void SymmetricRuizEquil
( DistSparseMatrix<F>& A,
DistMultiVec<Base<F>>& d,
Int maxIter, bool progress )
{
DEBUG_CSE
typedef Base<F> Real;
const Int n = A.Height();
mpi::Comm comm = A.Comm();
d.SetComm( comm );
Ones( d, n, 1 );
DistMultiVec<Real> scales(comm);
const Int indent = PushIndent();
for( Int iter=0; iter<maxIter; ++iter )
{
// Rescale the columns (and rows)
// ------------------------------
ColumnMaxNorms( A, scales );
EntrywiseMap( scales, function<Real(Real)>(DampScaling<Real>) );
EntrywiseMap( scales, function<Real(Real)>(SquareRootScaling<Real>) );
DiagonalScale( LEFT, NORMAL, scales, d );
SymmetricDiagonalSolve( scales, A );
}
SetIndent( indent );
}
void SymmetricRuizEquil
( DistSparseMatrix<Field>& A,
DistMultiVec<Base<Field>>& d,
Int maxIter, bool progress )
{
EL_DEBUG_CSE
typedef Base<Field> Real;
const Int n = A.Height();
const Grid& grid = A.Grid();
d.SetGrid( grid );
Ones( d, n, 1 );
DistMultiVec<Real> scales(grid);
const Int indent = PushIndent();
for( Int iter=0; iter<maxIter; ++iter )
{
// Rescale the columns (and rows)
// ------------------------------
ColumnMaxNorms( A, scales );
EntrywiseMap( scales, MakeFunction(DampScaling<Real>) );
EntrywiseMap( scales, MakeFunction(SquareRootScaling<Real>) );
DiagonalScale( LEFT, NORMAL, scales, d );
SymmetricDiagonalSolve( scales, A );
}
SetIndent( indent );
}
void EntrywiseMap( DistSparseMatrix<T>& A, function<T(T)> func )
{
DEBUG_CSE
T* vBuf = A.ValueBuffer();
const Int numLocalEntries = A.NumLocalEntries();
for( Int k=0; k<numLocalEntries; ++k )
vBuf[k] = func(vBuf[k]);
}
void StackedRuizEquil
( DistSparseMatrix<Field>& A,
DistSparseMatrix<Field>& B,
DistMultiVec<Base<Field>>& dRowA,
DistMultiVec<Base<Field>>& dRowB,
DistMultiVec<Base<Field>>& dCol,
bool progress )
{
EL_DEBUG_CSE
typedef Base<Field> Real;
const Int mA = A.Height();
const Int mB = B.Height();
const Int n = A.Width();
mpi::Comm comm = A.Comm();
dRowA.SetComm( comm );
dRowB.SetComm( comm );
dCol.SetComm( comm );
Ones( dRowA, mA, 1 );
Ones( dRowB, mB, 1 );
Ones( dCol, n, 1 );
// TODO(poulson): Expose to control structure
// For, simply hard-code a small number of iterations
const Int maxIter = 4;
DistMultiVec<Real> scales(comm), maxAbsValsB(comm);
auto& scalesLoc = scales.Matrix();
auto& maxAbsValsBLoc = maxAbsValsB.Matrix();
const Int localHeight = scalesLoc.Height();
const Int indent = PushIndent();
for( Int iter=0; iter<maxIter; ++iter )
{
// Rescale the columns
// -------------------
ColumnMaxNorms( A, scales );
ColumnMaxNorms( B, maxAbsValsB );
for( Int jLoc=0; jLoc<localHeight; ++jLoc )
scalesLoc(jLoc) = Max(scalesLoc(jLoc),maxAbsValsBLoc(jLoc));
EntrywiseMap( scales, MakeFunction(DampScaling<Real>) );
DiagonalScale( LEFT, NORMAL, scales, dCol );
DiagonalSolve( RIGHT, NORMAL, scales, A );
DiagonalSolve( RIGHT, NORMAL, scales, B );
// Rescale the rows
// ----------------
RowMaxNorms( A, scales );
EntrywiseMap( scales, MakeFunction(DampScaling<Real>) );
DiagonalScale( LEFT, NORMAL, scales, dRowA );
