inline void
Syr2kLT
( T alpha, const DistMatrix<T>& A, const DistMatrix<T>& B,
T beta, DistMatrix<T>& C )
{
#ifndef RELEASE
PushCallStack("internal::Syr2kLT");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
if( A.Width() != C.Height() ||
A.Width() != C.Width() ||
B.Width() != C.Height() ||
B.Width() != C.Width() ||
A.Height() != B.Height() )
{
std::ostringstream msg;
msg << "Nonconformal Syr2kLT:\n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " B ~ " << B.Height() << " x " << B.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
throw std::logic_error( msg.str().c_str() );
}
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<T> AT(g), A0(g),
AB(g), A1(g),
A2(g);
DistMatrix<T> BT(g), B0(g),
BB(g), B1(g),
B2(g);
// Temporary distributions
DistMatrix<T,MR, STAR> A1Trans_MR_STAR(g);
DistMatrix<T,MR, STAR> B1Trans_MR_STAR(g);
DistMatrix<T,STAR,VR > A1_STAR_VR(g);
DistMatrix<T,STAR,VR > B1_STAR_VR(g);
DistMatrix<T,STAR,MC > A1_STAR_MC(g);
DistMatrix<T,STAR,MC > B1_STAR_MC(g);
A1Trans_MR_STAR.AlignWith( C );
B1Trans_MR_STAR.AlignWith( C );
A1_STAR_MC.AlignWith( C );
B1_STAR_MC.AlignWith( C );
// Start the algorithm
ScaleTrapezoid( beta, LEFT, LOWER, 0, C );
LockedPartitionDown
( A, AT,
AB, 0 );
LockedPartitionDown
( B, BT,
BB, 0 );
while( AB.Height() > 0 )
{
LockedRepartitionDown
( AT, A0,
/**/ /**/
A1,
AB, A2 );
LockedRepartitionDown
( BT, B0,
/**/ /**/
B1,
BB, B2 );
//--------------------------------------------------------------------//
A1Trans_MR_STAR.TransposeFrom( A1 );
A1_STAR_VR.TransposeFrom( A1Trans_MR_STAR );
A1_STAR_MC = A1_STAR_VR;
B1Trans_MR_STAR.TransposeFrom( B1 );
B1_STAR_VR.TransposeFrom( B1Trans_MR_STAR );
B1_STAR_MC = B1_STAR_VR;
LocalTrr2k
( LOWER, TRANSPOSE, TRANSPOSE, TRANSPOSE, TRANSPOSE,
alpha, A1_STAR_MC, B1Trans_MR_STAR,
B1_STAR_MC, A1Trans_MR_STAR,
T(1), C );
//--------------------------------------------------------------------//
SlideLockedPartitionDown
( AT, A0,
A1,
/**/ /**/
AB, A2 );
SlideLockedPartitionDown
( BT, B0,
B1,
/**/ /**/
BB, B2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
inline void
internal::SymmRUA
( T alpha, const DistMatrix<T,MC,MR>& A,
const DistMatrix<T,MC,MR>& B,
T beta, DistMatrix<T,MC,MR>& C )
{
#ifndef RELEASE
PushCallStack("internal::SymmRUA");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
#endif
const Grid& g = A.Grid();
DistMatrix<T,MC,MR>
BT(g), B0(g),
BB(g), B1(g),
B2(g);
DistMatrix<T,MC,MR>
CT(g), C0(g),
CB(g), C1(g),
C2(g);
DistMatrix<T,MR, STAR> B1Trans_MR_STAR(g);
DistMatrix<T,VC, STAR> B1Trans_VC_STAR(g);
DistMatrix<T,STAR,MC > B1_STAR_MC(g);
DistMatrix<T,MC, STAR> Z1Trans_MC_STAR(g);
DistMatrix<T,MR, STAR> Z1Trans_MR_STAR(g);
DistMatrix<T,MC, MR > Z1Trans(g);
DistMatrix<T,MR, MC > Z1Trans_MR_MC(g);
