COMM_WORLD, &num_procs );
hypre_MPI_Comm_rank(hypre_MPI_COMM_WORLD, &myid );
/*-----------------------------------------------------------
* Set defaults
*-----------------------------------------------------------*/
dim = 3;
nx = 2;
ny = 2;
nz = 2;
P = num_procs;
Q = 1;
R = 1;
bx = 1;
by = 1;
bz = 1;
istart[0] = 1;
istart[1] = 1;
istart[2] = 1;
/*-----------------------------------------------------------
* Parse command line
*-----------------------------------------------------------*/
print_usage = 0;
arg_index = 1;
while (arg_index < argc)
{
if ( strcmp(argv[arg_index], "-n") == 0 )
{
arg_index++;
nx = atoi(argv[arg_index++]);
ny = atoi(argv[arg_index++]);
nz = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-istart") == 0 )
{
arg_index++;
istart[0] = atoi(argv[arg_index++]);
istart[1] = atoi(argv[arg_index++]);
istart[2] = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-P") == 0 )
{
arg_index++;
P = atoi(argv[arg_index++]);
Q = atoi(argv[arg_index++]);
R = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-b") == 0 )
{
arg_index++;
bx = atoi(argv[arg_index++]);
by = atoi(argv[arg_index++]);
bz = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-d") == 0 )
{
arg_index++;
dim = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-print") == 0 )
{
arg_index++;
print_system = 1;
}
else if ( strcmp(argv[arg_index], "-help") == 0 )
{
print_usage = 1;
break;
}
else
{
arg_index++;
}
}
/*-----------------------------------------------------------
* Print usage info
*-----------------------------------------------------------*/
if ( (print_usage) && (myid == 0) )
{
hypre_printf("\n");
hypre_printf("Usage: %s [<options>]\n", argv[0]);
hypre_printf("\n");
hypre_printf(" -n <nx> <ny> <nz> : problem size per block\n");
hypre_printf(" -istart <ix> <iy> <iz> : start of box\n");
hypre_printf(" -P <Px> <Py> <Pz> : processor topology\n");
hypre_printf(" -b <bx> <by> <bz> : blocking per processor\n");
hypre_printf(" -d <dim> : problem dimension (2 or 3)\n");
hypre_printf(" -print : print vectors\n");
hypre_printf("\n");
}
if ( print_usage )
{
exit(1);
}
/*-----------------------------------------------------------
* Check a few things
*-----------------------------------------------------------*/
if ((P*Q*R) > num_procs)
{
if (myid == 0)
{
hypre_printf("Error: PxQxR is more than the number of processors\n");
}
exit(1);
}
else if ((P*Q*R) < num_procs)
{
if (myid == 0)
{
hypre_printf("Warning: PxQxR is less than the number of processors\n");
}
}
/*-----------------------------------------------------------
* Print driver parameters
*-----------------------------------------------------------*/
if (myid == 0)
{
hypre_printf("Running with these driver parameters:\n");
hypre_printf(" (nx, ny, nz) = (%d, %d, %d)\n", nx, ny, nz);
hypre_printf(" (ix, iy, iz) = (%d, %d, %d)\n",
istart[0],istart[1],istart[2]);
hypre_printf(" (Px, Py, Pz) = (%d, %d, %d)\n", P, Q, R);
hypre_printf(" (bx, by, bz) = (%d, %d, %d)\n", bx, by, bz);
hypre_printf(" dim = %d\n", dim);
}
/*-----------------------------------------------------------
* Set up the stencil structure (7 points) when matrix is NOT read from file
* Set up the grid structure used when NO files are read
*-----------------------------------------------------------*/
switch (dim)
{
case 1:
