1 |
36 |
dgisselq |
////////////////////////////////////////////////////////////////////////////////
|
2 |
|
|
//
|
3 |
|
|
// Filename: hwbfly.v
|
4 |
|
|
//
|
5 |
|
|
// Project: A General Purpose Pipelined FFT Implementation
|
6 |
|
|
//
|
7 |
|
|
// Purpose: This routine is identical to the butterfly.v routine found
|
8 |
|
|
// in 'butterfly.v', save only that it uses the verilog
|
9 |
|
|
// operator '*' in hopes that the synthesizer would be able to optimize
|
10 |
|
|
// it with hardware resources.
|
11 |
|
|
//
|
12 |
|
|
// It is understood that a hardware multiply can complete its operation in
|
13 |
|
|
// a single clock.
|
14 |
|
|
//
|
15 |
|
|
// Operation:
|
16 |
|
|
//
|
17 |
|
|
// Given two inputs, A (i_left) and B (i_right), and a complex
|
18 |
|
|
// coefficient C (i_coeff), return two outputs, O1 and O2, where:
|
19 |
|
|
//
|
20 |
|
|
// O1 = A + B, and
|
21 |
|
|
// O2 = (A - B)*C
|
22 |
|
|
//
|
23 |
|
|
// This operation is commonly known as a Decimation in Frequency (DIF)
|
24 |
|
|
// Radix-2 Butterfly.
|
25 |
|
|
// O1 and O2 are rounded before being returned in (o_left) and o_right
|
26 |
|
|
// to OWIDTH bits. If SHIFT is one, an extra bit is dropped from these
|
27 |
|
|
// values during the rounding process.
|
28 |
|
|
//
|
29 |
|
|
// Further, since these outputs will take some number of clocks to
|
30 |
|
|
// calculate, we'll pipe a value (i_aux) through the system and return
|
31 |
|
|
// it with the results (o_aux), so you can synchronize to the outgoing
|
32 |
|
|
// output stream.
|
33 |
|
|
//
|
34 |
|
|
//
|
35 |
|
|
// Creator: Dan Gisselquist, Ph.D.
|
36 |
|
|
// Gisselquist Technology, LLC
|
37 |
|
|
//
|
38 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
39 |
|
|
//
|
40 |
|
|
// Copyright (C) 2015-2018, Gisselquist Technology, LLC
|
41 |
|
|
//
|
42 |
39 |
dgisselq |
// This file is part of the general purpose pipelined FFT project.
|
43 |
36 |
dgisselq |
//
|
44 |
39 |
dgisselq |
// The pipelined FFT project is free software (firmware): you can redistribute
|
45 |
|
|
// it and/or modify it under the terms of the GNU Lesser General Public License
|
46 |
|
|
// as published by the Free Software Foundation, either version 3 of the
|
47 |
|
|
// License, or (at your option) any later version.
|
48 |
36 |
dgisselq |
//
|
49 |
39 |
dgisselq |
// The pipelined FFT project is distributed in the hope that it will be useful,
|
50 |
|
|
// but WITHOUT ANY WARRANTY; without even the implied warranty of
|
51 |
|
|
// MERCHANTIBILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser
|
52 |
|
|
// General Public License for more details.
|
53 |
|
|
//
|
54 |
|
|
// You should have received a copy of the GNU Lesser General Public License
|
55 |
|
|
// along with this program. (It's in the $(ROOT)/doc directory. Run make
|
56 |
|
|
// with no target there if the PDF file isn't present.) If not, see
|
57 |
36 |
dgisselq |
// <http://www.gnu.org/licenses/> for a copy.
|
58 |
|
|
//
|
59 |
39 |
dgisselq |
// License: LGPL, v3, as defined and found on www.gnu.org,
|
60 |
|
|
// http://www.gnu.org/licenses/lgpl.html
|
61 |
36 |
dgisselq |
//
|
62 |
|
|
//
|
63 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
64 |
|
|
//
|
65 |
|
|
//
|
66 |
|
|
`default_nettype none
|
67 |
|
|
//
|
68 |
|
|
module hwbfly(i_clk, i_reset, i_ce, i_coef, i_left, i_right, i_aux,
|
69 |
|
|
o_left, o_right, o_aux);
|
70 |
|
|
// Public changeable parameters ...
|
71 |
|
|
// - IWIDTH, number of bits in each component of the input
|
72 |
|
|
// - CWIDTH, number of bits in each component of the twiddle factor
|
73 |
|
|
// - OWIDTH, number of bits in each component of the output
|
74 |
|
|
parameter IWIDTH=16,CWIDTH=IWIDTH+4,OWIDTH=IWIDTH+1;
|
75 |
|
|
// Drop an additional bit on the output?
|
76 |
|
|
parameter SHIFT=0;
|
77 |
|
|
// The number of clocks per clock enable, 1, 2, or 3.
