[vhdl-200x-ft] Meeting Minutes 20-Dec-04

VHDL-200X-FT Meeting Minutes 20-Dec-04
----------------------------------------------------------
Peter Ashenden
Chuck Swart
Jim Lewis
David Bishop

Next Meeting: January 10, 2pm - 4pm

Discussion of Status

1076 Package Integration Part 1:
   David wondered if we should include 1076.6 RTL
   attributes as part of 1076? No. Keep 1076.6 separate.
   Include VITAL packages also? No. Too much stuff in
   VITAL that are outside of the language.

Package Standard:
   David to update package standard to include
   boolean_vector, integer_vector, real_vector, time_vector
   in package standard (comming from ISAC approved issues).
   Will round up all package standard updates under FT30

Floating Point Standard
   Discussion about use model of floating point.
   For DSP type work are different sizes of floating
   point used within a single chip or in any given chip
   is only one size of floating point used?

   Currently the package architecture being considered
   is to constrain the width of the mantissa and exponent
   as package generics. With this approach one package
   would need to be instantiated per size of floating point
   being used in the design. Simple if only one size is
   being used. More interesting if a design uses multiple
   different sizes of floating point - in addition it would
   be a significantly different use model from types signed
   and unsigned in numeric_std.

   For generalized floating point, do different operations
   require a different number of guard bits (add/sub vs.
   multiply vs. divide). Do we need to specify these
   parameters separately?

   Discussion of call format for extended functions. Should
   it be a named function such as:
   function add (
     l, r : floating_point; -- floating point input
     constant round_style : round_type := round_nearest; -- rounding option
     constant guard : natural := 3; -- number of guard bits
     constant check_error : boolean := true; -- check for errors
     constant ieee_extend : boolean := true) -- Use IEEE extended FP
     return floating_point;

   or should the "+" be extended to allow optional parameters:
     function "+" ( L, R : FP;
                    round_style : round_type := default_round_style;
                    denormalize : BOOLEAN := default_denormalize;
                    check_error : BOOLEAN := default_check_error;
                    guard_bits : NATURAL := default_guard_bits )
                  return FP;

    in this case,
    signal A, B, Y : FP(18 downto -7) ;

    Y <= A + B ; -- use the defaults

    Y <= "+"(A, B, ...) ; -- set values.

1076 Package Integration Part 2:
   Chuck wondered whether fixed point and floating point
   packages should be part of 1076 or a separate standard?

   Discussion as to what belongs to the standard and why.
Received on Wed Dec 22 11:29:07 2004