nblocks = bx;
p = myid % P;
break;
case 2:
nblocks = bx*by;
p = myid % P;
q = (( myid - p)/P) % Q;
break;
case 3:
nblocks = bx*by*bz;
p = myid % P;
q = (( myid - p)/P) % Q;
r = ( myid - p - P*q)/( P*Q );
break;
}
if (myid >= (P*Q*R))
{
/* My processor has no data on it */
nblocks = bx = by = bz = 0;
}
/*-----------------------------------------------------------
* prepare space for the extents
*-----------------------------------------------------------*/
ilower = hypre_CTAlloc(HYPRE_Int*, nblocks);
iupper = hypre_CTAlloc(HYPRE_Int*, nblocks);
iupper2 = hypre_CTAlloc(HYPRE_Int*, nblocks);
for (i = 0; i < nblocks; i++)
{
ilower[i] = hypre_CTAlloc(HYPRE_Int, dim);
iupper[i] = hypre_CTAlloc(HYPRE_Int, dim);
iupper2[i] = hypre_CTAlloc(HYPRE_Int, dim);
}
ib = 0;
switch (dim)
{
case 1:
for (ix = 0; ix < bx; ix++)
{
ilower[ib][0] = istart[0]+ nx*(bx*p+ix);
iupper[ib][0] = istart[0]+ nx*(bx*p+ix+1) - 1;
iupper2[ib][0] = iupper[ib][0];
if ( (ix == (bx-1)) && (p < (P-1)) )
iupper2[ib][0] = iupper[ib][0] + 1;
ib++;
}
break;
case 2:
for (iy = 0; iy < by; iy++)
for (ix = 0; ix < bx; ix++)
{
ilower[ib][0] = istart[0]+ nx*(bx*p+ix);
iupper[ib][0] = istart[0]+ nx*(bx*p+ix+1) - 1;
ilower[ib][1] = istart[1]+ ny*(by*q+iy);
iupper[ib][1] = istart[1]+ ny*(by*q+iy+1) - 1;
iupper2[ib][0] = iupper[ib][0];
iupper2[ib][1] = iupper[ib][1];
if ( (ix == (bx-1)) && (p < (P-1)) )
iupper2[ib][0] = iupper[ib][0] + 1;
if ( (iy == (by-1)) && (q < (Q-1)) )
iupper2[ib][1] = iupper[ib][1] + 1;
ib++;
}
break;
case 3:
for (iz = 0; iz < bz; iz++)
for (iy = 0; iy < by; iy++)
for (ix = 0; ix < bx; ix++)
{
ilower[ib][0] = istart[0]+ nx*(bx*p+ix);
iupper[ib][0] = istart[0]+ nx*(bx*p+ix+1) - 1;
ilower[ib][1] = istart[1]+ ny*(by*q+iy);
iupper[ib][1] = istart[1]+ ny*(by*q+iy+1) - 1;
ilower[ib][2] = istart[2]+ nz*(bz*r+iz);
iupper[ib][2] = istart[2]+ nz*(bz*r+iz+1) - 1;
iupper2[ib][0] = iupper[ib][0];
iupper2[ib][1] = iupper[ib][1];
iupper2[ib][2] = iupper[ib][2];
if ( (ix == (bx-1)) && (p < (P-1)) )
iupper2[ib][0] = iupper[ib][0] + 1;
if ( (iy == (by-1)) && (q < (Q-1)) )
iupper2[ib][1] = iupper[ib][1] + 1;
if ( (iz == (bz-1)) && (r < (R-1)) )
iupper2[ib][2] = iupper[ib][2] + 1;
ib++;
}
break;
}
HYPRE_StructGridCreate(hypre_MPI_COMM_WORLD, dim, &from_grid);
HYPRE_StructGridCreate(hypre_MPI_COMM_WORLD, dim, &to_grid);
for (ib = 0; ib < nblocks; ib++)
{
HYPRE_StructGridSetExtents(from_grid, ilower[ib], iupper[ib]);
HYPRE_StructGridSetExtents(to_grid, ilower[ib], iupper2[ib]);
}
HYPRE_StructGridAssemble(from_grid);
HYPRE_StructGridAssemble(to_grid);
/*-----------------------------------------------------------
* Set up the vectors
*-----------------------------------------------------------*/
HYPRE_StructVectorCreate(hypre_MPI_COMM_WORLD, from_grid, &from_vector);
HYPRE_StructVectorInitialize(from_vector);
AddValuesVector(from_grid, from_vector, 1.0);
HYPRE_StructVectorAssemble(from_vector);
HYPRE_StructVectorCreate(hypre_MPI_COMM_WORLD, to_grid, &to_vector);
HYPRE_StructVectorInitialize(to_vector);
AddValuesVector(to_grid, to_vector, 0.0);
HYPRE_StructVectorAssemble(to_vector);
/* Vector used to check the migration */
HYPRE_StructVectorCreate(hypre_MPI_COMM_WORLD, to_grid, &check_vector);
HYPRE_StructVectorInitialize(check_vector);
AddValuesVector(to_grid, check_vector, 1.0);