|
78 |
|
|
parameter [1:0] CKPCE=1;
|
79 |
|
|
//
|
80 |
39 |
dgisselq |
input wire i_clk, i_reset, i_ce;
|
81 |
|
|
input wire [(2*CWIDTH-1):0] i_coef;
|
82 |
|
|
input wire [(2*IWIDTH-1):0] i_left, i_right;
|
83 |
|
|
input wire i_aux;
|
84 |
36 |
dgisselq |
output wire [(2*OWIDTH-1):0] o_left, o_right;
|
85 |
|
|
output reg o_aux;
|
86 |
|
|
|
87 |
|
|
|
88 |
|
|
reg [(2*IWIDTH-1):0] r_left, r_right;
|
89 |
|
|
reg r_aux, r_aux_2;
|
90 |
|
|
reg [(2*CWIDTH-1):0] r_coef;
|
91 |
|
|
wire signed [(IWIDTH-1):0] r_left_r, r_left_i, r_right_r, r_right_i;
|
92 |
|
|
assign r_left_r = r_left[ (2*IWIDTH-1):(IWIDTH)];
|
93 |
|
|
assign r_left_i = r_left[ (IWIDTH-1):0];
|
94 |
|
|
assign r_right_r = r_right[(2*IWIDTH-1):(IWIDTH)];
|
95 |
|
|
assign r_right_i = r_right[(IWIDTH-1):0];
|
96 |
|
|
reg signed [(CWIDTH-1):0] ir_coef_r, ir_coef_i;
|
97 |
|
|
|
98 |
|
|
reg signed [(IWIDTH):0] r_sum_r, r_sum_i, r_dif_r, r_dif_i;
|
99 |
|
|
|
100 |
|
|
reg [(2*IWIDTH+2):0] leftv, leftvv;
|
101 |
|
|
|
102 |
|
|
// Set up the input to the multiply
|
103 |
|
|
initial r_aux = 1'b0;
|
104 |
|
|
initial r_aux_2 = 1'b0;
|
105 |
|
|
always @(posedge i_clk)
|
106 |
|
|
if (i_reset)
|
107 |
|
|
begin
|
108 |
|
|
r_aux <= 1'b0;
|
109 |
|
|
r_aux_2 <= 1'b0;
|
110 |
|
|
end else if (i_ce)
|
111 |
|
|
begin
|
112 |
|
|
// One clock just latches the inputs
|
113 |
|
|
r_aux <= i_aux;
|
114 |
|
|
// Next clock adds/subtracts
|
115 |
|
|
// Other inputs are simply delayed on second clock
|
116 |
|
|
r_aux_2 <= r_aux;
|
117 |
|
|
end
|
118 |
|
|
always @(posedge i_clk)
|
119 |
|
|
if (i_ce)
|
120 |
|
|
begin
|
121 |
|
|
// One clock just latches the inputs
|
122 |
|
|
r_left <= i_left; // No change in # of bits
|
123 |
|
|
r_right <= i_right;
|
124 |
|
|
r_coef <= i_coef;
|
125 |
|
|
// Next clock adds/subtracts
|
126 |
|
|
r_sum_r <= r_left_r + r_right_r; // Now IWIDTH+1 bits
|
127 |
|
|
r_sum_i <= r_left_i + r_right_i;
|
128 |
|
|
r_dif_r <= r_left_r - r_right_r;
|
129 |
|
|
r_dif_i <= r_left_i - r_right_i;
|
130 |
|
|
// Other inputs are simply delayed on second clock
|
131 |
|
|
ir_coef_r <= r_coef[(2*CWIDTH-1):CWIDTH];
|
132 |
|
|
ir_coef_i <= r_coef[(CWIDTH-1):0];
|
133 |
|
|
end
|
134 |
|
|
|
135 |
|
|
|
136 |
|
|
// See comments in the butterfly.v source file for a discussion of
|
137 |
|
|
// these operations and the appropriate bit widths.
|
138 |
|
|
|
139 |
|
|
wire signed [((IWIDTH+1)+(CWIDTH)-1):0] p_one, p_two;
|
140 |
|
|
wire signed [((IWIDTH+2)+(CWIDTH+1)-1):0] p_three;
|
141 |
|
|
|
142 |
|
|
initial leftv = 0;
|
143 |
|
|
initial leftvv = 0;
|
144 |
|
|
always @(posedge i_clk)
|
145 |
|
|
if (i_reset)
|
146 |
|
|
begin
|
147 |
|
|
leftv <= 0;
|
148 |
|
|
leftvv <= 0;
|
149 |
|
|
end else if (i_ce)
|
150 |
|
|
begin
|
151 |
|
|
// Second clock, pipeline = 1
|
152 |
|
|
leftv <= { r_aux_2, r_sum_r, r_sum_i };
|
153 |
|
|
|
154 |
|
|
// Third clock, pipeline = 3
|
155 |
|
|
// As desired, each of these lines infers a DSP48
|
156 |
|
|
leftvv <= leftv;
|
157 |
|
|
end
|
158 |
|
|
|
159 |
|
|
generate if (CKPCE <= 1)
|
160 |
|
|
begin : CKPCE_ONE
|
161 |
|
|
// Coefficient multiply inputs
|
162 |
|
|
reg signed [(CWIDTH-1):0] p1c_in, p2c_in;
|
163 |
|
|
// Data multiply inputs
|
164 |
|
|
reg signed [(IWIDTH):0] p1d_in, p2d_in;
|
165 |
|
|
// Product 3, coefficient input
|
166 |
|
|
reg signed [(CWIDTH):0] p3c_in;
|
167 |
|
|
// Product 3, data input
|
168 |
|
|
reg signed [(IWIDTH+1):0] p3d_in;
|
