1 |
36 |
dgisselq |
////////////////////////////////////////////////////////////////////////////////
|
2 |
|
|
//
|
3 |
|
|
// Filename: butterfly.cpp
|
4 |
|
|
//
|
5 |
|
|
// Project: A General Purpose Pipelined FFT Implementation
|
6 |
|
|
//
|
7 |
37 |
dgisselq |
// Purpose: Builds one of two butterflies: either a butterfly implementation
|
8 |
|
|
// using hardware optimized multiplies, or one that uses a logic
|
9 |
|
|
// soft-multiply.
|
10 |
36 |
dgisselq |
//
|
11 |
|
|
// Creator: Dan Gisselquist, Ph.D.
|
12 |
|
|
// Gisselquist Technology, LLC
|
13 |
|
|
//
|
14 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
15 |
|
|
//
|
16 |
|
|
// Copyright (C) 2015-2018, Gisselquist Technology, LLC
|
17 |
|
|
//
|
18 |
37 |
dgisselq |
// This file is part of the general purpose pipelined FFT project.
|
19 |
36 |
dgisselq |
//
|
20 |
37 |
dgisselq |
// The pipelined FFT project is free software (firmware): you can redistribute
|
21 |
|
|
// it and/or modify it under the terms of the GNU Lesser General Public License
|
22 |
|
|
// as published by the Free Software Foundation, either version 3 of the
|
23 |
|
|
// License, or (at your option) any later version.
|
24 |
36 |
dgisselq |
//
|
25 |
37 |
dgisselq |
// The pipelined FFT project is distributed in the hope that it will be useful,
|
26 |
|
|
// but WITHOUT ANY WARRANTY; without even the implied warranty of
|
27 |
|
|
// MERCHANTIBILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser
|
28 |
|
|
// General Public License for more details.
|
29 |
|
|
//
|
30 |
|
|
// You should have received a copy of the GNU Lesser General Public License
|
31 |
|
|
// along with this program. (It's in the $(ROOT)/doc directory. Run make
|
32 |
|
|
// with no target there if the PDF file isn't present.) If not, see
|
33 |
36 |
dgisselq |
// <http://www.gnu.org/licenses/> for a copy.
|
34 |
|
|
//
|
35 |
37 |
dgisselq |
// License: LGPL, v3, as defined and found on www.gnu.org,
|
36 |
|
|
// http://www.gnu.org/licenses/lgpl.html
|
37 |
36 |
dgisselq |
//
|
38 |
|
|
//
|
39 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
40 |
|
|
//
|
41 |
|
|
//
|
42 |
|
|
#define _CRT_SECURE_NO_WARNINGS // ms vs 2012 doesn't like fopen
|
43 |
|
|
#include <stdio.h>
|
44 |
|
|
#include <stdlib.h>
|
45 |
|
|
|
46 |
|
|
#ifdef _MSC_VER // added for ms vs compatibility
|
47 |
|
|
|
48 |
|
|
#include <io.h>
|
49 |
|
|
#include <direct.h>
|
50 |
|
|
#define _USE_MATH_DEFINES
|
51 |
|
|
#define R_OK 4 /* Test for read permission. */
|
52 |
|
|
#define W_OK 2 /* Test for write permission. */
|
53 |
|
|
#define X_OK 0 /* !!!!!! execute permission - unsupported in windows*/
|
54 |
|
|
#define F_OK 0 /* Test for existence. */
|
55 |
|
|
|
56 |
|
|
#if _MSC_VER <= 1700
|
57 |
|
|
|
58 |
|
|
int lstat(const char *filename, struct stat *buf) { return 1; };
|
59 |
|
|
#define S_ISDIR(A) 0
|
60 |
|
|
|
61 |
|
|
#else
|
62 |
|
|
|
63 |
|
|
#define lstat _stat
|
64 |
|
|
#define S_ISDIR _S_IFDIR
|
65 |
|
|
|
66 |
|
|
#endif
|
67 |
|
|
|
68 |
|
|
#define mkdir(A,B) _mkdir(A)
|
69 |
|
|
|
70 |
|
|
#define access _access
|
71 |
|
|
|
72 |
|
|
#else
|
73 |
|
|
// And for G++/Linux environment
|
74 |
|
|
|
75 |
|
|
#include <unistd.h> // Defines the R_OK/W_OK/etc. macros
|
76 |
|
|
#include <sys/stat.h>
|
77 |
|
|
#endif
|
78 |
|
|
|
79 |
|
|
#include <string.h>
|
80 |
|
|
#include <string>
|
81 |
|
|
#include <math.h>
|
82 |
|
|
#include <ctype.h>
|
83 |
|
|
#include <assert.h>
|
84 |
|
|
|
85 |
|
|