HYPRE_StructVectorAssemble(check_vector);
/*-----------------------------------------------------------
* Migrate
*-----------------------------------------------------------*/
time_index = hypre_InitializeTiming("Struct Migrate");
hypre_BeginTiming(time_index);
HYPRE_StructVectorGetMigrateCommPkg(from_vector, to_vector, &comm_pkg);
HYPRE_StructVectorMigrate(comm_pkg, from_vector, to_vector);
HYPRE_CommPkgDestroy(comm_pkg);
hypre_EndTiming(time_index);
hypre_PrintTiming("Struct Migrate", hypre_MPI_COMM_WORLD);
hypre_FinalizeTiming(time_index);
/*-----------------------------------------------------------
* Check the migration and print the result
*-----------------------------------------------------------*/
hypre_StructAxpy(-1.0, to_vector, check_vector);
check = hypre_StructInnerProd (check_vector, check_vector);
if (myid == 0)
{
printf("\nCheck = %1.0f (success = 0)\n\n", check);
}
/*-----------------------------------------------------------
* Print out the vectors
*-----------------------------------------------------------*/
if (print_system)
{
HYPRE_StructVectorPrint("struct_migrate.out.xfr", from_vector, 0);
HYPRE_StructVectorPrint("struct_migrate.out.xto", to_vector, 0);
}
/*-----------------------------------------------------------
* Finalize things
*-----------------------------------------------------------*/
HYPRE_StructGridDestroy(from_grid);
HYPRE_StructGridDestroy(to_grid);
for (i = 0; i < nblocks; i++)
{
hypre_TFree(ilower[i]);
hypre_TFree(iupper[i]);
hypre_TFree(iupper2[i]);
}
hypre_TFree(ilower);
hypre_TFree(iupper);
hypre_TFree(iupper2);
HYPRE_StructVectorDestroy(from_vector);
HYPRE_StructVectorDestroy(to_vector);
HYPRE_StructVectorDestroy(check_vector);
/* Finalize MPI */
hypre_MPI_Finalize();
return (0);
}
HYPRE_Int
hypre_SMGSolve( void *smg_vdata,
hypre_StructMatrix *A,
hypre_StructVector *b,
hypre_StructVector *x )
{
hypre_SMGData *smg_data = smg_vdata;
double tol = (smg_data -> tol);
HYPRE_Int max_iter = (smg_data -> max_iter);
HYPRE_Int rel_change = (smg_data -> rel_change);
HYPRE_Int zero_guess = (smg_data -> zero_guess);
HYPRE_Int num_levels = (smg_data -> num_levels);
HYPRE_Int num_pre_relax = (smg_data -> num_pre_relax);
HYPRE_Int num_post_relax = (smg_data -> num_post_relax);
hypre_IndexRef base_index = (smg_data -> base_index);
hypre_IndexRef base_stride = (smg_data -> base_stride);
hypre_StructMatrix **A_l = (smg_data -> A_l);
hypre_StructMatrix **PT_l = (smg_data -> PT_l);
hypre_StructMatrix **R_l = (smg_data -> R_l);
hypre_StructVector **b_l = (smg_data -> b_l);
hypre_StructVector **x_l = (smg_data -> x_l);
hypre_StructVector **r_l = (smg_data -> r_l);
hypre_StructVector **e_l = (smg_data -> e_l);
void **relax_data_l = (smg_data -> relax_data_l);
void **residual_data_l = (smg_data -> residual_data_l);
void **restrict_data_l = (smg_data -> restrict_data_l);
void **interp_data_l = (smg_data -> interp_data_l);
HYPRE_Int logging = (smg_data -> logging);
double *norms = (smg_data -> norms);
double *rel_norms = (smg_data -> rel_norms);
double b_dot_b = 0, r_dot_r, eps = 0;
double e_dot_e = 0, x_dot_x = 1;
HYPRE_Int i, l;
#if DEBUG
char filename[255];
#endif
/*-----------------------------------------------------
* Initialize some things and deal with special cases
*-----------------------------------------------------*/