169 |
|
|
|
170 |
|
|
reg signed [((IWIDTH+1)+(CWIDTH)-1):0] rp_one, rp_two;
|
171 |
|
|
reg signed [((IWIDTH+2)+(CWIDTH+1)-1):0] rp_three;
|
172 |
|
|
|
173 |
|
|
always @(posedge i_clk)
|
174 |
|
|
if (i_ce)
|
175 |
|
|
begin
|
176 |
|
|
// Second clock, pipeline = 1
|
177 |
|
|
p1c_in <= ir_coef_r;
|
178 |
|
|
p2c_in <= ir_coef_i;
|
179 |
|
|
p1d_in <= r_dif_r;
|
180 |
|
|
p2d_in <= r_dif_i;
|
181 |
|
|
p3c_in <= ir_coef_i + ir_coef_r;
|
182 |
|
|
p3d_in <= r_dif_r + r_dif_i;
|
183 |
|
|
end
|
184 |
|
|
|
185 |
|
|
`ifndef FORMAL
|
186 |
|
|
always @(posedge i_clk)
|
187 |
|
|
if (i_ce)
|
188 |
|
|
begin
|
189 |
|
|
// Third clock, pipeline = 3
|
190 |
|
|
// As desired, each of these lines infers a DSP48
|
191 |
|
|
rp_one <= p1c_in * p1d_in;
|
192 |
|
|
rp_two <= p2c_in * p2d_in;
|
193 |
|
|
rp_three <= p3c_in * p3d_in;
|
194 |
|
|
end
|
195 |
|
|
`else
|
196 |
|
|
wire signed [((IWIDTH+1)+(CWIDTH)-1):0] pre_rp_one, pre_rp_two;
|
197 |
|
|
wire signed [((IWIDTH+2)+(CWIDTH+1)-1):0] pre_rp_three;
|
198 |
|
|
|
199 |
|
|
abs_mpy #(CWIDTH,IWIDTH+1,1'b1)
|
200 |
|
|
onei(p1c_in, p1d_in, pre_rp_one);
|
201 |
|
|
abs_mpy #(CWIDTH,IWIDTH+1,1'b1)
|
202 |
|
|
twoi(p2c_in, p2d_in, pre_rp_two);
|
203 |
|
|
abs_mpy #(CWIDTH+1,IWIDTH+2,1'b1)
|
204 |
|
|
threei(p3c_in, p3d_in, pre_rp_three);
|
205 |
|
|
|
206 |
|
|
always @(posedge i_clk)
|
207 |
|
|
if (i_ce)
|
208 |
|
|
begin
|
209 |
|
|
rp_one = pre_rp_one;
|
210 |
|
|
rp_two = pre_rp_two;
|
211 |
|
|
rp_three = pre_rp_three;
|
212 |
|
|
end
|
213 |
|
|
`endif // FORMAL
|
214 |
|
|
|
215 |
|
|
assign p_one = rp_one;
|
216 |
|
|
assign p_two = rp_two;
|
217 |
|
|
assign p_three = rp_three;
|
218 |
|
|
|
219 |
|
|
end else if (CKPCE <= 2)
|
220 |
|
|
begin : CKPCE_TWO
|
221 |
|
|
// Coefficient multiply inputs
|
222 |
|
|
reg [2*(CWIDTH)-1:0] mpy_pipe_c;
|
223 |
|
|
// Data multiply inputs
|
224 |
|
|
reg [2*(IWIDTH+1)-1:0] mpy_pipe_d;
|
225 |
|
|
wire signed [(CWIDTH-1):0] mpy_pipe_vc;
|
226 |
|
|
wire signed [(IWIDTH):0] mpy_pipe_vd;
|
227 |
|
|
//
|
228 |
|
|
reg signed [(CWIDTH+1)-1:0] mpy_cof_sum;
|
229 |
|
|
reg signed [(IWIDTH+2)-1:0] mpy_dif_sum;
|
230 |
|
|
|
231 |
|
|
assign mpy_pipe_vc = mpy_pipe_c[2*(CWIDTH)-1:CWIDTH];
|
232 |
|
|
assign mpy_pipe_vd = mpy_pipe_d[2*(IWIDTH+1)-1:IWIDTH+1];
|
233 |
|
|
|
234 |
|
|
reg mpy_pipe_v;
|
235 |
|
|
reg ce_phase;
|
236 |
|
|
|
237 |
|
|
reg signed [(CWIDTH+IWIDTH+1)-1:0] mpy_pipe_out;
|
238 |
|
|
reg signed [IWIDTH+CWIDTH+3-1:0] longmpy;
|
239 |
|
|
|
240 |
|
|
|
241 |
|
|
initial ce_phase = 1'b1;
|
242 |
|
|
always @(posedge i_clk)
|
243 |
|
|
if (i_reset)
|
244 |
|
|
ce_phase <= 1'b1;
|
245 |
|
|
else if (i_ce)
|
246 |
|
|
ce_phase <= 1'b0;
|
247 |
|
|
else
|
248 |
|
|
ce_phase <= 1'b1;
|
249 |
|
|
|
250 |
|
|
always @(*)
|
251 |
|
|
mpy_pipe_v = (i_ce)||(!ce_phase);
|
252 |
|
|
|
253 |
|
|
always @(posedge i_clk)
|
254 |
|
|
if (!ce_phase)
|
255 |
|
|
begin
|
256 |
|
|
// Pre-clock
|
257 |
|
|
mpy_pipe_c[2*CWIDTH-1:0] <=
|
258 |
|
|
{ ir_coef_r, ir_coef_i };
|
259 |
|
|
mpy_pipe_d[2*(IWIDTH+1)-1:0] <=
|
260 |
|
|
{ r_dif_r, r_dif_i };
|
261 |
|
|
|
262 |
|
|
mpy_cof_sum <= ir_coef_i + ir_coef_r;
|
263 |
|
|
mpy_dif_sum <= r_dif_r + r_dif_i;
|
264 |
|
|
|
265 |
|
|
end else if (i_ce)
|
266 |
|
|
begin
|
267 |
|
|
// First clock
|
268 |
|
|
mpy_pipe_c[2*(CWIDTH)-1:0] <= {
|
269 |
|
|
mpy_pipe_c[(CWIDTH)-1:0], {(CWIDTH){1'b0}} };
|
270 |
|
|
mpy_pipe_d[2*(IWIDTH+1)-1:0] <= {
|
271 |
|
|
