| 1 |
118 |
jguarin200 |
------------------------------------------------
|
| 2 |
|
|
--! @file ema32x2.vhd
|
| 3 |
|
|
--! @brief RayTrac Floating Point Adder
|
| 4 |
|
|
--! @author Julián Andrés Guarín Reyes
|
| 5 |
|
|
--------------------------------------------------
|
| 6 |
|
|
|
| 7 |
|
|
|
| 8 |
|
|
-- RAYTRAC (FP BRANCH)
|
| 9 |
|
|
-- Author Julian Andres Guarin
|
| 10 |
|
|
-- ema32x2.vhd
|
| 11 |
|
|
-- This file is part of raytrac.
|
| 12 |
|
|
--
|
| 13 |
|
|
-- raytrac is free software: you can redistribute it and/or modify
|
| 14 |
|
|
-- it under the terms of the GNU General Public License as published by
|
| 15 |
|
|
-- the Free Software Foundation, either version 3 of the License, or
|
| 16 |
|
|
-- (at your option) any later version.
|
| 17 |
|
|
--
|
| 18 |
|
|
-- raytrac is distributed in the hope that it will be useful,
|
| 19 |
|
|
-- but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 20 |
|
|
-- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
| 21 |
|
|
-- GNU General Public License for more details.
|
| 22 |
|
|
--
|
| 23 |
|
|
-- You should have received a copy of the GNU General Public License
|
| 24 |
|
|
-- along with raytrac. If not, see <http://www.gnu.org/licenses/>
|
| 25 |
|
|
|
| 26 |
|
|
library ieee;
|
| 27 |
|
|
use ieee.std_logic_1164.all;
|
| 28 |
|
|
use ieee.std_logic_unsigned.all;
|
| 29 |
|
|
use ieee.std_logic_arith.all;
|
| 30 |
|
|
|
| 31 |
|
|
|
| 32 |
|
|
--! Esta entidad recibe dos números en formato punto flotante IEEE 754, de precision simple y devuelve las mantissas signadas y corridas, y el exponente correspondiente al resultado antes de normalizarlo al formato float.
|
| 33 |
|
|
--!\nLas 2 mantissas y el exponente entran despues a la entidad add2 que suma las mantissas y entrega el resultado en formato IEEE 754.
|
| 34 |
|
|
entity ema32x2 is
|
| 35 |
|
|
port (
|
| 36 |
|
|
clk,dpc : in std_logic;
|
| 37 |
|
|
a32,b32 : in std_logic_vector (31 downto 0);
|
| 38 |
|
|
c32 : out std_logic_vector(31 downto 0)
|
| 39 |
|
|
);
|
| 40 |
|
|
end ema32x2;
|
| 41 |
|
|
|
| 42 |
|
|
architecture ema32x2_arch of ema32x2 is
|
| 43 |
|
|
|
| 44 |
|
|
component lpm_mult
|
| 45 |
|
|
generic (
|
| 46 |
|
|
lpm_hint : string;
|
| 47 |
|
|
lpm_representation : string;
|
| 48 |
|
|
lpm_type : string;
|
| 49 |
|
|
lpm_widtha : natural;
|
| 50 |
|
|
lpm_widthb : natural;
|
| 51 |
|
|
lpm_widthp : natural
|
| 52 |
|
|
);
|
| 53 |
|
|
port (
|
| 54 |
|
|
dataa : in std_logic_vector ( lpm_widtha-1 downto 0 );
|
| 55 |
|
|
datab : in std_logic_vector ( lpm_widthb-1 downto 0 );
|
| 56 |
|
|
result : out std_logic_vector( lpm_widthp-1 downto 0 )
|
| 57 |
|
|
);
|
| 58 |
|
|
end component;
|
| 59 |
|
|
|
| 60 |
|
|
signal s1zero : std_logic;
|
| 61 |
|
|
signal s1delta : std_logic_vector(5 downto 0);
|
| 62 |
|
|
signal s0delta,s1exp,s2exp,s3exp,s4exp,s5exp,s5factor,s6exp,s6factor: std_logic_vector(7 downto 0);