#include "defaults.h"
|
86 |
|
|
#include "legal.h"
|
87 |
|
|
#include "rounding.h"
|
88 |
|
|
#include "fftlib.h"
|
89 |
|
|
#include "bldstage.h"
|
90 |
|
|
#include "bitreverse.h"
|
91 |
|
|
#include "softmpy.h"
|
92 |
|
|
#include "butterfly.h"
|
93 |
|
|
|
94 |
|
|
void build_butterfly(const char *fname, int xtracbits, ROUND_T rounding,
|
95 |
|
|
int ckpce, const bool async_reset) {
|
96 |
|
|
FILE *fp = fopen(fname, "w");
|
97 |
|
|
if (NULL == fp) {
|
98 |
|
|
fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
|
99 |
|
|
perror("O/S Err was:");
|
100 |
|
|
return;
|
101 |
|
|
}
|
102 |
|
|
const char *rnd_string;
|
103 |
|
|
if (rounding == RND_TRUNCATE)
|
104 |
|
|
rnd_string = "truncate";
|
105 |
|
|
else if (rounding == RND_FROMZERO)
|
106 |
|
|
rnd_string = "roundfromzero";
|
107 |
|
|
else if (rounding == RND_HALFUP)
|
108 |
|
|
rnd_string = "roundhalfup";
|
109 |
|
|
else
|
110 |
|
|
rnd_string = "convround";
|
111 |
|
|
|
112 |
|
|
//if (ckpce >= 3)
|
113 |
|
|
//ckpce = 3;
|
114 |
|
|
if (ckpce <= 1)
|
115 |
|
|
ckpce = 1;
|
116 |
|
|
|
117 |
|
|
std::string resetw("i_reset");
|
118 |
|
|
if (async_reset)
|
119 |
|
|
resetw = std::string("i_areset_n");
|
120 |
|
|
|
121 |
|
|
|
122 |
|
|
fprintf(fp,
|
123 |
|
|
SLASHLINE
|
124 |
|
|
"//\n"
|
125 |
|
|
"// Filename:\tbutterfly.v\n"
|
126 |
|
|
"//\n"
|
127 |
|
|
"// Project:\t%s\n"
|
128 |
|
|
"//\n"
|
129 |
|
|
"// Purpose:\tThis routine caculates a butterfly for a decimation\n"
|
130 |
|
|
"// in frequency version of an FFT. Specifically, given\n"
|
131 |
|
|
"// complex Left and Right values together with a coefficient, the output\n"
|
132 |
|
|
"// of this routine is given by:\n"
|
133 |
|
|
"//\n"
|
134 |
|
|
"// L' = L + R\n"
|
135 |
|
|
"// R' = (L - R)*C\n"
|
136 |
|
|
"//\n"
|
137 |
|
|
"// The rest of the junk below handles timing (mostly), to make certain\n"
|
138 |
|
|
"// that L' and R' reach the output at the same clock. Further, just to\n"
|
139 |
|
|
"// make certain that is the case, an 'aux' input exists. This aux value\n"
|
140 |
|
|
"// will come out of this routine synchronized to the values it came in\n"
|
141 |
|
|
"// with. (i.e., both L', R', and aux all have the same delay.) Hence,\n"
|
142 |
|
|
"// a caller of this routine may set aux on the first input with valid\n"
|
143 |
|
|
"// data, and then wait to see aux set on the output to know when to find\n"
|
144 |
|
|
"// the first output with valid data.\n"
|
145 |
|
|
"//\n"
|
146 |
|
|
"// All bits are preserved until the very last clock, where any more bits\n"
|
147 |
|
|
"// than OWIDTH will be quietly discarded.\n"
|
148 |
|
|
"//\n"
|
149 |
|
|
"// This design features no overflow checking.\n"
|
150 |
|
|
"//\n"
|
151 |
|
|
"// Notes:\n"
|
152 |
|
|
"// CORDIC:\n"
|
153 |
|
|
"// Much as we might like, we can't use a cordic here.\n"
|
154 |
|
|
"// The goal is to accomplish an FFT, as defined, and a\n"
|
155 |
|
|
"// CORDIC places a scale factor onto the data. Removing\n"
|
156 |
|
|
"// the scale factor would cost two multiplies, which\n"
|
157 |
|
|
"// is precisely what we are trying to avoid.\n"
|
158 |
|
|
"//\n"
|
159 |
|
|
"//\n"
|
160 |
|
|
"// 3-MULTIPLIES:\n"
|
161 |
|
|
"// It should also be possible to do this with three multiplies\n"
|
162 |
|
|
"// and an extra two addition cycles.\n"
|
163 |
|
|
"//\n"
|
164 |
|
|
"// We want\n"
|
165 |
|
|
"// R+I = (a + jb) * (c + jd)\n"
|
166 |
|
|
"// R+I = (ac-bd) + j(ad+bc)\n"
|
167 |
|
|
"// We multiply\n"
