Esempio n. 1
0
inline void
LVar2( DistMatrix<F>& A )
{
#ifndef RELEASE
    CallStackEntry entry("cholesky::LVar2");
    if( A.Height() != A.Width() )
        LogicError("Can only compute Cholesky factor of square matrices");
#endif
    const Grid& g = A.Grid();

    // Matrix views
    DistMatrix<F> 
        ATL(g), ATR(g),   A00(g), A01(g), A02(g),
        ABL(g), ABR(g),   A10(g), A11(g), A12(g),
                          A20(g), A21(g), A22(g);

    // Temporary distributions
    DistMatrix<F,MR,  STAR> A10Adj_MR_STAR(g);
    DistMatrix<F,STAR,STAR> A11_STAR_STAR(g);
    DistMatrix<F,VC,  STAR> A21_VC_STAR(g);
    DistMatrix<F,MC,  STAR> X11_MC_STAR(g);
    DistMatrix<F,MC,  STAR> X21_MC_STAR(g);

    // Start the algorithm
    PartitionDownDiagonal
    ( A, ATL, ATR,
         ABL, ABR, 0 );
    while( ATL.Height() < A.Height() )
    {
        RepartitionDownDiagonal
        ( ATL, /**/ ATR,  A00, /**/ A01, A02,
         /*************/ /******************/
               /**/       A10, /**/ A11, A12,
          ABL, /**/ ABR,  A20, /**/ A21, A22 );

        A10Adj_MR_STAR.AlignWith( A10 );
        X11_MC_STAR.AlignWith( A10 );
        X21_MC_STAR.AlignWith( A20 );
        //--------------------------------------------------------------------//
        A10Adj_MR_STAR.AdjointFrom( A10 );
        LocalGemm( NORMAL, NORMAL, F(1), A10, A10Adj_MR_STAR, X11_MC_STAR );
        A11.SumScatterUpdate( F(-1), X11_MC_STAR );

        A11_STAR_STAR = A11;
        LocalCholesky( LOWER, A11_STAR_STAR );
        A11 = A11_STAR_STAR;

        LocalGemm( NORMAL, NORMAL, F(1), A20, A10Adj_MR_STAR, X21_MC_STAR );
        A21.SumScatterUpdate( F(-1), X21_MC_STAR );

        A21_VC_STAR = A21;
        LocalTrsm
        ( RIGHT, LOWER, ADJOINT, NON_UNIT, F(1), A11_STAR_STAR, A21_VC_STAR );
        A21 = A21_VC_STAR;
        //--------------------------------------------------------------------//

        SlidePartitionDownDiagonal
        ( ATL, /**/ ATR,  A00, A01, /**/ A02,
               /**/       A10, A11, /**/ A12,
         /*************/ /******************/
          ABL, /**/ ABR,  A20, A21, /**/ A22 );
    }
}
Esempio n. 2
0
inline void
CholeskyUVar3( DistMatrix<F>& A )
{
#ifndef RELEASE
    PushCallStack("internal::CholeskyUVar3");
    if( A.Height() != A.Width() )
        throw std::logic_error
        ("Can only compute Cholesky factor of square matrices");
#endif
    const Grid& g = A.Grid();

    // Matrix views
    DistMatrix<F> 
        ATL(g), ATR(g),  A00(g), A01(g), A02(g),
        ABL(g), ABR(g),  A10(g), A11(g), A12(g),
                         A20(g), A21(g), A22(g);

    // Temporary matrix distributions
    DistMatrix<F,STAR,STAR> A11_STAR_STAR(g);
    DistMatrix<F,STAR,VR  > A12_STAR_VR(g);
    DistMatrix<F,STAR,MC  > A12_STAR_MC(g);
    DistMatrix<F,STAR,MR  > A12_STAR_MR(g);

    // Start the algorithm
    PartitionDownDiagonal
    ( A, ATL, ATR,
         ABL, ABR, 0 ); 
    while( ABR.Height() > 0 )
    {
        RepartitionDownDiagonal
        ( ATL, /**/ ATR,  A00, /**/ A01, A02,
         /*************/ /******************/
               /**/       A10, /**/ A11, A12,
          ABL, /**/ ABR,  A20, /**/ A21, A22 );

        A12_STAR_MC.AlignWith( A22 );
        A12_STAR_MR.AlignWith( A22 );
        A12_STAR_VR.AlignWith( A22 );
        //--------------------------------------------------------------------//
        A11_STAR_STAR = A11;
        LocalCholesky( UPPER, A11_STAR_STAR );
        A11 = A11_STAR_STAR;

