DokuWiki

You've loaded an old revision of the document! If you save it, you will create a new version with this data.

https://schaumont.dyn.wpi.edu/ece4530f19/lectures/lecture13-notes.html

<code vhdl>
library ieee;
use ieee.std_logic_1164.all;
use ieee.numeric_std.all;

--------------------------------------------------------------------------------

package float32 is

--------------------------------------------------------------------------------

type fp32 is record
		sign : std_logic;
		exponent : unsigned(7 downto 0);
		mantissa : unsigned(22 downto 0);
	end record fp32;
	
	type fp32_array is array(natural range<>) of fp32;
	
--------------------------------------------------------------------------------
	
	constant fp32_EXPONENT_OFFSET : unsigned(7 downto 0) := "01111111";
	constant fp32_EXPONENT_OFFSET_INT : integer := 127;
	
	constant fp32_ZERO : fp32 := (
		sign => '0', 
		exponent => (others => '0'), 
		mantissa => (others => '0')
	);

constant fp32_INF  : fp32 := (
		sign => '0', 
		exponent => (others => '1'), 
		mantissa => (others => '0')
	);
	
--------------------------------------------------------------------------------
	
	function fp32_abs_eq (first, second : in fp32) return boolean;
	function fp32_is_nan (first, second : in fp32) return boolean;
	function fp32_gt (first, second: in fp32) return boolean;

--------------------------------------------------------------------------------	
	
	function fp32_from_slv (src : in std_logic_vector(31 downto 0)) return fp32;
	function slv_from_fp32 (src : in fp32) return std_logic_vector;

function fp32_add (first, second: in fp32) return fp32;
	function fp32_sub (first, second: in fp32) return fp32;
	function fp32_mul (first, second: in fp32) return fp32;
	--function fp32_div (first, second: in fp32) return fp32;
	
--------------------------------------------------------------------------------

end float32;

--------------------------------------------------------------------------------

package body float32 is

--------------------------------------------------------------------------------

function fp32_abs_eq (first, second : in fp32) return boolean is
	begin
		return (first.exponent = second.exponent and 
			first.mantissa = second.mantissa);
	end function fp32_abs_eq;
	
	function fp32_is_nan (first, second : in fp32) return boolean is
	begin
		return (first.exponent = fp32_INF.exponent and 
			first.mantissa /= fp32_INF.mantissa);
	end function fp32_is_nan;

--------------------------------------------------------------------------------

function fp32_gt (first, second: in fp32) return boolean is
		variable result : boolean := false;
	begin		
		if first.exponent = fp32_ZERO.exponent and second.exponent = fp32_ZERO.exponent then
		   -- zero > zero
		   result := false; 
	    elsif first.exponent = fp32_ZERO.exponent and second.exponent /= fp32_ZERO.exponent then
		   -- zero > nonzero
		   result := second.sign /= '0'; 
	    elsif first.exponent /= fp32_ZERO.exponent and second.exponent = fp32_ZERO.exponent then
		   -- nonzero > zero
		   result := first.sign = '0'; 
	    else -- if first.exponent /= 0 and second.exponent /= 0 then
		   -- both nonzero
  	       if first.sign = '1' and second.sign = '0' then
			   -- negative > positive
  		       result := false; 
  	       elsif first.sign = '0' and second.sign = '1' then
			   -- positive > negative
  		       result := true; 
		   else
		       -- same sign
  	           if first.exponent > second.exponent then
		          result := true;
  	           elsif first.exponent < second.exponent then
		          result := false;
  	           else
		          result := first.mantissa > second.mantissa;
   	           end if;

if first.sign = '1' then
				  -- both negative
			      result := not result; 
			   end if;
		   end if;
		end if;
		return result; 
	end function fp32_gt;
	 
--------------------------------------------------------------------------------

function fp32_from_slv (src : in std_logic_vector(31 downto 0)) return fp32 is
	begin
		return (sign => src(31), 
			exponent => unsigned(src(30 downto 23)),
			mantissa => unsigned(src(22 downto 0)));
	end function fp32_from_slv;

function slv_from_fp32 (src : in fp32) return std_logic_vector is
	begin
		return src.sign & std_logic_vector(src.exponent) & std_logic_vector(src.mantissa);
	end function slv_from_fp32;

function fp32_mul (first, second: in fp32) return fp32 is
		variable result : fp32 := fp32_ZERO;
		variable buf : unsigned(47 downto 0) := (others => '0');
		variable start : integer := 46;
	begin
	    if first.exponent /= 0 or second.exponent /= 0 then
		   result.sign := first.sign xor second.sign;
		   result.exponent := first.exponent + second.exponent - fp32_EXPONENT_OFFSET;
		   buf := unsigned('1' & first.mantissa) * unsigned('1' & second.mantissa);

-- normalize
		   for i in 47 downto 46 loop
			   if buf(i) = '1' then
				   start := i - 1;
				   exit;
			   end if;
		   end loop;

result.exponent := to_unsigned(to_integer(result.exponent) + start - 45, 8);
    	   result.mantissa := buf(start downto start-22);
		end if;
		return result;
	end function fp32_mul;

--------------------------------------------------------------------------------

function fp32_sum (first, second: in fp32) return fp32 is
        variable result : fp32 := fp32_ZERO;
        variable buf  : unsigned (24 downto 0) := (others => '0');
        variable buf1 : unsigned (24 downto 0) := (others => '0');
        variable buf2 : unsigned (24 downto 0) := (others => '0');
        variable delta : integer := 0;
        variable start : integer := 22;
	variable zero : boolean := true;
    begin
         result.sign := first.sign;
         result.exponent := first.exponent;

buf1 := unsigned("01" & first.mantissa);
         buf2 := unsigned("01" & second.mantissa);

delta := to_integer(first.exponent) - to_integer (second.exponent);
         if delta < 24 then
              if delta /= 0 then
                 buf2 (24-delta downto 0) := buf2 (24 downto delta);
                 buf2 (24 downto 24-delta+1) := (others => '0');
              end if;

if first.sign = second.sign then
                 buf := buf1 + buf2;
              else
                 buf := buf1 - buf2;
              end if;
         end if;

-- normalize
	 for i in 24 downto 0 loop
         	if buf(i) = '1' then
            		start := i;
			zero := false;
			exit;
         	end if;
	 end loop;
	
	if not zero then
        	result.exponent := result.exponent + start - 1 - 22;

if start = 24 then
			result.mantissa(22 downto 0) := buf(23 downto 1);
		-- elsif start = 23 then
		--	result.mantissa(22 downto 0) := buf(22 downto 0);
		elsif start > 0
			result.mantissa(22 downto 23-start) := buf(start-1 downto 0);
		end if;
	end if;
        return result;
    end function fp32_sum;

function fp32_add (first, second: in fp32) return fp32 is
    begin
        if first.exponent = 0 then
            return second;
        elsif second.exponent = 0 then
            return first;
        elsif first.exponent > second.exponent then
            return fp32_sum (first, second);
        elsif first.exponent < second.exponent then
            return fp32_sum (second, first);
	elsif first.mantissa >= second.mantissa then
		return fp32_sum (first, second);
	else
		return fp32_sum (second, first);
        end if;
    end function fp32_add;

function fp32_sub (first, second: in fp32) return fp32 is
        variable temp : fp32 := second;
    begin
        temp.sign := not temp.sign;
		return fp32_add (first, temp);
    end function fp32_sub;

--------------------------------------------------------------------------------

end float32;

--------------------------------------------------------------------------------
</code>

Edit summary

Note: By editing this page you agree to license your content under the following license: CC Attribution-Noncommercial-Share Alike 4.0 International