|
168 |
|
|
"// P1 = ac\n"
|
169 |
|
|
"// P2 = bd\n"
|
170 |
|
|
"// P3 = (a+b)(c+d)\n"
|
171 |
|
|
"// Then\n"
|
172 |
|
|
"// R+I=(P1-P2)+j(P3-P2-P1)\n"
|
173 |
|
|
"//\n"
|
174 |
|
|
"// WIDTHS:\n"
|
175 |
|
|
"// On multiplying an X width number by an\n"
|
176 |
|
|
"// Y width number, X>Y, the result should be (X+Y)\n"
|
177 |
|
|
"// bits, right?\n"
|
178 |
|
|
"// -2^(X-1) <= a <= 2^(X-1) - 1\n"
|
179 |
|
|
"// -2^(Y-1) <= b <= 2^(Y-1) - 1\n"
|
180 |
|
|
"// (2^(Y-1)-1)*(-2^(X-1)) <= ab <= 2^(X-1)2^(Y-1)\n"
|
181 |
|
|
"// -2^(X+Y-2)+2^(X-1) <= ab <= 2^(X+Y-2) <= 2^(X+Y-1) - 1\n"
|
182 |
|
|
"// -2^(X+Y-1) <= ab <= 2^(X+Y-1)-1\n"
|
183 |
|
|
"// YUP! But just barely. Do this and you'll really want\n"
|
184 |
|
|
"// to drop a bit, although you will risk overflow in so\n"
|
185 |
|
|
"// doing.\n"
|
186 |
|
|
"//\n"
|
187 |
|
|
"// 20150602 -- The sync logic lines have been completely redone. The\n"
|
188 |
|
|
"// synchronization lines no longer go through the FIFO with the\n"
|
189 |
|
|
"// left hand sum, but are kept out of memory. This allows the\n"
|
190 |
|
|
"// butterfly to use more optimal memory resources, while also\n"
|
191 |
|
|
"// guaranteeing that the sync lines can be properly reset upon\n"
|
192 |
|
|
"// any reset signal.\n"
|
193 |
|
|
"//\n"
|
194 |
|
|
"//\n%s"
|
195 |
|
|
"//\n", prjname, creator);
|
196 |
|
|
fprintf(fp, "%s", cpyleft);
|
197 |
|
|
fprintf(fp, "//\n//\n`default_nettype\tnone\n//\n");
|
198 |
|
|
|
199 |
|
|
fprintf(fp,
|
200 |
|
|
"module\tbutterfly(i_clk, %s, i_ce, i_coef, i_left, i_right, i_aux,\n"
|
201 |
|
|
"\t\to_left, o_right, o_aux);\n"
|
202 |
|
|
"\t// Public changeable parameters ...\n", resetw.c_str());
|
203 |
|
|
|
204 |
|
|
fprintf(fp,
|
205 |
|
|
"\tparameter IWIDTH=%d,", TST_BUTTERFLY_IWIDTH);
|
206 |
|
|
#ifdef TST_BUTTERFLY_CWIDTH
|
207 |
|
|
fprintf(fp, "CWIDTH=%d,", TST_BUTTERFLY_CWIDTH);
|
208 |
|
|
#else
|
209 |
|
|
fprintf(fp, "CWIDTH=IWIDTH+%d,", xtracbits);
|
210 |
|
|
#endif
|
211 |
|
|
#ifdef TST_BUTTERFLY_OWIDTH
|
212 |
|
|
fprintf(fp, "OWIDTH=%d;\n", TST_BUTTERFLY_OWIDTH);
|
213 |
|
|
// OWIDTH = TST_BUTTERFLY_OWIDTH;
|
214 |
|
|
#else
|
215 |
|
|
fprintf(fp, "OWIDTH=IWIDTH+1;\n");
|
216 |
|
|
#endif
|
217 |
|
|
fprintf(fp, "\tparameter\tSHIFT=0;\n");
|
218 |
|
|
|
219 |
|
|
fprintf(fp,
|
220 |
|
|
"\t// The number of clocks per each i_ce. The actual number can be\n"
|
221 |
|
|
"\t// more, but the algorithm depends upon at least this many for\n"
|
222 |
|
|
"\t// extra internal processing.\n"
|
223 |
|
|
"\tparameter CKPCE=%d;\n", ckpce);
|
224 |
|
|
|
225 |
|
|
fprintf(fp,
|
226 |
|
|
"\t//\n"
|
227 |
|
|
"\t// Local/derived parameters that are calculated from the above\n"
|
228 |
|
|
"\t// params. Apart from algorithmic changes below, these should not\n"
|
229 |
|
|
"\t// be adjusted\n"
|
230 |
|
|
"\t//\n"
|
231 |
|
|
"\t// The first step is to calculate how many clocks it takes our\n"
|
232 |
|
|
"\t// multiply to come back with an answer within. The time in the\n"
|
233 |
|
|
"\t// multiply depends upon the input value with the fewest number of\n"
|
234 |
|
|
"\t// bits--to keep the pipeline depth short. So, let's find the\n"
|
235 |
|
|
"\t// fewest number of bits here.\n"
|
236 |
|
|
"\tlocalparam MXMPYBITS = \n"
|
237 |
|
|
"\t\t((IWIDTH+2)>(CWIDTH+1)) ? (CWIDTH+1) : (IWIDTH + 2);\n"
|
238 |
|
|
"\t//\n"
|
239 |
|
|
"\t// Given this \"fewest\" number of bits, we can calculate the\n"
|
240 |
|
|
"\t// number of clocks the multiply itself will take.\n"