DiagonalSolve( LEFT, NORMAL, scales, A );
RowMaxNorms( B, scales );
EntrywiseMap( scales, MakeFunction(DampScaling<Real>) );
DiagonalScale( LEFT, NORMAL, scales, dRowB );
DiagonalSolve( LEFT, NORMAL, scales, B );
}
SetIndent( indent );
}
Int ZeroNorm( const DistSparseMatrix<T>& A, Base<T> tol )
{
DEBUG_ONLY(CSE cse("ZeroNorm"))
Int numNonzeros = 0;
const Int numLocalEntries = A.NumLocalEntries();
for( Int k=0; k<numLocalEntries; ++k )
if( Abs(A.Value(k)) > tol )
++numNonzeros;
return mpi::AllReduce( numNonzeros, A.Comm() );
}
void ShiftDiagonal
( DistSparseMatrix<T>& A, S alphaPre, Int offset, bool existingDiag )
{
EL_DEBUG_CSE
const Int mLocal = A.LocalHeight();
const Int n = A.Width();
const T alpha = T(alphaPre);
if( existingDiag )
{
T* valBuf = A.ValueBuffer();
for( Int iLoc=0; iLoc<mLocal; ++iLoc )
{
const Int i = A.GlobalRow(iLoc);
const Int e = A.Offset( iLoc, i+offset );
valBuf[e] += alpha;
}
}
else
{
A.Reserve( mLocal );
for( Int iLoc=0; iLoc<mLocal; ++iLoc )
{
const Int i = A.GlobalRow(iLoc);
if( i+offset >= 0 && i+offset < n )
A.QueueLocalUpdate( iLoc, i+offset, alpha );
}
A.ProcessLocalQueues();
}
}
void Helmholtz( DistSparseMatrix<F>& H, Int nx, Int ny, F shift )
{
DEBUG_CSE
typedef Base<F> Real;
const Int n = nx*ny;
Zeros( H, n, n );
const Real hxInv = nx+1;
const Real hyInv = ny+1;
const Real hxInvSquared = hxInv*hxInv;
const Real hyInvSquared = hyInv*hyInv;
const F mainTerm = 2*(hxInvSquared+hyInvSquared) - shift;
const Int localHeight = H.LocalHeight();
H.Reserve( 5*localHeight );
for( Int iLoc=0; iLoc<localHeight; ++iLoc )
{
const Int i = H.GlobalRow(iLoc);
const Int x = i % nx;
const Int y = i/nx;
H.QueueUpdate( i, i, mainTerm );
if( x != 0 )
H.QueueUpdate( i, i-1, -hxInvSquared );
if( x != nx-1 )
H.QueueUpdate( i, i+1, -hxInvSquared );
if( y != 0 )
H.QueueUpdate( i, i-nx, -hyInvSquared );
if( y != ny-1 )
H.QueueUpdate( i, i+nx, -hyInvSquared );
}
H.ProcessQueues();
}
Int NumOutside( const DistSparseMatrix<Real>& A )
{
EL_DEBUG_CSE
const Int numLocalEntries = A.NumLocalEntries();
const Real* valBuf = A.LockedValueBuffer();
Int numLocalNonPos = 0;
for( Int k=0; k<numLocalEntries; ++k )
if( valBuf[k] <= Real(0) )
++numLocalNonPos;
return mpi::AllReduce( numLocalNonPos, A.Comm() );
}
void KKT
( const DistSparseMatrix<Real>& A,
const DistSparseMatrix<Real>& G,
const DistMultiVec<Real>& s,
const DistMultiVec<Real>& z,
DistSparseMatrix<Real>& J,
bool onlyLower )
{
EL_DEBUG_CSE
const Int n = A.Width();
DistSparseMatrix<Real> Q(A.Grid());
Q.Resize( n, n );
qp::affine::KKT( Q, A, G, s, z, J, onlyLower );
}
void StaticKKT
( const DistSparseMatrix<Real>& A,
const DistSparseMatrix<Real>& G,
Real gamma,
Real delta,
Real beta,
DistSparseMatrix<Real>& J,
bool onlyLower )
{
EL_DEBUG_CSE
const Int n = A.Width();
DistSparseMatrix<Real> Q(A.Grid());
Q.Resize( n, n );
qp::affine::StaticKKT( Q, A, G, gamma, delta, beta, J, onlyLower );
}
void AugmentedKKT
( const DistSparseMatrix<Real>& A,
Real gamma,
Real delta,
const DistMultiVec<Real>& x,
