1 |
2 |
__alexs__ |
-- --------------------------------------------------------------------------
|
2 |
|
|
-- >>>>>>>>>>>>>>>>>>>>>>>>>>>> COPYRIGHT NOTICE <<<<<<<<<<<<<<<<<<<<<<<<<<<<
|
3 |
|
|
-- --------------------------------------------------------------------------
|
4 |
|
|
-- TITLE: Plasma FPU ALU
|
5 |
|
|
-- AUTHORS: Maximilian Reuter (maximilian.reuter@fs-etit.de)
|
6 |
|
|
-- Alex Schoenberger (Alex.Schoenberger@ies.tu-darmstadt.de)
|
7 |
|
|
-- COMMENT: This project is based on Plasma CPU core by Steve Rhoads
|
8 |
|
|
--
|
9 |
|
|
-- www.ies.tu-darmstadt.de
|
10 |
|
|
-- TU Darmstadt
|
11 |
|
|
-- Institute for Integrated Systems
|
12 |
|
|
-- Merckstr. 25
|
13 |
|
|
--
|
14 |
|
|
-- 64283 Darmstadt - GERMANY
|
15 |
|
|
-- --------------------------------------------------------------------------
|
16 |
|
|
-- PROJECT: Plasma CPU core with FPU
|
17 |
|
|
-- FILENAME: plasma_fpu_alu.vhd
|
18 |
|
|
-- --------------------------------------------------------------------------
|
19 |
|
|
-- COPYRIGHT:
|
20 |
|
|
-- This project is distributed by GPLv2.0
|
21 |
|
|
-- Software placed into the public domain by the author.
|
22 |
|
|
-- Software 'as is' without warranty. Author liable for nothing.
|
23 |
|
|
-- --------------------------------------------------------------------------
|
24 |
|
|
-- DESCRIPTION
|
25 |
|
|
-- implementation of FPU ALU operations using FLI interface of ModelSIM
|
26 |
|
|
--
|
27 |
|
|
-- NOT SYNTHESIZABLE
|
28 |
|
|
--
|
29 |
|
|
----------------------------------------------------------------------------
|
30 |
|
|
-- Revision History
|
31 |
|
|
-- --------------------------------------------------------------------------
|
32 |
|
|
-- Revision Date Author CHANGES
|
33 |
|
|
-- 1.0 4/2015 MR initial
|
34 |
|
|
-- 2.0 5/2015 AS changed to plasma coding style
|
35 |
|
|
-- --------------------------------------------------------------------------
|
36 |
|
|
library IEEE;
|
37 |
|
|
use IEEE.std_logic_1164.ALL;
|
38 |
|
|
use IEEE.numeric_std.ALL;
|
39 |
|
|
|
40 |
|
|
library FLOATFIXLIB;
|
41 |
|
|
use FLOATFIXLIB.fixed_float_types.all;
|
42 |
|
|
use FLOATFIXLIB.float_pkg.all;
|
43 |
|
|
|
44 |
|
|
library PLASMA;
|
45 |
|
|
use PLASMA.mips_instruction_set.ALL;
|
46 |
|
|
use PLASMA.plasma_pack.ALL;
|
47 |
|
|
|
48 |
|
|
entity plasma_fpu_alu is
|
49 |
|
|
port (
|
50 |
|
|
control : in t_main_control;
|
51 |
|
|
-- INPUT
|
52 |
|
|
alu_a_in : in t_plasma_dword;
|
53 |
|
|
alu_b_in : in t_plasma_dword;
|
54 |
|
|
|
55 |
|
|
-- CONTROL
|
56 |
|
|
fpu_ctrl : in t_fpu_ctrl;
|
57 |
|
|
round_mode : in t_fpu_rm;
|
58 |
|
|
|
59 |
|
|
-- STATUS
|
60 |
|
|
cause_e : out std_logic;
|
61 |
|
|
cause : out t_fpu_flags;
|
62 |
|
|
|
63 |
|
|
-- OUTPUT
|
64 |
|
|
alu_out : out t_plasma_dword
|
65 |
|
|
);
|
66 |
|
|
end entity plasma_fpu_alu;
|
67 |
|
|
|
68 |
|
|
|
69 |
|
|
-- ____ _ ____ ____ ___ ____ _ _ _ _ _ ___ ____ _ _ _ _ _
|
70 |
|
|