        A12_STAR_VR = A12;
        LocalTrsm
        ( LEFT, UPPER, ADJOINT, NON_UNIT, F(1), A11_STAR_STAR, A12_STAR_VR );

        A12_STAR_MC = A12_STAR_VR;
        A12_STAR_MR = A12_STAR_VR;
        LocalTrrk
        ( UPPER, ADJOINT, F(-1), A12_STAR_MC, A12_STAR_MR, F(1), A22 );
        A12 = A12_STAR_MR;
        //--------------------------------------------------------------------//
        A12_STAR_MC.FreeAlignments();
        A12_STAR_MR.FreeAlignments();
        A12_STAR_VR.FreeAlignments();

        SlidePartitionDownDiagonal
        ( ATL, /**/ ATR,  A00, A01, /**/ A02,
               /**/       A10, A11, /**/ A12,
         /*************/ /******************/
          ABL, /**/ ABR,  A20, A21, /**/ A22 );
    }
#ifndef RELEASE
    PopCallStack();
#endif
}
Esempio n. 3
0
inline void
HPDInverseLVar2( DistMatrix<F>& A )
{
#ifndef RELEASE
    PushCallStack("internal::HPDInverseLVar2");
    if( A.Height() != A.Width() )
        throw std::logic_error("Nonsquare matrices cannot be triangular");
#endif
    const Grid& g = A.Grid();

    // Matrix views
    DistMatrix<F> 
        ATL(g), ATR(g),  A00(g), A01(g), A02(g),
        ABL(g), ABR(g),  A10(g), A11(g), A12(g),
                         A20(g), A21(g), A22(g);

    // Temporary distributions
    DistMatrix<F,STAR,STAR> A11_STAR_STAR(g);
    DistMatrix<F,STAR,VR  > A10_STAR_VR(g);
    DistMatrix<F,VC,  STAR> A21_VC_STAR(g);
    DistMatrix<F,STAR,MC  > A10_STAR_MC(g);
    DistMatrix<F,STAR,MR  > A10_STAR_MR(g);
    DistMatrix<F,STAR,MC  > A21Trans_STAR_MC(g);
    DistMatrix<F,VR,  STAR> A21_VR_STAR(g);
    DistMatrix<F,STAR,MR  > A21Adj_STAR_MR(g);

    // Start the algorithm
    PartitionDownDiagonal
    ( A, ATL, ATR,
         ABL, ABR, 0 );
    while( ATL.Height() < A.Height() )
    {
        RepartitionDownDiagonal
        ( ATL, /**/ ATR,  A00, /**/ A01, A02,
         /*************/ /******************/
               /**/       A10, /**/ A11, A12,
          ABL, /**/ ABR,  A20, /**/ A21, A22 );

        A10_STAR_VR.AlignWith( A00 );
        A21_VC_STAR.AlignWith( A20 );
        A10_STAR_MC.AlignWith( A00 );
        A10_STAR_MR.AlignWith( A00 );
        A21Trans_STAR_MC.AlignWith( A20 );
        A21_VR_STAR.AlignWith( A22 );
        A21Adj_STAR_MR.AlignWith( A22 );
        //--------------------------------------------------------------------//
        A11_STAR_STAR = A11;
        LocalCholesky( LOWER, A11_STAR_STAR );

        A10_STAR_VR = A10;
        LocalTrsm
        ( LEFT, LOWER, NORMAL, NON_UNIT, F(1), A11_STAR_STAR, A10_STAR_VR );

        A21_VC_STAR = A21;
        LocalTrsm
        ( RIGHT, LOWER, ADJOINT, NON_UNIT, F(1), A11_STAR_STAR, A21_VC_STAR );

        A10_STAR_MC = A10_STAR_VR;
        A10_STAR_MR = A10_STAR_VR;
        LocalTrrk
        ( LOWER, ADJOINT,
          F(1), A10_STAR_MC, A10_STAR_MR, F(1), A00 );

        A21Trans_STAR_MC.TransposeFrom( A21_VC_STAR );
        LocalGemm
        ( TRANSPOSE, NORMAL, F(-1), A21Trans_STAR_MC, A10_STAR_MR, F(1), A20 );

        A21_VR_STAR = A21_VC_STAR;
        A21Adj_STAR_MR.AdjointFrom( A21_VR_STAR );
        LocalTrrk
        ( LOWER, TRANSPOSE,
          F(-1), A21Trans_STAR_MC, A21Adj_STAR_MR, F(1), A22 );

        LocalTrsm
        ( LEFT, LOWER, ADJOINT, NON_UNIT, F(1), A11_STAR_STAR, A10_STAR_VR );