Matrix<T> Z1Local;
Scal( beta, C );
LockedPartitionDown
( B, BT,
BB, 0 );
PartitionDown
( C, CT,
CB, 0 );
while( CT.Height() < C.Height() )
{
LockedRepartitionDown
( BT, B0,
/**/ /**/
B1,
BB, B2 );
RepartitionDown
( CT, C0,
/**/ /**/
C1,
CB, C2 );
B1Trans_MR_STAR.AlignWith( A );
B1Trans_VC_STAR.AlignWith( A );
B1_STAR_MC.AlignWith( A );
Z1Trans_MC_STAR.AlignWith( A );
Z1Trans_MR_STAR.AlignWith( A );
Z1Trans_MR_MC.AlignWith( C1 );
Z1Trans_MC_STAR.ResizeTo( C1.Width(), C1.Height() );
Z1Trans_MR_STAR.ResizeTo( C1.Width(), C1.Height() );
//--------------------------------------------------------------------//
B1Trans_MR_STAR.TransposeFrom( B1 );
B1Trans_VC_STAR = B1Trans_MR_STAR;
B1_STAR_MC.TransposeFrom( B1Trans_VC_STAR );
Zero( Z1Trans_MC_STAR );
Zero( Z1Trans_MR_STAR );
internal::LocalSymmetricAccumulateRU
( TRANSPOSE, alpha, A, B1_STAR_MC, B1Trans_MR_STAR,
Z1Trans_MC_STAR, Z1Trans_MR_STAR );
Z1Trans.SumScatterFrom( Z1Trans_MC_STAR );
Z1Trans_MR_MC = Z1Trans;
Z1Trans_MR_MC.SumScatterUpdate( (T)1, Z1Trans_MR_STAR );
Transpose( Z1Trans_MR_MC.LockedLocalMatrix(), Z1Local );
Axpy( (T)1, Z1Local, C1.LocalMatrix() );
//--------------------------------------------------------------------//
B1Trans_MR_STAR.FreeAlignments();
B1Trans_VC_STAR.FreeAlignments();
B1_STAR_MC.FreeAlignments();
Z1Trans_MC_STAR.FreeAlignments();
Z1Trans_MR_STAR.FreeAlignments();
Z1Trans_MR_MC.FreeAlignments();
SlideLockedPartitionDown
( BT, B0,
B1,
/**/ /**/
BB, B2 );
SlidePartitionDown
( CT, C0,
C1,
/**/ /**/
CB, C2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
inline void
internal::GemmTNC
( Orientation orientationOfA,
T alpha, const DistMatrix<T,MC,MR>& A,
const DistMatrix<T,MC,MR>& B,
T beta, DistMatrix<T,MC,MR>& C )
{
#ifndef RELEASE
PushCallStack("internal::GemmTNC");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
if( orientationOfA == NORMAL )
throw std::logic_error("GemmTNC assumes A is (Conjugate)Transposed");
if( A.Width() != C.Height() ||
B.Width() != C.Width() ||
A.Height() != B.Height() )
{
std::ostringstream msg;
msg << "Nonconformal GemmTNC: \n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " B ~ " << B.Height() << " x " << B.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
throw std::logic_error( msg.str().c_str() );
}
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<T,MC,MR> AT(g), A0(g),
AB(g), A1(g),
A2(g);
DistMatrix<T,MC,MR> BT(g), B0(g),
BB(g), B1(g),
B2(g);
// Temporary distributions
DistMatrix<T,STAR,MC> A1_STAR_MC(g);
DistMatrix<T,STAR,MR> B1_STAR_MR(g);
// Start the algorithm
Scal( beta, C );
LockedPartitionDown
( A, AT,
AB, 0 );
LockedPartitionDown
( B, BT,
BB, 0 );
while( AB.Height() > 0 )
{
LockedRepartitionDown
( AT, A0,
/**/ /**/
A1,
AB, A2 );
LockedRepartitionDown