-- |___ | | | |__| | |___ | \/ | | |__] | | |\ | | \_/
|
71 |
|
|
-- | |___ |__| | | | | | _/\_ |___ | |__] |__| | \| |___ |
|
72 |
|
|
--
|
73 |
|
|
-- use only floatfixlibrary
|
74 |
|
|
--
|
75 |
|
|
architecture fphdl_plasma_fpu_alu of plasma_fpu_alu is
|
76 |
|
|
--
|
77 |
|
|
-- intern calculation result
|
78 |
|
|
--
|
79 |
|
|
signal i_result : t_plasma_dword := (others => '0');
|
80 |
|
|
|
81 |
|
|
--
|
82 |
|
|
-- single precision parts
|
83 |
|
|
--
|
84 |
|
|
alias alu_a_in_S : t_plasma_word is alu_a_in(PLASMA_DATA_WIDTH - 1 downto 0);
|
85 |
|
|
alias alu_b_in_S : t_plasma_word is alu_b_in(PLASMA_DATA_WIDTH - 1 downto 0);
|
86 |
|
|
|
87 |
|
|
alias i_result_S : t_plasma_word is i_result(PLASMA_DATA_WIDTH - 1 downto 0);
|
88 |
|
|
|
89 |
|
|
--
|
90 |
|
|
-- rouund type
|
91 |
|
|
--
|
92 |
|
|
signal rm : round_type;
|
93 |
|
|
|
94 |
|
|
begin
|
95 |
|
|
--
|
96 |
|
|
-- define round mode
|
97 |
|
|
--
|
98 |
|
|
with round_mode select
|
99 |
|
|
rm <= round_zero when "01",
|
100 |
|
|
round_inf when "10",
|
101 |
|
|
round_neginf when "11",
|
102 |
|
|
round_nearest when others;
|
103 |
|
|
|
104 |
|
|
-- ____ ____ _ ____ _ _ _ ____ ___ ____
|
105 |
|
|
-- | |__| | | | | | |__| | |___
|
106 |
|
|
-- |___ | | |___ |___ |__| |___ | | | |___
|
107 |
|
|
--
|
108 |
|
|
-- decode operation and call extern function
|
109 |
|
|
--
|
110 |
|
|
cacl_proc:process( fpu_ctrl, alu_a_in, alu_b_in, rm )
|
111 |
|
|
--cacl_proc:process( control.clk )
|
112 |
|
|
--
|
113 |
|
|
-- FLI INTERFACE
|
114 |
|
|
--
|
115 |
|
|
variable func : integer;
|
116 |
|
|
variable double : integer;
|
117 |
|
|
|
118 |
|
|
variable real_a_in : real;
|
119 |
|
|
variable real_b_in : real;
|
120 |
|
|
variable real_out : real;
|
121 |
|
|
|
122 |
|
|
--
|
123 |
|
|
-- special values
|
124 |
|
|
--
|
125 |
|
|
variable class_a : valid_fpstate;
|
126 |
|
|
variable class_b : valid_fpstate;
|
127 |
|
|
variable class_out : valid_fpstate;
|
128 |
|
|
|
129 |
|
|
begin
|
130 |
|
|
-- if rising_edge( control.clk ) then
|
131 |
|
|
--
|
132 |
|
|
-- DEFAULT VALUES
|
133 |
|
|
--
|
134 |
|
|
-- result
|
135 |
|
|
i_result(63 downto 32) <= PLASMA_ZERO_WORD;
|
136 |
|
|
i_result_S <= alu_a_in_S;
|
137 |
|
|
|
138 |
|
|
--
|
139 |
|
|
-- unimplemented instruction flag
|
140 |
|
|
--
|
141 |
|
|
cause_e <= '0'; -- not active
|
142 |
|
|
|
143 |
|
|
--
|
144 |
|
|
-- DECODE
|
145 |
|
|
--
|
146 |
|
|
case fpu_ctrl.operation is
|
147 |
|
|
|
148 |
|
|
-- ############# PASS INPUT ############
|
149 |
|
|
when MIPS_FUNC_FMT_MOV =>
|
150 |
|
|
|
151 |
|
|
-- ############# CONVERT TO SINGLE #####
|
152 |
|
|
when MIPS_FUNC_FMT_CVTS =>
|
153 |
|
|
|
154 |
|
|
if fpu_ctrl.fix = '1' then -- integer -> single
|
155 |
|
|
i_result_S <= to_slv( to_Float( arg => signed( alu_a_in_S ),
|
156 |
|
|
exponent_width => 8,
|
157 |
|
|
fraction_width => 23) );
|
158 |
|
|
else -- double -> single
|
159 |
|
|
i_result_S <= to_slv( to_Float32( Float64(alu_a_in) ) );
|
160 |
|
