static PetscErrorCode KSPFGMRESBuildSoln(PetscScalar *nrs,Vec vguess,Vec vdest,KSP ksp,PetscInt it)
{
PetscScalar tt;
PetscErrorCode ierr;
PetscInt ii,k,j;
KSP_FGMRES *fgmres = (KSP_FGMRES*)(ksp->data);
PetscFunctionBegin;
/* Solve for solution vector that minimizes the residual */
/* If it is < 0, no fgmres steps have been performed */
if (it < 0) {
ierr = VecCopy(vguess,vdest);CHKERRQ(ierr); /* VecCopy() is smart, exists immediately if vguess == vdest */
PetscFunctionReturn(0);
}
/* so fgmres steps HAVE been performed */
/* solve the upper triangular system - RS is the right side and HH is
the upper triangular matrix - put soln in nrs */
if (*HH(it,it) != 0.0) {
nrs[it] = *RS(it) / *HH(it,it);
} else {
nrs[it] = 0.0;
}
for (ii=1; ii<=it; ii++) {
k = it - ii;
tt = *RS(k);
for (j=k+1; j<=it; j++) tt = tt - *HH(k,j) * nrs[j];
nrs[k] = tt / *HH(k,k);
}
/* Accumulate the correction to the soln of the preconditioned prob. in
VEC_TEMP - note that we use the preconditioned vectors */
ierr = VecSet(VEC_TEMP,0.0);CHKERRQ(ierr); /* set VEC_TEMP components to 0 */
ierr = VecMAXPY(VEC_TEMP,it+1,nrs,&PREVEC(0));CHKERRQ(ierr);
/* put updated solution into vdest.*/
if (vdest != vguess) {
ierr = VecCopy(VEC_TEMP,vdest);CHKERRQ(ierr);
ierr = VecAXPY(vdest,1.0,vguess);CHKERRQ(ierr);
} else { /* replace guess with solution */
ierr = VecAXPY(vdest,1.0,VEC_TEMP);CHKERRQ(ierr);
}
PetscFunctionReturn(0);
}
/*
KSPPIPEFGMRESBuildSoln - create the solution from the starting vector and the
current iterates.
Input parameters:
nrs - work area of size it + 1.
vguess - index of initial guess
vdest - index of result. Note that vguess may == vdest (replace
guess with the solution).
it - HH upper triangular part is a block of size (it+1) x (it+1)
This is an internal routine that knows about the PIPEFGMRES internals.
*/
static PetscErrorCode KSPPIPEFGMRESBuildSoln(PetscScalar *nrs,Vec vguess,Vec vdest,KSP ksp,PetscInt it)
{
PetscScalar tt;
PetscErrorCode ierr;
PetscInt k,j;
KSP_PIPEFGMRES *pipefgmres = (KSP_PIPEFGMRES*)(ksp->data);
PetscFunctionBegin;
/* Solve for solution vector that minimizes the residual */
if (it < 0) { /* no pipefgmres steps have been performed */
ierr = VecCopy(vguess,vdest);CHKERRQ(ierr); /* VecCopy() is smart, exits immediately if vguess == vdest */
PetscFunctionReturn(0);
}
/* solve the upper triangular system - RS is the right side and HH is
the upper triangular matrix - put soln in nrs */
if (*HH(it,it) != 0.0) nrs[it] = *RS(it) / *HH(it,it);
else nrs[it] = 0.0;
for (k=it-1; k>=0; k--) {
tt = *RS(k);
for (j=k+1; j<=it; j++) tt -= *HH(k,j) * nrs[j];
nrs[k] = tt / *HH(k,k);
}
/* Accumulate the correction to the solution of the preconditioned problem in VEC_TEMP */
ierr = VecZeroEntries(VEC_TEMP);CHKERRQ(ierr);
ierr = VecMAXPY(VEC_TEMP,it+1,nrs,&PREVEC(0));CHKERRQ(ierr);
/* add solution to previous solution */
if (vdest == vguess) {
ierr = VecAXPY(vdest,1.0,VEC_TEMP);CHKERRQ(ierr);
} else {
ierr = VecWAXPY(vdest,1.0,VEC_TEMP,vguess);CHKERRQ(ierr);
}
PetscFunctionReturn(0);
}
PetscErrorCode KSPFGMRESCycle(PetscInt *itcount,KSP ksp)
{
KSP_FGMRES *fgmres = (KSP_FGMRES*)(ksp->data);
PetscReal res_norm;
PetscReal hapbnd,tt;
PetscBool hapend = PETSC_FALSE; /* indicates happy breakdown ending */
PetscErrorCode ierr;
PetscInt loc_it; /* local count of # of dir. in Krylov space */
PetscInt max_k = fgmres->max_k; /* max # of directions Krylov space */
Mat Amat,Pmat;
MatStructure pflag;
PetscFunctionBegin;
/* Number of pseudo iterations since last restart is the number
of prestart directions */
loc_it = 0;
/* note: (fgmres->it) is always set one less than (loc_it) It is used in
KSPBUILDSolution_FGMRES, where it is passed to KSPFGMRESBuildSoln.