|
241 |
|
|
"\tlocalparam MPYDELAY=((MXMPYBITS+1)/2)+2;\n"
|
242 |
|
|
"\t//\n"
|
243 |
|
|
"\t// In an environment when CKPCE > 1, the multiply delay isn\'t\n"
|
244 |
|
|
"\t// necessarily the delay felt by this algorithm--measured in\n"
|
245 |
|
|
"\t// i_ce\'s. In particular, if the multiply can operate with more\n"
|
246 |
|
|
"\t// operations per clock, it can appear to finish \"faster\".\n"
|
247 |
|
|
"\t// Since most of the logic in this core operates on the slower\n"
|
248 |
|
|
"\t// clock, we'll need to map that speed into the number of slower\n"
|
249 |
|
|
"\t// clock ticks that it takes.\n"
|
250 |
|
|
"\tlocalparam LCLDELAY = (CKPCE == 1) ? MPYDELAY\n"
|
251 |
|
|
"\t\t: (CKPCE == 2) ? (MPYDELAY/2+2)\n"
|
252 |
|
|
"\t\t: (MPYDELAY/3 + 2);\n"
|
253 |
|
|
"\tlocalparam LGDELAY = (MPYDELAY>64) ? 7\n"
|
254 |
|
|
"\t\t\t: (MPYDELAY > 32) ? 6\n"
|
255 |
|
|
"\t\t\t: (MPYDELAY > 16) ? 5\n"
|
256 |
|
|
"\t\t\t: (MPYDELAY > 8) ? 4\n"
|
257 |
|
|
"\t\t\t: (MPYDELAY > 4) ? 3\n"
|
258 |
|
|
"\t\t\t: 2;\n"
|
259 |
|
|
"\tlocalparam AUXLEN=(LCLDELAY+3);\n"
|
260 |
|
|
"\tlocalparam MPYREMAINDER = MPYDELAY - CKPCE*(MPYDELAY/CKPCE);\n"
|
261 |
|
|
"\n\n");
|
262 |
|
|
|
263 |
|
|
|
264 |
|
|
fprintf(fp,
|
265 |
37 |
dgisselq |
"\tinput\twire\ti_clk, %s, i_ce;\n"
|
266 |
|
|
"\tinput\twire\t[(2*CWIDTH-1):0] i_coef;\n"
|
267 |
|
|
"\tinput\twire\t[(2*IWIDTH-1):0] i_left, i_right;\n"
|
268 |
|
|
"\tinput\twire\ti_aux;\n"
|
269 |
36 |
dgisselq |
"\toutput\twire [(2*OWIDTH-1):0] o_left, o_right;\n"
|
270 |
|
|
"\toutput\treg\to_aux;\n\n", resetw.c_str());
|
271 |
37 |
dgisselq |
|
272 |
|
|
if (formal_property_flag) fprintf(fp,
|
273 |
|
|
"`ifdef FORMAL\n"
|
274 |
|
|
"\tlocalparam F_LGDEPTH = (AUXLEN > 64) ? 7\n"
|
275 |
|
|
"\t\t\t: (AUXLEN > 32) ? 6\n"
|
276 |
|
|
"\t\t\t: (AUXLEN > 16) ? 5\n"
|
277 |
|
|
"\t\t\t: (AUXLEN > 8) ? 4\n"
|
278 |
|
|
"\t\t\t: (AUXLEN > 4) ? 3 : 2;\n"
|
279 |
|
|
"\n"
|
280 |
|
|
"\tlocalparam F_DEPTH = AUXLEN;\n"
|
281 |
|
|
"\tlocalparam [F_LGDEPTH-1:0] F_D = F_DEPTH[F_LGDEPTH-1:0]-1;\n"
|
282 |
|
|
"\n"
|
283 |
|
|
"\treg signed [IWIDTH-1:0] f_dlyleft_r [0:F_DEPTH-1];\n"
|
284 |
|
|
"\treg signed [IWIDTH-1:0] f_dlyleft_i [0:F_DEPTH-1];\n"
|
285 |
|
|
"\treg signed [IWIDTH-1:0] f_dlyright_r [0:F_DEPTH-1];\n"
|
286 |
|
|
"\treg signed [IWIDTH-1:0] f_dlyright_i [0:F_DEPTH-1];\n"
|
287 |
|
|
"\treg signed [CWIDTH-1:0] f_dlycoeff_r [0:F_DEPTH-1];\n"
|
288 |
|
|
"\treg signed [CWIDTH-1:0] f_dlycoeff_i [0:F_DEPTH-1];\n"
|
289 |
|
|
"\treg signed [F_DEPTH-1:0] f_dlyaux;\n"
|
290 |
|
|
"\n"
|
291 |
|
|
"\twire signed [IWIDTH:0] f_predifr, f_predifi;\n"
|
292 |
|
|
"\twire signed [IWIDTH+CWIDTH+3-1:0] f_predifrx, f_predifix;\n"
|
293 |
|
|
"\twire signed [CWIDTH:0] f_sumcoef;\n"
|
294 |
|
|
"\twire signed [IWIDTH+1:0] f_sumdiff;\n"
|
295 |
|
|
"\twire signed [IWIDTH:0] f_sumr, f_sumi;\n"
|
296 |
|
|
"\twire signed [IWIDTH+CWIDTH+3-1:0] f_sumrx, f_sumix;\n"
|
297 |
|
|
"\twire signed [IWIDTH:0] f_difr, f_difi;\n"
|
298 |
|
|
"\twire signed [IWIDTH+CWIDTH+3-1:0] f_difrx, f_difix;\n"
|
299 |
|
|
"\twire signed [IWIDTH+CWIDTH+3-1:0] f_widecoeff_r, f_widecoeff_i;\n"
|
300 |
|
|
"\n"
|
301 |
|
|
"\twire [(CWIDTH):0] fp_one_ic, fp_two_ic, fp_three_ic, f_p3c_in;\n"
|
302 |
|
|
"\twire [(IWIDTH+1):0] fp_one_id, fp_two_id, fp_three_id, f_p3d_in;\n"
|
303 |
|
|
"`endif\n\n");
|
304 |
|
|
|
305 |
36 |
dgisselq |
fprintf(fp,
|
306 |
|
|
"\treg\t[(2*IWIDTH-1):0]\tr_left, r_right;\n"