( BT, B0,
/**/ /**/
B1,
BB, B2 );
A1_STAR_MC.AlignWith( C );
B1_STAR_MR.AlignWith( C );
//--------------------------------------------------------------------//
A1_STAR_MC = A1; // A1[*,MC] <- A1[MC,MR]
B1_STAR_MR = B1; // B1[*,MR] <- B1[MC,MR]
// C[MC,MR] += alpha (A1[*,MC])^T B1[*,MR]
// = alpha (A1^T)[MC,*] B1[*,MR]
internal::LocalGemm
( orientationOfA, NORMAL, alpha, A1_STAR_MC, B1_STAR_MR, (T)1, C );
//--------------------------------------------------------------------//
A1_STAR_MC.FreeAlignments();
B1_STAR_MR.FreeAlignments();
SlideLockedPartitionDown
( AT, A0,
A1,
/**/ /**/
AB, A2 );
SlideLockedPartitionDown
( BT, B0,
B1,
/**/ /**/
BB, B2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
void Trr2kNNTN
( UpperOrLower uplo,
Orientation orientationOfC,
T alpha, const DistMatrix<T>& A, const DistMatrix<T>& B,
const DistMatrix<T>& C, const DistMatrix<T>& D,
T beta, DistMatrix<T>& E )
{
#ifndef RELEASE
CallStackEntry entry("internal::Trr2kNNTN");
if( E.Height() != E.Width() || A.Width() != C.Height() ||
A.Height() != E.Height() || C.Width() != E.Height() ||
B.Width() != E.Width() || D.Width() != E.Width() ||
A.Width() != B.Height() || C.Height() != D.Height() )
throw std::logic_error("Nonconformal Trr2kNNTN");
#endif
const Grid& g = E.Grid();
DistMatrix<T> AL(g), AR(g),
A0(g), A1(g), A2(g);
DistMatrix<T> BT(g), B0(g),
BB(g), B1(g),
B2(g);
DistMatrix<T> CT(g), C0(g),
CB(g), C1(g),
C2(g);
DistMatrix<T> DT(g), D0(g),
DB(g), D1(g),
D2(g);
DistMatrix<T,MC, STAR> A1_MC_STAR(g);
DistMatrix<T,MR, STAR> B1Trans_MR_STAR(g);
DistMatrix<T,STAR,MC > C1_STAR_MC(g);
DistMatrix<T,MR, STAR> D1Trans_MR_STAR(g);
A1_MC_STAR.AlignWith( E );
B1Trans_MR_STAR.AlignWith( E );
C1_STAR_MC.AlignWith( E );
D1Trans_MR_STAR.AlignWith( E );
LockedPartitionRight( A, AL, AR, 0 );
LockedPartitionDown
( B, BT,
BB, 0 );
LockedPartitionDown
( C, CT,
CB, 0 );
LockedPartitionDown
( D, DT,
DB, 0 );
while( AL.Width() < A.Width() )
{
LockedRepartitionRight
( AL, /**/ AR,
A0, /**/ A1, A2 );
LockedRepartitionDown
( BT, B0,
/**/ /**/
B1,
BB, B2 );
LockedRepartitionDown
( CT, C0,
/**/ /**/
C1,
CB, C2 );
LockedRepartitionDown
( DT, D0,
/**/ /**/
D1,
DB, D2 );
//--------------------------------------------------------------------//
A1_MC_STAR = A1;
C1_STAR_MC = C1;
B1Trans_MR_STAR.TransposeFrom( B1 );
D1Trans_MR_STAR.TransposeFrom( D1 );
LocalTrr2k
( uplo, TRANSPOSE, orientationOfC, TRANSPOSE,
alpha, A1_MC_STAR, B1Trans_MR_STAR,
C1_STAR_MC, D1Trans_MR_STAR,
beta, E );
//--------------------------------------------------------------------//
SlideLockedPartitionDown
( DT, D0,
D1,
/**/ /**/
DB, D2 );
SlideLockedPartitionDown
( CT, C0,
C1,
/**/ /**/
CB, C2 );
SlideLockedPartitionDown
( BT, B0,
B1,
/**/ /**/
BB, B2 );