mpy_pipe_d[(IWIDTH+1)-1:0], {(IWIDTH+1){1'b0}} };
|
272 |
|
|
end
|
273 |
|
|
|
274 |
|
|
`ifndef FORMAL
|
275 |
|
|
always @(posedge i_clk)
|
276 |
|
|
if (i_ce) // First clock
|
277 |
|
|
longmpy <= mpy_cof_sum * mpy_dif_sum;
|
278 |
|
|
|
279 |
|
|
always @(posedge i_clk)
|
280 |
|
|
if (mpy_pipe_v)
|
281 |
|
|
mpy_pipe_out <= mpy_pipe_vc * mpy_pipe_vd;
|
282 |
|
|
`else
|
283 |
|
|
wire signed [IWIDTH+CWIDTH+3-1:0] pre_longmpy;
|
284 |
|
|
wire signed [(CWIDTH+IWIDTH+1)-1:0] pre_mpy_pipe_out;
|
285 |
|
|
|
286 |
|
|
abs_mpy #(CWIDTH+1,IWIDTH+2,1)
|
287 |
|
|
longmpyi(mpy_cof_sum, mpy_dif_sum, pre_longmpy);
|
288 |
|
|
|
289 |
|
|
always @(posedge i_clk)
|
290 |
|
|
if (i_ce)
|
291 |
|
|
longmpy <= pre_longmpy;
|
292 |
|
|
|
293 |
|
|
|
294 |
|
|
abs_mpy #(CWIDTH,IWIDTH+1,1)
|
295 |
|
|
mpy_pipe_outi(mpy_pipe_vc, mpy_pipe_vd, pre_mpy_pipe_out);
|
296 |
|
|
|
297 |
|
|
always @(posedge i_clk)
|
298 |
|
|
if (mpy_pipe_v)
|
299 |
|
|
mpy_pipe_out <= pre_mpy_pipe_out;
|
300 |
|
|
`endif
|
301 |
|
|
|
302 |
|
|
reg signed [((IWIDTH+1)+(CWIDTH)-1):0] rp_one,
|
303 |
|
|
rp2_one, rp_two;
|
304 |
|
|
reg signed [((IWIDTH+2)+(CWIDTH+1)-1):0] rp_three;
|
305 |
|
|
|
306 |
|
|
always @(posedge i_clk)
|
307 |
|
|
if (!ce_phase) // 1.5 clock
|
308 |
|
|
rp_one <= mpy_pipe_out;
|
309 |
|
|
always @(posedge i_clk)
|
310 |
|
|
if (i_ce) // two clocks
|
311 |
|
|
rp_two <= mpy_pipe_out;
|
312 |
|
|
always @(posedge i_clk)
|
313 |
|
|
if (i_ce) // Second clock
|
314 |
|
|
rp_three<= longmpy;
|
315 |
|
|
always @(posedge i_clk)
|
316 |
|
|
if (i_ce)
|
317 |
|
|
rp2_one<= rp_one;
|
318 |
|
|
|
319 |
|
|
assign p_one = rp2_one;
|
320 |
|
|
assign p_two = rp_two;
|
321 |
|
|
assign p_three= rp_three;
|
322 |
|
|
|
323 |
|
|
end else if (CKPCE <= 2'b11)
|
324 |
|
|
begin : CKPCE_THREE
|
325 |
|
|
// Coefficient multiply inputs
|
326 |
|
|
reg [3*(CWIDTH+1)-1:0] mpy_pipe_c;
|
327 |
|
|
// Data multiply inputs
|
328 |
|
|
reg [3*(IWIDTH+2)-1:0] mpy_pipe_d;
|
329 |
|
|
wire signed [(CWIDTH):0] mpy_pipe_vc;
|
330 |
|
|
wire signed [(IWIDTH+1):0] mpy_pipe_vd;
|
331 |
|
|
|
332 |
|
|
assign mpy_pipe_vc = mpy_pipe_c[3*(CWIDTH+1)-1:2*(CWIDTH+1)];
|
333 |
|
|
assign mpy_pipe_vd = mpy_pipe_d[3*(IWIDTH+2)-1:2*(IWIDTH+2)];
|
334 |
|
|
|
335 |
|
|
reg mpy_pipe_v;
|
336 |
|
|
reg [2:0] ce_phase;
|
337 |
|
|
|
338 |
|
|
reg signed [ (CWIDTH+IWIDTH+3)-1:0] mpy_pipe_out;
|
339 |
|
|
|
340 |
|
|
initial ce_phase = 3'b011;
|
341 |
|
|
always @(posedge i_clk)
|
342 |
|
|
if (i_reset)
|
343 |
|
|
ce_phase <= 3'b011;
|
344 |
|
|
else if (i_ce)
|
345 |
|
|
ce_phase <= 3'b000;
|
346 |
|
|
else if (ce_phase != 3'b011)
|
347 |
|
|
ce_phase <= ce_phase + 1'b1;
|
348 |
|
|
|
349 |
|
|
always @(*)
|
350 |
|
|
mpy_pipe_v = (i_ce)||(ce_phase < 3'b010);
|
351 |
|
|
|
352 |
|
|
always @(posedge i_clk)
|
353 |
|
|
if (ce_phase == 3'b000)
|
354 |
|
|
begin
|
355 |
|
|
// Second clock
|
356 |
|
|
mpy_pipe_c[3*(CWIDTH+1)-1:(CWIDTH+1)] <= {
|
357 |
|
|
ir_coef_r[CWIDTH-1], ir_coef_r,
|
358 |
|
|
ir_coef_i[CWIDTH-1], ir_coef_i };
|
359 |
|
|
mpy_pipe_c[CWIDTH:0] <= ir_coef_i + ir_coef_r;
|
360 |
|
|
mpy_pipe_d[3*(IWIDTH+2)-1:(IWIDTH+2)] <= {
|
361 |
|
|
r_dif_r[IWIDTH], r_dif_r,
|
362 |
|
|
r_dif_i[IWIDTH], r_dif_i };
|
363 |
|
|
mpy_pipe_d[(IWIDTH+2)-1:0] <= r_dif_r + r_dif_i;
|
364 |
|
|
|
365 |
|
|
end else if (mpy_pipe_v)
|
366 |
|
|
begin
|
367 |
|
|
mpy_pipe_c[3*(CWIDTH+1)-1:0] <= {
|
368 |
|
|
mpy_pipe_c[2*(CWIDTH+1)-1:0], {(CWIDTH+1){1'b0}} };
