void DenseLinAlgPack::M_StMtInvM(DMatrixSlice* gms_lhs, value_type alpha, const DMatrixSlice& gms_rhs1
, BLAS_Cpp::Transp trans_rhs1, const DMatrixSliceTri& tri_rhs2
, BLAS_Cpp::Transp trans_rhs2)
{
Mp_MtM_assert_sizes( gms_lhs->rows(), gms_lhs->cols(), BLAS_Cpp::no_trans
, gms_rhs1.rows(), gms_rhs1.cols(), trans_rhs1
, tri_rhs2.gms().rows(), tri_rhs2.gms().cols(), trans_rhs2 );
i_trsm_alt(BLAS_Cpp::right,alpha,tri_rhs2,trans_rhs2,gms_rhs1,trans_rhs1,gms_lhs);
}
void DenseLinAlgPack::M_StInvMtM(DMatrix* gm_lhs, value_type alpha, const DMatrixSliceTri& tri_rhs1
, BLAS_Cpp::Transp trans_rhs1, const DMatrixSlice& gms_rhs2
, BLAS_Cpp::Transp trans_rhs2)
{
MtM_assert_sizes( tri_rhs1.gms().rows(), tri_rhs1.gms().cols(), trans_rhs1
, gms_rhs2.rows(), gms_rhs2.cols(), trans_rhs2 );
gm_lhs->resize( rows(tri_rhs1.gms().rows(), tri_rhs1.gms().cols(), trans_rhs1)
, cols(gms_rhs2.rows(), gms_rhs2.cols(), trans_rhs2) );
i_trsm_alt(BLAS_Cpp::left,alpha,tri_rhs1,trans_rhs1,gms_rhs2,trans_rhs2,&(*gm_lhs)());
}
void DenseLinAlgPack::V_InvMtV(DVectorSlice* vs_lhs, const DMatrixSliceTri& tri_rhs1, BLAS_Cpp::Transp trans_rhs1
, const DVectorSlice& vs_rhs2)
{
assert_gms_square(tri_rhs1.gms());
MtV_assert_sizes(tri_rhs1.gms().rows(), tri_rhs1.gms().cols(), trans_rhs1, vs_rhs2.dim());
Vp_V_assert_sizes( vs_lhs->dim(), tri_rhs1.gms().rows() );
(*vs_lhs) = vs_rhs2;
BLAS_Cpp::trsv(tri_rhs1.uplo(),trans_rhs1,tri_rhs1.diag(),tri_rhs1.gms().rows()
,tri_rhs1.gms().col_ptr(1),tri_rhs1.gms().max_rows(), vs_lhs->raw_ptr(),vs_lhs->stride());
}
void DenseLinAlgPack::Vp_StMtV(DVectorSlice* vs_lhs, value_type alpha, const DMatrixSliceTri& tri_rhs1
, BLAS_Cpp::Transp trans_rhs1, const DVectorSlice& vs_rhs2, value_type beta)
{
Vp_MtV_assert_sizes(vs_lhs->dim(),tri_rhs1.gms().rows(),tri_rhs1.gms().cols()
,trans_rhs1,vs_rhs2.dim() );
// If op(gms_rhs2) == gms_lhs and beta = 0.0 then this is a direct call to the BLAS.
if( vs_lhs->overlap(vs_rhs2) == SAME_MEM && beta == 0.0 )
{
V_MtV(vs_lhs, tri_rhs1, trans_rhs1, vs_rhs2);
if(alpha != 1.0) Vt_S(vs_lhs,alpha);
}
else {
// This is something else so the alias problem is not handled.
if(beta != 1.0) Vt_S(vs_lhs,beta);
DVector tmp;
V_MtV(&tmp, tri_rhs1, trans_rhs1, vs_rhs2);
Vp_StV(vs_lhs,alpha,tmp());
}
}
void DenseLinAlgPack::Mp_StM(DMatrixSlice* gms_lhs, value_type alpha, const DMatrixSliceTri& tri_rhs
, BLAS_Cpp::Transp trans_rhs)
{
using BLAS_Cpp::trans;
using BLAS_Cpp::no_trans;
using BLAS_Cpp::lower;
using BLAS_Cpp::upper;
Mp_M_assert_sizes(gms_lhs->rows(), gms_lhs->cols(), BLAS_Cpp::no_trans
, tri_rhs.rows(), tri_rhs.cols(), trans_rhs );
// diagonal
if( tri_rhs.diag() == BLAS_Cpp::nonunit )
Vp_StV( &gms_lhs->diag(), alpha, tri_rhs.gms().diag() );
else
Vp_S( &gms_lhs->diag(), alpha );
// upper or lower triangle
const size_type n = gms_lhs->rows(); // same as cols
if( n == 1 )
return; // There is not lower or upper triangular region
const DMatrixSliceTriEle
tri = ( tri_rhs.uplo() == lower
? tri_ele(tri_rhs.gms()(2,n,1,n-1),lower)
: tri_ele(tri_rhs.gms()(1,n-1,2,n),upper) );
const BLAS_Cpp::Uplo
as_uplo = ( ( tri_rhs.uplo() == lower && trans_rhs == no_trans )
|| ( tri_rhs.uplo() == upper && trans_rhs == trans )
? lower : upper );
switch(as_uplo) {
case lower:
Mp_StM( const_cast<DenseLinAlgPack::DMatrixSliceTriEle*>(
&tri_ele((*gms_lhs)(2,n,1,n-1),lower) )
, alpha, tri );
break;
case upper:
Mp_StM( const_cast<DenseLinAlgPack::DMatrixSliceTriEle*>(
&tri_ele((*gms_lhs)(1,n-1,2,n),upper) )
, alpha, tri );
break;
}
}