SlideLockedPartitionRight
( AL, /**/ AR,
A0, A1, /**/ A2 );
}
}
inline void
SUMMA_NNC
( T alpha, const DistMatrix<T>& A,
const DistMatrix<T>& B,
T beta, DistMatrix<T>& C )
{
#ifndef RELEASE
CallStackEntry entry("gemm::SUMMA_NNC");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
LogicError("{A,B,C} must be distributed over the same grid");
if( A.Height() != C.Height() ||
B.Width() != C.Width() ||
A.Width() != B.Height() )
{
std::ostringstream msg;
msg << "Nonconformal matrices: \n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " B ~ " << B.Height() << " x " << B.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
LogicError( msg.str() );
}
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<T> AL(g), AR(g),
A0(g), A1(g), A2(g);
DistMatrix<T> BT(g), B0(g),
BB(g), B1(g),
B2(g);
// Temporary distributions
DistMatrix<T,MC,STAR> A1_MC_STAR(g);
DistMatrix<T,MR,STAR> B1Trans_MR_STAR(g);
A1_MC_STAR.AlignWith( C );
B1Trans_MR_STAR.AlignWith( C );
// Start the algorithm
Scale( beta, C );
LockedPartitionRight( A, AL, AR, 0 );
LockedPartitionDown
( B, BT,
BB, 0 );
while( AR.Width() > 0 )
{
LockedRepartitionRight( AL, /**/ AR,
A0, /**/ A1, A2 );
LockedRepartitionDown( BT, B0,
/**/ /**/
B1,
BB, B2 );
//--------------------------------------------------------------------//
A1_MC_STAR = A1;
B1Trans_MR_STAR.TransposeFrom( B1 );
// C[MC,MR] += alpha A1[MC,*] (B1^T[MR,*])^T
// = alpha A1[MC,*] B1[*,MR]
LocalGemm
( NORMAL, TRANSPOSE, alpha, A1_MC_STAR, B1Trans_MR_STAR, T(1), C );
//--------------------------------------------------------------------//
SlideLockedPartitionRight( AL, /**/ AR,
A0, A1, /**/ A2 );
SlideLockedPartitionDown( BT, B0,
B1,
/**/ /**/
BB, B2 );
}
}
void test_basic_types() {
bcon basic_types[] = {"string", BS("a string"), "f(double)", BF(3.14159), "boolean", BB(1), "time", BT(time(0)), "null", BNULL, "symbol", BX("a symbol"), "int", BI(123), "long", BL(456789L), BEND};
test_bson_from_bcon( basic_types, BCON_OK, BSON_VALID );
}
inline void
internal::Trr2kTNTT
( UpperOrLower uplo,
Orientation orientationOfA,
Orientation orientationOfC, Orientation orientationOfD,
T alpha, const DistMatrix<T,MC,MR>& A, const DistMatrix<T,MC,MR>& B,
const DistMatrix<T,MC,MR>& C, const DistMatrix<T,MC,MR>& D,
T beta, DistMatrix<T,MC,MR>& E )
{
#ifndef RELEASE
PushCallStack("internal::Trr2kTNTT");
if( E.Height() != E.Width() || A.Height() != C.Height() ||
A.Width() != E.Height() || C.Width() != E.Height() ||
B.Width() != E.Width() || D.Height() != E.Width() ||
A.Height() != B.Height() || C.Height() != D.Width() )
throw std::logic_error("Nonconformal Trr2kNNTT");
#endif
const Grid& g = E.Grid();
DistMatrix<T,MC,MR> AT(g), A0(g),
AB(g), A1(g),
A2(g);
DistMatrix<T,MC,MR> BT(g), B0(g),
BB(g), B1(g),
B2(g);