        LocalTrsm
        ( RIGHT, LOWER, NORMAL, NON_UNIT, F(-1), A11_STAR_STAR, A21_VC_STAR );

        LocalTriangularInverse( LOWER, NON_UNIT, A11_STAR_STAR );

        LocalTrtrmm( ADJOINT, LOWER, A11_STAR_STAR );

        A11 = A11_STAR_STAR;
        A10 = A10_STAR_VR;
        A21 = A21_VC_STAR;
        //--------------------------------------------------------------------//
        A10_STAR_VR.FreeAlignments();
        A21_VC_STAR.FreeAlignments();
        A10_STAR_MC.FreeAlignments();
        A10_STAR_MR.FreeAlignments();
        A21Trans_STAR_MC.FreeAlignments();
        A21_VR_STAR.FreeAlignments();
        A21Adj_STAR_MR.FreeAlignments();

        SlidePartitionDownDiagonal
        ( ATL, /**/ ATR,  A00, A01, /**/ A02,
               /**/       A10, A11, /**/ A12,
         /*************/ /******************/
          ABL, /**/ ABR,  A20, A21, /**/ A22 );
    }
#ifndef RELEASE
    PopCallStack();
#endif
}
Esempio n. 4
0
inline void
CholeskyLVar2Naive( DistMatrix<F>& A )
{
#ifndef RELEASE
    PushCallStack("internal::CholeskyLVar2Naive");
    if( A.Height() != A.Width() )
        throw std::logic_error
        ("Can only compute Cholesky factor of square matrices");
    if( A.Grid().VCRank() == 0 )
    {
        std::cout 
            << "CholeskyLVar2Naive exists solely for academic purposes. Please "
               "use CholeskyLVar2 in real applications." << std::endl;
    }
#endif
    const Grid& g = A.Grid();

    // Matrix views
    DistMatrix<F> 
        ATL(g), ATR(g),   A00(g), A01(g), A02(g),
        ABL(g), ABR(g),   A10(g), A11(g), A12(g),
                          A20(g), A21(g), A22(g);

    // Temporary distributions
    DistMatrix<F,STAR,MR  > A10_STAR_MR(g);
    DistMatrix<F,STAR,STAR> A11_STAR_STAR(g);
    DistMatrix<F,VC,  STAR> A21_VC_STAR(g);
    DistMatrix<F,MC,  STAR> X11_MC_STAR(g);
    DistMatrix<F,MC,  STAR> X21_MC_STAR(g);

    // Start the algorithm
    PartitionDownDiagonal
    ( A, ATL, ATR,
         ABL, ABR, 0 );
    while( ATL.Height() < A.Height() )
    {
        RepartitionDownDiagonal
        ( ATL, /**/ ATR,  A00, /**/ A01, A02,
         /*************/ /******************/
               /**/       A10, /**/ A11, A12,
          ABL, /**/ ABR,  A20, /**/ A21, A22 );

        A10_STAR_MR.AlignWith( A10 );
        X11_MC_STAR.AlignWith( A10 );
        X21_MC_STAR.AlignWith( A20 );
        X11_MC_STAR.ResizeTo( A11.Height(), A11.Width() );
        X21_MC_STAR.ResizeTo( A21.Height(), A21.Width() );
        //--------------------------------------------------------------------//
        A10_STAR_MR = A10;
        LocalGemm( NORMAL, ADJOINT, F(1), A10, A10_STAR_MR, F(0), X11_MC_STAR );
        A11.SumScatterUpdate( F(-1), X11_MC_STAR );

        A11_STAR_STAR = A11;
        LocalCholesky( LOWER, A11_STAR_STAR );
        A11 = A11_STAR_STAR;

        LocalGemm( NORMAL, ADJOINT, F(1), A20, A10_STAR_MR, F(0), X21_MC_STAR );
        A21.SumScatterUpdate( F(-1), X21_MC_STAR );

        A21_VC_STAR = A21;
        LocalTrsm
        ( RIGHT, LOWER, ADJOINT, NON_UNIT, F(1), A11_STAR_STAR, A21_VC_STAR );
        A21 = A21_VC_STAR;
        //--------------------------------------------------------------------//
        A10_STAR_MR.FreeAlignments();
        X11_MC_STAR.FreeAlignments();
        X21_MC_STAR.FreeAlignments();

        SlidePartitionDownDiagonal
        ( ATL, /**/ ATR,  A00, A01, /**/ A02,
               /**/       A10, A11, /**/ A12,
         /*************/ /******************/
          ABL, /**/ ABR,  A20, A21, /**/ A22 );
    }
#ifndef RELEASE
    PopCallStack();
#endif
}