FLA_Error FLA_Obj_create_copy_of( FLA_Trans trans, FLA_Obj obj_cur, FLA_Obj *obj_new )
{
// Create a new object conformal to the current object.
FLA_Obj_create_conf_to( trans, obj_cur, obj_new );
#ifdef FLA_ENABLE_SCC
if ( !FLA_is_owner() )
return FLA_SUCCESS;
#endif
// Copy the contents of the current object to the new object.
FLA_Copyt_external( trans, obj_cur, *obj_new );
return FLA_SUCCESS;
}
FLA_Error FLA_Copyt( FLA_Trans trans, FLA_Obj A, FLA_Obj B )
{
FLA_Error r_val;
#ifdef FLA_ENABLE_BLAS1_FRONT_END_CNTL_TREES
// Check parameters.
if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
FLA_Copyt_check( trans, A, B );
// Invoke FLA_Copyt_internal() with flat control tree that simply calls
// external wrapper.
r_val = FLA_Copyt_internal( trans, A, B, fla_copyt_cntl_blas );
#else
r_val = FLA_Copyt_external( trans, A, B );
#endif
return r_val;
}
FLA_Error FLA_Copy_object_to_buffer( FLA_Trans trans, dim_t i, dim_t j, FLA_Obj A, dim_t m, dim_t n, void* B_buffer, dim_t rs, dim_t cs )
{
FLA_Obj B;
FLA_Obj ATL, ATR,
ABL, Aij;
if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
FLA_Copy_object_to_buffer_check( trans, i, j, A, m, n, B_buffer, rs, cs );
FLA_Part_2x2( A, &ATL, &ATR,
&ABL, &Aij, i, j, FLA_TL );
FLA_Obj_create_without_buffer( FLA_Obj_datatype( A ), m, n, &B );
FLA_Obj_attach_buffer( B_buffer, rs, cs, &B );
FLA_Copyt_external( trans, Aij, B );
FLA_Obj_free_without_buffer( &B );
return FLA_SUCCESS;
}
FLA_Error FLA_Copy_buffer_to_object( FLA_Trans trans, dim_t m, dim_t n, void* A_buffer, dim_t rs, dim_t cs, dim_t i, dim_t j, FLA_Obj B )
{
FLA_Obj A;
FLA_Obj BTL, BTR,
BBL, Bij;
if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
FLA_Copy_buffer_to_object_check( trans, m, n, A_buffer, rs, cs, i, j, B );
FLA_Part_2x2( B, &BTL, &BTR,
&BBL, &Bij, i, j, FLA_TL );
FLA_Obj_create_without_buffer( FLA_Obj_datatype( B ), m, n, &A );
FLA_Obj_attach_buffer( A_buffer, rs, cs, &A );
FLA_Copyt_external( trans, A, Bij );
FLA_Obj_free_without_buffer( &A );
return FLA_SUCCESS;
}
FLA_Error FLA_LQ_UT_unb_var2( FLA_Obj A, FLA_Obj T )
{
FLA_Obj ATL, ATR, A00, a01, A02,
ABL, ABR, a10t, alpha11, a12t,
A20, a21, A22;
FLA_Obj TTL, TTR, T00, t01, T02,
TBL, TBR, t10t, tau11, t12t,
T20, t21, T22;
FLA_Part_2x2( A, &ATL, &ATR,
&ABL, &ABR, 0, 0, FLA_TL );
FLA_Part_2x2( T, &TTL, &TTR,
&TBL, &TBR, 0, 0, FLA_TL );
while ( FLA_Obj_min_dim( ABR ) > 0 ){
FLA_Repart_2x2_to_3x3( ATL, /**/ ATR, &A00, /**/ &a01, &A02,
/* ************* */ /* ************************** */
&a10t, /**/ &alpha11, &a12t,
ABL, /**/ ABR, &A20, /**/ &a21, &A22,
1, 1, FLA_BR );
FLA_Repart_2x2_to_3x3( TTL, /**/ TTR, &T00, /**/ &t01, &T02,
/* ************* */ /* ************************ */
&t10t, /**/ &tau11, &t12t,
TBL, /**/ TBR, &T20, /**/ &t21, &T22,
1, 1, FLA_BR );
/*------------------------------------------------------------*/
// Compute tau11 and u12t from alpha11 and a12t such that tau11 and u12t
// determine a Householder transform H such that applying H from the
// right to the row vector consisting of alpha11 and a12t annihilates
// the entries in a12t (and updates alpha11).