DistMatrix<T,MC,MR> CT(g), C0(g),
CB(g), C1(g),
C2(g);
DistMatrix<T,MC,MR> DL(g), DR(g),
D0(g), D1(g), D2(g);
DistMatrix<T,STAR,MC > A1_STAR_MC(g);
DistMatrix<T,MR, STAR> B1Trans_MR_STAR(g);
DistMatrix<T,STAR,MC > C1_STAR_MC(g);
DistMatrix<T,VR, STAR> D1_VR_STAR(g);
DistMatrix<T,STAR,MR > D1AdjOrTrans_STAR_MR(g);
LockedPartitionDown
( A, AT,
AB, 0 );
LockedPartitionDown
( B, BT,
BB, 0 );
LockedPartitionDown
( C, CT,
CB, 0 );
LockedPartitionRight( D, DL, DR, 0 );
while( AT.Height() < A.Height() )
{
LockedRepartitionDown
( AT, A0,
/**/ /**/
A1,
AB, A2 );
LockedRepartitionDown
( BT, B0,
/**/ /**/
B1,
BB, B2 );
LockedRepartitionDown
( CT, C0,
/**/ /**/
C1,
CB, C2 );
LockedRepartitionRight
( DL, /**/ DR,
D0, /**/ D1, D2 );
A1_STAR_MC.AlignWith( E );
B1Trans_MR_STAR.AlignWith( E );
C1_STAR_MC.AlignWith( E );
D1_VR_STAR.AlignWith( E );
D1AdjOrTrans_STAR_MR.AlignWith( E );
//--------------------------------------------------------------------//
A1_STAR_MC = A1;
C1_STAR_MC = C1;
B1Trans_MR_STAR.TransposeFrom( B1 );
D1_VR_STAR = D1;
if( orientationOfD == ADJOINT )
D1AdjOrTrans_STAR_MR.AdjointFrom( D1_VR_STAR );
else
D1AdjOrTrans_STAR_MR.TransposeFrom( D1_VR_STAR );
internal::LocalTrr2k
( uplo, orientationOfA, TRANSPOSE, orientationOfC,
alpha, A1_STAR_MC, B1Trans_MR_STAR,
C1_STAR_MC, D1AdjOrTrans_STAR_MR,
beta, E );
//--------------------------------------------------------------------//
D1AdjOrTrans_STAR_MR.FreeAlignments();
D1_VR_STAR.FreeAlignments();
C1_STAR_MC.FreeAlignments();
B1Trans_MR_STAR.FreeAlignments();
A1_STAR_MC.FreeAlignments();
SlideLockedPartitionRight
( DL, /**/ DR,
D0, D1, /**/ D2 );
SlideLockedPartitionDown
( CT, C0,
C1,
/**/ /**/
CB, C2 );
SlideLockedPartitionDown
( BT, B0,
B1,
/**/ /**/
BB, B2 );
SlideLockedPartitionDown
( AT, A0,
A1,
/**/ /**/
AB, A2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
inline void
RowEchelon( DistMatrix<F>& A, DistMatrix<F>& B )
{
#ifndef RELEASE
CallStackEntry entry("RowEchelon");
if( A.Grid() != B.Grid() )
LogicError("{A,B} must be distributed over the same grid");
if( A.Height() != B.Height() )
LogicError("A and B must be the same height");
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<F>
ATL(g), ATR(g), A00(g), A01(g), A02(g), APan(g),
ABL(g), ABR(g), A10(g), A11(g), A12(g),
A20(g), A21(g), A22(g);
DistMatrix<F>
BT(g), B0(g),
BB(g), B1(g),
B2(g);
// Temporary distributions
DistMatrix<F,STAR,STAR> A11_STAR_STAR(g);
DistMatrix<F,STAR,VR > A12_STAR_VR(g);
DistMatrix<F,STAR,MR > A12_STAR_MR(g);
DistMatrix<F,MC, STAR> A21_MC_STAR(g);
DistMatrix<F,STAR,VR > B1_STAR_VR(g);
DistMatrix<F,STAR,MR > B1_STAR_MR(g);
DistMatrix<Int,STAR,STAR> p1_STAR_STAR(g);
// In case B's columns are not aligned with A's
const bool BAligned = ( B.ColShift() == A.ColShift() );