Note that when KSPFGMRESBuildSoln is called from this function,
(loc_it -1) is passed, so the two are equivalent */
fgmres->it = (loc_it - 1);
/* initial residual is in VEC_VV(0) - compute its norm*/
ierr = VecNorm(VEC_VV(0),NORM_2,&res_norm);CHKERRQ(ierr);
/* first entry in right-hand-side of hessenberg system is just
the initial residual norm */
*RS(0) = res_norm;
ksp->rnorm = res_norm;
ierr = KSPLogResidualHistory(ksp,res_norm);CHKERRQ(ierr);
ierr = KSPMonitor(ksp,ksp->its,res_norm);CHKERRQ(ierr);
/* check for the convergence - maybe the current guess is good enough */
ierr = (*ksp->converged)(ksp,ksp->its,res_norm,&ksp->reason,ksp->cnvP);CHKERRQ(ierr);
if (ksp->reason) {
if (itcount) *itcount = 0;
PetscFunctionReturn(0);
}
/* scale VEC_VV (the initial residual) */
ierr = VecScale(VEC_VV(0),1.0/res_norm);CHKERRQ(ierr);
/* MAIN ITERATION LOOP BEGINNING*/
/* keep iterating until we have converged OR generated the max number
of directions OR reached the max number of iterations for the method */
while (!ksp->reason && loc_it < max_k && ksp->its < ksp->max_it) {
if (loc_it) {
ierr = KSPLogResidualHistory(ksp,res_norm);CHKERRQ(ierr);
ierr = KSPMonitor(ksp,ksp->its,res_norm);CHKERRQ(ierr);
}
fgmres->it = (loc_it - 1);
/* see if more space is needed for work vectors */
if (fgmres->vv_allocated <= loc_it + VEC_OFFSET + 1) {
ierr = KSPFGMRESGetNewVectors(ksp,loc_it+1);CHKERRQ(ierr);
/* (loc_it+1) is passed in as number of the first vector that should
be allocated */
}
/* CHANGE THE PRECONDITIONER? */
/* ModifyPC is the callback function that can be used to
change the PC or its attributes before its applied */
(*fgmres->modifypc)(ksp,ksp->its,loc_it,res_norm,fgmres->modifyctx);
/* apply PRECONDITIONER to direction vector and store with
preconditioned vectors in prevec */
ierr = KSP_PCApply(ksp,VEC_VV(loc_it),PREVEC(loc_it));CHKERRQ(ierr);
ierr = PCGetOperators(ksp->pc,&Amat,&Pmat,&pflag);CHKERRQ(ierr);
/* Multiply preconditioned vector by operator - put in VEC_VV(loc_it+1) */
ierr = MatMult(Amat,PREVEC(loc_it),VEC_VV(1+loc_it));CHKERRQ(ierr);
/* update hessenberg matrix and do Gram-Schmidt - new direction is in
VEC_VV(1+loc_it)*/
ierr = (*fgmres->orthog)(ksp,loc_it);CHKERRQ(ierr);
/* new entry in hessenburg is the 2-norm of our new direction */
ierr = VecNorm(VEC_VV(loc_it+1),NORM_2,&tt);CHKERRQ(ierr);
*HH(loc_it+1,loc_it) = tt;
*HES(loc_it+1,loc_it) = tt;
/* Happy Breakdown Check */
hapbnd = PetscAbsScalar((tt) / *RS(loc_it));
/* RS(loc_it) contains the res_norm from the last iteration */
hapbnd = PetscMin(fgmres->haptol,hapbnd);
if (tt > hapbnd) {
/* scale new direction by its norm */
ierr = VecScale(VEC_VV(loc_it+1),1.0/tt);CHKERRQ(ierr);
} else {
/* This happens when the solution is exactly reached. */
/* So there is no new direction... */
ierr = VecSet(VEC_TEMP,0.0);CHKERRQ(ierr); /* set VEC_TEMP to 0 */
hapend = PETSC_TRUE;
}
/* note that for FGMRES we could get HES(loc_it+1, loc_it) = 0 and the
current solution would not be exact if HES was singular. Note that
HH non-singular implies that HES is no singular, and HES is guaranteed
to be nonsingular when PREVECS are linearly independent and A is