|
307 |
|
|
"\treg\t[(2*CWIDTH-1):0]\tr_coef, r_coef_2;\n"
|
308 |
|
|
"\twire\tsigned\t[(IWIDTH-1):0]\tr_left_r, r_left_i, r_right_r, r_right_i;\n"
|
309 |
|
|
"\tassign\tr_left_r = r_left[ (2*IWIDTH-1):(IWIDTH)];\n"
|
310 |
|
|
"\tassign\tr_left_i = r_left[ (IWIDTH-1):0];\n"
|
311 |
|
|
"\tassign\tr_right_r = r_right[(2*IWIDTH-1):(IWIDTH)];\n"
|
312 |
|
|
"\tassign\tr_right_i = r_right[(IWIDTH-1):0];\n"
|
313 |
|
|
"\n"
|
314 |
|
|
"\treg\tsigned\t[(IWIDTH):0]\tr_sum_r, r_sum_i, r_dif_r, r_dif_i;\n"
|
315 |
|
|
"\n"
|
316 |
|
|
"\treg [(LGDELAY-1):0] fifo_addr;\n"
|
317 |
|
|
"\twire [(LGDELAY-1):0] fifo_read_addr;\n"
|
318 |
|
|
"\tassign\tfifo_read_addr = fifo_addr - LCLDELAY[(LGDELAY-1):0];\n"
|
319 |
|
|
"\treg [(2*IWIDTH+1):0] fifo_left [ 0:((1<<LGDELAY)-1)];\n"
|
320 |
|
|
"\n");
|
321 |
|
|
fprintf(fp,
|
322 |
|
|
"\t// Set up the input to the multiply\n"
|
323 |
|
|
"\talways @(posedge i_clk)\n"
|
324 |
37 |
dgisselq |
"\tif (i_ce)\n"
|
325 |
|
|
"\tbegin\n"
|
326 |
|
|
"\t\t// One clock just latches the inputs\n"
|
327 |
|
|
"\t\tr_left <= i_left; // No change in # of bits\n"
|
328 |
|
|
"\t\tr_right <= i_right;\n"
|
329 |
|
|
"\t\tr_coef <= i_coef;\n"
|
330 |
|
|
"\t\t// Next clock adds/subtracts\n"
|
331 |
|
|
"\t\tr_sum_r <= r_left_r + r_right_r; // Now IWIDTH+1 bits\n"
|
332 |
|
|
"\t\tr_sum_i <= r_left_i + r_right_i;\n"
|
333 |
|
|
"\t\tr_dif_r <= r_left_r - r_right_r;\n"
|
334 |
|
|
"\t\tr_dif_i <= r_left_i - r_right_i;\n"
|
335 |
|
|
"\t\t// Other inputs are simply delayed on second clock\n"
|
336 |
|
|
"\t\tr_coef_2<= r_coef;\n"
|
337 |
|
|
"\tend\n"
|
338 |
36 |
dgisselq |
"\n");
|
339 |
|
|
fprintf(fp,
|
340 |
|
|
"\t// Don\'t forget to record the even side, since it doesn\'t need\n"
|
341 |
|
|
"\t// to be multiplied, but yet we still need the results in sync\n"
|
342 |
|
|
"\t// with the answer when it is ready.\n"
|
343 |
|
|
"\tinitial fifo_addr = 0;\n");
|
344 |
|
|
if (async_reset)
|
345 |
37 |
dgisselq |
fprintf(fp, "\talways @(posedge i_clk, negedge i_areset_n)\n\tif (!i_areset_n)\n");
|
346 |
36 |
dgisselq |
else
|
347 |
37 |
dgisselq |
fprintf(fp, "\talways @(posedge i_clk)\n\tif (i_reset)\n");
|
348 |
36 |
dgisselq |
fprintf(fp,
|
349 |
37 |
dgisselq |
"\t\tfifo_addr <= 0;\n"
|
350 |
|
|
"\telse if (i_ce)\n"
|
351 |
|
|
"\t\t// Need to delay the sum side--nothing else happens\n"
|
352 |
|
|
"\t\t// to it, but it needs to stay synchronized with the\n"
|
353 |
|
|
"\t\t// right side.\n"
|
354 |
|
|
"\t\tfifo_addr <= fifo_addr + 1;\n"
|
355 |
36 |
dgisselq |
"\n"
|
356 |
|
|
"\talways @(posedge i_clk)\n"
|
357 |
37 |
dgisselq |
"\tif (i_ce)\n"
|
358 |
|
|
"\t\tfifo_left[fifo_addr] <= { r_sum_r, r_sum_i };\n"
|
359 |
36 |
dgisselq |
"\n"
|
360 |
|
|
"\twire\tsigned\t[(CWIDTH-1):0] ir_coef_r, ir_coef_i;\n"
|
361 |
|
|
"\tassign\tir_coef_r = r_coef_2[(2*CWIDTH-1):CWIDTH];\n"
|
362 |
|
|
"\tassign\tir_coef_i = r_coef_2[(CWIDTH-1):0];\n"
|
363 |
|
|
"\twire\tsigned\t[((IWIDTH+2)+(CWIDTH+1)-1):0]\tp_one, p_two, p_three;\n"
|
364 |
|
|
"\n"
|
365 |
|
|
"\n");
|
366 |
|
|
fprintf(fp,
|
367 |
|
|
"\t// Multiply output is always a width of the sum of the widths of\n"
|
368 |
|
|
"\t// the two inputs. ALWAYS. This is independent of the number of\n"
|
369 |
|
|
"\t// bits in p_one, p_two, or p_three. These values needed to\n"
|
370 |
|
|
"\t// accumulate a bit (or two) each. However, this approach to a\n"
|
371 |
|
|
"\t// three multiply complex multiply cannot increase the total\n"
|
372 |
