PetscErrorCode TSSetFromOptions_Sundials_Nonlinear(TS ts)
{
TS_Sundials *cvode = (TS_Sundials*)ts->data;
PetscErrorCode ierr;
int indx;
PetscTruth flag;
PetscFunctionBegin;
ierr = PetscOptionsHead("SUNDIALS ODE solver options");CHKERRQ(ierr);
ierr = PetscOptionsEList("-ts_sundials_type","Scheme","TSSundialsSetType",TSSundialsLmmTypes,3,TSSundialsLmmTypes[cvode->cvode_type],&indx,&flag);CHKERRQ(ierr);
if (flag) {
ierr = TSSundialsSetType(ts,(TSSundialsLmmType)indx);CHKERRQ(ierr);
}
ierr = PetscOptionsEList("-ts_sundials_gramschmidt_type","Type of orthogonalization","TSSundialsSetGramSchmidtType",TSSundialsGramSchmidtTypes,3,TSSundialsGramSchmidtTypes[cvode->gtype],&indx,&flag);CHKERRQ(ierr);
if (flag) {
ierr = TSSundialsSetGramSchmidtType(ts,(TSSundialsGramSchmidtType)indx);CHKERRQ(ierr);
}
ierr = PetscOptionsReal("-ts_sundials_atol","Absolute tolerance for convergence","TSSundialsSetTolerance",cvode->abstol,&cvode->abstol,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsReal("-ts_sundials_rtol","Relative tolerance for convergence","TSSundialsSetTolerance",cvode->reltol,&cvode->reltol,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsReal("-ts_sundials_linear_tolerance","Convergence tolerance for linear solve","TSSundialsSetLinearTolerance",cvode->linear_tol,&cvode->linear_tol,&flag);CHKERRQ(ierr);
ierr = PetscOptionsInt("-ts_sundials_gmres_restart","Number of GMRES orthogonalization directions","TSSundialsSetGMRESRestart",cvode->restart,&cvode->restart,&flag);CHKERRQ(ierr);
ierr = PetscOptionsTruth("-ts_sundials_exact_final_time","Interpolate output to stop exactly at the final time","TSSundialsSetExactFinalTime",cvode->exact_final_time,&cvode->exact_final_time,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsTruth("-ts_sundials_monitor_steps","Monitor SUNDIALS internel steps","TSSundialsMonitorInternalSteps",cvode->monitorstep,&cvode->monitorstep,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsTail();CHKERRQ(ierr);
PetscFunctionReturn(0);
}
PetscErrorCode KSPSetFromOptions_BCGSL(KSP ksp)
{
KSP_BCGSL *bcgsl = (KSP_BCGSL *)ksp->data;
PetscErrorCode ierr;
PetscInt this_ell;
PetscReal delta;
PetscTruth flga = PETSC_FALSE, flg;
PetscFunctionBegin;
/* PetscOptionsBegin/End are called in KSPSetFromOptions. They
don't need to be called here.
*/
ierr = PetscOptionsHead("KSP BiCGStab(L) Options");
CHKERRQ(ierr);
/* Set number of search directions */
ierr = PetscOptionsInt("-ksp_bcgsl_ell","Number of Krylov search directions","KSPBCGSLSetEll",bcgsl->ell,&this_ell,&flg);
CHKERRQ(ierr);
if (flg) {
ierr = KSPBCGSLSetEll(ksp, this_ell);
CHKERRQ(ierr);
}
/* Set polynomial type */
ierr = PetscOptionsTruth("-ksp_bcgsl_cxpoly", "Polynomial part of BiCGStabL is MinRes + OR", "KSPBCGSLSetPol", flga,&flga,PETSC_NULL);
CHKERRQ(ierr);
if (flga) {
ierr = KSPBCGSLSetPol(ksp, PETSC_TRUE);
CHKERRQ(ierr);
} else {
flg = PETSC_FALSE;
ierr = PetscOptionsTruth("-ksp_bcgsl_mrpoly", "Polynomial part of BiCGStabL is MinRes", "KSPBCGSLSetPol", flg,&flg,PETSC_NULL);
CHKERRQ(ierr);
ierr = KSPBCGSLSetPol(ksp, PETSC_FALSE);
CHKERRQ(ierr);
}
/* Will computed residual be refreshed? */
ierr = PetscOptionsReal("-ksp_bcgsl_xres", "Threshold used to decide when to refresh computed residuals", "KSPBCGSLSetXRes", bcgsl->delta, &delta, &flg);
CHKERRQ(ierr);
if (flg) {
ierr = KSPBCGSLSetXRes(ksp, delta);
CHKERRQ(ierr);
}
ierr = PetscOptionsTail();
CHKERRQ(ierr);
PetscFunctionReturn(0);
}
PetscErrorCode MatPartitioningSetFromOptions_Party(MatPartitioning part)
