1 |
2 |
rudi |
/////////////////////////////////////////////////////////////////////
|
2 |
|
|
//// ////
|
3 |
|
|
//// Pre Normalize ////
|
4 |
|
|
//// Pre Normalization Unit for Add/Sub Operations ////
|
5 |
|
|
//// ////
|
6 |
|
|
//// Author: Rudolf Usselmann ////
|
7 |
|
|
//// rudi@asics.ws ////
|
8 |
|
|
//// ////
|
9 |
|
|
/////////////////////////////////////////////////////////////////////
|
10 |
|
|
//// ////
|
11 |
|
|
//// Copyright (C) 2000 Rudolf Usselmann ////
|
12 |
|
|
//// rudi@asics.ws ////
|
13 |
|
|
//// ////
|
14 |
|
|
//// This source file may be used and distributed without ////
|
15 |
|
|
//// restriction provided that this copyright statement is not ////
|
16 |
|
|
//// removed from the file and that any derivative work contains ////
|
17 |
|
|
//// the original copyright notice and the associated disclaimer.////
|
18 |
|
|
//// ////
|
19 |
|
|
//// THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY ////
|
20 |
|
|
//// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED ////
|
21 |
|
|
//// TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS ////
|
22 |
|
|
//// FOR A PARTICULAR PURPOSE. IN NO EVENT SHALL THE AUTHOR ////
|
23 |
|
|
//// OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, ////
|
24 |
|
|
//// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES ////
|
25 |
|
|
//// (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE ////
|
26 |
|
|
//// GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR ////
|
27 |
|
|
//// BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF ////
|
28 |
|
|
//// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT ////
|
29 |
|
|
//// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT ////
|
30 |
|
|
//// OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE ////
|
31 |
|
|
//// POSSIBILITY OF SUCH DAMAGE. ////
|
32 |
|
|
//// ////
|
33 |
|
|
/////////////////////////////////////////////////////////////////////
|
34 |
|
|
|
35 |
|
|
`timescale 1ns / 100ps
|
36 |
|
|
|
37 |
|
|
|
38 |
|
|
module pre_norm(clk, rmode, add, opa, opb, opa_nan, opb_nan, fracta_out,
|
39 |
|
|
fractb_out, exp_dn_out, sign, nan_sign, result_zero_sign,
|
40 |
|
|
fasu_op);
|
41 |
|
|
input clk;
|
42 |
|
|
input [1:0] rmode;
|
43 |
|
|
input add;
|
44 |
|
|
input [31:0] opa, opb;
|
45 |
|
|
input opa_nan, opb_nan;
|
46 |
|
|
output [26:0] fracta_out, fractb_out;
|
47 |
|
|
output [7:0] exp_dn_out;
|
48 |
|
|
output sign;
|
49 |
|
|
output nan_sign, result_zero_sign;
|
50 |
|
|
output fasu_op; // Operation Output
|
51 |
|
|
|
52 |
|
|
////////////////////////////////////////////////////////////////////////
|
53 |
|
|
//
|
54 |
|
|
// Local Wires and registers
|
55 |
|
|
//
|
56 |
|
|
|
57 |
|
|
wire signa, signb; // alias to opX sign
|
58 |
|
|
wire [7:0] expa, expb; // alias to opX exponent
|
59 |
|
|
wire [22:0] fracta, fractb; // alias to opX fraction
|
60 |
|
|