|
| 63 |
|
|
signal s1shifter,s5factorhot9 : std_logic_vector(8 downto 0);
|
| 64 |
|
|
signal s1pl,s5pl : std_logic_vector(17 downto 0);
|
| 65 |
|
|
signal s5postshift,s6postshift : std_logic_vector(22 downto 0);
|
| 66 |
|
|
signal s1umantshift,s1umantfixed,s1postshift,s1xorslab,s2xorslab : std_logic_vector(23 downto 0);
|
| 67 |
|
|
signal s2umantshift,s2mantfixed,s3mantfixed,s3mantshift,s4xorslab : std_logic_vector(24 downto 0);
|
| 68 |
|
|
signal s5factorhot25 : std_logic_vector(24 downto 0);
|
| 69 |
|
|
signal s4sresult,s5result,s6result : std_logic_vector(25 downto 0); -- Signed mantissa result
|
| 70 |
|
|
signal s1ph,s5ph : std_logic_vector(26 downto 0);
|
| 71 |
|
|
signal s0a,s0b : std_logic_vector(31 downto 0); -- Float 32 bit
|
| 72 |
|
|
|
| 73 |
|
|
begin
|
| 74 |
|
|
|
| 75 |
|
|
process (clk)
|
| 76 |
|
|
begin
|
| 77 |
|
|
if clk'event and clk='1' then
|
| 78 |
|
|
|
| 79 |
|
|
--!Registro de entrada
|
| 80 |
|
|
s0a <= a32;
|
| 81 |
|
|
s0b(31) <= dpc xor b32(31); --! Importante: Integrar el signo en el operando B
|
| 82 |
|
|
s0b(30 downto 0) <= b32(30 downto 0);
|
| 83 |
|
|
|
| 84 |
|
|
--!Etapa 0,Escoger el mayor exponente que sera el resultado desnormalizado, calcula cuanto debe ser el corrimiento de la mantissa con menor exponente y reorganiza los operandos, si el mayor es b, intercambia las posición si el mayor es a las posiciones la mantiene. Zero check.
|
| 85 |
|
|
--!signo,exponente,mantissa
|
| 86 |
|
|
if (s0b(30 downto 23)&s0a(30 downto 23))=x"0000" then
|
| 87 |
|
|
s1zero <= '0';
|
| 88 |
|
|
else
|
| 89 |
|
|
s1zero <= '1';
|
| 90 |
|
|
end if;
|
| 91 |
|
|
s1delta <= s0delta(7) & (s0delta(7) xor s0delta(4))&(s0delta(7) xor s0delta(4)) & s0delta(2 downto 0);
|
| 92 |
|
|
case s0delta(7) is
|
| 93 |
|
|
when '1' =>
|
| 94 |
|
|
s1exp <= s0b(30 downto 23);
|
| 95 |
|
|
s1umantshift <= s0a(31)&s0a(22 downto 0);
|
| 96 |
|
|
s1umantfixed <= s0b(31)&s0b(22 downto 0);
|
| 97 |
|
|
when others =>
|
| 98 |
|
|
s1exp <= s0a(30 downto 23);
|
| 99 |
|
|
s1umantshift <= s0b(31)&s0b(22 downto 0);
|
| 100 |
|
|
s1umantfixed <= s0a(31)&s0a(22 downto 0);
|
| 101 |
|
|
end case;
|
| 102 |
|
|
|
| 103 |
|
|
--! Etapa 1: Denormalización de la mantissas.
|
| 104 |
|
|
case s1delta(4 downto 3) is
|
| 105 |
|
|
when "00" => s2umantshift <= s1umantshift(23)&s1postshift(23 downto 0);
|
| 106 |
|
|
when "01" => s2umantshift <= s1umantshift(23)&x"00"&s1postshift(23 downto 8);
|
| 107 |
|
|
when "10" => s2umantshift <= s1umantshift(23)&x"0000"&s1postshift(23 downto 16);
|
| 108 |
|
|
when others => s2umantshift <= (others => '0');
|
| 109 |
|
|
end case;
|
| 110 |
|
|
s2mantfixed <= s1umantfixed(23) & ( ( ('1'&s1umantfixed(22 downto 0)) xor s1xorslab) + ( x"00000"&"000"&s1umantfixed(23) ) );
|
| 111 |
|
|
s2exp <= s1exp;
|
| 112 |
|
|
|
| 113 |
|
|
--! Etapa2: Signar la mantissa denormalizada.