|
end if;
|
161 |
|
|
|
162 |
|
|
|
163 |
|
|
-- ############# CONVERT TO DOUBLE #####
|
164 |
|
|
when MIPS_FUNC_FMT_CVTD =>
|
165 |
|
|
|
166 |
|
|
if fpu_ctrl.fix = '1' then -- integer -> double
|
167 |
|
|
i_result <= to_slv( to_Float( arg => signed( alu_a_in_S ),
|
168 |
|
|
exponent_width => 11,
|
169 |
|
|
fraction_width => 52) );
|
170 |
|
|
else -- single -> double
|
171 |
|
|
i_result <= to_slv( to_Float64( Float32(alu_a_in_S) ) );
|
172 |
|
|
end if;
|
173 |
|
|
|
174 |
|
|
-- ############# CONVERT TO WORD #######
|
175 |
|
|
when MIPS_FUNC_FMT_CVTW =>
|
176 |
|
|
|
177 |
|
|
if fpu_ctrl.double = '1' then -- double -> integer
|
178 |
|
|
i_result_S <= std_logic_vector( to_signed( to_integer(Float64(alu_a_in)),
|
179 |
|
|
PLASMA_DATA_WIDTH) );
|
180 |
|
|
else -- single -> integer
|
181 |
|
|
i_result_S <= std_logic_vector( to_signed( to_integer(Float32(alu_a_in_S)),
|
182 |
|
|
PLASMA_DATA_WIDTH) );
|
183 |
|
|
end if;
|
184 |
|
|
|
185 |
|
|
-- ############# CACLUATION ############
|
186 |
|
|
when MIPS_FUNC_FMT_ADD =>
|
187 |
|
|
|
188 |
|
|
if fpu_ctrl.double = '1' then
|
189 |
|
|
i_result <= to_slv( add( l => Float64(alu_a_in), r => Float64(alu_b_in), round_style => rm ));
|
190 |
|
|
else
|
191 |
|
|
i_result_S <= to_slv( add( l => Float32(alu_a_in_S), r => Float32(alu_b_in_S), round_style => rm ));
|
192 |
|
|
end if;
|
193 |
|
|
|
194 |
|
|
when MIPS_FUNC_FMT_SUB =>
|
195 |
|
|
|
196 |
|
|
if fpu_ctrl.double = '1' then
|
197 |
|
|
i_result <= to_slv( subtract( l => Float64(alu_a_in), r => Float64(alu_b_in), round_style => rm ));
|
198 |
|
|
else
|
199 |
|
|
i_result_S <= to_slv( subtract( l => Float32(alu_a_in_S), r => Float32(alu_b_in_S), round_style => rm ));
|
200 |
|
|
end if;
|
201 |
|
|
|
202 |
|
|
when MIPS_FUNC_FMT_MUL =>
|
203 |
|
|
|
204 |
|
|
if fpu_ctrl.double = '1' then
|
205 |
|
|
i_result <= to_slv( multiply( l => Float64(alu_a_in), r => Float64(alu_b_in), round_style => rm ));
|
206 |
|
|
else
|
207 |
|
|
i_result_S <= to_slv( multiply( l => Float32(alu_a_in_S), r => Float32(alu_b_in_S), round_style => rm ));
|
208 |
|
|
end if;
|
209 |
|
|
|
210 |
|
|
when MIPS_FUNC_FMT_DIV =>
|
211 |
|
|
|
212 |
|
|
if fpu_ctrl.double = '1' then
|
213 |
|
|
i_result <= to_slv( divide( l => Float64(alu_a_in), r => Float64(alu_b_in), round_style => rm ));
|
214 |
|
|
else
|
215 |
|
|
i_result_S <= to_slv( divide( l => Float32(alu_a_in_S), r => Float32(alu_b_in_S), round_style => rm ));
|
216 |
|
|
end if;
|
217 |
|
|
|
218 |
|
|
when MIPS_FUNC_FMT_SQRT =>
|
219 |
|
|
|
220 |
|
|
if fpu_ctrl.double = '1' then
|
221 |
|
|
i_result <= to_slv( sqrt( arg => Float64(alu_a_in), round_style => rm ));
|
222 |
|
|
else
|
223 |
|
|
i_result_S <= to_slv( sqrt( arg => Float32(alu_a_in_S), round_style => rm ));
|
224 |
|
|
end if;
|
225 |
|
|
|
226 |
|
|
when MIPS_FUNC_FMT_ABS =>
|
227 |
|
|
|
228 |
|
|
if fpu_ctrl.double = '1' then
|
229 |
|
|
i_result <= to_slv( abs( Float64(alu_a_in) ));
|