hypre_BeginTiming(smg_data -> time_index);
hypre_StructMatrixDestroy(A_l[0]);
hypre_StructVectorDestroy(b_l[0]);
hypre_StructVectorDestroy(x_l[0]);
A_l[0] = hypre_StructMatrixRef(A);
b_l[0] = hypre_StructVectorRef(b);
x_l[0] = hypre_StructVectorRef(x);
(smg_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(smg_data -> time_index);
return hypre_error_flag;
}
/* part of convergence check */
if (tol > 0.0)
{
/* eps = (tol^2) */
b_dot_b = hypre_StructInnerProd(b_l[0], b_l[0]);
eps = tol*tol;
/* if rhs is zero, return a zero solution */
if (b_dot_b == 0.0)
{
hypre_StructVectorSetConstantValues(x, 0.0);
if (logging > 0)
{
norms[0] = 0.0;
rel_norms[0] = 0.0;
}
hypre_EndTiming(smg_data -> time_index);
return hypre_error_flag;
}
}
/*-----------------------------------------------------
* Do V-cycles:
* For each index l, "fine" = l, "coarse" = (l+1)
*-----------------------------------------------------*/
for (i = 0; i < max_iter; i++)
{
/*--------------------------------------------------
* Down cycle
*--------------------------------------------------*/
/* fine grid pre-relaxation */
if (num_levels > 1)
{
hypre_SMGRelaxSetRegSpaceRank(relax_data_l[0], 0, 0);
hypre_SMGRelaxSetRegSpaceRank(relax_data_l[0], 1, 1);
}
hypre_SMGRelaxSetMaxIter(relax_data_l[0], num_pre_relax);
hypre_SMGRelaxSetZeroGuess(relax_data_l[0], zero_guess);
hypre_SMGRelax(relax_data_l[0], A_l[0], b_l[0], x_l[0]);
zero_guess = 0;
/* compute fine grid residual (b - Ax) */
hypre_SMGResidual(residual_data_l[0], A_l[0], x_l[0], b_l[0], r_l[0]);
/* convergence check */
if (tol > 0.0)
{
r_dot_r = hypre_StructInnerProd(r_l[0], r_l[0]);
if (logging > 0)
{
norms[i] = sqrt(r_dot_r);
if (b_dot_b > 0)
rel_norms[i] = sqrt(r_dot_r/b_dot_b);
else
rel_norms[i] = 0.0;
}
/* always do at least 1 V-cycle */
if ((r_dot_r/b_dot_b < eps) && (i > 0))
{
if (rel_change)
{
if ((e_dot_e/x_dot_x) < eps)
break;
}
else
{
break;
}
}
}
if (num_levels > 1)
{
/* restrict fine grid residual */
hypre_SemiRestrict(restrict_data_l[0], R_l[0], r_l[0], b_l[1]);
#if DEBUG
if(hypre_StructStencilDim(hypre_StructMatrixStencil(A)) == 3)
{
hypre_sprintf(filename, "zout_xdown.%02d", 0);
hypre_StructVectorPrint(filename, x_l[0], 0);
hypre_sprintf(filename, "zout_rdown.%02d", 0);
hypre_StructVectorPrint(filename, r_l[0], 0);
hypre_sprintf(filename, "zout_b.%02d", 1);
hypre_StructVectorPrint(filename, b_l[1], 0);
}
#endif
for (l = 1; l <= (num_levels - 2); l++)
{
/* pre-relaxation */
hypre_SMGRelaxSetRegSpaceRank(relax_data_l[l], 0, 0);
hypre_SMGRelaxSetRegSpaceRank(relax_data_l[l], 1, 1);
hypre_SMGRelaxSetMaxIter(relax_data_l[l], num_pre_relax);
hypre_SMGRelaxSetZeroGuess(relax_data_l[l], 1);
hypre_SMGRelax(relax_data_l[l], A_l[l], b_l[l], x_l[l]);
/* compute residual (b - Ax) */
hypre_SMGResidual(residual_data_l[l],
A_l[l], x_l[l], b_l[l], r_l[l]);
/* restrict residual */
hypre_SemiRestrict(restrict_data_l[l], R_l[l], r_l[l], b_l[l+1]);
#if DEBUG
if(hypre_StructStencilDim(hypre_StructMatrixStencil(A)) == 3)
{
hypre_sprintf(filename, "zout_xdown.%02d", l);
hypre_StructVectorPrint(filename, x_l[l], 0);
hypre_sprintf(filename, "zout_rdown.%02d", l);
hypre_StructVectorPrint(filename, r_l[l], 0);
hypre_sprintf(filename, "zout_b.%02d", l+1);
hypre_StructVectorPrint(filename, b_l[l+1], 0);
}