const DistMultiVec<Real>& z,
DistSparseMatrix<Real>& J,
bool onlyLower )
{
EL_DEBUG_CSE
const Int n = A.Width();
DistSparseMatrix<Real> Q(A.Comm());
Zeros( Q, n, n );
qp::direct::AugmentedKKT( Q, A, gamma, delta, x, z, J, onlyLower );
}
void KKT
( const DistSparseMatrix<Real>& A,
Real gamma,
Real delta,
Real beta,
const DistMultiVec<Real>& x,
const DistMultiVec<Real>& z,
DistSparseMatrix<Real>& J, bool onlyLower )
{
EL_DEBUG_CSE
const Int n = A.Width();
DistSparseMatrix<Real> Q(A.Comm());
Q.Resize( n, n );
qp::direct::KKT( Q, A, gamma, delta, beta, x, z, J, onlyLower );
}
Base<Field>
MinAbsNonzero( const DistSparseMatrix<Field>& A, Base<Field> upperBound )
{
EL_DEBUG_CSE
typedef Base<Field> Real;
const Int numEntries = A.NumLocalEntries();
Real minLocAbs = upperBound;
for( Int e=0; e<numEntries; ++e )
{
const Real absVal = Abs(A.Value(e));
if( absVal > Real(0) )
minLocAbs = Min(minLocAbs,absVal);
}
return mpi::AllReduce( minLocAbs, mpi::MIN, A.Comm() );
}
void Mehrotra
( const DistSparseMatrix<Real>& A,
const DistMultiVec<Real>& b,
const DistMultiVec<Real>& c,
const DistMultiVec<Int>& orders,
const DistMultiVec<Int>& firstInds,
DistMultiVec<Real>& x,
DistMultiVec<Real>& y,
DistMultiVec<Real>& z,
const MehrotraCtrl<Real>& ctrl )
{
EL_DEBUG_CSE
const Int n = c.Height();
const Grid& grid = A.Grid();
DistSparseMatrix<Real> G(grid);
Identity( G, n, n );
G *= -1;
DistMultiVec<Real> h(grid);
Zeros( h, n, 1 );
MehrotraCtrl<Real> affineCtrl = ctrl;
affineCtrl.primalInit = false;
affineCtrl.dualInit = false;
DistMultiVec<Real> s(grid);
socp::affine::Mehrotra(A,G,b,c,h,orders,firstInds,x,y,z,s,affineCtrl);
}
pair<Base<F>,Base<F>>
HermitianHelper( const DistSparseMatrix<F>& A, Int basisSize )
{
typedef Base<F> Real;
Grid grid( A.Comm() );
DistMatrix<Real,STAR,STAR> T(grid);
Lanczos( A, T, basisSize );
const Int k = T.Height();
if( k == 0 )
return pair<Real,Real>(0,0);
auto d = GetDiagonal( T.Matrix() );
auto dSub = GetDiagonal( T.Matrix(), -1 );
Matrix<Real> w;
HermitianTridiagEig( d, dSub, w );
pair<Real,Real> extremal;
extremal.second = MaxNorm(w);
extremal.first = extremal.second;
for( Int i=0; i<k; ++i )
extremal.first = Min(extremal.first,Abs(w.Get(i,0)));
return extremal;
}
pair<Real,Real>
HermitianHelper( const DistSparseMatrix<Complex<Real>>& A, Int basisSize )
{
DistSparseMatrix<Complex<double>> ADbl(A.Comm());
Copy( A, ADbl );
auto pairDbl = HermitianHelper( ADbl, basisSize );
return std::make_pair( Real(pairDbl.first), Real(pairDbl.second) );
}
void Lanczos
( const DistSparseMatrix<F>& A,
ElementalMatrix<Base<F>>& T,
Int basisSize )
{
DEBUG_CSE
const Int n = A.Height();
if( n != A.Width() )
LogicError("A was not square");
auto applyA =
[&]( const DistMultiVec<F>& X, DistMultiVec<F>& Y )
{
Zeros( Y, n, X.Width() );
Multiply( NORMAL, F(1), A, X, F(0), Y );
};
Lanczos<F>( n, applyA, T, basisSize );
}
void LAV
( const DistSparseMatrix<Real>& A,
const DistMultiVec<Real>& b,
DistMultiVec<Real>& x,
const lp::affine::Ctrl<Real>& ctrl )
