FLA_LU_piv_blk_var5.c File Reference

(r)

Functions
FLA_Error	FLA_LU_piv_blk_var5 (FLA_Obj A, FLA_Obj p, fla_lu_t *cntl)

Function Documentation

FLA_LU_piv_blk_var5()

FLA_Error FLA_LU_piv_blk_var5	(	FLA_Obj	A,
		FLA_Obj	p,
		fla_lu_t *	cntl
	)

{
  FLA_Error r_val = FLA_SUCCESS, r_val_sub = FLA_SUCCESS;
  FLA_Obj ATL,   ATR,      A00, A01, A02, 
          ABL,   ABR,      A10, A11, A12,
                           A20, A21, A22;
 
  FLA_Obj pT,              p0,
          pB,              p1,
                           p2;
 
  FLA_Obj AB0, AB1, AB2;
 
  dim_t b;
 
 
  FLA_Part_2x2( A,    &ATL, &ATR,
                      &ABL, &ABR,     0, 0, FLA_TL );
 
  FLA_Part_2x1( p,    &pT, 
                      &pB,            0, FLA_TOP );
 
  while ( FLA_Obj_length( ATL ) < FLA_Obj_length( A ) &&
          FLA_Obj_width( ATL ) < FLA_Obj_width( A )){
 
    b = FLA_Determine_blocksize( ABR, FLA_BR, FLA_Cntl_blocksize( cntl ) );
 
    FLA_Repart_2x2_to_3x3( ATL, /**/ ATR,       &A00, /**/ &A01, &A02,
                        /* ************* */   /* ******************** */
                                                &A10, /**/ &A11, &A12,
                           ABL, /**/ ABR,       &A20, /**/ &A21, &A22,
                           b, b, FLA_BR );
 
    FLA_Repart_2x1_to_3x1( pT,                &p0, 
                        /* ** */            /* ** */
                                              &p1, 
                           pB,                &p2,        b, FLA_BOTTOM );
 
    /*------------------------------------------------------------*/
 
    // AB1 = / A11 \
    //       \ A21 /
    FLA_Merge_2x1( A11,
                   A21,      &AB1 );
 
    // AB1, p1 = LU_piv( AB1 )
    r_val_sub = FLA_LU_piv_internal( AB1, p1, 
                     FLA_Cntl_sub_lu( cntl ) );
 
    // If the unblocked algorithm returns a null pivot, 
    // update the pivot index and return it.
    if ( r_val == FLA_SUCCESS && r_val_sub >= 0 )
    {
        r_val = FLA_Obj_length( A01 ) + r_val_sub;
    }
 
    // AB0 = / A10 \
    //       \ A20 /
    FLA_Merge_2x1( A10,
                   A20,      &AB0 );
 
    // Apply computed pivots to AB0
    FLA_Apply_pivots_internal( FLA_LEFT, FLA_NO_TRANSPOSE, p1, AB0,
                               FLA_Cntl_sub_appiv1( cntl ) );
 
    // AB2 = / A12 \
    //       \ A22 /
    FLA_Merge_2x1( A12,
                   A22,      &AB2 );
 
    // Apply computed pivots to AB2
    FLA_Apply_pivots_internal( FLA_LEFT, FLA_NO_TRANSPOSE, p1, AB2,
                               FLA_Cntl_sub_appiv1( cntl ) );
 
    // A12 = trilu( A11 ) \ A12
    FLA_Trsm_internal( FLA_LEFT, FLA_LOWER_TRIANGULAR, 
                       FLA_NO_TRANSPOSE, FLA_UNIT_DIAG,
                       FLA_ONE, A11, A12,
                       FLA_Cntl_sub_trsm1( cntl ) );
 
    // A22 = A22 - A21 * A12
    FLA_Gemm_internal( FLA_NO_TRANSPOSE, FLA_NO_TRANSPOSE,
                       FLA_MINUS_ONE, A21, A12, FLA_ONE, A22,
                       FLA_Cntl_sub_gemm1( cntl ) );
 
    /*------------------------------------------------------------*/
 
    FLA_Cont_with_3x3_to_2x2( &ATL, /**/ &ATR,       A00, A01, /**/ A02,
                                                     A10, A11, /**/ A12,
                            /* ************** */  /* ****************** */
                              &ABL, /**/ &ABR,       A20, A21, /**/ A22,
                              FLA_TL );
 
    FLA_Cont_with_3x1_to_2x1( &pT,                p0, 
                                                  p1, 
                            /* ** */           /* ** */
                              &pB,                p2,     FLA_TOP );
 
  }
 
  return r_val;
}

References FLA_Apply_pivots_internal(), FLA_Cont_with_3x1_to_2x1(), FLA_Cont_with_3x3_to_2x2(), FLA_Determine_blocksize(), FLA_Gemm_internal(), FLA_LU_piv_internal(), FLA_Merge_2x1(), FLA_MINUS_ONE, FLA_Obj_length(), FLA_Obj_width(), FLA_ONE, FLA_Part_2x1(), FLA_Part_2x2(), FLA_Repart_2x1_to_3x1(), FLA_Repart_2x2_to_3x3(), FLA_Trsm_internal(), and i.

Referenced by FLA_LU_piv_internal().