|
|
"\t// number of bits in our final output. We\'ll take care of\n"
|
373 |
|
|
"\t// dropping back down to the proper width, OWIDTH, in our routine\n"
|
374 |
|
|
"\t// below.\n"
|
375 |
|
|
"\n"
|
376 |
|
|
"\n");
|
377 |
|
|
fprintf(fp,
|
378 |
|
|
"\t// We accomplish here \"Karatsuba\" multiplication. That is,\n"
|
379 |
|
|
"\t// by doing three multiplies we accomplish the work of four.\n"
|
380 |
|
|
"\t// Let\'s prove to ourselves that this works ... We wish to\n"
|
381 |
|
|
"\t// multiply: (a+jb) * (c+jd), where a+jb is given by\n"
|
382 |
|
|
"\t//\ta + jb = r_dif_r + j r_dif_i, and\n"
|
383 |
|
|
"\t//\tc + jd = ir_coef_r + j ir_coef_i.\n"
|
384 |
|
|
"\t// We do this by calculating the intermediate products P1, P2,\n"
|
385 |
|
|
"\t// and P3 as\n"
|
386 |
|
|
"\t//\tP1 = ac\n"
|
387 |
|
|
"\t//\tP2 = bd\n"
|
388 |
|
|
"\t//\tP3 = (a + b) * (c + d)\n"
|
389 |
|
|
"\t// and then complete our final answer with\n"
|
390 |
|
|
"\t//\tac - bd = P1 - P2 (this checks)\n"
|
391 |
|
|
"\t//\tad + bc = P3 - P2 - P1\n"
|
392 |
|
|
"\t//\t = (ac + bc + ad + bd) - bd - ac\n"
|
393 |
|
|
"\t//\t = bc + ad (this checks)\n"
|
394 |
|
|
"\n"
|
395 |
|
|
"\n");
|
396 |
|
|
fprintf(fp,
|
397 |
|
|
"\t// This should really be based upon an IF, such as in\n"
|
398 |
|
|
"\t// if (IWIDTH < CWIDTH) then ...\n"
|
399 |
|
|
"\t// However, this is the only (other) way I know to do it.\n"
|
400 |
|
|
"\tgenerate if (CKPCE <= 1)\n"
|
401 |
|
|
"\tbegin\n"
|
402 |
|
|
"\n"
|
403 |
|
|
"\t\twire\t[(CWIDTH):0]\tp3c_in;\n"
|
404 |
|
|
"\t\twire\t[(IWIDTH+1):0]\tp3d_in;\n"
|
405 |
|
|
"\t\tassign\tp3c_in = ir_coef_i + ir_coef_r;\n"
|
406 |
|
|
"\t\tassign\tp3d_in = r_dif_r + r_dif_i;\n"
|
407 |
|
|
"\n"
|
408 |
|
|
"\t\t// We need to pad these first two multiplies by an extra\n"
|
409 |
|
|
"\t\t// bit just to keep them aligned with the third,\n"
|
410 |
|
|
"\t\t// simpler, multiply.\n"
|
411 |
|
|
"\t\tlongbimpy #(CWIDTH+1,IWIDTH+2) p1(i_clk, i_ce,\n"
|
412 |
|
|
"\t\t\t\t{ir_coef_r[CWIDTH-1],ir_coef_r},\n"
|
413 |
37 |
dgisselq |
"\t\t\t\t{r_dif_r[IWIDTH],r_dif_r}, p_one");
|
414 |
|
|
if (formal_property_flag) fprintf(fp,
|
415 |
|
|
"\n`ifdef\tFORMAL\n"
|
416 |
|
|
"\t\t\t\t, fp_one_ic, fp_one_id\n"
|
417 |
|
|
"`endif\n"
|
418 |
|
|
"\t\t\t");
|
419 |
|
|
fprintf(fp, ");\n"
|
420 |
36 |
dgisselq |
"\t\tlongbimpy #(CWIDTH+1,IWIDTH+2) p2(i_clk, i_ce,\n"
|
421 |
|
|
"\t\t\t\t{ir_coef_i[CWIDTH-1],ir_coef_i},\n"
|
422 |
37 |
dgisselq |
"\t\t\t\t{r_dif_i[IWIDTH],r_dif_i}, p_two");
|
423 |
|
|
if (formal_property_flag) fprintf(fp,
|
424 |
|
|
"\n`ifdef\tFORMAL\n"
|
425 |
|
|
"\t\t\t\t, fp_two_ic, fp_two_id\n"
|
426 |
|
|
"`endif\n"
|
427 |
|
|
"\t\t\t");
|
428 |
|
|
fprintf(fp, ");\n"
|
429 |
36 |
dgisselq |
"\t\tlongbimpy #(CWIDTH+1,IWIDTH+2) p3(i_clk, i_ce,\n"
|
430 |
37 |
dgisselq |
"\t\t\t\tp3c_in, p3d_in, p_three");
|
431 |
|
|
if (formal_property_flag) fprintf(fp,
|
432 |
|
|
"\n`ifdef\tFORMAL\n"
|
433 |
|
|
"\t\t\t\t, fp_three_ic, fp_three_id\n"
|
434 |
|
|
"`endif\n"
|
435 |
|
|
"\t\t\t");
|
436 |
|
|
fprintf(fp, ");\n"
|
437 |
36 |
dgisselq |
"\n");
|
438 |
|
|
|
439 |
|
|
///////////////////////////////////////////
|
440 |
|
|
///
|
441 |
|
|
/// Two clocks per CE, so CE, no-ce, CE, no-ce, etc
|
442 |
|
|
///
|
443 |
|
|
fprintf(fp,
|
444 |
|
|
"\tend else if (CKPCE == 2)\n"
|
445 |
|
|
"\tbegin : CKPCE_TWO\n"
|
446 |
|
|
"\t\t// Coefficient multiply inputs\n"
|
447 |
|
|
"\t\treg [2*(CWIDTH)-1:0] mpy_pipe_c;\n"
|
448 |
|
|