nonsingular (in GMRES, the nonsingularity of A implies the nonsingularity
of HES). So we should really add a check to verify that HES is nonsingular.*/
/* Now apply rotations to new col of hessenberg (and right side of system),
calculate new rotation, and get new residual norm at the same time*/
ierr = KSPFGMRESUpdateHessenberg(ksp,loc_it,hapend,&res_norm);CHKERRQ(ierr);
if (ksp->reason) break;
loc_it++;
fgmres->it = (loc_it-1); /* Add this here in case it has converged */
ierr = PetscObjectSAWsTakeAccess((PetscObject)ksp);CHKERRQ(ierr);
ksp->its++;
ksp->rnorm = res_norm;
ierr = PetscObjectSAWsGrantAccess((PetscObject)ksp);CHKERRQ(ierr);
ierr = (*ksp->converged)(ksp,ksp->its,res_norm,&ksp->reason,ksp->cnvP);CHKERRQ(ierr);
/* Catch error in happy breakdown and signal convergence and break from loop */
if (hapend) {
if (!ksp->reason) {
if (ksp->errorifnotconverged) SETERRQ1(PetscObjectComm((PetscObject)ksp),PETSC_ERR_NOT_CONVERGED,"You reached the happy break down, but convergence was not indicated. Residual norm = %G",res_norm);
else {
ksp->reason = KSP_DIVERGED_BREAKDOWN;
break;
}
}
}
}
/* END OF ITERATION LOOP */
ierr = KSPLogResidualHistory(ksp,res_norm);CHKERRQ(ierr);
/*
Monitor if we know that we will not return for a restart */
if (loc_it && (ksp->reason || ksp->its >= ksp->max_it)) {
ierr = KSPMonitor(ksp,ksp->its,res_norm);CHKERRQ(ierr);
}
if (itcount) *itcount = loc_it;
/*
Down here we have to solve for the "best" coefficients of the Krylov
columns, add the solution values together, and possibly unwind the
preconditioning from the solution
*/
/* Form the solution (or the solution so far) */
/* Note: must pass in (loc_it-1) for iteration count so that KSPFGMRESBuildSoln
properly navigates */
ierr = KSPFGMRESBuildSoln(RS(0),ksp->vec_sol,ksp->vec_sol,ksp,loc_it-1);CHKERRQ(ierr);
PetscFunctionReturn(0);
}
static PetscErrorCode KSPPIPEFGMRESCycle(PetscInt *itcount,KSP ksp)
{
KSP_PIPEFGMRES *pipefgmres = (KSP_PIPEFGMRES*)(ksp->data);
PetscReal res_norm;
PetscReal hapbnd,tt;
PetscScalar *hh,*hes,*lhh,shift = pipefgmres->shift;
PetscBool hapend = PETSC_FALSE; /* indicates happy breakdown ending */
PetscErrorCode ierr;
PetscInt loc_it; /* local count of # of dir. in Krylov space */
PetscInt max_k = pipefgmres->max_k; /* max # of directions Krylov space */
PetscInt i,j,k;
Mat Amat,Pmat;
Vec Q,W; /* Pipelining vectors */
Vec *redux = pipefgmres->redux; /* workspace for single reduction */
PetscFunctionBegin;
if (itcount) *itcount = 0;
/* Assign simpler names to these vectors, allocated as pipelining workspace */
Q = VEC_Q;
W = VEC_W;
/* Allocate memory for orthogonalization work (freed in the GMRES Destroy routine)*/
/* Note that we add an extra value here to allow for a single reduction */
if (!pipefgmres->orthogwork) { ierr = PetscMalloc1(pipefgmres->max_k + 2 ,&pipefgmres->orthogwork);CHKERRQ(ierr);
}
lhh = pipefgmres->orthogwork;
/* Number of pseudo iterations since last restart is the number
of prestart directions */
loc_it = 0;
/* note: (pipefgmres->it) is always set one less than (loc_it) It is used in
KSPBUILDSolution_PIPEFGMRES, where it is passed to KSPPIPEFGMRESBuildSoln.