{
PetscErrorCode ierr;
PetscTruth flag, b;
char value[15];
PetscReal r;
PetscFunctionBegin;
ierr = PetscOptionsHead("Set Party partitioning options");CHKERRQ(ierr);
ierr = PetscOptionsString("-mat_partitioning_party_global",
"Global method to use", "MatPartitioningPartySetGlobal", "gcf,gbf",
value, 15, &flag);CHKERRQ(ierr);
if (flag)
ierr = MatPartitioningPartySetGlobal(part, value);CHKERRQ(ierr);
ierr = PetscOptionsString("-mat_partitioning_party_local",
"Local method to use", "MatPartitioningPartySetLocal", "kl", value, 15,
&flag);CHKERRQ(ierr);
if (flag)
ierr = MatPartitioningPartySetLocal(part, value);CHKERRQ(ierr);
ierr = PetscOptionsReal("-mat_partitioning_party_coarse_level",
"Coarse level", "MatPartitioningPartySetCoarseLevel", 0, &r,
&flag);CHKERRQ(ierr);
if (flag)
ierr = MatPartitioningPartySetCoarseLevel(part, r);CHKERRQ(ierr);
ierr = PetscOptionsTruth("-mat_partitioning_party_match_optimization",
"Matching optimization on/off (boolean)",
"MatPartitioningPartySetMatchOptimization", PETSC_TRUE, &b, &flag);CHKERRQ(ierr);
if (flag)
ierr = MatPartitioningPartySetMatchOptimization(part, b);CHKERRQ(ierr);
ierr = PetscOptionsTruth("-mat_partitioning_party_bipart",
"Bipartitioning option on/off (boolean)",
"MatPartitioningPartySetBipart", PETSC_TRUE, &b, &flag);CHKERRQ(ierr);
if (flag)
ierr = MatPartitioningPartySetBipart(part, b);CHKERRQ(ierr);
ierr = PetscOptionsTail();CHKERRQ(ierr);
PetscFunctionReturn(0);
}
static PetscErrorCode PCSetFromOptions_KSP(PC pc)
{
PetscErrorCode ierr;
PetscTruth flg = PETSC_FALSE;
PetscFunctionBegin;
ierr = PetscOptionsHead("KSP preconditioner options");CHKERRQ(ierr);
ierr = PetscOptionsTruth("-pc_ksp_true","Use true matrix to define inner linear system, not preconditioner matrix","PCKSPSetUseTrue",flg,&flg,PETSC_NULL);CHKERRQ(ierr);
if (flg) {
ierr = PCKSPSetUseTrue(pc);CHKERRQ(ierr);
}
ierr = PetscOptionsTail();CHKERRQ(ierr);
PetscFunctionReturn(0);
}
static PetscErrorCode PCSetFromOptions_Eisenstat(PC pc)
{
PC_Eisenstat *eis = (PC_Eisenstat*)pc->data;
PetscErrorCode ierr;
PetscTruth flg = PETSC_FALSE;
PetscFunctionBegin;
ierr = PetscOptionsHead("Eisenstat SSOR options");CHKERRQ(ierr);
ierr = PetscOptionsReal("-pc_eisenstat_omega","Relaxation factor 0 < omega < 2","PCEisenstatSetOmega",eis->omega,&eis->omega,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsTruth("-pc_eisenstat_no_diagonal_scaling","Do not use standard diagonal scaling","PCEisenstatNoDiagonalScaling",flg,&flg,PETSC_NULL);CHKERRQ(ierr);
if (flg) {
ierr = PCEisenstatNoDiagonalScaling(pc);CHKERRQ(ierr);
}
ierr = PetscOptionsTail();CHKERRQ(ierr);
PetscFunctionReturn(0);
}
PetscErrorCode PETSC_DLLEXPORT PetscSSEIsEnabled(MPI_Comm comm,PetscTruth *lflag,PetscTruth *gflag) {
PetscErrorCode ierr;
PetscTruth disabled_option;
PetscFunctionBegin;
if (petsc_sse_local_is_untested && petsc_sse_global_is_untested) {
disabled_option = PETSC_FALSE;
ierr = PetscOptionsTruth("-disable_sse",
"Disable use of hand tuned Intel SSE implementations <true,false>.",
"PetscSSEIsEnabled",disabled_option,&disabled_option,PETSC_NULL);CHKERRQ(ierr);
if (disabled_option) {
petsc_sse_local_is_untested = PETSC_FALSE;
petsc_sse_enabled_local = PETSC_FALSE;
petsc_sse_global_is_untested = PETSC_FALSE;
petsc_sse_enabled_global = PETSC_FALSE;
}
if (petsc_sse_local_is_untested) {
ierr = PetscSSEHardwareTest(&petsc_sse_enabled_local);CHKERRQ(ierr);
if (petsc_sse_enabled_local) {