"\t\t// Data multiply inputs\n"
|
449 |
|
|
"\t\treg [2*(IWIDTH+1)-1:0] mpy_pipe_d;\n"
|
450 |
|
|
"\t\twire signed [(CWIDTH-1):0] mpy_pipe_vc;\n"
|
451 |
|
|
"\t\twire signed [(IWIDTH):0] mpy_pipe_vd;\n"
|
452 |
|
|
"\t\t//\n"
|
453 |
|
|
"\t\treg signed [(CWIDTH+1)-1:0] mpy_cof_sum;\n"
|
454 |
|
|
"\t\treg signed [(IWIDTH+2)-1:0] mpy_dif_sum;\n"
|
455 |
|
|
"\n"
|
456 |
|
|
"\t\tassign mpy_pipe_vc = mpy_pipe_c[2*(CWIDTH)-1:CWIDTH];\n"
|
457 |
|
|
"\t\tassign mpy_pipe_vd = mpy_pipe_d[2*(IWIDTH+1)-1:IWIDTH+1];\n"
|
458 |
|
|
"\n"
|
459 |
|
|
"\t\treg mpy_pipe_v;\n"
|
460 |
|
|
"\t\treg ce_phase;\n"
|
461 |
|
|
"\n"
|
462 |
|
|
"\t\treg signed [(CWIDTH+IWIDTH+3)-1:0] mpy_pipe_out;\n"
|
463 |
|
|
"\t\treg signed [IWIDTH+CWIDTH+3-1:0] longmpy;\n"
|
464 |
37 |
dgisselq |
"\n");
|
465 |
|
|
if (formal_property_flag) fprintf(fp,
|
466 |
|
|
"`ifdef FORMAL\n"
|
467 |
|
|
"\t\twire [CWIDTH:0] f_past_ic;\n"
|
468 |
|
|
"\t\twire [IWIDTH+1:0] f_past_id;\n"
|
469 |
|
|
"\t\twire [CWIDTH:0] f_past_mux_ic;\n"
|
470 |
|
|
"\t\twire [IWIDTH+1:0] f_past_mux_id;\n"
|
471 |
36 |
dgisselq |
"\n"
|
472 |
37 |
dgisselq |
"\t\treg [CWIDTH:0] f_rpone_ic, f_rptwo_ic, f_rpthree_ic,\n"
|
473 |
|
|
"\t\t\t\t\tf_rp2one_ic, f_rp2two_ic, f_rp2three_ic;\n"
|
474 |
|
|
"\t\treg [IWIDTH+1:0] f_rpone_id, f_rptwo_id, f_rpthree_id,\n"
|
475 |
|
|
"\t\t\t\t\tf_rp2one_id, f_rp2two_id, f_rp2three_id;\n"
|
476 |
|
|
"`endif\n\n");
|
477 |
|
|
|
478 |
|
|
fprintf(fp,
|
479 |
36 |
dgisselq |
"\n"
|
480 |
|
|
"\t\tinitial ce_phase = 1'b0;\n"
|
481 |
|
|
"\t\talways @(posedge i_clk)\n"
|
482 |
|
|
"\t\tif (i_reset)\n"
|
483 |
|
|
"\t\t\tce_phase <= 1'b0;\n"
|
484 |
|
|
"\t\telse if (i_ce)\n"
|
485 |
|
|
"\t\t\tce_phase <= 1'b1;\n"
|
486 |
|
|
"\t\telse\n"
|
487 |
|
|
"\t\t\tce_phase <= 1'b0;\n"
|
488 |
|
|
"\n"
|
489 |
|
|
"\t\talways @(*)\n"
|
490 |
|
|
"\t\t\tmpy_pipe_v = (i_ce)||(ce_phase);\n"
|
491 |
|
|
"\n"
|
492 |
|
|
"\t\talways @(posedge i_clk)\n"
|
493 |
|
|
"\t\tif (ce_phase)\n"
|
494 |
|
|
"\t\tbegin\n"
|
495 |
|
|
"\t\t\tmpy_pipe_c[2*CWIDTH-1:0] <=\n"
|
496 |
|
|
"\t\t\t\t\t{ ir_coef_r, ir_coef_i };\n"
|
497 |
|
|
"\t\t\tmpy_pipe_d[2*(IWIDTH+1)-1:0] <=\n"
|
498 |
|
|
"\t\t\t\t\t{ r_dif_r, r_dif_i };\n"
|
499 |
|
|
"\n"
|
500 |
|
|
"\t\t\tmpy_cof_sum <= ir_coef_i + ir_coef_r;\n"
|
501 |
|
|
"\t\t\tmpy_dif_sum <= r_dif_r + r_dif_i;\n"
|
502 |
|
|
"\n"
|
503 |
|
|
"\t\tend else if (i_ce)\n"
|
504 |
|
|
"\t\tbegin\n"
|
505 |
|
|
"\t\t\tmpy_pipe_c[2*(CWIDTH)-1:0] <= {\n"
|
506 |
|
|
"\t\t\t\tmpy_pipe_c[(CWIDTH)-1:0], {(CWIDTH){1'b0}} };\n"
|
507 |
|
|
"\t\t\tmpy_pipe_d[2*(IWIDTH+1)-1:0] <= {\n"
|
508 |
|
|
"\t\t\t\tmpy_pipe_d[(IWIDTH+1)-1:0], {(IWIDTH+1){1'b0}} };\n"
|
509 |
|
|
"\t\tend\n"
|
510 |
|
|
"\n");
|
511 |
|
|
fprintf(fp,
|
512 |
|
|
"\t\tlongbimpy #(CWIDTH+1,IWIDTH+2) mpy0(i_clk, mpy_pipe_v,\n"
|
513 |
37 |
dgisselq |
"\t\t\t\tmpy_cof_sum, mpy_dif_sum, longmpy\n");
|
514 |
|
|
if (formal_property_flag) fprintf(fp,
|
515 |
|
|
"`ifdef FORMAL\n"
|
516 |
|
|
"\t\t\t\t, f_past_ic, f_past_id\n"
|
517 |
|
|
"`endif\n");
|
518 |
|
|
fprintf(fp,"\t\t\t);\n"
|
519 |
36 |
dgisselq |
"\n");
|
520 |
|
|
|
521 |
|
|
fprintf(fp,
|
522 |
|
|
"\t\tlongbimpy #(CWIDTH+1,IWIDTH+2) mpy1(i_clk, mpy_pipe_v,\n"
|
523 |
|
|
"\t\t\t\t{ mpy_pipe_vc[CWIDTH-1], mpy_pipe_vc },\n"
|
524 |
|
|
"\t\t\t\t{ mpy_pipe_vd[IWIDTH ], mpy_pipe_vd },\n"
|
525 |
37 |
dgisselq |
"\t\t\t\tmpy_pipe_out\n");
|
526 |
|
|
if (formal_property_flag) fprintf(fp,
|
527 |
|
|
"`ifdef FORMAL\n"