Note that when KSPPIPEFGMRESBuildSoln is called from this function,
(loc_it -1) is passed, so the two are equivalent */
pipefgmres->it = (loc_it - 1);
/* initial residual is in VEC_VV(0) - compute its norm*/
ierr = VecNorm(VEC_VV(0),NORM_2,&res_norm);CHKERRQ(ierr);
/* first entry in right-hand-side of hessenberg system is just
the initial residual norm */
*RS(0) = res_norm;
ksp->rnorm = res_norm;
ierr = KSPLogResidualHistory(ksp,res_norm);CHKERRQ(ierr);
ierr = KSPMonitor(ksp,ksp->its,res_norm);CHKERRQ(ierr);
/* check for the convergence - maybe the current guess is good enough */
ierr = (*ksp->converged)(ksp,ksp->its,res_norm,&ksp->reason,ksp->cnvP);CHKERRQ(ierr);
if (ksp->reason) {
if (itcount) *itcount = 0;
PetscFunctionReturn(0);
}
/* scale VEC_VV (the initial residual) */
ierr = VecScale(VEC_VV(0),1.0/res_norm);CHKERRQ(ierr);
/* Fill the pipeline */
ierr = KSP_PCApply(ksp,VEC_VV(loc_it),PREVEC(loc_it));CHKERRQ(ierr);
ierr = PCGetOperators(ksp->pc,&Amat,&Pmat);CHKERRQ(ierr);
ierr = KSP_MatMult(ksp,Amat,PREVEC(loc_it),ZVEC(loc_it));CHKERRQ(ierr);
ierr = VecAXPY(ZVEC(loc_it),-shift,VEC_VV(loc_it));CHKERRQ(ierr); /* Note shift */
/* MAIN ITERATION LOOP BEGINNING*/
/* keep iterating until we have converged OR generated the max number
of directions OR reached the max number of iterations for the method */
while (!ksp->reason && loc_it < max_k && ksp->its < ksp->max_it) {
if (loc_it) {
ierr = KSPLogResidualHistory(ksp,res_norm);CHKERRQ(ierr);
ierr = KSPMonitor(ksp,ksp->its,res_norm);CHKERRQ(ierr);
}
pipefgmres->it = (loc_it - 1);
/* see if more space is needed for work vectors */
if (pipefgmres->vv_allocated <= loc_it + VEC_OFFSET + 1) {
ierr = KSPPIPEFGMRESGetNewVectors(ksp,loc_it+1);CHKERRQ(ierr);
/* (loc_it+1) is passed in as number of the first vector that should
be allocated */
}
/* Note that these inner products are with "Z" now, so
in particular, lhh[loc_it] is the 'barred' or 'shifted' value,
not the value from the equivalent FGMRES run (even in exact arithmetic)
That is, the H we need for the Arnoldi relation is different from the
coefficients we use in the orthogonalization process,because of the shift */
/* Do some local twiddling to allow for a single reduction */
for(i=0;i<loc_it+1;i++){
redux[i] = VEC_VV(i);
}
redux[loc_it+1] = ZVEC(loc_it);
/* note the extra dot product which ends up in lh[loc_it+1], which computes ||z||^2 */
ierr = VecMDotBegin(ZVEC(loc_it),loc_it+2,redux,lhh);CHKERRQ(ierr);
/* Start the split reduction (This actually calls the MPI_Iallreduce, otherwise, the reduction is simply delayed until the "end" call)*/
ierr = PetscCommSplitReductionBegin(PetscObjectComm((PetscObject)ZVEC(loc_it)));CHKERRQ(ierr);
/* The work to be overlapped with the inner products follows.
