Exemple #1
0
inline void
ApplyPackedReflectorsLUHB
( Conjugation conjugation, int offset, 
  const DistMatrix<Complex<R> >& H,
  const DistMatrix<Complex<R>,MD,STAR>& t,
        DistMatrix<Complex<R> >& A )
{
#ifndef RELEASE
    PushCallStack("internal::ApplyPackedReflectorsLUHB");
    if( H.Grid() != t.Grid() || t.Grid() != A.Grid() )
        throw std::logic_error
        ("{H,t,A} must be distributed over the same grid");
    if( offset < 0 || offset > H.Width() )
        throw std::logic_error("Transforms out of bounds");
    if( H.Width() != A.Height() )
        throw std::logic_error
        ("Width of transforms must equal height of target matrix");
    if( t.Height() != H.DiagonalLength( offset ) )
        throw std::logic_error("t must be the same length as H's offset diag");
    if( !t.AlignedWithDiagonal( H, offset ) )
        throw std::logic_error("t must be aligned with H's offset diagonal");
#endif
    typedef Complex<R> C;
    const Grid& g = H.Grid();

    DistMatrix<C>
        HTL(g), HTR(g),  H00(g), H01(g), H02(g),  HPan(g), HPanCopy(g),
        HBL(g), HBR(g),  H10(g), H11(g), H12(g),
                         H20(g), H21(g), H22(g);
    DistMatrix<C> ABottom(g);
    DistMatrix<C,MD,STAR>
        tT(g),  t0(g),
        tB(g),  t1(g),
                t2(g);

    DistMatrix<C,STAR,VR  > HPan_STAR_VR(g);
    DistMatrix<C,STAR,MC  > HPan_STAR_MC(g);
    DistMatrix<C,STAR,STAR> t1_STAR_STAR(g);
    DistMatrix<C,STAR,STAR> SInv_STAR_STAR(g);
    DistMatrix<C,STAR,MR  > Z_STAR_MR(g);
    DistMatrix<C,STAR,VR  > Z_STAR_VR(g);

    LockedPartitionUpDiagonal
    ( H, HTL, HTR,
         HBL, HBR, 0 );
    LockedPartitionUp
    ( t, tT,
         tB, 0 );
    while( HBR.Height() < H.Height() && HBR.Width() < H.Width() )
    {
        LockedRepartitionUpDiagonal
        ( HTL, /**/ HTR,  H00, H01, /**/ H02,
               /**/       H10, H11, /**/ H12,
         /*************/ /******************/
          HBL, /**/ HBR,  H20, H21, /**/ H22 );
    
        const int HPanWidth = H11.Width() + H12.Width();
        const int HPanHeight = 
            std::min( H11.Height(), std::max(HPanWidth-offset,0) );
        const int leftover = A.Height()-HPanWidth;
        HPan.LockedView( H, H00.Height(), H00.Width(), HPanHeight, HPanWidth );

        LockedRepartitionUp
        ( tT,  t0,
               t1,
         /**/ /**/
          tB,  t2, HPanHeight );

        ABottom.View( A, leftover, 0, HPanWidth, A.Width() );

        HPan_STAR_MC.AlignWith( ABottom );
        Z_STAR_MR.AlignWith( ABottom );
        Z_STAR_VR.AlignWith( ABottom );
        Zeros( HPanHeight, ABottom.Width(), Z_STAR_MR );
        Zeros( HPanHeight, HPanHeight, SInv_STAR_STAR );
        //--------------------------------------------------------------------//
        HPanCopy = HPan;
        MakeTrapezoidal( LEFT, UPPER, offset, HPanCopy );
        SetDiagonalToOne( LEFT, offset, HPanCopy );

        HPan_STAR_VR = HPanCopy;
        Herk
        ( UPPER, NORMAL,
          C(1), HPan_STAR_VR.LockedLocalMatrix(),
          C(0), SInv_STAR_STAR.LocalMatrix() );
        SInv_STAR_STAR.SumOverGrid();
        t1_STAR_STAR = t1;
        FixDiagonal( conjugation, t1_STAR_STAR, SInv_STAR_STAR );

        HPan_STAR_MC = HPan_STAR_VR;
        LocalGemm
        ( NORMAL, NORMAL, C(1), HPan_STAR_MC, ABottom, C(0), Z_STAR_MR );
        Z_STAR_VR.SumScatterFrom( Z_STAR_MR );

