コード例 #1
0
ファイル: FLA_Trsv.c プロジェクト: anaptyxis/libflame
FLA_Error FLA_Trsv( FLA_Uplo uplo, FLA_Trans trans, FLA_Diag diag, FLA_Obj A, FLA_Obj x )
{
  FLA_Error r_val;

  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Trsv_check( uplo, trans, diag, A, x );

#ifdef FLA_ENABLE_BLAS2_FRONT_END_CNTL_TREES
  // Invoke FLA_Trsv_internal() with flat control tree that simply calls
  // external wrapper.
  r_val = FLA_Trsv_internal( uplo, trans, diag, A, x, fla_trsv_cntl_blas );

#else
  r_val = FLA_Trsv_external( uplo, trans, diag, A, x );
#endif

  return r_val;
}
コード例 #2
0
ファイル: FLASH_Trsv.c プロジェクト: pgawron/tlash
FLA_Error FLASH_Trsv( FLA_Uplo uplo, FLA_Trans trans, FLA_Diag diag, FLA_Obj A, FLA_Obj x )
{
  FLA_Error r_val;
  FLA_Bool  enable_supermatrix;

  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Trsv_check( uplo, trans, diag, A, x );

  // Find the status of SuperMatrix.
  enable_supermatrix = FLASH_Queue_get_enabled();

  // Temporarily disable SuperMatrix.
  FLASH_Queue_disable();

  // Execute tasks.
  r_val = FLA_Trsv_internal( uplo, trans, diag, A, x, flash_trsv_cntl );

  // Restore SuperMatrix to its previous status.
  if ( enable_supermatrix )
     FLASH_Queue_enable();
  
  return r_val;
}
コード例 #3
0
FLA_Error FLA_Trsv_lc_blk_var1( FLA_Diag diagA, FLA_Obj A, FLA_Obj x, fla_trsv_t* cntl )
{
  FLA_Obj ATL,   ATR,      A00, A01, A02, 
          ABL,   ABR,      A10, A11, A12,
                           A20, A21, A22;

  FLA_Obj xT,              x0,
          xB,              x1,
                           x2;

  dim_t b;

  FLA_Part_2x2( A,    &ATL, &ATR,
                      &ABL, &ABR,     0, 0, FLA_BR );

  FLA_Part_2x1( x,    &xT, 
                      &xB,            0, FLA_BOTTOM );

  while ( FLA_Obj_length( ABR ) < FLA_Obj_length( A ) ){

    b = FLA_Determine_blocksize( ATL, FLA_TL, 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_TL );

    FLA_Repart_2x1_to_3x1( xT,                &x0, 
                                              &x1, 
                        /* ** */            /* ** */
                           xB,                &x2,        b, FLA_TOP );

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

    /* x1 = x1 - A21' * x2; */
    FLA_Gemv_internal( FLA_CONJ_TRANSPOSE, 
                       FLA_MINUS_ONE, A21, x2, FLA_ONE, x1,
                       FLA_Cntl_sub_gemv( cntl ) );

    /* x1 = tril( A11' ) \ x1; */
    FLA_Trsv_internal( FLA_LOWER_TRIANGULAR, FLA_CONJ_TRANSPOSE, diagA,
                       A11, x1,
                       FLA_Cntl_sub_trsv( cntl ) );

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

    FLA_Cont_with_3x3_to_2x2( &ATL, /**/ &ATR,       A00, /**/ A01, A02,
                            /* ************** */  /* ****************** */
                                                     A10, /**/ A11, A12,
                              &ABL, /**/ &ABR,       A20, /**/ A21, A22,
                              FLA_BR );

    FLA_Cont_with_3x1_to_2x1( &xT,                x0, 
                            /* ** */           /* ** */
                                                  x1, 
                              &xB,                x2,     FLA_BOTTOM );

  }

  return FLA_SUCCESS;
}