ierr = PetscSSEOSEnabledTest(&petsc_sse_enabled_local);CHKERRQ(ierr);
}
petsc_sse_local_is_untested = PETSC_FALSE;
}
if (gflag && petsc_sse_global_is_untested) {
ierr = MPI_Allreduce(&petsc_sse_enabled_local,&petsc_sse_enabled_global,1,MPI_INT,MPI_LAND,comm);CHKERRQ(ierr);
petsc_sse_global_is_untested = PETSC_FALSE;
}
}
if (lflag) {
*lflag = petsc_sse_enabled_local;
}
if (gflag) {
*gflag = petsc_sse_enabled_global;
}
PetscFunctionReturn(0);
}
int main(int argc,char **argv)
{
PetscErrorCode ierr;
SNES snes; /* nonlinear solver */
Vec Hu,r; /* solution, residual vectors */
Mat J; /* Jacobian matrix */
AppCtx user; /* user-defined work context */
PetscInt its, i, tmpxs, tmpxm; /* iteration count, index, etc. */
PetscReal tmp1, tmp2, tmp3, tmp4, tmp5,
errnorms[2], descaleNode[2];
PetscTruth eps_set = PETSC_FALSE, dump = PETSC_FALSE, exactinitial = PETSC_FALSE,
snes_mf_set, snes_fd_set;
MatFDColoring matfdcoloring = 0;
ISColoring iscoloring;
SNESConvergedReason reason; /* Check convergence */
PetscInitialize(&argc,&argv,(char *)0,help);
ierr = MPI_Comm_rank(PETSC_COMM_WORLD, &user.rank); CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,
"BODVARDSSON solves for thickness and velocity in 1D, steady ice stream\n"
" [run with -help for info and options]\n");CHKERRQ(ierr);
user.n = 3.0; /* Glen flow law exponent */
user.secpera = 31556926.0;
user.rho = 910.0; /* kg m^-3 */
user.rhow = 1028.0; /* kg m^-3 */
user.g = 9.81; /* m s^-2 */
/* ask Test N for its parameters, but only those we need to solve */
ierr = params_exactN(&(user.H0), &tmp1, &(user.xc), &tmp2, &tmp3, &tmp4, &tmp5,
&(user.Txc)); CHKERRQ(ierr);
/* regularize using strain rate of 1/xc per year */
user.epsilon = (1.0 / user.secpera) / user.xc;
/* tools for non-dimensionalizing to improve equation scaling */
user.scaleNode[0] = 1000.0; user.scaleNode[1] = 100.0 / user.secpera;
ierr = PetscOptionsTruth("-snes_mf","","",PETSC_FALSE,&snes_mf_set,NULL);CHKERRQ(ierr);
ierr = PetscOptionsTruth("-snes_fd","","",PETSC_FALSE,&snes_fd_set,NULL);CHKERRQ(ierr);
if (!snes_mf_set && !snes_fd_set) {
PetscPrintf(PETSC_COMM_WORLD,
"\n***ERROR: bodvardsson needs one or zero of '-snes_mf', '-snes_fd'***\n\n"
"USAGE FOLLOWS ...\n\n%s",help);
PetscEnd();
}
if (snes_fd_set) {
ierr = PetscPrintf(PETSC_COMM_WORLD,
" using approximate Jacobian; finite-differencing using coloring\n");
CHKERRQ(ierr);
} else if (snes_mf_set) {
ierr = PetscPrintf(PETSC_COMM_WORLD,
" matrix free; no preconditioner\n"); CHKERRQ(ierr);
} else {
ierr = PetscPrintf(PETSC_COMM_WORLD,
" true Jacobian\n"); CHKERRQ(ierr);
}
ierr = PetscOptionsBegin(PETSC_COMM_WORLD,NULL,
"bodvardsson program options",__FILE__);CHKERRQ(ierr);
{
ierr = PetscOptionsTruth("-bod_up_one","","",PETSC_FALSE,&user.upwind1,NULL);CHKERRQ(ierr);
ierr = PetscOptionsTruth("-bod_exact_init","","",PETSC_FALSE,&exactinitial,NULL);CHKERRQ(ierr);
ierr = PetscOptionsTruth("-bod_dump",
"dump out exact and approximate solution and residual, as asci","",
dump,&dump,NULL);CHKERRQ(ierr);
ierr = PetscOptionsReal("-bod_epsilon","regularization (a strain rate in units of 1/a)","",
user.epsilon * user.secpera,&user.epsilon,&eps_set);CHKERRQ(ierr);
if (eps_set) user.epsilon *= 1.0 / user.secpera;
}
ierr = PetscOptionsEnd();CHKERRQ(ierr);
/* Create machinery for parallel grid management (DA), nonlinear solver (SNES),
and Vecs for fields (solution, RHS). Note default Mx=46 grid points means
dx=10 km. Also degrees of freedom = 2 (thickness and velocity