|
369 |
|
|
mpy_pipe_d[3*(IWIDTH+2)-1:0] <= {
|
370 |
|
|
mpy_pipe_d[2*(IWIDTH+2)-1:0], {(IWIDTH+2){1'b0}} };
|
371 |
|
|
end
|
372 |
|
|
|
373 |
|
|
`ifndef FORMAL
|
374 |
|
|
always @(posedge i_clk)
|
375 |
|
|
if (mpy_pipe_v)
|
376 |
|
|
mpy_pipe_out <= mpy_pipe_vc * mpy_pipe_vd;
|
377 |
|
|
|
378 |
|
|
`else // FORMAL
|
379 |
|
|
wire signed [ (CWIDTH+IWIDTH+3)-1:0] pre_mpy_pipe_out;
|
380 |
|
|
|
381 |
|
|
abs_mpy #(CWIDTH+1,IWIDTH+2,1)
|
382 |
|
|
mpy_pipe_outi(mpy_pipe_vc, mpy_pipe_vd, pre_mpy_pipe_out);
|
383 |
|
|
always @(posedge i_clk)
|
384 |
|
|
if (mpy_pipe_v)
|
385 |
|
|
mpy_pipe_out <= pre_mpy_pipe_out;
|
386 |
|
|
`endif // FORMAL
|
387 |
|
|
|
388 |
|
|
reg signed [((IWIDTH+1)+(CWIDTH)-1):0] rp_one, rp_two,
|
389 |
|
|
rp2_one, rp2_two;
|
390 |
|
|
reg signed [((IWIDTH+2)+(CWIDTH+1)-1):0] rp_three, rp2_three;
|
391 |
|
|
|
392 |
|
|
always @(posedge i_clk)
|
393 |
|
|
if(i_ce)
|
394 |
|
|
rp_one <= mpy_pipe_out[(CWIDTH+IWIDTH):0];
|
395 |
|
|
always @(posedge i_clk)
|
396 |
|
|
if(ce_phase == 3'b000)
|
397 |
|
|
rp_two <= mpy_pipe_out[(CWIDTH+IWIDTH):0];
|
398 |
|
|
always @(posedge i_clk)
|
399 |
|
|
if(ce_phase == 3'b001)
|
400 |
|
|
rp_three <= mpy_pipe_out;
|
401 |
|
|
always @(posedge i_clk)
|
402 |
|
|
if (i_ce)
|
403 |
|
|
begin
|
404 |
|
|
rp2_one<= rp_one;
|
405 |
|
|
rp2_two<= rp_two;
|
406 |
|
|
rp2_three<= rp_three;
|
407 |
|
|
end
|
408 |
|
|
assign p_one = rp2_one;
|
409 |
|
|
assign p_two = rp2_two;
|
410 |
|
|
assign p_three = rp2_three;
|
411 |
|
|
|
412 |
|
|
end endgenerate
|
413 |
|
|
wire signed [((IWIDTH+2)+(CWIDTH+1)-1):0] w_one, w_two;
|
414 |
|
|
assign w_one = { {(2){p_one[((IWIDTH+1)+(CWIDTH)-1)]}}, p_one };
|
415 |
|
|
assign w_two = { {(2){p_two[((IWIDTH+1)+(CWIDTH)-1)]}}, p_two };
|
416 |
|
|
|
417 |
|
|
// These values are held in memory and delayed during the
|
418 |
|
|
// multiply. Here, we recover them. During the multiply,
|
419 |
|
|
// values were multiplied by 2^(CWIDTH-2)*exp{-j*2*pi*...},
|
420 |
|
|
// therefore, the left_x values need to be right shifted by
|
421 |
|
|
// CWIDTH-2 as well. The additional bits come from a sign
|
422 |
|
|
// extension.
|
423 |
|
|
wire aux_s;
|
424 |
|
|
wire signed [(IWIDTH+CWIDTH):0] left_si, left_sr;
|
425 |
|
|
reg [(2*IWIDTH+2):0] left_saved;
|
426 |
|
|
assign left_sr = { {2{left_saved[2*(IWIDTH+1)-1]}}, left_saved[(2*(IWIDTH+1)-1):(IWIDTH+1)], {(CWIDTH-2){1'b0}} };
|
427 |
|
|
assign left_si = { {2{left_saved[(IWIDTH+1)-1]}}, left_saved[((IWIDTH+1)-1):0], {(CWIDTH-2){1'b0}} };
|
428 |
|
|
assign aux_s = left_saved[2*IWIDTH+2];
|
429 |
|
|
|
430 |
|
|
(* use_dsp48="no" *)
|
431 |
|
|
reg signed [(CWIDTH+IWIDTH+3-1):0] mpy_r, mpy_i;
|
432 |
|
|
|
433 |
|
|
initial left_saved = 0;
|
434 |
|
|
initial o_aux = 1'b0;
|
435 |
|
|
always @(posedge i_clk)
|
436 |
|
|
if (i_reset)
|
437 |
|
|
begin
|
438 |
|
|
left_saved <= 0;
|
439 |
|
|
o_aux <= 1'b0;
|
440 |
|
|
end else if (i_ce)
|
441 |
|
|
begin
|
442 |
|
|
// First clock, recover all values
|
443 |
|
|
left_saved <= leftvv;
|
444 |
|
|
|
445 |
|
|
// Second clock, round and latch for final clock
|
446 |
|
|
o_aux <= aux_s;
|
447 |
|
|
end
|
448 |
|
|
always @(posedge i_clk)
|
449 |
|
|
if (i_ce)
|
450 |
|
|
begin
|
451 |
|
|
// These values are IWIDTH+CWIDTH+3 bits wide
|
452 |
|
|
// although they only need to be (IWIDTH+1)
|
453 |
|
|
// + (CWIDTH) bits wide. (We've got two
|
454 |
|
|
// extra bits we need to get rid of.)