FLA_Househ2_UT( FLA_RIGHT, alpha11, a12t,
tau11 );
// ( a21 A22 ) = ( a21 A22 ) H
//
// where H is formed from tau11 and u12t.
FLA_Apply_H2_UT( FLA_RIGHT, tau11, a12t, a21, A22 );
// t01 = conj(a01) + conj(A02) * u12t^T;
FLA_Copyt_external( FLA_CONJ_NO_TRANSPOSE, a01, t01 );
FLA_Gemvc_external( FLA_CONJ_NO_TRANSPOSE, FLA_NO_CONJUGATE, FLA_ONE, A02, a12t, FLA_ONE, t01 );
/*------------------------------------------------------------*/
FLA_Cont_with_3x3_to_2x2( &ATL, /**/ &ATR, A00, a01, /**/ A02,
a10t, alpha11, /**/ a12t,
/* ************** */ /* ************************ */
&ABL, /**/ &ABR, A20, a21, /**/ A22,
FLA_TL );
FLA_Cont_with_3x3_to_2x2( &TTL, /**/ &TTR, T00, t01, /**/ T02,
t10t, tau11, /**/ t12t,
/* ************** */ /* ********************** */
&TBL, /**/ &TBR, T20, t21, /**/ T22,
FLA_TL );
}
return FLA_SUCCESS;
}
FLA_Error FLA_QR_UT_unb_var2( FLA_Obj A, FLA_Obj T )
{
FLA_Obj ATL, ATR, A00, a01, A02,
ABL, ABR, a10t, alpha11, a12t,
A20, a21, A22;
FLA_Obj TTL, TTR, T00, t01, T02,
TBL, TBR, t10t, tau11, t12t,
T20, t21, T22;
FLA_Part_2x2( A, &ATL, &ATR,
&ABL, &ABR, 0, 0, FLA_TL );
FLA_Part_2x2( T, &TTL, &TTR,
&TBL, &TBR, 0, 0, FLA_TL );
while ( FLA_Obj_min_dim( ABR ) > 0 ){
FLA_Repart_2x2_to_3x3( ATL, /**/ ATR, &A00, /**/ &a01, &A02,
/* ************* */ /* ************************** */
&a10t, /**/ &alpha11, &a12t,
ABL, /**/ ABR, &A20, /**/ &a21, &A22,
1, 1, FLA_BR );
FLA_Repart_2x2_to_3x3( TTL, /**/ TTR, &T00, /**/ &t01, &T02,
/* ************* */ /* ************************ */
&t10t, /**/ &tau11, &t12t,
TBL, /**/ TBR, &T20, /**/ &t21, &T22,
1, 1, FLA_BR );
/*------------------------------------------------------------*/
// Compute tau11 and u21 from alpha11 and a21 such that tau11 and u21
// determine a Householder transform H such that applying H from the
// left to the column vector consisting of alpha11 and a21 annihilates
// the entries in a21 (and updates alpha11).
FLA_Househ2_UT( FLA_LEFT,
alpha11,
a21, tau11 );
// / a12t \ = H / a12t \
// \ A22 / \ A22 /
//
// where H is formed from tau11 and u21.
FLA_Apply_H2_UT( FLA_LEFT, tau11, a21, a12t,
A22 );
// t01 = a10t' + A20' * u21;
FLA_Copyt_external( FLA_CONJ_TRANSPOSE, a10t, t01 );
FLA_Gemv_external( FLA_CONJ_TRANSPOSE, FLA_ONE, A20, a21, FLA_ONE, t01 );
/*------------------------------------------------------------*/
FLA_Cont_with_3x3_to_2x2( &ATL, /**/ &ATR, A00, a01, /**/ A02,
a10t, alpha11, /**/ a12t,
/* ************** */ /* ************************ */
&ABL, /**/ &ABR, A20, a21, /**/ A22,
FLA_TL );
FLA_Cont_with_3x3_to_2x2( &TTL, /**/ &TTR, T00, t01, /**/ T02,
t10t, tau11, /**/ t12t,
/* ************** */ /* ********************** */
&TBL, /**/ &TBR, T20, t21, /**/ T22,
FLA_TL );
}
return FLA_SUCCESS;
}