#endif
}
/*--------------------------------------------------
* Bottom
*--------------------------------------------------*/
hypre_SMGRelaxSetZeroGuess(relax_data_l[l], 1);
hypre_SMGRelax(relax_data_l[l], A_l[l], b_l[l], x_l[l]);
#if DEBUG
if(hypre_StructStencilDim(hypre_StructMatrixStencil(A)) == 3)
{
hypre_sprintf(filename, "zout_xbottom.%02d", l);
hypre_StructVectorPrint(filename, x_l[l], 0);
}
#endif
/*--------------------------------------------------
* Up cycle
*--------------------------------------------------*/
for (l = (num_levels - 2); l >= 1; l--)
{
/* interpolate error and correct (x = x + Pe_c) */
hypre_SemiInterp(interp_data_l[l], PT_l[l], x_l[l+1], e_l[l]);
hypre_StructAxpy(1.0, e_l[l], x_l[l]);
#if DEBUG
if(hypre_StructStencilDim(hypre_StructMatrixStencil(A)) == 3)
{
hypre_sprintf(filename, "zout_eup.%02d", l);
hypre_StructVectorPrint(filename, e_l[l], 0);
hypre_sprintf(filename, "zout_xup.%02d", l);
hypre_StructVectorPrint(filename, x_l[l], 0);
}
#endif
/* post-relaxation */
hypre_SMGRelaxSetRegSpaceRank(relax_data_l[l], 0, 1);
hypre_SMGRelaxSetRegSpaceRank(relax_data_l[l], 1, 0);
hypre_SMGRelaxSetMaxIter(relax_data_l[l], num_post_relax);
hypre_SMGRelaxSetZeroGuess(relax_data_l[l], 0);
hypre_SMGRelax(relax_data_l[l], A_l[l], b_l[l], x_l[l]);
}
/* interpolate error and correct on fine grid (x = x + Pe_c) */
hypre_SemiInterp(interp_data_l[0], PT_l[0], x_l[1], e_l[0]);
hypre_SMGAxpy(1.0, e_l[0], x_l[0], base_index, base_stride);
#if DEBUG
if(hypre_StructStencilDim(hypre_StructMatrixStencil(A)) == 3)
{
hypre_sprintf(filename, "zout_eup.%02d", 0);
hypre_StructVectorPrint(filename, e_l[0], 0);
hypre_sprintf(filename, "zout_xup.%02d", 0);
hypre_StructVectorPrint(filename, x_l[0], 0);
}
#endif
}
/* part of convergence check */
if ((tol > 0.0) && (rel_change))
{
if (num_levels > 1)
{
e_dot_e = hypre_StructInnerProd(e_l[0], e_l[0]);
x_dot_x = hypre_StructInnerProd(x_l[0], x_l[0]);
}
else
{
e_dot_e = 0.0;
x_dot_x = 1.0;
}
}
/* fine grid post-relaxation */
if (num_levels > 1)
{
hypre_SMGRelaxSetRegSpaceRank(relax_data_l[0], 0, 1);
hypre_SMGRelaxSetRegSpaceRank(relax_data_l[0], 1, 0);
}
hypre_SMGRelaxSetMaxIter(relax_data_l[0], num_post_relax);
hypre_SMGRelaxSetZeroGuess(relax_data_l[0], 0);
hypre_SMGRelax(relax_data_l[0], A_l[0], b_l[0], x_l[0]);
(smg_data -> num_iterations) = (i + 1);
}
hypre_EndTiming(smg_data -> time_index);
return hypre_error_flag;
}
HYPRE_Int
hypre_SparseMSGSolve( void *smsg_vdata,
hypre_StructMatrix *A,
hypre_StructVector *b,
hypre_StructVector *x )
{
hypre_SparseMSGData *smsg_data = smsg_vdata;
HYPRE_Real tol = (smsg_data -> tol);
HYPRE_Int max_iter = (smsg_data -> max_iter);
HYPRE_Int rel_change = (smsg_data -> rel_change);
HYPRE_Int zero_guess = (smsg_data -> zero_guess);
HYPRE_Int jump = (smsg_data -> jump);
HYPRE_Int num_pre_relax = (smsg_data -> num_pre_relax);
HYPRE_Int num_post_relax = (smsg_data -> num_post_relax);
HYPRE_Int num_fine_relax = (smsg_data -> num_fine_relax);
HYPRE_Int *num_grids = (smsg_data -> num_grids);
HYPRE_Int num_all_grids = (smsg_data -> num_all_grids);
HYPRE_Int num_levels = (smsg_data -> num_levels);
hypre_StructMatrix **A_array = (smsg_data -> A_array);
hypre_StructMatrix **Px_array = (smsg_data -> Px_array);
hypre_StructMatrix **Py_array = (smsg_data -> Py_array);