|
| 114 |
|
|
s3mantfixed <= s2mantfixed;
|
| 115 |
|
|
s3mantshift <= s2umantshift(24)& ( ( s2umantshift(23 downto 0) xor s2xorslab) + ( x"00000"&"000"&s2umantshift(24) ) );
|
| 116 |
|
|
s3exp <= s2exp;
|
| 117 |
|
|
|
| 118 |
|
|
--! Etapa 3: Etapa 3 Realizar la suma, quitar el signo de la mantissa y codificar el corrimiento hacia la izquierda.
|
| 119 |
|
|
s4sresult <= (s3mantshift(24)&s3mantshift)+(s3mantfixed(24)&s3mantfixed);
|
| 120 |
|
|
s4exp <= s3exp;
|
| 121 |
|
|
|
| 122 |
|
|
--! Etapa 4: Quitar el signo a la mantissa resultante.
|
| 123 |
|
|
s5result <= s4sresult(25)&((s4sresult(24 downto 0) xor s4xorslab) +(x"000000"&s4sresult(25)));
|
| 124 |
|
|
s5exp <= s4exp;
|
| 125 |
|
|
|
| 126 |
|
|
|
| 127 |
|
|
--! Etapa 5: Codificar el corrimiento para la normalizacion de la mantissa resultante.
|
| 128 |
|
|
s6result <= s5result;
|
| 129 |
|
|
s6exp <= s5exp;
|
| 130 |
|
|
s6factor <= s5factor;
|
| 131 |
|
|
s6postshift <= s5postshift;
|
| 132 |
|
|
|
| 133 |
|
|
--! Etapa 6: Entregar el resultado.
|
| 134 |
|
|
c32(31) <= s6result(25);
|
| 135 |
|
|
c32(30 downto 23) <= s6exp+s6factor+x"ff";
|
| 136 |
|
|
case s6factor(4 downto 3) is
|
| 137 |
|
|
when "01" => c32(22 downto 0) <= s6postshift(14 downto 00)&x"00";
|
| 138 |
|
|
when "10" => c32(22 downto 0) <= s6postshift(06 downto 00)&x"0000";
|
| 139 |
|
|
when others => c32(22 downto 0) <= s6postshift;
|
| 140 |
|
|
end case;
|
| 141 |
|
|
end if;
|
| 142 |
|
|
end process;
|
| 143 |
|
|
--! Combinatorial gremlin, Etapa 0 el corrimiento de la mantissa con menor exponente y reorganiza los operandos,\n
|
| 144 |
|
|
--! si el mayor es b, intercambia las posición si el mayor es a las posiciones la mantiene.
|
| 145 |
|
|
s0delta <= s0a(30 downto 23)-s0b(30 downto 23);
|
| 146 |
|
|
--! Combinatorial Gremlin, Etapa 1 Codificar el factor de corrimiento de denormalizacion y denormalizar la mantissa no fija. Signar la mantissa que se queda fija.
|
| 147 |
|
|
decodeshiftfactor:
|
| 148 |
|
|
process (s1delta(2 downto 0))
|
| 149 |
|
|
begin
|
| 150 |
|
|
case s1delta(2 downto 0) is
|
| 151 |
|
|
when "111" => s1shifter(8 downto 0) <= '0'&s1delta(5)&"00000"¬(s1delta(5))&'0';
|
| 152 |
|
|
when "110" => s1shifter(8 downto 0) <= "00"&s1delta(5)&"000"¬(s1delta(5))&"00";
|
| 153 |
|
|
when "101" => s1shifter(8 downto 0) <= "000"&s1delta(5)&'0'¬(s1delta(5))&"000";
|
| 154 |
|
|
when "100" => s1shifter(8 downto 0) <= '0'&x"10";
|
| 155 |
|
|
when "011" => s1shifter(8 downto 0) <= "000"¬(s1delta(5))&'0'&s1delta(5)&"000";
|
| 156 |
|
|
when "010" => s1shifter(8 downto 0) <= "00"¬(s1delta(5))&"000"&s1delta(5)&"00";
|
| 157 |
|
|
when "001" => s1shifter(8 downto 0) <= '0'¬(s1delta(5))&"00000"&s1delta(5)&'0';
|
| 158 |