DistMatrix<F,MC,STAR> A21_MC_STAR_B(g);
// Pivot composition
std::vector<Int> image, preimage;
// Start the algorithm
PartitionDownDiagonal
( A, ATL, ATR,
ABL, ABR, 0 );
PartitionDown
( B, BT,
BB, 0 );
while( ATL.Height() < A.Height() && ATL.Width() < A.Width() )
{
RepartitionDownDiagonal
( ATL, /**/ ATR, A00, /**/ A01, A02,
/*************/ /******************/
/**/ A10, /**/ A11, A12,
ABL, /**/ ABR, A20, /**/ A21, A22 );
RepartitionDown
( BT, B0,
/**/ /**/
B1,
BB, B2 );
View2x1
( APan, A12,
A22 );
A12_STAR_VR.AlignWith( A22 );
A12_STAR_MR.AlignWith( A22 );
A21_MC_STAR.AlignWith( A22 );
B1_STAR_VR.AlignWith( B1 );
B1_STAR_MR.AlignWith( B1 );
if( ! BAligned )
A21_MC_STAR_B.AlignWith( B2 );
p1_STAR_STAR.ResizeTo( A11.Height(), 1 );
//--------------------------------------------------------------------//
A11_STAR_STAR = A11;
A21_MC_STAR = A21;
lu::Panel( A11_STAR_STAR, A21_MC_STAR, p1_STAR_STAR, A00.Height() );
ComposePivots( p1_STAR_STAR, A00.Height(), image, preimage );
ApplyRowPivots( APan, image, preimage );
ApplyRowPivots( BB, image, preimage );
A12_STAR_VR = A12;
B1_STAR_VR = B1;
LocalTrsm
( LEFT, LOWER, NORMAL, UNIT, F(1), A11_STAR_STAR, A12_STAR_VR );
LocalTrsm( LEFT, LOWER, NORMAL, UNIT, F(1), A11_STAR_STAR, B1_STAR_VR );
A12_STAR_MR = A12_STAR_VR;
B1_STAR_MR = B1_STAR_VR;
LocalGemm( NORMAL, NORMAL, F(-1), A21_MC_STAR, A12_STAR_MR, F(1), A22 );
if( BAligned )
{
LocalGemm
( NORMAL, NORMAL, F(-1), A21_MC_STAR, B1_STAR_MR, F(1), B2 );
}
else
{
A21_MC_STAR_B = A21_MC_STAR;
LocalGemm
( NORMAL, NORMAL, F(-1), A21_MC_STAR_B, B1_STAR_MR, F(1), B2 );
}
A11 = A11_STAR_STAR;
A12 = A12_STAR_MR;
B1 = B1_STAR_MR;
//--------------------------------------------------------------------//
SlidePartitionDownDiagonal
( ATL, /**/ ATR, A00, A01, /**/ A02,
/**/ A10, A11, /**/ A12,
/*************/ /******************/
ABL, /**/ ABR, A20, A21, /**/ A22 );
SlidePartitionDown
( BT, B0,
B1,
/**/ /**/
BB, B2 );
}
}
void IsotropicElasticTerm::get(MElement *ele, int npts, IntPt *GP, fullMatrix<double> &m) const
{
if(ele->getParent()) ele = ele->getParent();
if(sym)
{
int nbFF = BilinearTerm<SVector3, SVector3>::space1.getNumKeys(ele);
double jac[3][3];
fullMatrix<double> B(6, nbFF);
fullMatrix<double> BTH(nbFF, 6);
fullMatrix<double> BT(nbFF, 6);
m.resize(nbFF, nbFF);
m.setAll(0.);
//std::cout << m.size1() << " " << m.size2() << std::endl;
for(int i = 0; i < npts; i++)
{
const double u = GP[i].pt[0]; const double v = GP[i].pt[1]; const double w = GP[i].pt[2];
const double weight = GP[i].weight; const double detJ = ele->getJacobian(u, v, w, jac);
std::vector<TensorialTraits<SVector3>::GradType> Grads;
BilinearTerm<SVector3, SVector3>::space1.gradf(ele, u, v, w, Grads); // a optimiser ??