This is application of the preconditioner and the operator
to compute intermediate quantites which will be combined (locally)
with the results of the inner products.
*/
ierr = KSP_PCApply(ksp,ZVEC(loc_it),Q);CHKERRQ(ierr);
ierr = PCGetOperators(ksp->pc,&Amat,&Pmat);CHKERRQ(ierr);
ierr = KSP_MatMult(ksp,Amat,Q,W);CHKERRQ(ierr);
/* Compute inner products of the new direction with previous directions,
and the norm of the to-be-orthogonalized direction "Z".
This information is enough to build the required entries
of H. The inner product with VEC_VV(it_loc) is
*different* than in the standard FGMRES and need to be dealt with specially.
That is, for standard FGMRES the orthogonalization coefficients are the same
as the coefficients used in the Arnoldi relation to reconstruct, but here this
is not true (albeit only for the one entry of H which we "unshift" below. */
/* Finish the dot product, retrieving the extra entry */
ierr = VecMDotEnd(ZVEC(loc_it),loc_it+2,redux,lhh);CHKERRQ(ierr);
tt = PetscRealPart(lhh[loc_it+1]);
/* Hessenberg entries, and entries for (naive) classical Graham-Schmidt
Note that the Hessenberg entries require a shift, as these are for the
relation AU = VH, which is wrt unshifted basis vectors */
hh = HH(0,loc_it); hes=HES(0,loc_it);
for (j=0; j<loc_it; j++) {
hh[j] = lhh[j];
hes[j] = lhh[j];
}
hh[loc_it] = lhh[loc_it] + shift;
hes[loc_it] = lhh[loc_it] + shift;
/* we delay applying the shift here */
for (j=0; j<=loc_it; j++) {
lhh[j] = -lhh[j]; /* flip sign */
}
/* Compute the norm of the un-normalized new direction using the rearranged formula
Note that these are shifted ("barred") quantities */
for(k=0;k<=loc_it;k++) tt -= ((PetscReal)(PetscAbsScalar(lhh[k]) * PetscAbsScalar(lhh[k])));
/* On AVX512 this is accumulating roundoff errors for eg: tt=-2.22045e-16 */
if ((tt < 0.0) && tt > -PETSC_SMALL) tt = 0.0 ;
if (tt < 0.0) {
/* If we detect square root breakdown in the norm, we must restart the algorithm.
Here this means we simply break the current loop and reconstruct the solution
using the basis we have computed thus far. Note that by breaking immediately,
we do not update the iteration count, so computation done in this iteration
should be disregarded.