{
EL_DEBUG_CSE
const Int m = A.Height();
const Int n = A.Width();
const Grid& grid = A.Grid();
DistSparseMatrix<Real> AHat(grid), G(grid);
DistMultiVec<Real> c(grid), h(grid);
// c := [0;1;1]
// ============
Zeros( c, n+2*m, 1 );
for( Int iLoc=0; iLoc<c.LocalHeight(); ++iLoc )
if( c.GlobalRow(iLoc) >= n )
c.SetLocal( iLoc, 0, Real(1) );
// \hat A := [A, I, -I]
// ====================
Zeros( AHat, m, n+2*m );
const Int numLocalEntriesA = A.NumLocalEntries();
AHat.Reserve( numLocalEntriesA + 2*AHat.LocalHeight() );
for( Int e=0; e<numLocalEntriesA; ++e )
AHat.QueueUpdate( A.Row(e), A.Col(e), A.Value(e) );
for( Int iLoc=0; iLoc<AHat.LocalHeight(); ++iLoc )
{
const Int i = AHat.GlobalRow(iLoc);
AHat.QueueLocalUpdate( iLoc, i+n, Real( 1) );
AHat.QueueLocalUpdate( iLoc, i+n+m, Real(-1) );
}
AHat.ProcessLocalQueues();
// G := | 0 -I 0 |
// | 0 0 -I |
// ================
Zeros( G, 2*m, n+2*m );
G.Reserve( G.LocalHeight() );
for( Int iLoc=0; iLoc<G.LocalHeight(); ++iLoc )
G.QueueLocalUpdate( iLoc, G.GlobalRow(iLoc)+n, Real(-1) );
G.ProcessLocalQueues();
// h := | 0 |
// | 0 |
// ==========
Zeros( h, 2*m, 1 );
// Solve the affine QP
// ===================
DistMultiVec<Real> xHat(grid), y(grid), z(grid), s(grid);
LP( AHat, G, b, c, h, xHat, y, z, s, ctrl );
// Extract x
// =========
x = xHat( IR(0,n), ALL );
}
Base<Field> ProductLanczosDecomp
( const DistSparseMatrix<Field>& A,
DistMultiVec<Field>& V,
AbstractDistMatrix<Base<Field>>& T,
DistMultiVec<Field>& v,
Int basisSize )
{
EL_DEBUG_CSE
const Int m = A.Height();
const Int n = A.Width();
mpi::Comm comm = A.Comm();
DistMultiVec<Field> S(comm);
// Cache the adjoint
// -----------------
DistSparseMatrix<Field> AAdj(comm);
Adjoint( A, AAdj );
if( m >= n )
{
auto applyA =
[&]( const DistMultiVec<Field>& X, DistMultiVec<Field>& Y )
{
Zeros( S, m, X.Width() );
Multiply( NORMAL, Field(1), A, X, Field(0), S );
Zeros( Y, n, X.Width() );
Multiply( NORMAL, Field(1), AAdj, S, Field(0), Y );
};
return LanczosDecomp( n, applyA, V, T, v, basisSize );
}
else
{
auto applyA =
[&]( const DistMultiVec<Field>& X, DistMultiVec<Field>& Y )
{
Zeros( S, n, X.Width() );
Multiply( NORMAL, Field(1), AAdj, X, Field(0), S );
Zeros( Y, m, X.Width() );
Multiply( NORMAL, Field(1), A, S, Field(0), Y );
};
return LanczosDecomp( m, applyA, V, T, v, basisSize );
}
}
Base<Field> LanczosDecomp
( const DistSparseMatrix<Field>& A,
DistMultiVec<Field>& V,
AbstractDistMatrix<Base<Field>>& T,
DistMultiVec<Field>& v,
Int basisSize )
{
EL_DEBUG_CSE
const Int n = A.Height();
if( n != A.Width() )
LogicError("A was not square");
auto applyA =
[&]( const DistMultiVec<Field>& X, DistMultiVec<Field>& Y )
{
Zeros( Y, n, X.Width() );
Multiply( NORMAL, Field(1), A, X, Field(0), Y );
};
return LanczosDecomp( n, applyA, V, T, v, basisSize );
}
void QP
( const DistSparseMatrix<Real>& A,
const DistMultiVec<Real>& B,
DistMultiVec<Real>& X,
const qp::direct::Ctrl<Real>& ctrl )
{
DEBUG_CSE
const Int m = A.Height();
const Int n = A.Width();