hypre_StructMatrix **Pz_array = (smsg_data -> Pz_array);
hypre_StructMatrix **RTx_array = (smsg_data -> RTx_array);
hypre_StructMatrix **RTy_array = (smsg_data -> RTy_array);
hypre_StructMatrix **RTz_array = (smsg_data -> RTz_array);
hypre_StructVector **b_array = (smsg_data -> b_array);
hypre_StructVector **x_array = (smsg_data -> x_array);
hypre_StructVector **t_array = (smsg_data -> t_array);
hypre_StructVector **r_array = (smsg_data -> r_array);
hypre_StructVector **e_array = (smsg_data -> e_array);
hypre_StructVector **visitx_array = (smsg_data -> visitx_array);
hypre_StructVector **visity_array = (smsg_data -> visity_array);
hypre_StructVector **visitz_array = (smsg_data -> visitz_array);
HYPRE_Int *grid_on = (smsg_data -> grid_on);
void **relax_array = (smsg_data -> relax_array);
void **matvec_array = (smsg_data -> matvec_array);
void **restrictx_array = (smsg_data -> restrictx_array);
void **restricty_array = (smsg_data -> restricty_array);
void **restrictz_array = (smsg_data -> restrictz_array);
void **interpx_array = (smsg_data -> interpx_array);
void **interpy_array = (smsg_data -> interpy_array);
void **interpz_array = (smsg_data -> interpz_array);
HYPRE_Int logging = (smsg_data -> logging);
HYPRE_Real *norms = (smsg_data -> norms);
HYPRE_Real *rel_norms = (smsg_data -> rel_norms);
HYPRE_Int *restrict_count;
HYPRE_Real b_dot_b, r_dot_r, eps;
HYPRE_Real e_dot_e, x_dot_x;
HYPRE_Int i, l, lx, ly, lz;
HYPRE_Int lymin, lymax, lzmin, lzmax;
HYPRE_Int fi, ci;
HYPRE_Int ierr = 0;
#if DEBUG
char filename[255];
#endif
/*-----------------------------------------------------
* Initialize some things and deal with special cases
*-----------------------------------------------------*/
hypre_BeginTiming(smsg_data -> time_index);
hypre_StructMatrixDestroy(A_array[0]);
hypre_StructVectorDestroy(b_array[0]);
hypre_StructVectorDestroy(x_array[0]);
A_array[0] = hypre_StructMatrixRef(A);
b_array[0] = hypre_StructVectorRef(b);
x_array[0] = hypre_StructVectorRef(x);
(smsg_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(smsg_data -> time_index);
return ierr;
}
/* part of convergence check */
if (tol > 0.0)
{
/* eps = (tol^2) */
b_dot_b = hypre_StructInnerProd(b_array[0], b_array[0]);
eps = tol*tol;
/* if rhs is zero, return a zero solution */
if (b_dot_b == 0.0)
{
hypre_StructVectorSetConstantValues(x, 0.0);
if (logging > 0)
{
norms[0] = 0.0;
rel_norms[0] = 0.0;
}
hypre_EndTiming(smsg_data -> time_index);
return ierr;
}
}
restrict_count = hypre_TAlloc(HYPRE_Int, num_all_grids);
/*-----------------------------------------------------
* Do V-cycles:
* For each index l, "fine" = l, "coarse" = (l+1)
*-----------------------------------------------------*/
for (i = 0; i < max_iter; i++)
{
/*--------------------------------------------------
* Down cycle:
* Note that r = b = x through the jump region
*--------------------------------------------------*/
/* fine grid pre-relaxation */
hypre_PFMGRelaxSetPreRelax(relax_array[0]);
hypre_PFMGRelaxSetMaxIter(relax_array[0], num_fine_relax);
hypre_PFMGRelaxSetZeroGuess(relax_array[0], zero_guess);
hypre_PFMGRelax(relax_array[0], A_array[0], b_array[0], x_array[0]);
zero_guess = 0;
/* compute fine grid residual (b - Ax) */
hypre_StructCopy(b_array[0], r_array[0]);
hypre_StructMatvecCompute(matvec_array[0],