*/
ierr = PetscInfo2(ksp,"Restart due to square root breakdown at it = %D, tt=%g\n",ksp->its,(double)tt);CHKERRQ(ierr);
break;
} else {
tt = PetscSqrtReal(tt);
}
/* new entry in hessenburg is the 2-norm of our new direction */
hh[loc_it+1] = tt;
hes[loc_it+1] = tt;
/* The recurred computation for the new direction
The division by tt is delayed to the happy breakdown check later
Note placement BEFORE the unshift
*/
ierr = VecCopy(ZVEC(loc_it),VEC_VV(loc_it+1));CHKERRQ(ierr);
ierr = VecMAXPY(VEC_VV(loc_it+1),loc_it+1,lhh,&VEC_VV(0));CHKERRQ(ierr);
/* (VEC_VV(loc_it+1) is not normalized yet) */
/* The recurred computation for the preconditioned vector (u) */
ierr = VecCopy(Q,PREVEC(loc_it+1));CHKERRQ(ierr);
ierr = VecMAXPY(PREVEC(loc_it+1),loc_it+1,lhh,&PREVEC(0));CHKERRQ(ierr);
ierr = VecScale(PREVEC(loc_it+1),1.0/tt);CHKERRQ(ierr);
/* Unshift an entry in the GS coefficients ("removing the bar") */
lhh[loc_it] -= shift;
/* The recurred computation for z (Au)
Note placement AFTER the "unshift" */
ierr = VecCopy(W,ZVEC(loc_it+1));CHKERRQ(ierr);
ierr = VecMAXPY(ZVEC(loc_it+1),loc_it+1,lhh,&ZVEC(0));CHKERRQ(ierr);
ierr = VecScale(ZVEC(loc_it+1),1.0/tt);CHKERRQ(ierr);
/* Happy Breakdown Check */
hapbnd = PetscAbsScalar((tt) / *RS(loc_it));
/* RS(loc_it) contains the res_norm from the last iteration */
hapbnd = PetscMin(pipefgmres->haptol,hapbnd);
if (tt > hapbnd) {
/* scale new direction by its norm */
ierr = VecScale(VEC_VV(loc_it+1),1.0/tt);CHKERRQ(ierr);
} else {
/* This happens when the solution is exactly reached. */
/* So there is no new direction... */
ierr = VecSet(VEC_TEMP,0.0);CHKERRQ(ierr); /* set VEC_TEMP to 0 */
hapend = PETSC_TRUE;
}
/* note that for pipefgmres we could get HES(loc_it+1, loc_it) = 0 and the
current solution would not be exact if HES was singular. Note that
HH non-singular implies that HES is not singular, and HES is guaranteed
to be nonsingular when PREVECS are linearly independent and A is
nonsingular (in GMRES, the nonsingularity of A implies the nonsingularity
of HES). So we should really add a check to verify that HES is nonsingular.*/
/* Note that to be thorough, in debug mode, one could call a LAPACK routine
here to check that the hessenberg matrix is indeed non-singular (since
FGMRES does not guarantee this) */
/* Now apply rotations to new col of hessenberg (and right side of system),
calculate new rotation, and get new residual norm at the same time*/
ierr = KSPPIPEFGMRESUpdateHessenberg(ksp,loc_it,&hapend,&res_norm);CHKERRQ(ierr);
if (ksp->reason) break;
loc_it++;
pipefgmres->it = (loc_it-1); /* Add this here in case it has converged */
ierr = PetscObjectSAWsTakeAccess((PetscObject)ksp);CHKERRQ(ierr);
ksp->its++;
ksp->rnorm = res_norm;
ierr = PetscObjectSAWsGrantAccess((PetscObject)ksp);CHKERRQ(ierr);
ierr = (*ksp->converged)(ksp,ksp->its,res_norm,&ksp->reason,ksp->cnvP);CHKERRQ(ierr);
/* Catch error in happy breakdown and signal convergence and break from loop */
if (hapend) {
if (!ksp->reason) {
if (ksp->errorifnotconverged) SETERRQ1(PetscObjectComm((PetscObject)ksp),PETSC_ERR_NOT_CONVERGED,"You reached the happy break down, but convergence was not indicated. Residual norm = %g",(double)res_norm);
else {
ksp->reason = KSP_DIVERGED_BREAKDOWN;
break;
}
}
}
}
/* END OF ITERATION LOOP */
ierr = KSPLogResidualHistory(ksp,res_norm);CHKERRQ(ierr);
/*
Monitor if we know that we will not return for a restart */
if (loc_it && (ksp->reason || ksp->its >= ksp->max_it)) {
ierr = KSPMonitor(ksp,ksp->its,res_norm);CHKERRQ(ierr);
}
if (itcount) *itcount = loc_it;
/*
Down here we have to solve for the "best" coefficients of the Krylov
columns, add the solution values together, and possibly unwind the
preconditioning from the solution
*/
/* Form the solution (or the solution so far) */
/* Note: must pass in (loc_it-1) for iteration count so that KSPPIPEGMRESIIBuildSoln
properly navigates */
ierr = KSPPIPEFGMRESBuildSoln(RS(0),ksp->vec_sol,ksp->vec_sol,ksp,loc_it-1);CHKERRQ(ierr);
PetscFunctionReturn(0);
}