const Int k = B.Width();
mpi::Comm comm = A.Comm();
DistSparseMatrix<Real> Q(comm), AHat(comm);
DistMultiVec<Real> bHat(comm), c(comm);
Herk( LOWER, ADJOINT, Real(1), A, Q );
MakeHermitian( LOWER, Q );
Zeros( AHat, 0, n );
Zeros( bHat, 0, 1 );
Zeros( X, n, k );
DistMultiVec<Real> q(comm), y(comm), z(comm);
auto& qLoc = q.Matrix();
auto& XLoc = X.Matrix();
auto& BLoc = B.LockedMatrix();
for( Int j=0; j<k; ++j )
{
auto xLoc = XLoc( ALL, IR(j) );
auto bLoc = BLoc( ALL, IR(j) );
Zeros( c, n, 1 );
Zeros( q, m, 1 );
qLoc = bLoc;
Multiply( ADJOINT, Real(-1), A, q, Real(0), c );
Zeros( q, n, 1 );
qLoc = xLoc;
El::QP( Q, AHat, bHat, c, q, y, z, ctrl );
xLoc = qLoc;
}
}
void RowMaxNorms( const DistSparseMatrix<F>& A, DistMultiVec<Base<F>>& norms )
{
DEBUG_CSE
typedef Base<F> Real;
const Int localHeight = A.LocalHeight();
const F* valBuf = A.LockedValueBuffer();
const Int* offsetBuf = A.LockedOffsetBuffer();
norms.SetComm( A.Comm() );
norms.Resize( A.Height(), 1 );
auto& normsLoc = norms.Matrix();
for( Int iLoc=0; iLoc<localHeight; ++iLoc )
{
Real rowMax = 0;
const Int offset = offsetBuf[iLoc];
const Int numConn = offsetBuf[iLoc+1] - offset;
for( Int e=offset; e<offset+numConn; ++e )
rowMax = Max(rowMax,Abs(valBuf[e]));
normsLoc(iLoc) = rowMax;
}
}
void RuizEquil
( DistSparseMatrix<Field>& A,
DistMultiVec<Base<Field>>& dRow,
DistMultiVec<Base<Field>>& dCol,
bool progress )
{
EL_DEBUG_CSE
typedef Base<Field> Real;
const Int m = A.Height();
const Int n = A.Width();
mpi::Comm comm = A.Comm();
dRow.SetComm( comm );
dCol.SetComm( comm );
Ones( dRow, m, 1 );
Ones( dCol, n, 1 );
// TODO(poulson): Expose to control structure
// For, simply hard-code a small number of iterations
const Int maxIter = 4;
DistMultiVec<Real> scales(comm);
const Int indent = PushIndent();
for( Int iter=0; iter<maxIter; ++iter )
{
// Rescale the columns
// -------------------
ColumnMaxNorms( A, scales );
EntrywiseMap( scales, MakeFunction(DampScaling<Real>) );
DiagonalScale( LEFT, NORMAL, scales, dCol );
DiagonalSolve( RIGHT, NORMAL, scales, A );
// Rescale the rows
// ----------------
RowMaxNorms( A, scales );
EntrywiseMap( scales, MakeFunction(DampScaling<Real>) );
DiagonalScale( LEFT, NORMAL, scales, dRow );
DiagonalSolve( LEFT, NORMAL, scales, A );
}
SetIndent( indent );
}
void RowTwoNorms( const DistSparseMatrix<F>& A, DistMultiVec<Base<F>>& norms )
{
DEBUG_CSE
typedef Base<F> Real;
const Int localHeight = A.LocalHeight();
const F* valBuf = A.LockedValueBuffer();
const Int* offsetBuf = A.LockedOffsetBuffer();
norms.SetComm( A.Comm() );
norms.Resize( A.Height(), 1 );
auto& normLoc = norms.Matrix();
for( Int iLoc=0; iLoc<localHeight; ++iLoc )
{
Real scale = 0;
Real scaledSquare = 1;
const Int offset = offsetBuf[iLoc];
const Int numConn = offsetBuf[iLoc+1] - offset;
for( Int e=offset; e<offset+numConn; ++e )
UpdateScaledSquare( valBuf[e], scale, scaledSquare );
normLoc(iLoc) = scale*Sqrt(scaledSquare);
}
}
DistSparseMatrix<Ring>::DistSparseMatrix( const DistSparseMatrix<Ring>& A )