at each point) and stencil radius = ghost width = 2 for 2nd-order upwinding. */
user.solnghostwidth = 2;
ierr = DACreate1d(PETSC_COMM_WORLD,DA_NONPERIODIC,-46,2,user.solnghostwidth,PETSC_NULL,&user.da);
CHKERRQ(ierr);
ierr = DASetUniformCoordinates(user.da,0.0,user.xc,
PETSC_NULL,PETSC_NULL,PETSC_NULL,PETSC_NULL);CHKERRQ(ierr);
ierr = DASetFieldName(user.da,0,"ice thickness [non-dimensional]"); CHKERRQ(ierr);
ierr = DASetFieldName(user.da,1,"ice velocity [non-dimensional]"); CHKERRQ(ierr);
ierr = DAGetInfo(user.da,PETSC_IGNORE,&user.Mx,PETSC_IGNORE,PETSC_IGNORE,PETSC_IGNORE,PETSC_IGNORE,
PETSC_IGNORE,PETSC_IGNORE,PETSC_IGNORE,PETSC_IGNORE,PETSC_IGNORE);
ierr = DAGetCorners(user.da,&user.xs,PETSC_NULL,PETSC_NULL,&user.xm,PETSC_NULL,PETSC_NULL);
CHKERRQ(ierr);
user.dx = user.xc / (PetscReal)(user.Mx-1);
/* another DA for scalar parameters, with same length */
ierr = DACreate1d(PETSC_COMM_WORLD,DA_NONPERIODIC,user.Mx,1,1,PETSC_NULL,&user.scalarda);CHKERRQ(ierr);
ierr = DASetUniformCoordinates(user.scalarda,0.0,user.xc,
PETSC_NULL,PETSC_NULL,PETSC_NULL,PETSC_NULL);CHKERRQ(ierr);
/* check that parallel layout of scalar DA is same as dof=2 DA */
ierr = DAGetCorners(user.scalarda,&tmpxs,PETSC_NULL,PETSC_NULL,&tmpxm,PETSC_NULL,PETSC_NULL);
CHKERRQ(ierr);
if ((tmpxs != user.xs) || (tmpxm != user.xm)) {
PetscPrintf(PETSC_COMM_SELF,
"\n***ERROR: rank %d gets different ownership range for the two DAs! ENDING ...***\n\n",
user.rank);
PetscEnd();
}
ierr = PetscPrintf(PETSC_COMM_WORLD,
" Mx = %D points, dx = %.3f m\n H0 = %.2f m, xc = %.2f km, Txc = %.5e Pa m\n",
user.Mx, user.dx, user.H0, user.xc/1000.0, user.Txc);CHKERRQ(ierr);
/* Extract/allocate global vectors from DAs and duplicate for remaining same types */
ierr = DACreateGlobalVector(user.da,&Hu);CHKERRQ(ierr);
ierr = VecSetBlockSize(Hu,2);CHKERRQ(ierr);
ierr = VecDuplicate(Hu,&r);CHKERRQ(ierr); /* inherits block size */
ierr = VecDuplicate(Hu,&user.Huexact);CHKERRQ(ierr); /* ditto */
ierr = DACreateGlobalVector(user.scalarda,&user.M);CHKERRQ(ierr);
ierr = VecDuplicate(user.M,&user.Bstag);CHKERRQ(ierr);
ierr = VecDuplicate(user.M,&user.beta);CHKERRQ(ierr);
ierr = DASetLocalFunction(user.da,(DALocalFunction1)scshell);CHKERRQ(ierr);
ierr = DASetLocalJacobian(user.da,(DALocalFunction1)BodJacobianMatrixLocal);CHKERRQ(ierr);
ierr = SNESCreate(PETSC_COMM_WORLD,&snes);CHKERRQ(ierr);
ierr = SNESSetFunction(snes,r,SNESDAFormFunction,&user);CHKERRQ(ierr);
/* setting up a matrix is only actually needed for -snes_fd case */
ierr = DAGetMatrix(user.da,MATAIJ,&J);CHKERRQ(ierr);
if (snes_fd_set) {
/* tools needed so DA can use sparse matrix for its F.D. Jacobian approx */
ierr = DAGetColoring(user.da,IS_COLORING_GLOBAL,MATAIJ,&iscoloring);CHKERRQ(ierr);
ierr = MatFDColoringCreate(J,iscoloring,&matfdcoloring);CHKERRQ(ierr);
ierr = ISColoringDestroy(iscoloring);CHKERRQ(ierr);
ierr = MatFDColoringSetFunction(matfdcoloring,
(PetscErrorCode (*)(void))SNESDAFormFunction,&user);CHKERRQ(ierr);
ierr = MatFDColoringSetFromOptions(matfdcoloring);CHKERRQ(ierr);
ierr = SNESSetJacobian(snes,J,J,SNESDefaultComputeJacobianColor,matfdcoloring);CHKERRQ(ierr);
} else {
ierr = SNESSetJacobian(snes,J,J,SNESDAComputeJacobian,&user);CHKERRQ(ierr);
}
ierr = SNESSetFromOptions(snes);CHKERRQ(ierr);
/* the the Bodvardsson (1955) exact solution allows setting M(x), B(x), beta(x), T(xc) */
ierr = FillDistributedParams(&user);CHKERRQ(ierr);
/* the exact thickness and exact ice velocity (user.uHexact) are known from Bodvardsson (1955) */