230 |
|
|
else
|
231 |
|
|
i_result_S <= to_slv( abs( Float32(alu_a_in_S) ));
|
232 |
|
|
end if;
|
233 |
|
|
|
234 |
|
|
when MIPS_FUNC_FMT_NEG =>
|
235 |
|
|
|
236 |
|
|
if fpu_ctrl.double = '1' then
|
237 |
|
|
i_result <= to_slv( -( Float64(alu_a_in) ));
|
238 |
|
|
else
|
239 |
|
|
i_result_S <= to_slv( -( Float32(alu_a_in_S) ));
|
240 |
|
|
end if;
|
241 |
|
|
|
242 |
|
|
when others =>
|
243 |
|
|
|
244 |
|
|
if not std_match( fpu_ctrl.operation, MIPS_FUNC_FMT_COND ) then -- check for comparator instruction
|
245 |
|
|
cause_e <= '1'; -- if not -> unimplemented instruction
|
246 |
|
|
end if;
|
247 |
|
|
|
248 |
|
|
end case;
|
249 |
|
|
|
250 |
|
|
-- ALU Status Bits
|
251 |
|
|
--e: Unimplemented -> set before
|
252 |
|
|
--v: Invalid
|
253 |
|
|
--z: Divide by zero
|
254 |
|
|
--o: Overflow
|
255 |
|
|
--u: Underflow
|
256 |
|
|
--i: Inexact result (denormalized)
|
257 |
|
|
--
|
258 |
|
|
-- default values
|
259 |
|
|
--
|
260 |
|
|
cause <= ('0', '0', '0', '0', '0');
|
261 |
|
|
|
262 |
|
|
if fpu_ctrl.double = '1' then
|
263 |
|
|
class_a := classFP( Float64( alu_a_in ) );
|
264 |
|
|
class_b := classFP( Float64( alu_b_in ) );
|
265 |
|
|
class_out := classFP( to_Float( arg => real_out, exponent_width => 11, fraction_width => 52) );
|
266 |
|
|
else
|
267 |
|
|
class_a := classFP( Float32( alu_a_in_S ) );
|
268 |
|
|
class_b := classFP( Float32( alu_b_in_S ) );
|
269 |
|
|
class_out := classFP( to_Float( arg => real_out, exponent_width => 8, fraction_width => 23) );
|
270 |
|
|
end if;
|
271 |
|
|
|
272 |
|
|
--
|
273 |
|
|
-- CLASS OF 1. INPUT
|
274 |
|
|
--
|
275 |
|
|
case class_a is
|
276 |
|
|
when neg_zero | pos_zero => if fpu_ctrl.operation = MIPS_FUNC_FMT_DIV then cause.z <= '1'; end if;
|
277 |
|
|
when nan | quiet_nan => cause.v <= '1';
|
278 |
|
|
if fpu_ctrl.double = '1' then i_result <= to_slv(nanfp(11,52));
|
279 |
|
|
else i_result_S <= to_slv(nanfp( 8,23)); end if;
|
280 |
|
|
when others =>
|
281 |
|
|
end case;
|
282 |
|
|
|
283 |
|
|
--
|
284 |
|
|
-- CLASS OF 2. INPUT
|
285 |
|
|
--
|
286 |
|
|
case class_b is
|
287 |
|
|
when neg_zero | pos_zero => if fpu_ctrl.operation = MIPS_FUNC_FMT_DIV then cause.z <= '1'; end if;
|
288 |
|
|
when nan | quiet_nan => cause.v <= '1';
|
289 |
|
|
if fpu_ctrl.double = '1' then i_result <= to_slv(nanfp(11,52));
|
290 |
|
|
else i_result_S <= to_slv(nanfp( 8,23)); end if;
|
291 |
|
|
when others =>
|
292 |
|
|
end case;
|
293 |
|
|
|
294 |
|
|
--
|
295 |
|
|
-- CLASS OF OUTPUT
|
296 |
|
|
--
|
297 |
|
|
case class_out is
|
298 |
|
|
when pos_inf | neg_inf => cause.o <= '1'; cause.i <= '1';
|
299 |
|
|
when pos_denormal | neg_denormal => cause.u <= '1'; cause.i <= '1';
|
300 |
|
|
when others =>
|
301 |
|
|
end case;
|
302 |
|
|
|
303 |
|
|
-- end if;
|
304 |
|
|
end process;
|
305 |
|
|
|
306 |
|
|
--
|
307 |
|
|
-- OUTPUT
|
308 |
|
|
--
|
309 |
|
|
alu_out <= i_result;
|
310 |
|
|
|
311 |
|
|
end architecture fphdl_plasma_fpu_alu;
|