|
|
when others => s1shifter(8 downto 0) <= not(s1delta(5))&"0000000"&s1delta(5);
|
| 159 |
|
|
end case;
|
| 160 |
|
|
end process;
|
| 161 |
|
|
denormhighshiftermult:lpm_mult
|
| 162 |
|
|
generic map ("DEDICATED_MULTIPLIER_CIRCUITRY=YES,MAXIMIZE_SPEED=9","UNSIGNED","LPM_MULT",9,18,27)
|
| 163 |
|
|
port map (s1shifter,s1zero&s1umantshift(22 downto 06),s1ph);
|
| 164 |
|
|
denormlowshiftermult:lpm_mult
|
| 165 |
|
|
generic map ("DEDICATED_MULTIPLIER_CIRCUITRY=YES,MAXIMIZE_SPEED=9","UNSIGNED","LPM_MULT",9,9,18)
|
| 166 |
|
|
port map (s1shifter,s1umantshift(5 downto 0)&"000",s1pl);
|
| 167 |
|
|
|
| 168 |
|
|
s1postshift(23 downto 7) <= s1ph(25 downto 9);
|
| 169 |
|
|
s1postshift(06 downto 0) <= s1ph(08 downto 2) or s1pl(17 downto 11);
|
| 170 |
|
|
s1xorslab(23 downto 0) <= (others => s1umantfixed(23));
|
| 171 |
|
|
|
| 172 |
|
|
--! Combinatorial Gremlin, Etapa 2: Signar la mantissa denormalizada.
|
| 173 |
|
|
s2xorslab <= (others => s2umantshift(24));
|
| 174 |
|
|
|
| 175 |
|
|
--! Combinatorial Gremlin, Etapa 4: Quitar el signo de la mantissa resultante.
|
| 176 |
|
|
s4xorslab <= (others => s4sresult(25));
|
| 177 |
|
|
|
| 178 |
|
|
--! Combinatorial Gremlin, Etapa 5: Codificar el factor de normalizacion de la mantissa resultante.
|
| 179 |
|
|
normalizerdecodeshift:
|
| 180 |
|
|
process (s5result,s5factorhot25)
|
| 181 |
|
|
begin
|
| 182 |
|
|
s5factor<=(others => '0');
|
| 183 |
|
|
s5factorhot25 <= (others => '0');
|
| 184 |
|
|
for i in 24 downto 0 loop
|
| 185 |
|
|
if s5result(i)='1' then
|
| 186 |
|
|
s5factor <= conv_std_logic_vector(24-i,8);
|
| 187 |
|
|
s5factorhot25(24-i) <= '1';
|
| 188 |
|
|
exit;
|
| 189 |
|
|
end if;
|
| 190 |
|
|
end loop;
|
| 191 |
|
|
s5factorhot9 <= (s5factorhot25(8 downto 1)or s5factorhot25(16 downto 9)or s5factorhot25(24 downto 17)) & s5factorhot25(0);
|
| 192 |
|
|
end process;
|
| 193 |
|
|
normhighshiftermult:lpm_mult
|
| 194 |
|
|
generic map ("DEDICATED_MULTIPLIER_CIRCUITRY=YES,MAXIMIZE_SPEED=9","UNSIGNED","LPM_MULT",9,18,27)
|
| 195 |
|
|
port map (s5factorhot9,s5result(24 downto 7),s5ph);
|
| 196 |
|
|
normlowshiftermult:lpm_mult
|
| 197 |
|
|
generic map ("DEDICATED_MULTIPLIER_CIRCUITRY=YES,MAXIMIZE_SPEED=9","UNSIGNED","LPM_MULT",9,9,18)
|
| 198 |
|
|
port map (s5factorhot9,s5result(06 downto 0)&"00",s5pl);
|
| 199 |
|
|
s5postshift(22 downto 15) <= s5ph(16 downto 09);
|
| 200 |
|
|
s5postshift(14 downto 06) <= s5ph(08 downto 00); --! Activar este pedazo si se requiere extrema precision or s5pl(17 downto 9);
|
| 201 |
|
|
s5postshift(05 downto 00) <= s5pl(08 downto 03);
|
| 202 |
|
|
|
| 203 |
|
|
|
| 204 |
|
|
|
| 205 |
|
|
|
| 206 |
|
|
|
| 207 |
|
|
end ema32x2_arch;
|
| 208 |
|
|
|
| 209 |
|
|
|