|
455 |
|
|
|
456 |
|
|
// These two lines also infer DSP48's.
|
457 |
|
|
// To keep from using extra DSP48 resources,
|
458 |
|
|
// they are prevented from using DSP48's
|
459 |
|
|
// by the (* use_dsp48 ... *) comment above.
|
460 |
|
|
mpy_r <= w_one - w_two;
|
461 |
|
|
mpy_i <= p_three - w_one - w_two;
|
462 |
|
|
end
|
463 |
|
|
|
464 |
|
|
// Round the results
|
465 |
|
|
wire signed [(OWIDTH-1):0] rnd_left_r, rnd_left_i, rnd_right_r, rnd_right_i;
|
466 |
|
|
|
467 |
|
|
convround #(CWIDTH+IWIDTH+1,OWIDTH,SHIFT+2) do_rnd_left_r(i_clk, i_ce,
|
468 |
|
|
left_sr, rnd_left_r);
|
469 |
|
|
|
470 |
|
|
convround #(CWIDTH+IWIDTH+1,OWIDTH,SHIFT+2) do_rnd_left_i(i_clk, i_ce,
|
471 |
|
|
left_si, rnd_left_i);
|
472 |
|
|
|
473 |
|
|
convround #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_right_r(i_clk, i_ce,
|
474 |
|
|
mpy_r, rnd_right_r);
|
475 |
|
|
|
476 |
|
|
convround #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_right_i(i_clk, i_ce,
|
477 |
|
|
mpy_i, rnd_right_i);
|
478 |
|
|
|
479 |
|
|
// As a final step, we pack our outputs into two packed two's
|
480 |
|
|
// complement numbers per output word, so that each output word
|
481 |
|
|
// has (2*OWIDTH) bits in it, with the top half being the real
|
482 |
|
|
// portion and the bottom half being the imaginary portion.
|
483 |
|
|
assign o_left = { rnd_left_r, rnd_left_i };
|
484 |
|
|
assign o_right= { rnd_right_r,rnd_right_i};
|
485 |
|
|
|
486 |
|
|
`ifdef FORMAL
|
487 |
|
|
localparam F_LGDEPTH = 3;
|
488 |
|
|
localparam F_DEPTH = 5;
|
489 |
|
|
localparam [F_LGDEPTH-1:0] F_D = F_DEPTH-1;
|
490 |
|
|
|
491 |
|
|
reg signed [IWIDTH-1:0] f_dlyleft_r [0:F_DEPTH-1];
|
492 |
|
|
reg signed [IWIDTH-1:0] f_dlyleft_i [0:F_DEPTH-1];
|
493 |
|
|
reg signed [IWIDTH-1:0] f_dlyright_r [0:F_DEPTH-1];
|
494 |
|
|
reg signed [IWIDTH-1:0] f_dlyright_i [0:F_DEPTH-1];
|
495 |
|
|
reg signed [CWIDTH-1:0] f_dlycoeff_r [0:F_DEPTH-1];
|
496 |
|
|
reg signed [CWIDTH-1:0] f_dlycoeff_i [0:F_DEPTH-1];
|
497 |
|
|
reg signed [F_DEPTH-1:0] f_dlyaux;
|
498 |
|
|
|
499 |
|
|
always @(posedge i_clk)
|
500 |
|
|
if (i_reset)
|
501 |
|
|
f_dlyaux <= 0;
|
502 |
|
|
else if (i_ce)
|
503 |
|
|
f_dlyaux <= { f_dlyaux[F_DEPTH-2:0], i_aux };
|
504 |
|
|
|
505 |
|
|
always @(posedge i_clk)
|
506 |
|
|
if (i_ce)
|
507 |
|
|
begin
|
508 |
|
|
f_dlyleft_r[0] <= i_left[ (2*IWIDTH-1):IWIDTH];
|
509 |
|
|
f_dlyleft_i[0] <= i_left[ ( IWIDTH-1):0];
|
510 |
|
|
f_dlyright_r[0] <= i_right[(2*IWIDTH-1):IWIDTH];
|
511 |
|
|
f_dlyright_i[0] <= i_right[( IWIDTH-1):0];
|
512 |
|
|
f_dlycoeff_r[0] <= i_coef[ (2*CWIDTH-1):CWIDTH];
|
513 |
|
|
f_dlycoeff_i[0] <= i_coef[ ( CWIDTH-1):0];
|
514 |
|
|
end
|
515 |
|
|
|
516 |
|
|
genvar k;
|
517 |
|
|
generate for(k=1; k<F_DEPTH; k=k+1)
|
518 |
|
|
|
519 |
|
|
always @(posedge i_clk)
|
520 |
|
|
if (i_ce)
|
521 |
|
|
begin
|
522 |
|
|
f_dlyleft_r[k] <= f_dlyleft_r[ k-1];
|
523 |
|
|
f_dlyleft_i[k] <= f_dlyleft_i[ k-1];
|
524 |
|
|
f_dlyright_r[k] <= f_dlyright_r[k-1];
|
525 |
|
|
f_dlyright_i[k] <= f_dlyright_i[k-1];
|
526 |
|
|
f_dlycoeff_r[k] <= f_dlycoeff_r[k-1];
|
527 |
|
|
f_dlycoeff_i[k] <= f_dlycoeff_i[k-1];
|
528 |
|
|