-1.0, A_array[0], x_array[0], 1.0, r_array[0]);
/* convergence check */
if (tol > 0.0)
{
r_dot_r = hypre_StructInnerProd(r_array[0], r_array[0]);
if (logging > 0)
{
norms[i] = sqrt(r_dot_r);
if (b_dot_b > 0)
rel_norms[i] = sqrt(r_dot_r/b_dot_b);
else
rel_norms[i] = 0.0;
}
/* RDF */
#if 0
hypre_printf("iter = %d, rel_norm = %e\n", i, rel_norms[i]);
#endif
/* always do at least 1 V-cycle */
if ((r_dot_r/b_dot_b < eps) && (i > 0))
{
if (rel_change)
{
if ((e_dot_e/x_dot_x) < eps)
break;
}
else
{
break;
}
}
}
if (num_levels > 1)
{
/* initialize restrict_count */
for (fi = 0; fi < num_all_grids; fi++)
{
restrict_count[fi] = 0;
}
for (l = 0; l <= (num_levels - 2); l++)
{
lzmin = hypre_max((l - num_grids[1] - num_grids[0] + 2), 0);
lzmax = hypre_min((l), (num_grids[2] - 1));
for (lz = lzmin; lz <= lzmax; lz++)
{
lymin = hypre_max((l - lz - num_grids[0] + 1), 0);
lymax = hypre_min((l - lz), (num_grids[1] - 1));
for (ly = lymin; ly <= lymax; ly++)
{
lx = l - lz - ly;
hypre_SparseMSGMapIndex(lx, ly, lz, num_grids, fi);
if (!grid_on[fi])
{
break;
}
if (restrict_count[fi] > 1)
{
hypre_StructScale((1.0/restrict_count[fi]), b_array[fi]);
}
if (l > jump)
{
/* pre-relaxation */
hypre_PFMGRelaxSetPreRelax(relax_array[fi]);
hypre_PFMGRelaxSetMaxIter(relax_array[fi], num_pre_relax);
hypre_PFMGRelaxSetZeroGuess(relax_array[fi], 1);
hypre_PFMGRelax(relax_array[fi], A_array[fi], b_array[fi],
x_array[fi]);
/* compute residual (b - Ax) */
hypre_StructCopy(b_array[fi], r_array[fi]);
hypre_StructMatvecCompute(matvec_array[fi],
-1.0, A_array[fi], x_array[fi],
1.0, r_array[fi]);
}
if ((lx+1) < num_grids[0])
{
/* restrict to ((lx+1), ly, lz) */
hypre_SparseMSGMapIndex((lx+1), ly, lz, num_grids, ci);
if (grid_on[ci])
{
if (restrict_count[ci])
{
hypre_SparseMSGRestrict(restrictx_array[fi],
RTx_array[lx], r_array[fi],
t_array[ci]);
hypre_StructAxpy(1.0, t_array[ci], b_array[ci]);
}
else
{
hypre_SparseMSGRestrict(restrictx_array[fi],
RTx_array[lx], r_array[fi],
b_array[ci]);
}
restrict_count[ci]++;
}
}
if ((ly+1) < num_grids[1])
{
/* restrict to (lx, (ly+1), lz) */
hypre_SparseMSGMapIndex(lx, (ly+1), lz, num_grids, ci);
if (grid_on[ci])
{
if (restrict_count[ci])
{
hypre_SparseMSGRestrict(restricty_array[fi],
RTy_array[ly], r_array[fi],
t_array[ci]);
hypre_StructAxpy(1.0, t_array[ci], b_array[ci]);
}
else
{
hypre_SparseMSGRestrict(restricty_array[fi],
RTy_array[ly], r_array[fi],
b_array[ci]);
}
restrict_count[ci]++;
}
}
if ((lz+1) < num_grids[2])
{
/* restrict to (lx, ly, (lz+1)) */
hypre_SparseMSGMapIndex(lx, ly, (lz+1), num_grids, ci);
if (grid_on[ci])
{
if (restrict_count[ci])
{
hypre_SparseMSGRestrict(restrictz_array[fi],
RTz_array[lz], r_array[fi],
t_array[ci]);
hypre_StructAxpy(1.0, t_array[ci], b_array[ci]);
}
else
{
hypre_SparseMSGRestrict(restrictz_array[fi],
RTz_array[lz], r_array[fi],
b_array[ci]);
}
restrict_count[ci]++;
}
}
#if DEBUG
hypre_sprintf(filename, "zoutSMSG_bdown.%d.%d.%d", lx, ly, lz);
hypre_StructVectorPrint(filename, b_array[fi], 0);
hypre_sprintf(filename, "zoutSMSG_xdown.%d.%d.%d", lx, ly, lz);
hypre_StructVectorPrint(filename, x_array[fi], 0);
hypre_sprintf(filename, "zoutSMSG_rdown.%d.%d.%d", lx, ly, lz);
hypre_StructVectorPrint(filename, r_array[fi], 0);
#endif
}
}
}
/*--------------------------------------------------