for(int j = 0; j < nbFF; j++)
{
BT(j, 0) = B(0, j) = Grads[j](0, 0);
BT(j, 1) = B(1, j) = Grads[j](1, 1);
BT(j, 2) = B(2, j) = Grads[j](2, 2);
BT(j, 3) = B(3, j) = Grads[j](0, 1) + Grads[j](1, 0);
BT(j, 4) = B(4, j) = Grads[j](1, 2) + Grads[j](2, 1);
BT(j, 5) = B(5, j) = Grads[j](0, 2) + Grads[j](2, 0);
}
BTH.setAll(0.);
BTH.gemm(BT, H);
m.gemm(BTH, B, weight * detJ, 1.);
}
}
else
{
int nbFF1 = BilinearTerm<SVector3, SVector3>::space1.getNumKeys(ele);
int nbFF2 = BilinearTerm<SVector3, SVector3>::space2.getNumKeys(ele);
double jac[3][3];
fullMatrix<double> B(6, nbFF2);
fullMatrix<double> BTH(nbFF2, 6);
fullMatrix<double> BT(nbFF1, 6);
m.resize(nbFF1, nbFF2);
m.setAll(0.);
// Sum on Gauss Points i
for(int i = 0; i < npts; i++)
{
const double u = GP[i].pt[0]; const double v = GP[i].pt[1]; const double w = GP[i].pt[2];
const double weight = GP[i].weight; const double detJ = ele->getJacobian(u, v, w, jac);
std::vector<TensorialTraits<SVector3>::GradType> Grads;// tableau de matrices...
std::vector<TensorialTraits<SVector3>::GradType> GradsT;// tableau de matrices...
BilinearTerm<SVector3, SVector3>::space1.gradf(ele, u, v, w, Grads);
BilinearTerm<SVector3, SVector3>::space2.gradf(ele, u, v, w, GradsT);
for(int j = 0; j < nbFF1; j++)
{
BT(j, 0) = Grads[j](0, 0);
BT(j, 1) = Grads[j](1, 1);
BT(j, 2) = Grads[j](2, 2);
BT(j, 3) = Grads[j](0, 1) + Grads[j](1, 0);
BT(j, 4) = Grads[j](1, 2) + Grads[j](2, 1);
BT(j, 5) = Grads[j](0, 2) + Grads[j](2, 0);
}
for(int j = 0; j < nbFF2; j++)
{
B(0, j) = GradsT[j](0, 0);
B(1, j) = GradsT[j](1, 1);
B(2, j) = GradsT[j](2, 2);
B(3, j) = GradsT[j](0, 1) + GradsT[j](1, 0);
B(4, j) = GradsT[j](1, 2) + GradsT[j](2, 1);
B(5, j) = GradsT[j](0, 2) + GradsT[j](2, 0);
}
BTH.setAll(0.);
BTH.gemm(BT, H);
// gemm add the product to m so there is a sum on gauss' points here
m.gemm(BTH, B, weight * detJ, 1.);
}
}
}