wire expa_lt_expb; // expa is larger than expb indicator
|
61 |
|
|
wire fractb_lt_fracta; // fractb is larger than fracta indicator
|
62 |
|
|
reg [7:0] exp_dn_out; // de normalized exponent output
|
63 |
|
|
wire [7:0] exp_small, exp_large;
|
64 |
|
|
wire [7:0] exp_diff; // Numeric difference of the two exponents
|
65 |
|
|
wire [22:0] adj_op; // Fraction adjustment: input
|
66 |
|
|
wire [26:0] adj_op_tmp;
|
67 |
|
|
wire [26:0] adj_op_out; // Fraction adjustment: output
|
68 |
|
|
wire [26:0] fracta_n, fractb_n; // Fraction selection after normalizing
|
69 |
|
|
wire [26:0] fracta_s, fractb_s; // Fraction Sorting out
|
70 |
|
|
reg [26:0] fracta_out, fractb_out; // Fraction Output
|
71 |
|
|
reg sign, sign_d; // Sign Output
|
72 |
|
|
reg add_d; // operation (add/sub)
|
73 |
|
|
reg fasu_op; // operation (add/sub) register
|
74 |
|
|
wire expa_dn, expb_dn;
|
75 |
|
|
reg sticky;
|
76 |
|
|
reg result_zero_sign;
|
77 |
|
|
reg add_r, signa_r, signb_r;
|
78 |
|
|
wire [4:0] exp_diff_sft;
|
79 |
|
|
wire exp_lt_27;
|
80 |
|
|
wire op_dn;
|
81 |
|
|
wire [26:0] adj_op_out_sft;
|
82 |
|
|
reg fracta_lt_fractb, fracta_eq_fractb;
|
83 |
|
|
wire nan_sign1;
|
84 |
|
|
reg nan_sign;
|
85 |
|
|
|
86 |
|
|
////////////////////////////////////////////////////////////////////////
|
87 |
|
|
//
|
88 |
|
|
// Aliases
|
89 |
|
|
//
|
90 |
|
|
|
91 |
|
|
assign signa = opa[31];
|
92 |
|
|
assign signb = opb[31];
|
93 |
|
|
assign expa = opa[30:23];
|
94 |
|
|
assign expb = opb[30:23];
|
95 |
|
|
assign fracta = opa[22:0];
|
96 |
|
|
assign fractb = opb[22:0];
|
97 |
|
|
|
98 |
|
|
////////////////////////////////////////////////////////////////////////
|
99 |
|
|
//
|
100 |
|
|
// Pre-Normalize exponents (and fractions)
|
101 |
|
|
//
|
102 |
|
|
|
103 |
|
|
assign expa_lt_expb = expa > expb; // expa is larger than expb
|
104 |
|
|
|
105 |
|
|
// ---------------------------------------------------------------------
|
106 |
|
|
// Normalize
|
107 |
|
|
|
108 |
|
|
assign expa_dn = !(|expa); // opa denormalized
|
109 |
|
|
assign expb_dn = !(|expb); // opb denormalized
|
110 |
|
|
|
111 |
|
|
// ---------------------------------------------------------------------
|
112 |
|
|
// Calculate the difference between the smaller and larger exponent
|
113 |
|
|
|
114 |
|
|
wire [7:0] exp_diff1, exp_diff1a, exp_diff2;
|
115 |
|
|
|
116 |
|
|
assign exp_small = expa_lt_expb ? expb : expa;
|
117 |
|
|
assign exp_large = expa_lt_expb ? expa : expb;
|
118 |
|
|
assign exp_diff1 = exp_large - exp_small;
|
119 |
|
|
assign exp_diff1a = exp_diff1-1;
|
120 |
|
|
assign exp_diff2 = (expa_dn | expb_dn) ? exp_diff1a : exp_diff1;
|
121 |
|
|
assign exp_diff = (expa_dn & expb_dn) ? 8'h0 : exp_diff2;
|
122 |
|
|
|
123 |
|
|
always @(posedge clk) // If numbers are equal we should return zero
|
124 |
|
|
exp_dn_out <= #1 (!add_d & expa==expb & fracta==fractb) ? 8'h0 : exp_large;
|
125 |
|
|
|
126 |
