Bug fixes and consistency updates to numeric_std, fixed_generic_pkg, and float_generic_pkg.

Proposal Information

• Who Updates: DavidBishop, JimLewis, TristanGingold, ...
• Date Proposed: 2013-11-13
• Date Last Updated: 2013-11-13
• Priority:
• Complexity:
• Focus: Testbench

Numeric_Std

Bugzilla 15: : Extend Resize

Should numeric_bit/std have resize with size_res parameter? David proposed it for consistency with the fixed and floating point packages

Bugzilla 17: : add_carry for unsigned/signed

In the fixed point package, but not floating point package. Not in numeric_std. Add for consistency?

Add new function overload to_unsigned and to_signed

Add overloads for functions to_unsigned(x) and to_signed(x) to support usage as defined by proposal Function Knows Return Vector Size.

Fixed_generic_pkg

Modify the add_carry procedures:

procedure add_carry ( L, R : in UNRESOLVED_ufixed; c_in : in STD_ULOGIC; result : out UNRESOLVED_ufixed; c_out : out STD_ULOGIC)

procedure add_carry ( L, R : in UNRESOLVED_sfixed; c_in : in STD_ULOGIC; result : out UNRESOLVED_sfixed; c_out : out STD_ULOGIC)

in each of these modify line assigning "c_out" to be as follows: c_out := result_slv(left_index-right_index);

Fixed point documentation: Fixed the size returned by the signed reciprocal in Table xxx

Bugzilla 284 Appendix G.4.3 gives the sizing rules for fixed point numbers. The rules for Signed reciprocal and unsigned reciprocal are reversed.

Bugzilla 262: : Range Checking

Fixed Generic Pkg and checking valid range

Conversion error in to_sulv

The parameter arg is directly converted to STD_ULOGIC_VECTOR, but this is not correct as there are negative indexes for arg which aren't in the range of the index type of std_ulogic_vector. Better to use a target type with defined index ranges, like that:

  subtype result_subtype is STD_ULOGIC_VECTOR (arg'length-1 downto 0);
  variable result : result_subtype;
...
   result := result_subtype (arg);

Overflow in remainder

The bounds of lresize are correctly protected against overflow, but those of rresize aren't. The issue is that r'low can be integer'first, so r'low - guard_bits may overflow.

Procedure skip_whitespace assumes l'left = 1

The procedure skip_whitespace incorrectly assumes that left bound of the string accessed by L is 1. Better to use more portable code:

      c := l (l'left);
      if (c = ' ' or c = NBSP or c = HT) then

Float_generic_pkg

Modify to_float for unsigned/signed:

As detailed in Bugzilla 281

function to_float ( arg : UNSIGNED;
                    constant exponent_width : NATURAL := float_exponent_width; -- length of FP output exponent
                    constant fraction_width : NATURAL := float_fraction_width; -- length of FP output fraction
                    constant round_style : round_type := float_round_style) -- rounding option
                    return UNRESOLVED_float

function to_float ( arg : SIGNED;
                    constant exponent_width : NATURAL := float_exponent_width; -- length of FP output exponent
                    constant fraction_width : NATURAL := float_fraction_width; -- length of FP output fraction
                    constant round_style : round_type := float_round_style) -- rounding option
                    return UNRESOLVED_float

Mofify the "arg" argument in both functions to be "UNRESOLVED_UNSIGNED" and "UNRESOLVED_SIGN". The type of the "XARG" variable must be modified accordingly.

Modify to_float for sfixed:

function to_float ( arg : UNRESOLVED_sfixed; -- signed fixed point constant 
                    exponent_width : NATURAL := float_exponent_width; -- length of FP output exponent constant 
                    fraction_width : NATURAL := float_fraction_width; -- length of FP output fraction constant 
                    round_style : round_type := float_round_style; -- rounding constant 
                    denormalize : BOOLEAN := float_denormalize) -- rounding option 
         return UNRESOLVED_float

Modify the variable "arg_int" to be as follows:

variable arg_int : UNSIGNED(integer_width - in_fraction_width downto 0); -- unsigned version of argument 

Modify the variables "exp" and "exptmp" as follows:

variable exp, exptmp : SIGNED (exponent_width + 1 downto 0);

Modify the assignment of "arg_int" on line 2924 from "argx := -argx;" to:

arg_int := UNSIGNED(to_x01(not STD_LOGIC_VECTOR (argx))) + 1; -- Make it positive with two's complement 

Modify the assignment of "arg_int" on line 2927 to the following:

arg_int := UNSIGNED(to_x01(STD_LOGIC_VECTOR (argx))); -- new line: direct conversion to unsigned

Conversion error in to_sulv

Same issue as in the fixed_generic_pkg package.

Procedure skip_whitespace assumes l'left = 1

Same issue as in the fixed_generic_pkg_package.

Add new function overload to_unsigned and to_signed

Add overloads for functions to_unsigned(x) and to_signed(x) to support usage as defined by proposal Function Knows Return Vector Size.