ierr = FillExactSoln(&user); CHKERRQ(ierr);
if (exactinitial) {
ierr = PetscPrintf(PETSC_COMM_WORLD," using exact solution as initial guess\n");
CHKERRQ(ierr);
/* the initial guess is the exact continuum solution */
ierr = VecCopy(user.Huexact,Hu); CHKERRQ(ierr);
} else {
ierr = FillInitial(&user, &Hu); CHKERRQ(ierr);
}
/************ SOLVE NONLINEAR SYSTEM ************/
/* recall that RHS r is used internally by KSP, and is set by the SNES */
for (i = 0; i < 2; i++) descaleNode[i] = 1.0 / user.scaleNode[i];
ierr = VecStrideScaleAll(Hu,descaleNode); CHKERRQ(ierr); /* de-dimensionalize initial guess */
ierr = SNESSolve(snes,PETSC_NULL,Hu);CHKERRQ(ierr);
ierr = VecStrideScaleAll(Hu,user.scaleNode); CHKERRQ(ierr); /* put back in "real" scale */
ierr = SNESGetIterationNumber(snes,&its);CHKERRQ(ierr);
ierr = SNESGetConvergedReason(snes,&reason);CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,
" %s Number of Newton iterations = %D\n",
SNESConvergedReasons[reason],its);CHKERRQ(ierr);
if (dump) {
ierr = PetscPrintf(PETSC_COMM_WORLD,
" viewing combined result Hu\n");CHKERRQ(ierr);
ierr = VecView(Hu,PETSC_VIEWER_STDOUT_WORLD); CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,
" viewing combined exact result Huexact\n");CHKERRQ(ierr);
ierr = VecView(user.Huexact,PETSC_VIEWER_STDOUT_WORLD); CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,
" viewing final combined residual at Hu\n");CHKERRQ(ierr);
ierr = VecView(r,PETSC_VIEWER_STDOUT_WORLD); CHKERRQ(ierr);
}
/* evaluate error relative to exact solution */
ierr = VecAXPY(Hu,-1.0,user.Huexact); CHKERRQ(ierr); /* Hu = - Huexact + Hu */
ierr = VecStrideNormAll(Hu,NORM_INFINITY,errnorms); CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,
"(dx,errHinf,erruinf) %.3f %.4e %.4e\n",
user.dx,errnorms[0],errnorms[1]*user.secpera);CHKERRQ(ierr);
ierr = VecDestroy(Hu);CHKERRQ(ierr);
ierr = VecDestroy(r);CHKERRQ(ierr);
ierr = VecDestroy(user.Huexact);CHKERRQ(ierr);
ierr = VecDestroy(user.M);CHKERRQ(ierr);
ierr = VecDestroy(user.Bstag);CHKERRQ(ierr);
ierr = VecDestroy(user.beta);CHKERRQ(ierr);
ierr = MatDestroy(J); CHKERRQ(ierr);
ierr = SNESDestroy(snes);CHKERRQ(ierr);
ierr = DADestroy(user.da);CHKERRQ(ierr);
ierr = DADestroy(user.scalarda);CHKERRQ(ierr);
ierr = PetscFinalize();CHKERRQ(ierr);
return 0;
}
int main(int argc,char *argv[])
{
char mat_type[256] = "aij"; /* default matrix type */
PetscErrorCode ierr;
MPI_Comm comm;
PetscMPIInt rank,size;
Sliced slice;
PetscInt i,bs=1,N=5,n,m,rstart,ghosts[2],*d_nnz,*o_nnz,dfill[4]={1,0,0,1},ofill[4]={1,1,1,1};
PetscReal alpha=1,K=1,rho0=1,u0=0,sigma=0.2;
PetscTruth useblock=PETSC_TRUE;
PetscScalar *xx;
Mat A;
Vec x,b,lf;
ierr = PetscInitialize(&argc,&argv,0,help);CHKERRQ(ierr);
comm = PETSC_COMM_WORLD;
ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr);
ierr = MPI_Comm_rank(comm,&rank);CHKERRQ(ierr);
ierr = PetscOptionsBegin(comm,0,"Options for Sliced test",0);CHKERRQ(ierr);
{
ierr = PetscOptionsInt("-n","Global number of nodes","",N,&N,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsInt("-bs","Block size (1 or 2)","",bs,&bs,PETSC_NULL);CHKERRQ(ierr);
if (bs != 1) {
if (bs != 2) SETERRQ(1,"Block size must be 1 or 2");