|
|
// ---------------------------------------------------------------------
|
127 |
|
|
// Adjust the smaller fraction
|
128 |
|
|
|
129 |
|
|
|
130 |
|
|
assign op_dn = expa_lt_expb ? expb_dn : expa_dn;
|
131 |
|
|
assign adj_op = expa_lt_expb ? fractb : fracta;
|
132 |
|
|
assign adj_op_tmp = { ~op_dn, adj_op, 3'b0 }; // recover hidden bit (op_dn)
|
133 |
|
|
|
134 |
|
|
// adj_op_out is 27 bits wide, so can only be shifted 27 bits to the right
|
135 |
|
|
assign exp_lt_27 = exp_diff > 8'd27;
|
136 |
|
|
assign exp_diff_sft = exp_lt_27 ? 5'd27 : exp_diff[4:0];
|
137 |
|
|
assign adj_op_out_sft = adj_op_tmp >> exp_diff_sft;
|
138 |
|
|
assign adj_op_out = {adj_op_out_sft[26:1], adj_op_out_sft[0] | sticky };
|
139 |
|
|
|
140 |
|
|
// ---------------------------------------------------------------------
|
141 |
|
|
// Get truncated portion (sticky bit)
|
142 |
|
|
|
143 |
|
|
always @(exp_diff_sft or adj_op_tmp)
|
144 |
|
|
case(exp_diff_sft) // synopsys full_case parallel_case
|
145 |
|
|
00: sticky = 1'h0;
|
146 |
|
|
01: sticky = adj_op_tmp[0];
|
147 |
|
|
02: sticky = |adj_op_tmp[01:0];
|
148 |
|
|
03: sticky = |adj_op_tmp[02:0];
|
149 |
|
|
04: sticky = |adj_op_tmp[03:0];
|
150 |
|
|
05: sticky = |adj_op_tmp[04:0];
|
151 |
|
|
06: sticky = |adj_op_tmp[05:0];
|
152 |
|
|
07: sticky = |adj_op_tmp[06:0];
|
153 |
|
|
08: sticky = |adj_op_tmp[07:0];
|
154 |
|
|
09: sticky = |adj_op_tmp[08:0];
|
155 |
|
|
10: sticky = |adj_op_tmp[09:0];
|
156 |
|
|
11: sticky = |adj_op_tmp[10:0];
|
157 |
|
|
12: sticky = |adj_op_tmp[11:0];
|
158 |
|
|
13: sticky = |adj_op_tmp[12:0];
|
159 |
|
|
14: sticky = |adj_op_tmp[13:0];
|
160 |
|
|
15: sticky = |adj_op_tmp[14:0];
|
161 |
|
|
16: sticky = |adj_op_tmp[15:0];
|
162 |
|
|
17: sticky = |adj_op_tmp[16:0];
|
163 |
|
|
18: sticky = |adj_op_tmp[17:0];
|
164 |
|
|
19: sticky = |adj_op_tmp[18:0];
|
165 |
|
|
20: sticky = |adj_op_tmp[19:0];
|
166 |
|
|
21: sticky = |adj_op_tmp[20:0];
|
167 |
|
|
22: sticky = |adj_op_tmp[21:0];
|
168 |
|
|
23: sticky = |adj_op_tmp[22:0];
|
169 |
|
|
24: sticky = |adj_op_tmp[23:0];
|
170 |
|
|
25: sticky = |adj_op_tmp[24:0];
|
171 |
|
|
26: sticky = |adj_op_tmp[25:0];
|
172 |
|
|
27: sticky = |adj_op_tmp[26:0];
|
173 |
|
|
endcase
|
174 |
|
|
|
175 |
|
|
// ---------------------------------------------------------------------
|
176 |
|
|
// Select operands for add/sub (recover hidden bit)
|
177 |
|
|
|
178 |
|
|
assign fracta_n = expa_lt_expb ? {~expa_dn, fracta, 3'b0} : adj_op_out;
|
179 |
|
|
assign fractb_n = expa_lt_expb ? adj_op_out : {~expb_dn, fractb, 3'b0};
|
180 |
|
|
|
181 |
|
|
// ---------------------------------------------------------------------
|
182 |
|
|
// Sort operands (for sub only)
|
183 |
|
|
|
184 |
|
|
assign fractb_lt_fracta = fractb_n > fracta_n; // fractb is larger than fracta
|
185 |
|
|
assign fracta_s = fractb_lt_fracta ? fractb_n : fracta_n;
|
186 |
|
|