end
|
529 |
|
|
|
530 |
|
|
endgenerate
|
531 |
|
|
|
532 |
|
|
`ifdef VERILATOR
|
533 |
|
|
`else
|
534 |
|
|
always @(posedge i_clk)
|
535 |
|
|
if ((!$past(i_ce))&&(!$past(i_ce,2))&&(!$past(i_ce,3))
|
536 |
|
|
&&(!$past(i_ce,4)))
|
537 |
|
|
assume(i_ce);
|
538 |
|
|
|
539 |
|
|
generate if (CKPCE <= 1)
|
540 |
|
|
begin
|
541 |
|
|
|
542 |
|
|
// i_ce is allowed to be anything in this mode
|
543 |
|
|
|
544 |
|
|
end else if (CKPCE == 2)
|
545 |
|
|
begin : F_CKPCE_TWO
|
546 |
|
|
|
547 |
|
|
always @(posedge i_clk)
|
548 |
|
|
if ($past(i_ce))
|
549 |
|
|
assume(!i_ce);
|
550 |
|
|
|
551 |
|
|
end else if (CKPCE == 3)
|
552 |
|
|
begin : F_CKPCE_THREE
|
553 |
|
|
|
554 |
|
|
always @(posedge i_clk)
|
555 |
|
|
if (($past(i_ce))||($past(i_ce,2)))
|
556 |
|
|
assume(!i_ce);
|
557 |
|
|
|
558 |
|
|
end endgenerate
|
559 |
|
|
`endif
|
560 |
|
|
reg [F_LGDEPTH-1:0] f_startup_counter;
|
561 |
|
|
initial f_startup_counter = 0;
|
562 |
|
|
always @(posedge i_clk)
|
563 |
|
|
if (i_reset)
|
564 |
|
|
f_startup_counter <= 0;
|
565 |
|
|
else if ((i_ce)&&(!(&f_startup_counter)))
|
566 |
|
|
f_startup_counter <= f_startup_counter + 1;
|
567 |
|
|
|
568 |
|
|
wire signed [IWIDTH:0] f_sumr, f_sumi;
|
569 |
|
|
always @(*)
|
570 |
|
|
begin
|
571 |
|
|
f_sumr = f_dlyleft_r[F_D] + f_dlyright_r[F_D];
|
572 |
|
|
f_sumi = f_dlyleft_i[F_D] + f_dlyright_i[F_D];
|
573 |
|
|
end
|
574 |
|
|
|
575 |
|
|
wire signed [IWIDTH+CWIDTH:0] f_sumrx, f_sumix;
|
576 |
|
|
assign f_sumrx = { {(2){f_sumr[IWIDTH]}}, f_sumr, {(CWIDTH-2){1'b0}} };
|
577 |
|
|
assign f_sumix = { {(2){f_sumi[IWIDTH]}}, f_sumi, {(CWIDTH-2){1'b0}} };
|
578 |
|
|
|
579 |
|
|
wire signed [IWIDTH:0] f_difr, f_difi;
|
580 |
|
|
always @(*)
|
581 |
|
|
begin
|
582 |
|
|
f_difr = f_dlyleft_r[F_D] - f_dlyright_r[F_D];
|
583 |
|
|
f_difi = f_dlyleft_i[F_D] - f_dlyright_i[F_D];
|
584 |
|
|
end
|
585 |
|
|
|
586 |
|
|
wire signed [IWIDTH+CWIDTH+3-1:0] f_difrx, f_difix;
|
587 |
|
|
assign f_difrx = { {(CWIDTH+2){f_difr[IWIDTH]}}, f_difr };
|
588 |
|
|
assign f_difix = { {(CWIDTH+2){f_difi[IWIDTH]}}, f_difi };
|
589 |
|
|
|
590 |
|
|
wire signed [IWIDTH+CWIDTH+3-1:0] f_widecoeff_r, f_widecoeff_i;
|
591 |
|
|
assign f_widecoeff_r = {{(IWIDTH+3){f_dlycoeff_r[F_D][CWIDTH-1]}},
|
592 |
|
|
f_dlycoeff_r[F_D] };
|
593 |
|
|
assign f_widecoeff_i = {{(IWIDTH+3){f_dlycoeff_i[F_D][CWIDTH-1]}},
|
594 |
|
|
f_dlycoeff_i[F_D] };
|
595 |
|
|
|
596 |
|
|
always @(posedge i_clk)
|
597 |
|
|
if (f_startup_counter > F_D)
|
598 |
|
|
begin
|
599 |
|
|
assert(left_sr == f_sumrx);
|
600 |
|
|
assert(left_si == f_sumix);
|
601 |
|
|
assert(aux_s == f_dlyaux[F_D]);
|
602 |
|
|
|
603 |
|
|
if ((f_difr == 0)&&(f_difi == 0))
|
604 |
|
|
begin
|
605 |
|
|
assert(mpy_r == 0);
|
606 |
|
|
assert(mpy_i == 0);
|
607 |
|
|
end else if ((f_dlycoeff_r[F_D] == 0)
|
608 |
|
|
&&(f_dlycoeff_i[F_D] == 0))
|
609 |
|
|
begin
|
610 |
|
|
assert(mpy_r == 0);
|
611 |
|
|
assert(mpy_i == 0);
|
612 |
|
|
end
|
613 |
|
|
|
614 |
|
|
if ((f_dlycoeff_r[F_D] == 1)&&(f_dlycoeff_i[F_D] == 0))
|
615 |
|
|
begin
|
616 |
|
|
assert(mpy_r == f_difrx);
|
617 |
|
|
assert(mpy_i == f_difix);
|
618 |
|
|
end
|
619 |