        LocalTrsm
        ( LEFT, UPPER, NORMAL, NON_UNIT, C(1), SInv_STAR_STAR, Z_STAR_VR );

        Z_STAR_MR = Z_STAR_VR;
        LocalGemm
        ( ADJOINT, NORMAL, C(-1), HPan_STAR_MC, Z_STAR_MR, C(1), ABottom );
        //--------------------------------------------------------------------//
        HPan_STAR_MC.FreeAlignments();
        Z_STAR_MR.FreeAlignments();
        Z_STAR_VR.FreeAlignments();

        SlideLockedPartitionUpDiagonal
        ( HTL, /**/ HTR,  H00, /**/ H01, H02,
         /*************/ /******************/
               /**/       H10, /**/ H11, H12,
          HBL, /**/ HBR,  H20, /**/ H21, H22 );

        SlideLockedPartitionUp
        ( tT,  t0,
         /**/ /**/
               t1,
          tB,  t2 );
    }
#ifndef RELEASE
    PopCallStack();
#endif
}
Exemple #2
0
inline void
RLVB
( Conjugation conjugation, int offset, 
  const DistMatrix<Complex<R> >& H,
  const DistMatrix<Complex<R>,MD,STAR>& t,
        DistMatrix<Complex<R> >& A )
{
#ifndef RELEASE
    PushCallStack("apply_packed_reflectors::RLVB");
    if( H.Grid() != t.Grid() || t.Grid() != A.Grid() )
        throw std::logic_error
        ("{H,t,A} must be distributed over the same grid");
    if( offset > 0 || offset < -H.Height() )
        throw std::logic_error("Transforms out of bounds");
    if( H.Height() != A.Width() )
        throw std::logic_error
        ("Height of transforms must equal width of target matrix");
    if( t.Height() != H.DiagonalLength( offset ) )
        throw std::logic_error("t must be the same length as H's offset diag");
    if( !t.AlignedWithDiagonal( H, offset ) )
        throw std::logic_error("t must be aligned with H's 'offset' diagonal");
#endif
    typedef Complex<R> C;
    const Grid& g = H.Grid();

    DistMatrix<C>
        HTL(g), HTR(g),  H00(g), H01(g), H02(g),  HPan(g), HPanCopy(g),
        HBL(g), HBR(g),  H10(g), H11(g), H12(g),
                         H20(g), H21(g), H22(g);
    DistMatrix<C> ARight(g);
    DistMatrix<C,MD,STAR>
        tT(g),  t0(g),
        tB(g),  t1(g),
                t2(g);
    
    DistMatrix<C,VC,  STAR> HPan_VC_STAR(g);
    DistMatrix<C,MR,  STAR> HPan_MR_STAR(g);
    DistMatrix<C,STAR,STAR> t1_STAR_STAR(g);
    DistMatrix<C,STAR,STAR> SInv_STAR_STAR(g);
    DistMatrix<C,STAR,MC  > ZAdj_STAR_MC(g);
    DistMatrix<C,STAR,VC  > ZAdj_STAR_VC(g);

    LockedPartitionUpDiagonal
    ( H, HTL, HTR,
         HBL, HBR, 0 );
    LockedPartitionUp
    ( t, tT,
         tB, 0 );
    while( HBR.Height() < H.Height() && HBR.Width() < H.Width() )
    {
        LockedRepartitionUpDiagonal
        ( HTL, /**/ HTR,  H00, H01, /**/ H02,
               /**/       H10, H11, /**/ H12,
         /*************/ /******************/
          HBL, /**/ HBR,  H20, H21, /**/ H22 );

        const int HPanHeight = H11.Height() + H21.Height();
        const int HPanWidth = 
            std::min( H11.Width(), std::max(HPanHeight+offset,0) );
        const int leftover = A.Width()-HPanHeight;
        LockedView( HPan, H, H00.Height(), H00.Width(), HPanHeight, HPanWidth );

        LockedRepartitionUp
        ( tT,  t0,
               t1,
         /**/ /**/
          tB,  t2, HPanWidth );

        View( ARight, A, 0, leftover, A.Height(), HPanHeight );

        HPan_MR_STAR.AlignWith( ARight );
        ZAdj_STAR_MC.AlignWith( ARight );
        ZAdj_STAR_VC.AlignWith( ARight );
        Zeros( HPan.Width(), ARight.Height(), ZAdj_STAR_MC );
        Zeros( HPan.Width(), HPan.Width(), SInv_STAR_STAR );
        //--------------------------------------------------------------------//
        HPanCopy = HPan;
        MakeTrapezoidal( LEFT, LOWER, offset, HPanCopy );
        SetDiagonal( LEFT, offset, HPanCopy, C(1) );