{
EL_DEBUG_CSE
distGraph_.numSources_ = -1;
distGraph_.numTargets_ = -1;
distGraph_.grid_ = &A.Grid();
if( &A != this )
*this = A;
EL_DEBUG_ONLY(
else
LogicError("Tried to construct DistMultiVec via itself");
)
}
void EntrywiseMap
( const DistSparseMatrix<S>& A,
DistSparseMatrix<T>& B,
function<T(S)> func )
{
DEBUG_CSE
const Int numEntries = A.vals_.size();
const Int numRemoteEntries = A.remoteVals_.size();
B.distGraph_ = A.distGraph_;
B.vals_.resize( numEntries );
for( Int k=0; k<numEntries; ++k )
B.vals_[k] = func(A.vals_[k]);
B.remoteVals_.resize( numRemoteEntries );
for( Int k=0; k<numRemoteEntries; ++k )
B.remoteVals_[k] = func(A.remoteVals_[k]);
B.ProcessQueues();
}
void Fill( DistSparseMatrix<T>& A, T alpha )
{
EL_DEBUG_CSE
const Int m = A.Height();
const Int n = A.Width();
A.Resize( m, n );
Zero( A );
if( alpha != T(0) )
{
const Int localHeight = A.LocalHeight();
A.Reserve( localHeight*n );
for( Int iLoc=0; iLoc<localHeight; ++iLoc )
for( Int j=0; j<n; ++j )
A.QueueLocalUpdate( iLoc, j, alpha );
A.ProcessLocalQueues();
}
}
void JordanCholesky( DistSparseMatrix<T>& A, Int n )
{
DEBUG_ONLY(CSE cse("JordanCholesky"))
Zeros( A, n, n );
const Int localHeight = A.LocalHeight();
A.Reserve( 3*localHeight );
for( Int iLoc=0; iLoc<localHeight; ++iLoc )
{
const Int i = A.GlobalRow(iLoc);
if( i == 0 )
A.QueueUpdate( i, i, T(1) );
else
A.QueueUpdate( i, i, T(5) );
if( i > 0 )
A.QueueUpdate( i, i-1, T(2) );
if( i < n-1 )
A.QueueUpdate( i, i+1, T(2) );
}
A.ProcessQueues();
}
pair<Base<F>,Base<F>>
Helper( const DistSparseMatrix<F>& A, Int basisSize )
{
typedef Base<F> Real;
Grid grid( A.Comm() );
DistMatrix<Real,STAR,STAR> T(grid);
ProductLanczos( A, T, basisSize );
const Int k = T.Height();
if( k == 0 )
return pair<Real,Real>(0,0);
auto d = GetDiagonal( T.Matrix() );
auto dSub = GetDiagonal( T.Matrix(), -1 );
Matrix<Real> w;
HermitianTridiagEig( d, dSub, w, ASCENDING );
pair<Real,Real> extremal;
extremal.first = Sqrt( Max(w.Get(0,0),Real(0)) );
extremal.second = Sqrt( Max(w.Get(k-1,0),Real(0)) );
return extremal;
}
void GetMappedDiagonal
( const DistSparseMatrix<T>& A,
DistMultiVec<S>& d,
function<S(const T&)> func,
Int offset )
{
EL_DEBUG_CSE
const Int m = A.Height();
const Int n = A.Width();
const T* valBuf = A.LockedValueBuffer();
const Int* colBuf = A.LockedTargetBuffer();
if( m != n )
LogicError("DistSparseMatrix GetMappedDiagonal assumes square matrix");
if( offset != 0 )
LogicError("DistSparseMatrix GetMappedDiagonal assumes offset=0");
d.SetGrid( A.Grid() );
d.Resize( El::DiagonalLength(m,n,offset), 1 );
Fill( d, S(1) );
S* dBuf = d.Matrix().Buffer();
const Int dLocalHeight = d.LocalHeight();
for( Int iLoc=0; iLoc<dLocalHeight; ++iLoc )
{
const Int i = d.GlobalRow(iLoc);
const Int thisOff = A.RowOffset(iLoc);
const Int nextOff = A.RowOffset(iLoc+1);
auto it = std::lower_bound( colBuf+thisOff, colBuf+nextOff, i );
if( *it == i )
{
const Int e = it-colBuf;
dBuf[iLoc] = func(valBuf[e]);
}
else
dBuf[iLoc] = func(0);
}
}