ierr = PetscOptionsReal("-alpha","Inverse time step for wave operator","",alpha,&alpha,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsReal("-K","Bulk modulus of compressibility","",K,&K,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsReal("-rho0","Reference density","",rho0,&rho0,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsReal("-u0","Reference velocity","",u0,&u0,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsReal("-sigma","Width of Gaussian density perturbation","",sigma,&sigma,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsTruth("-block","Use block matrix assembly","",useblock,&useblock,PETSC_NULL);CHKERRQ(ierr);
}
ierr = PetscOptionsString("-sliced_mat_type","Matrix type to use (aij or baij)","",mat_type,mat_type,sizeof mat_type,PETSC_NULL);CHKERRQ(ierr);
}
ierr = PetscOptionsEnd();CHKERRQ(ierr);
/* Split ownership, set up periodic grid in 1D */
n = PETSC_DECIDE;
ierr = PetscSplitOwnership(comm,&n,&N);CHKERRQ(ierr);
rstart = 0;
ierr = MPI_Scan(&n,&rstart,1,MPIU_INT,MPI_SUM,comm);CHKERRQ(ierr);
rstart -= n;
ghosts[0] = (N+rstart-1)%N;
ghosts[1] = (rstart+n)%N;
ierr = SlicedCreate(comm,&slice);CHKERRQ(ierr);
ierr = SlicedSetGhosts(slice,bs,n,2,ghosts);CHKERRQ(ierr);
ierr = PetscMalloc2(n,PetscInt,&d_nnz,n,PetscInt,&o_nnz);CHKERRQ(ierr);
for (i=0; i<n; i++) {
if (size > 1 && (i==0 || i==n-1)) {
d_nnz[i] = 2;
o_nnz[i] = 1;
} else {
d_nnz[i] = 3;
o_nnz[i] = 0;
}
}
ierr = SlicedSetPreallocation(slice,0,d_nnz,0,o_nnz);CHKERRQ(ierr); /* Currently does not copy X_nnz so we can't free them until after SlicedGetMatrix */
if (!useblock) {ierr = SlicedSetBlockFills(slice,dfill,ofill);CHKERRQ(ierr);} /* Irrelevant for baij formats */
ierr = SlicedGetMatrix(slice,mat_type,&A);CHKERRQ(ierr);
ierr = PetscFree2(d_nnz,o_nnz);CHKERRQ(ierr);
ierr = MatSetOption(A,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_TRUE);CHKERRQ(ierr);
ierr = SlicedCreateGlobalVector(slice,&x);CHKERRQ(ierr);
ierr = VecDuplicate(x,&b);CHKERRQ(ierr);
ierr = VecGhostGetLocalForm(x,&lf);CHKERRQ(ierr);
ierr = VecGetSize(lf,&m);CHKERRQ(ierr);
if (m != (n+2)*bs) SETERRQ2(1,"size of local form %D, expected %D",m,(n+2)*bs);
ierr = VecGetArray(lf,&xx);CHKERRQ(ierr);
for (i=0; i<n; i++) {
PetscInt row[2],col[9],im,ip;
PetscScalar v[12];
const PetscReal xref = 2.0*(rstart+i)/N - 1; /* [-1,1] */
const PetscReal h = 1.0/N; /* grid spacing */
im = (i==0) ? n : i-1;
ip = (i==n-1) ? n+1 : i+1;
switch (bs) {
case 1: /* Laplacian with periodic boundaries */
col[0] = im; col[1] = i; col[2] = ip;
v[0] = -h; v[1] = 2*h; v[2] = -h;
ierr = MatSetValuesLocal(A,1,&i,3,col,v,INSERT_VALUES);CHKERRQ(ierr);
xx[i] = sin(xref*PETSC_PI);
break;
case 2: /* Linear acoustic wave operator in variables [rho, u], central differences, periodic, timestep 1/alpha */
v[0] = -0.5*u0; v[1] = -0.5*K; v[2] = alpha; v[3] = 0; v[4] = 0.5*u0; v[5] = 0.5*K;
v[6] = -0.5/rho0; v[7] = -0.5*u0; v[8] = 0; v[9] = alpha; v[10] = 0.5/rho0; v[11] = 0.5*u0;
if (useblock) {
row[0] = i; col[0] = im; col[1] = i; col[2] = ip;
ierr = MatSetValuesBlockedLocal(A,1,row,3,col,v,INSERT_VALUES);CHKERRQ(ierr);
} else {
row[0] = 2*i; row[1] = 2*i+1;
col[0] = 2*im; col[1] = 2*im+1; col[2] = 2*i; col[3] = 2*ip; col[4] = 2*ip+1;
v[3] = v[4]; v[4] = v[5]; /* pack values in first row */
ierr = MatSetValuesLocal(A,1,row,5,col,v,INSERT_VALUES);CHKERRQ(ierr);
col[2] = 2*i+1;
v[8] = v[9]; v[9] = v[10]; v[10] = v[11]; /* pack values in second row */
ierr = MatSetValuesLocal(A,1,row+1,5,col,v+6,INSERT_VALUES);CHKERRQ(ierr);
}
/* Set current state (gaussian density perturbation) */
xx[2*i] = 0.2*exp(-PetscSqr(xref)/(2*PetscSqr(sigma)));
xx[2*i+1] = 0;
break;
default: SETERRQ1(1,"not implemented for block size %D",bs);