        HPan_VC_STAR = HPanCopy;
        Herk
        ( LOWER, ADJOINT, 
          C(1), HPan_VC_STAR.LockedMatrix(),
          C(0), SInv_STAR_STAR.Matrix() );     
        SInv_STAR_STAR.SumOverGrid();
        t1_STAR_STAR = t1;
        FixDiagonal( conjugation, t1_STAR_STAR, SInv_STAR_STAR );

        HPan_MR_STAR = HPan_VC_STAR;
        LocalGemm
        ( ADJOINT, ADJOINT,
          C(1), HPan_MR_STAR, ARight, C(0), ZAdj_STAR_MC );
        ZAdj_STAR_VC.SumScatterFrom( ZAdj_STAR_MC );
 
        LocalTrsm
        ( LEFT, LOWER, ADJOINT, NON_UNIT, C(1), SInv_STAR_STAR, ZAdj_STAR_VC );

        ZAdj_STAR_MC = ZAdj_STAR_VC;
        LocalGemm
        ( ADJOINT, ADJOINT, C(-1), ZAdj_STAR_MC, HPan_MR_STAR, C(1), ARight );
        //--------------------------------------------------------------------//
        HPan_MR_STAR.FreeAlignments();
        ZAdj_STAR_MC.FreeAlignments();
        ZAdj_STAR_VC.FreeAlignments();

        SlideLockedPartitionUpDiagonal
        ( HTL, /**/ HTR,  H00, /**/ H01, H02,
         /*************/ /******************/
               /**/       H10, /**/ H11, H12,
          HBL, /**/ HBR,  H20, /**/ H21, H22 );

        SlideLockedPartitionUp
        ( tT,  t0,
         /**/ /**/
               t1,
          tB,  t2 );
    }
#ifndef RELEASE
    PopCallStack();
#endif
}
Exemple #3
0
inline void
ApplyPackedReflectorsLUHB
( Conjugation conjugation, int offset, 
  const Matrix<Complex<R> >& H,
  const Matrix<Complex<R> >& t,
        Matrix<Complex<R> >& A )
{
#ifndef RELEASE
    PushCallStack("internal::ApplyPackedReflectorsLUHB");
    if( offset < 0 || offset > H.Width() )
        throw std::logic_error("Transforms out of bounds");
    if( H.Width() != A.Height() )
        throw std::logic_error
        ("Width of transforms must equal height of target matrix");
    if( t.Height() != H.DiagonalLength( offset ) )
        throw std::logic_error("t must be the same length as H's offset diag");
#endif
    typedef Complex<R> C;

    Matrix<C>
        HTL, HTR,  H00, H01, H02,  HPan, HPanCopy,
        HBL, HBR,  H10, H11, H12,
                   H20, H21, H22;
    Matrix<C> ABottom;
    Matrix<C>
        tT,  t0,
        tB,  t1,
             t2;

    Matrix<C> SInv, Z;

    LockedPartitionUpDiagonal
    ( H, HTL, HTR,
         HBL, HBR, 0 );
    LockedPartitionUp
    ( t, tT,
         tB, 0 );
    while( HBR.Height() < H.Height() && HBR.Width() < H.Width() )
    {
        LockedRepartitionUpDiagonal
        ( HTL, /**/ HTR,  H00, H01, /**/ H02,
               /**/       H10, H11, /**/ H12,
         /*************/ /******************/
          HBL, /**/ HBR,  H20, H21, /**/ H22 );
    
        const int HPanWidth = H11.Width() + H12.Width();
        const int HPanHeight = 
            std::min( H11.Height(), std::max(HPanWidth-offset,0) );
        const int leftover = A.Height()-HPanWidth;
        HPan.LockedView( H, H00.Height(), H00.Width(), HPanHeight, HPanWidth );

        LockedRepartitionUp
        ( tT,  t0,
               t1,
         /**/ /**/
          tB,  t2, HPanHeight );

        ABottom.View( A, leftover, 0, HPanWidth, A.Width() );

        Zeros( HPanHeight, ABottom.Width(), Z );
        Zeros( HPanHeight, HPanHeight, SInv );
        //--------------------------------------------------------------------//
        HPanCopy = HPan;
        MakeTrapezoidal( LEFT, UPPER, offset, HPanCopy );
        SetDiagonalToOne( LEFT, offset, HPanCopy );