assign fractb_s = fractb_lt_fracta ? fracta_n : fractb_n;
|
187 |
|
|
|
188 |
|
|
always @(posedge clk)
|
189 |
|
|
fracta_out <= #1 fracta_s;
|
190 |
|
|
|
191 |
|
|
always @(posedge clk)
|
192 |
|
|
fractb_out <= #1 fractb_s;
|
193 |
|
|
|
194 |
|
|
// ---------------------------------------------------------------------
|
195 |
|
|
// Determine sign for the output
|
196 |
|
|
|
197 |
|
|
// sign: 0=Positive Number; 1=Negative Number
|
198 |
|
|
always @(signa or signb or add or fractb_lt_fracta)
|
199 |
|
|
case({signa, signb, add}) // synopsys full_case parallel_case
|
200 |
|
|
|
201 |
|
|
// Add
|
202 |
|
|
3'b0_0_1: sign_d = 0;
|
203 |
|
|
3'b0_1_1: sign_d = fractb_lt_fracta;
|
204 |
|
|
3'b1_0_1: sign_d = !fractb_lt_fracta;
|
205 |
|
|
3'b1_1_1: sign_d = 1;
|
206 |
|
|
|
207 |
|
|
// Sub
|
208 |
|
|
3'b0_0_0: sign_d = fractb_lt_fracta;
|
209 |
|
|
3'b0_1_0: sign_d = 0;
|
210 |
|
|
3'b1_0_0: sign_d = 1;
|
211 |
|
|
3'b1_1_0: sign_d = !fractb_lt_fracta;
|
212 |
|
|
endcase
|
213 |
|
|
|
214 |
|
|
always @(posedge clk)
|
215 |
|
|
sign <= #1 sign_d;
|
216 |
|
|
|
217 |
|
|
// Fix sign for ZERO result
|
218 |
|
|
always @(posedge clk)
|
219 |
|
|
signa_r <= #1 signa;
|
220 |
|
|
|
221 |
|
|
always @(posedge clk)
|
222 |
|
|
signb_r <= #1 signb;
|
223 |
|
|
|
224 |
|
|
always @(posedge clk)
|
225 |
|
|
add_r <= #1 add;
|
226 |
|
|
|
227 |
|
|
always @(posedge clk)
|
228 |
|
|
result_zero_sign <= #1 ( add_r & signa_r & signb_r) |
|
229 |
|
|
(!add_r & signa_r & !signb_r) |
|
230 |
|
|
( add_r & (signa_r | signb_r) & (rmode==3)) |
|
231 |
|
|
(!add_r & (signa_r == signb_r) & (rmode==3));
|
232 |
|
|
|
233 |
|
|
// Fix sign for NAN result
|
234 |
|
|
always @(posedge clk)
|
235 |
|
|
fracta_lt_fractb <= #1 fracta < fractb;
|
236 |
|
|
|
237 |
|
|
always @(posedge clk)
|
238 |
|
|
fracta_eq_fractb <= #1 fracta == fractb;
|
239 |
|
|
|
240 |
|
|
assign nan_sign1 = fracta_eq_fractb ? (signa_r & signb_r) : fracta_lt_fractb ? signb_r : signa_r;
|
241 |
|
|
|
242 |
|
|
always @(posedge clk)
|
243 |
|
|
nan_sign <= #1 (opa_nan & opb_nan) ? nan_sign1 : opb_nan ? signb_r : signa_r;
|
244 |
|
|
|
245 |
|
|
////////////////////////////////////////////////////////////////////////
|
246 |
|
|
//
|
247 |
|
|
// Decode Add/Sub operation
|
248 |
|
|
//
|
249 |
|
|
|
250 |
|
|
// add: 1=Add; 0=Subtract
|
251 |
|
|
always @(signa or signb or add)
|
252 |
|
|
case({signa, signb, add}) // synopsys full_case parallel_case
|
253 |
|
|
|
254 |
|
|
// Add
|
255 |
|
|
3'b0_0_1: add_d = 1;
|
256 |
|
|
3'b0_1_1: add_d = 0;
|
257 |
|
|
3'b1_0_1: add_d = 0;
|
258 |
|
|
3'b1_1_1: add_d = 1;
|
259 |
|
|
|
260 |
|
|
// Sub
|
261 |
|
|
3'b0_0_0: add_d = 0;
|
262 |
|
|
3'b0_1_0: add_d = 1;
|
263 |
|
|
3'b1_0_0: add_d = 1;
|
264 |
|
|
3'b1_1_0: add_d = 0;
|
265 |
|
|
endcase
|
266 |
|
|
|
267 |
|
|
always @(posedge clk)
|
268 |
|
|
fasu_op <= #1 add_d;
|
269 |
|
|
|
270 |
|
|
endmodule
|