}
}
ierr = VecRestoreArray(lf,&xx);CHKERRQ(ierr);
ierr = VecGhostRestoreLocalForm(x,&lf);CHKERRQ(ierr);
ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatMult(A,x,b);CHKERRQ(ierr);
ierr = MatView(A,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
ierr = VecView(x,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
ierr = VecView(b,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
/* Update the ghosted values, view the result on rank 0. */
ierr = VecGhostUpdateBegin(b,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecGhostUpdateEnd(b,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
if (!rank) {
ierr = VecGhostGetLocalForm(b,&lf);CHKERRQ(ierr);
ierr = PetscViewerASCIIPrintf(PETSC_VIEWER_STDOUT_SELF,"Local form of b on rank 0, last two nodes are ghost nodes\n");CHKERRQ(ierr);
ierr = VecView(lf,PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);
ierr = VecGhostRestoreLocalForm(b,&lf);CHKERRQ(ierr);
}
ierr = SlicedDestroy(slice);CHKERRQ(ierr);
ierr = VecDestroy(x);CHKERRQ(ierr);
ierr = VecDestroy(b);CHKERRQ(ierr);
ierr = MatDestroy(A);CHKERRQ(ierr);
ierr = PetscFinalize();CHKERRQ(ierr);
return 0;
}
PetscErrorCode SetSpoolesOptions(Mat A, Spooles_options *options)
{
PetscErrorCode ierr;
int indx;
const char *ordertype[]={"BestOfNDandMS","MMD","MS","ND"};
PetscTruth flg;
PetscFunctionBegin;
/* set default input parameters */
#if defined(PETSC_USE_COMPLEX)
options->typeflag = SPOOLES_COMPLEX;
#else
options->typeflag = SPOOLES_REAL;
#endif
options->msglvl = 0;
options->msgFile = 0;
options->tau = 100.;
options->seed = 10101;
options->ordering = 1; /* MMD */
options->maxdomainsize = 500;
options->maxzeros = 1000;
options->maxsize = 96;
options->FrontMtxInfo = PETSC_FALSE;
if ( options->symflag == SPOOLES_SYMMETRIC ) { /* || SPOOLES_HERMITIAN */
options->patchAndGoFlag = 0; /* no patch */
options->storeids = 1;
options->storevalues = 1;
options->toosmall = 1.e-9;
options->fudge = 1.e-9;
}
/* get runtime input parameters */
ierr = PetscOptionsBegin(((PetscObject)A)->comm,((PetscObject)A)->prefix,"Spooles Options","Mat");CHKERRQ(ierr);
ierr = PetscOptionsReal("-mat_spooles_tau","tau (used for pivoting; \n\
all entries in L and U have magnitude no more than tau)","None",
options->tau,&options->tau,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsInt("-mat_spooles_seed","random number seed, used for ordering","None",
options->seed,&options->seed,PETSC_NULL);CHKERRQ(ierr);
if (PetscLogPrintInfo) options->msglvl = 1;
ierr = PetscOptionsInt("-mat_spooles_msglvl","msglvl","None",
options->msglvl,&options->msglvl,0);CHKERRQ(ierr);
if (options->msglvl > 0) {
options->msgFile = fopen("spooles.msgFile", "a");
PetscPrintf(PETSC_COMM_SELF,"\n Spooles' output is written into the file 'spooles.msgFile' \n\n");
}
ierr = PetscOptionsEList("-mat_spooles_ordering","ordering type","None",ordertype,4,ordertype[1],&indx,&flg);CHKERRQ(ierr);
if (flg) options->ordering = indx;
ierr = PetscOptionsInt("-mat_spooles_maxdomainsize","maxdomainsize","None",\
options->maxdomainsize,&options->maxdomainsize,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsInt("-mat_spooles_maxzeros ","maxzeros","None",\
options->maxzeros,&options->maxzeros,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsInt("-mat_spooles_maxsize","maxsize","None",\
options->maxsize,&options->maxsize,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsTruth("-mat_spooles_FrontMtxInfo","FrontMtxInfo","None",PETSC_FALSE,&flg,0);CHKERRQ(ierr);