|
|
|
620 |
|
|
if ((f_dlycoeff_r[F_D] == 0)&&(f_dlycoeff_i[F_D] == 1))
|
621 |
|
|
begin
|
622 |
|
|
assert(mpy_r == -f_difix);
|
623 |
|
|
assert(mpy_i == f_difrx);
|
624 |
|
|
end
|
625 |
|
|
|
626 |
|
|
if ((f_difr == 1)&&(f_difi == 0))
|
627 |
|
|
begin
|
628 |
|
|
assert(mpy_r == f_widecoeff_r);
|
629 |
|
|
assert(mpy_i == f_widecoeff_i);
|
630 |
|
|
end
|
631 |
|
|
|
632 |
|
|
if ((f_difr == 0)&&(f_difi == 1))
|
633 |
|
|
begin
|
634 |
|
|
assert(mpy_r == -f_widecoeff_i);
|
635 |
|
|
assert(mpy_i == f_widecoeff_r);
|
636 |
|
|
end
|
637 |
|
|
end
|
638 |
|
|
|
639 |
|
|
// Let's see if we can improve our performance at all by
|
640 |
|
|
// moving our test one clock earlier. If nothing else, it should
|
641 |
|
|
// help induction finish one (or more) clocks ealier than
|
642 |
|
|
// otherwise
|
643 |
|
|
|
644 |
|
|
|
645 |
|
|
wire signed [IWIDTH:0] f_predifr, f_predifi;
|
646 |
|
|
always @(*)
|
647 |
|
|
begin
|
648 |
|
|
f_predifr = f_dlyleft_r[F_D-1] - f_dlyright_r[F_D-1];
|
649 |
|
|
f_predifi = f_dlyleft_i[F_D-1] - f_dlyright_i[F_D-1];
|
650 |
|
|
end
|
651 |
|
|
|
652 |
|
|
wire signed [IWIDTH+CWIDTH+1-1:0] f_predifrx, f_predifix;
|
653 |
|
|
assign f_predifrx = { {(CWIDTH){f_predifr[IWIDTH]}}, f_predifr };
|
654 |
|
|
assign f_predifix = { {(CWIDTH){f_predifi[IWIDTH]}}, f_predifi };
|
655 |
|
|
|
656 |
|
|
wire signed [CWIDTH:0] f_sumcoef;
|
657 |
|
|
wire signed [IWIDTH+1:0] f_sumdiff;
|
658 |
|
|
always @(*)
|
659 |
|
|
begin
|
660 |
|
|
f_sumcoef = f_dlycoeff_r[F_D-1] + f_dlycoeff_i[F_D-1];
|
661 |
|
|
f_sumdiff = f_predifr + f_predifi;
|
662 |
|
|
end
|
663 |
|
|
|
664 |
|
|
// Induction helpers
|
665 |
|
|
always @(posedge i_clk)
|
666 |
|
|
if (f_startup_counter >= F_D)
|
667 |
|
|
begin
|
668 |
|
|
if (f_dlycoeff_r[F_D-1] == 0)
|
669 |
|
|
assert(p_one == 0);
|
670 |
|
|
if (f_dlycoeff_i[F_D-1] == 0)
|
671 |
|
|
assert(p_two == 0);
|
672 |
|
|
|
673 |
|
|
if (f_dlycoeff_r[F_D-1] == 1)
|
674 |
|
|
assert(p_one == f_predifrx);
|
675 |
|
|
if (f_dlycoeff_i[F_D-1] == 1)
|
676 |
|
|
assert(p_two == f_predifix);
|
677 |
|
|
|
678 |
|
|
if (f_predifr == 0)
|
679 |
|
|
assert(p_one == 0);
|
680 |
|
|
if (f_predifi == 0)
|
681 |
|
|
assert(p_two == 0);
|
682 |
|
|
|
683 |
|
|
// verilator lint_off WIDTH
|
684 |
|
|
if (f_predifr == 1)
|
685 |
|
|
assert(p_one == f_dlycoeff_r[F_D-1]);
|
686 |
|
|
if (f_predifi == 1)
|
687 |
|
|
assert(p_two == f_dlycoeff_i[F_D-1]);
|
688 |
|
|
// verilator lint_on WIDTH
|
689 |
|
|
|
690 |
|
|
if (f_sumcoef == 0)
|
691 |
|
|
assert(p_three == 0);
|
692 |
|
|
if (f_sumdiff == 0)
|
693 |
|
|
assert(p_three == 0);
|
694 |
|
|
// verilator lint_off WIDTH
|
695 |
|
|
if (f_sumcoef == 1)
|
696 |
|
|
assert(p_three == f_sumdiff);
|
697 |
|
|
if (f_sumdiff == 1)
|
698 |
|
|
assert(p_three == f_sumcoef);
|
699 |
|
|
// verilator lint_on WIDTH
|
700 |
|
|
`ifdef VERILATOR
|
701 |
|
|
assert(p_one == f_predifr * f_dlycoeff_r[F_D-1]);
|
702 |
|
|
assert(p_two == f_predifi * f_dlycoeff_i[F_D-1]);
|
703 |
|
|
assert(p_three == f_sumdiff * f_sumcoef);
|
704 |
|
|
`endif // VERILATOR
|
705 |
|
|
end
|
706 |
|
|
|
707 |
|
|
`endif // FORMAL
|
708 |
|
|
endmodule
|