libflame  12600
Functions
FLA_Gemv_n_blk_var2.c File Reference

(r12600)

Functions

FLA_Error FLA_Gemv_n_blk_var2 (FLA_Obj alpha, FLA_Obj A, FLA_Obj x, FLA_Obj beta, FLA_Obj y, fla_gemv_t *cntl)

Function Documentation

FLA_Error FLA_Gemv_n_blk_var2 ( FLA_Obj  alpha,
FLA_Obj  A,
FLA_Obj  x,
FLA_Obj  beta,
FLA_Obj  y,
fla_gemv_t cntl 
)

References FLA_Cont_with_3x1_to_2x1(), FLA_Determine_blocksize(), FLA_Gemv_internal(), FLA_Obj_length(), FLA_Part_2x1(), and FLA_Repart_2x1_to_3x1().

Referenced by FLA_Gemv_n().

{
  FLA_Obj AT,              A0,
          AB,              A1,
                           A2;

  FLA_Obj yT,              y0,
          yB,              y1,
                           y2;

  dim_t b;

  FLA_Part_2x1( A,    &AT, 
                      &AB,            0, FLA_BOTTOM );

  FLA_Part_2x1( y,    &yT, 
                      &yB,            0, FLA_BOTTOM );

  while ( FLA_Obj_length( AB ) < FLA_Obj_length( A ) ){

    b = FLA_Determine_blocksize( AT, FLA_TOP, FLA_Cntl_blocksize( cntl ) );

    FLA_Repart_2x1_to_3x1( AT,                &A0, 
                                              &A1, 
                        /* ** */            /* ** */
                           AB,                &A2,        b, FLA_TOP );

    FLA_Repart_2x1_to_3x1( yT,                &y0, 
                                              &y1, 
                        /* ** */            /* ** */
                           yB,                &y2,        b, FLA_TOP );

    /*------------------------------------------------------------*/

    /*   y1 = alpha * A1 * x + y1;   */
    FLA_Gemv_internal( FLA_NO_TRANSPOSE, 
                       alpha, A1, x, beta, y1, 
                       FLA_Cntl_sub_gemv( cntl ) );

    /*------------------------------------------------------------*/

    FLA_Cont_with_3x1_to_2x1( &AT,                A0, 
                            /* ** */           /* ** */
                                                  A1, 
                              &AB,                A2,     FLA_BOTTOM );

    FLA_Cont_with_3x1_to_2x1( &yT,                y0, 
                            /* ** */           /* ** */
                                                  y1, 
                              &yB,                y2,     FLA_BOTTOM );

  }

  return FLA_SUCCESS;
}