* Bottom
*--------------------------------------------------*/
fi = num_all_grids - 1;
if (restrict_count[fi] > 1)
{
hypre_StructScale((1.0/restrict_count[fi]), b_array[fi]);
}
hypre_PFMGRelaxSetZeroGuess(relax_array[fi], 1);
hypre_PFMGRelax(relax_array[fi], A_array[fi], b_array[fi],
x_array[fi]);
#if DEBUG
hypre_sprintf(filename, "zoutSMSG_bbottom.%d.%d.%d", lx, ly, lz);
hypre_StructVectorPrint(filename, b_array[fi], 0);
hypre_sprintf(filename, "zoutSMSG_xbottom.%d.%d.%d", lx, ly, lz);
hypre_StructVectorPrint(filename, x_array[fi], 0);
#endif
/*--------------------------------------------------
* Up cycle
* Note that r = b = x through the jump region
*--------------------------------------------------*/
for (l = (num_levels - 2); l >= 0; l--)
{
lzmin = hypre_max((l - num_grids[1] - num_grids[0] + 2), 0);
lzmax = hypre_min((l), (num_grids[2] - 1));
for (lz = lzmax; lz >= lzmin; lz--)
{
lymin = hypre_max((l - lz - num_grids[0] + 1), 0);
lymax = hypre_min((l - lz), (num_grids[1] - 1));
for (ly = lymax; ly >= lymin; ly--)
{
lx = l - lz - ly;
hypre_SparseMSGMapIndex(lx, ly, lz, num_grids, fi);
if (!grid_on[fi])
{
break;
}
if ((l >= 1) && (l <= jump))
{
hypre_StructVectorSetConstantValues(x_array[fi], 0.0);
}
if ((lx+1) < num_grids[0])
{
/* interpolate from ((lx+1), ly, lz) */
hypre_SparseMSGMapIndex((lx+1), ly, lz, num_grids, ci);
if (grid_on[ci])
{
hypre_SparseMSGInterp(interpx_array[fi],
Px_array[lx], x_array[ci],
e_array[fi]);
hypre_SparseMSGFilter(visitx_array[fi], e_array[fi],
lx, ly, lz, jump);
hypre_StructAxpy(1.0, e_array[fi], x_array[fi]);
}
}
if ((ly+1) < num_grids[1])
{
/* interpolate from (lx, (ly+1), lz) */
hypre_SparseMSGMapIndex(lx, (ly+1), lz, num_grids, ci);
if (grid_on[ci])
{
hypre_SparseMSGInterp(interpy_array[fi],
Py_array[ly], x_array[ci],
e_array[fi]);
hypre_SparseMSGFilter(visity_array[fi], e_array[fi],
lx, ly, lz, jump);
hypre_StructAxpy(1.0, e_array[fi], x_array[fi]);
}
}
if ((lz+1) < num_grids[2])
{
/* interpolate from (lx, ly, (lz+1)) */
hypre_SparseMSGMapIndex(lx, ly, (lz+1), num_grids, ci);
if (grid_on[ci])
{
hypre_SparseMSGInterp(interpz_array[fi],
Pz_array[lz], x_array[ci],
e_array[fi]);
hypre_SparseMSGFilter(visitz_array[fi], e_array[fi],
lx, ly, lz, jump);
hypre_StructAxpy(1.0, e_array[fi], x_array[fi]);
}
}
#if DEBUG
hypre_sprintf(filename, "zoutSMSG_xup.%d.%d.%d", lx, ly, lz);
hypre_StructVectorPrint(filename, x_array[fi], 0);
#endif
if (l > jump)
{
/* post-relaxation */
hypre_PFMGRelaxSetPostRelax(relax_array[fi]);
hypre_PFMGRelaxSetMaxIter(relax_array[fi],
num_post_relax);
hypre_PFMGRelaxSetZeroGuess(relax_array[fi], 0);
hypre_PFMGRelax(relax_array[fi], A_array[fi], b_array[fi],
x_array[fi]);
}
}
}
}
}
/* part of convergence check */
if ((tol > 0.0) && (rel_change))
{
if (num_levels > 1)
{
e_dot_e = hypre_StructInnerProd(e_array[0], e_array[0]);
x_dot_x = hypre_StructInnerProd(x_array[0], x_array[0]);
}
else
{
e_dot_e = 0.0;
x_dot_x = 1.0;
}
}
/* fine grid post-relaxation */
hypre_PFMGRelaxSetPostRelax(relax_array[0]);
hypre_PFMGRelaxSetMaxIter(relax_array[0], num_fine_relax);
hypre_PFMGRelaxSetZeroGuess(relax_array[0], 0);
hypre_PFMGRelax(relax_array[0], A_array[0], b_array[0], x_array[0]);
(smsg_data -> num_iterations) = (i + 1);
}
hypre_EndTiming(smsg_data -> time_index);
return ierr;
}