        Herk( UPPER, NORMAL, C(1), HPanCopy, C(0), SInv );
        FixDiagonal( conjugation, t1, SInv );

        Gemm( NORMAL, NORMAL, C(1), HPanCopy, ABottom, C(0), Z );
        Trsm( LEFT, UPPER, NORMAL, NON_UNIT, C(1), SInv, Z );
        Gemm( ADJOINT, NORMAL, C(-1), HPanCopy, Z, C(1), ABottom );
        //--------------------------------------------------------------------//

        SlideLockedPartitionUpDiagonal
        ( HTL, /**/ HTR,  H00, /**/ H01, H02,
         /*************/ /******************/
               /**/       H10, /**/ H11, H12,
          HBL, /**/ HBR,  H20, /**/ H21, H22 );

        SlideLockedPartitionUp
        ( tT,  t0,
         /**/ /**/
               t1,
          tB,  t2 );
    }
#ifndef RELEASE
    PopCallStack();
#endif
}
Exemple #4
0
inline void
RLVB
( Conjugation conjugation, int offset, 
  const Matrix<Complex<R> >& H,
  const Matrix<Complex<R> >& t,
        Matrix<Complex<R> >& A )
{
#ifndef RELEASE
    PushCallStack("apply_packed_reflectors::RLVB");
    if( offset > 0 || offset < -H.Height() )
        throw std::logic_error("Transforms out of bounds");
    if( H.Height() != A.Width() )
        throw std::logic_error
        ("Height of transforms must equal width of target matrix");
    if( t.Height() != H.DiagonalLength( offset ) )
        throw std::logic_error("t must be the same length as H's offset diag");
#endif
    typedef Complex<R> C;

    Matrix<C>
        HTL, HTR,  H00, H01, H02,  HPan, HPanCopy,
        HBL, HBR,  H10, H11, H12,
                   H20, H21, H22;
    Matrix<C> ARight;
    Matrix<C>
        tT,  t0,
        tB,  t1,
             t2;
    
    Matrix<C> SInv, Z;

    LockedPartitionUpDiagonal
    ( H, HTL, HTR,
         HBL, HBR, 0 );
    LockedPartitionUp
    ( t, tT,
         tB, 0 );
    while( HBR.Height() < H.Height() && HBR.Width() < H.Width() )
    {
        LockedRepartitionUpDiagonal
        ( HTL, /**/ HTR,  H00, H01, /**/ H02,
               /**/       H10, H11, /**/ H12,
         /*************/ /******************/
          HBL, /**/ HBR,  H20, H21, /**/ H22 );

        const int HPanHeight = H11.Height() + H21.Height();
        const int HPanWidth = 
            std::min( H11.Width(), std::max(HPanHeight+offset,0) );
        const int leftover = A.Width()-HPanHeight;
        LockedView( HPan, H, H00.Height(), H00.Width(), HPanHeight, HPanWidth );

        LockedRepartitionUp
        ( tT,  t0,
               t1,
         /**/ /**/
          tB,  t2, HPanWidth );

        View( ARight, A, 0, leftover, A.Height(), HPanHeight );

        Zeros( ARight.Height(), HPan.Width(), Z );
        Zeros( HPan.Width(), HPan.Width(), SInv );
        //--------------------------------------------------------------------//
        HPanCopy = HPan;
        MakeTrapezoidal( LEFT, LOWER, offset, HPanCopy );
        SetDiagonal( LEFT, offset, HPanCopy, C(1) );

        Herk( LOWER, ADJOINT, C(1), HPanCopy, C(0), SInv );
        FixDiagonal( conjugation, t1, SInv );

        Gemm( NORMAL, NORMAL, C(1), ARight, HPanCopy, C(0), Z );
        Trsm( RIGHT, LOWER, NORMAL, NON_UNIT, C(1), SInv, Z );
        Gemm( NORMAL, ADJOINT, C(-1), Z, HPanCopy, C(1), ARight );
        //--------------------------------------------------------------------//

        SlideLockedPartitionUpDiagonal
        ( HTL, /**/ HTR,  H00, /**/ H01, H02,
         /*************/ /******************/
               /**/       H10, /**/ H11, H12,
          HBL, /**/ HBR,  H20, /**/ H21, H22 );

        SlideLockedPartitionUp
        ( tT,  t0,
         /**/ /**/
               t1,
          tB,  t2 );
    }
#ifndef RELEASE
    PopCallStack();
#endif
}