if (flg) options->FrontMtxInfo = PETSC_TRUE;
if ( options->symflag == SPOOLES_SYMMETRIC ) {
ierr = PetscOptionsInt("-mat_spooles_symmetryflag","matrix type","None", \
options->symflag,&options->symflag,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsInt("-mat_spooles_patchAndGoFlag","patchAndGoFlag","None", \
options->patchAndGoFlag,&options->patchAndGoFlag,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsReal("-mat_spooles_fudge","fudge","None", \
options->fudge,&options->fudge,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsReal("-mat_spooles_toosmall","toosmall","None", \
options->toosmall,&options->toosmall,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsInt("-mat_spooles_storeids","storeids","None", \
options->storeids,&options->storeids,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsInt("-mat_spooles_storevalues","storevalues","None", \
options->storevalues,&options->storevalues,PETSC_NULL);CHKERRQ(ierr);
}
PetscOptionsEnd();
PetscFunctionReturn(0);
}
PetscErrorCode MatPartitioningSetFromOptions_Scotch(MatPartitioning part)
{
PetscErrorCode ierr;
PetscTruth flag;
char name[PETSC_MAX_PATH_LEN];
int i;
PetscReal r;
const char *global[] = { "greedy", "gps", "gr_gps" };
const char *local[] = { "kernighan-lin", "none" };
PetscFunctionBegin;
ierr = PetscOptionsHead("Set Scotch partitioning options");CHKERRQ(ierr);
ierr = PetscOptionsEList("-mat_partitioning_scotch_global",
"Global method to use", "MatPartitioningScotchSetGlobal", global, 3,
global[0], &i, &flag);CHKERRQ(ierr);
if (flag) {
ierr = MatPartitioningScotchSetGlobal(part, (MPScotchGlobalType)i);CHKERRQ(ierr);
}
ierr = PetscOptionsEList("-mat_partitioning_scotch_local",
"Local method to use", "MatPartitioningScotchSetLocal", local, 2,
local[0], &i, &flag);CHKERRQ(ierr);
if (flag) {
ierr = MatPartitioningScotchSetLocal(part, (MPScotchLocalType)i);CHKERRQ(ierr);
}
flag = PETSC_FALSE;
ierr = PetscOptionsTruth("-mat_partitioning_scotch_mapping", "Use mapping","MatPartitioningScotchSetMapping", flag,&flag,PETSC_NULL);CHKERRQ(ierr);
if (flag) {
ierr = MatPartitioningScotchSetMapping(part);CHKERRQ(ierr);
}
ierr = PetscOptionsString("-mat_partitioning_scotch_arch",
"architecture file in scotch format", "MatPartitioningScotchSetArch",
"archgraph.src", name, PETSC_MAX_PATH_LEN, &flag);CHKERRQ(ierr);
if (flag)
ierr = MatPartitioningScotchSetArch(part, name);CHKERRQ(ierr);
ierr = PetscOptionsString("-mat_partitioning_scotch_hosts",
"host list filename", "MatPartitioningScotchSetHostList",
"host_list", name, PETSC_MAX_PATH_LEN, &flag);CHKERRQ(ierr);
if (flag)
ierr = MatPartitioningScotchSetHostList(part, name);CHKERRQ(ierr);
ierr = PetscOptionsReal("-mat_partitioning_scotch_coarse_level",
"coarse level", "MatPartitioningScotchSetCoarseLevel", 0, &r,
&flag);CHKERRQ(ierr);
if (flag)
ierr = MatPartitioningScotchSetCoarseLevel(part, r);CHKERRQ(ierr);
flag = PETSC_FALSE;
ierr = PetscOptionsTruth("-mat_partitioning_scotch_mul", "Use coarse level","MatPartitioningScotchSetMultilevel", flag,&flag,PETSC_NULL);CHKERRQ(ierr);
if (flag) {
ierr = MatPartitioningScotchSetMultilevel(part);CHKERRQ(ierr);
}
ierr = PetscOptionsString("-mat_partitioning_scotch_strategy",
"Scotch strategy string",
"MatPartitioningScotchSetStrategy", "", name, PETSC_MAX_PATH_LEN,
&flag);CHKERRQ(ierr);
if (flag)
ierr = MatPartitioningScotchSetStrategy(part, name);CHKERRQ(ierr);
ierr = PetscOptionsTail();CHKERRQ(ierr);
PetscFunctionReturn(0);
}