|
528 |
|
|
"\t\t\t\t, f_past_mux_ic, f_past_mux_id\n"
|
529 |
|
|
"`endif\n");
|
530 |
|
|
fprintf(fp,"\t\t\t);\n"
|
531 |
|
|
"\n");
|
532 |
36 |
dgisselq |
|
533 |
|
|
fprintf(fp,
|
534 |
|
|
"\t\treg\tsigned\t[((IWIDTH+2)+(CWIDTH+1)-1):0]\n"
|
535 |
|
|
"\t\t\t\t\trp_one, rp_two, rp_three,\n"
|
536 |
|
|
"\t\t\t\t\trp2_one, rp2_two, rp2_three;\n"
|
537 |
|
|
"\n"
|
538 |
|
|
"\t\talways @(posedge i_clk)\n"
|
539 |
|
|
"\t\tif (((i_ce)&&(!MPYDELAY[0]))\n"
|
540 |
|
|
"\t\t\t||((ce_phase)&&(MPYDELAY[0])))\n"
|
541 |
37 |
dgisselq |
"\t\tbegin\n"
|
542 |
|
|
"\t\t\trp_one <= mpy_pipe_out;\n");
|
543 |
|
|
if (formal_property_flag) fprintf(fp,
|
544 |
|
|
"`ifdef FORMAL\n"
|
545 |
|
|
"\t\t\tf_rpone_ic <= f_past_mux_ic;\n"
|
546 |
|
|
"\t\t\tf_rpone_id <= f_past_mux_id;\n"
|
547 |
|
|
"`endif\n");
|
548 |
|
|
fprintf(fp,
|
549 |
|
|
"\t\tend\n\n");
|
550 |
|
|
fprintf(fp,
|
551 |
36 |
dgisselq |
"\t\talways @(posedge i_clk)\n"
|
552 |
|
|
"\t\tif (((i_ce)&&(MPYDELAY[0]))\n"
|
553 |
|
|
"\t\t\t||((ce_phase)&&(!MPYDELAY[0])))\n"
|
554 |
37 |
dgisselq |
"\t\tbegin\n"
|
555 |
|
|
"\t\t\trp_two <= mpy_pipe_out;\n");
|
556 |
|
|
if (formal_property_flag) fprintf(fp,
|
557 |
|
|
"`ifdef FORMAL\n"
|
558 |
|
|
"\t\t\tf_rptwo_ic <= f_past_mux_ic;\n"
|
559 |
|
|
"\t\t\tf_rptwo_id <= f_past_mux_id;\n"
|
560 |
|
|
"`endif\n");
|
561 |
|
|
fprintf(fp,
|
562 |
|
|
"\t\tend\n\n");
|
563 |
|
|
fprintf(fp,
|
564 |
36 |
dgisselq |
"\t\talways @(posedge i_clk)\n"
|
565 |
|
|
"\t\tif (i_ce)\n"
|
566 |
37 |
dgisselq |
"\t\tbegin\n"
|
567 |
|
|
"\t\t\trp_three <= longmpy;\n");
|
568 |
|
|
if (formal_property_flag) fprintf(fp,
|
569 |
|
|
"`ifdef FORMAL\n"
|
570 |
|
|
"\t\t\tf_rpthree_ic <= f_past_ic;\n"
|
571 |
|
|
"\t\t\tf_rpthree_id <= f_past_id;\n"
|
572 |
|
|
"`endif\n");
|
573 |
|
|
fprintf(fp,
|
574 |
|
|
"\t\tend\n"
|
575 |
|
|
"\n\n");
|
576 |
|
|
|
577 |
|
|
|
578 |
|
|
fprintf(fp,
|
579 |
36 |
dgisselq |
"\t\t// Our outputs *MUST* be set on a clock where i_ce is\n"
|
580 |
|
|
"\t\t// true for the following logic to work. Make that\n"
|
581 |
|
|
"\t\t// happen here.\n"
|
582 |
|
|
"\t\talways @(posedge i_clk)\n"
|
583 |
|
|
"\t\tif (i_ce)\n"
|
584 |
37 |
dgisselq |
"\t\tbegin\n"
|
585 |
36 |
dgisselq |
"\t\t\trp2_one<= rp_one;\n"
|
586 |
|
|
"\t\t\trp2_two <= rp_two;\n"
|
587 |
37 |
dgisselq |
"\t\t\trp2_three<= rp_three;\n");
|
588 |
|
|
if (formal_property_flag) fprintf(fp,
|
589 |
|
|
"`ifdef FORMAL\n"
|
590 |
|
|
"\t\t\tf_rp2one_ic <= f_rpone_ic;\n"
|
591 |
|
|
"\t\t\tf_rp2one_id <= f_rpone_id;\n"
|
592 |
36 |
dgisselq |
"\n"
|
593 |
37 |
dgisselq |
|
594 |
|
|
"\t\t\tf_rp2two_ic <= f_rptwo_ic;\n"
|
595 |
|
|
"\t\t\tf_rp2two_id <= f_rptwo_id;\n"
|
596 |
|
|
"\n"
|
597 |
|
|
|
598 |
|
|
"\t\t\tf_rp2three_ic <= f_rpthree_ic;\n"
|
599 |
|
|
"\t\t\tf_rp2three_id <= f_rpthree_id;\n"
|
600 |
|
|
"`endif\n");
|
601 |
|
|
fprintf(fp,
|
602 |
|
|
"\t\tend\n"
|
603 |
|
|
"\n"
|
604 |
36 |
dgisselq |
"\t\tassign p_one = rp2_one;\n"
|
605 |
|
|
"\t\tassign p_two = (!MPYDELAY[0])? rp2_two : rp_two;\n"
|
606 |
|
|
"\t\tassign p_three = ( MPYDELAY[0])? rp_three : rp2_three;\n"
|
607 |
|
|
"\n"
|
608 |
|
|
"\t\t// verilator lint_off UNUSED\n"
|
609 |
|
|
"\t\twire\t[2*(IWIDTH+CWIDTH+3)-1:0]\tunused;\n"
|
610 |
|
|
"\t\tassign\tunused = { rp2_two, rp2_three };\n"
|
611 |
|
|
"\t\t// verilator lint_on UNUSED\n"
|
612 |
|
|
"\n");
|
613 |
37 |
dgisselq |
if (formal_property_flag) fprintf(fp,
|
614 |
|
|
"`ifdef FORMAL\n"
|
615 |
|
|
"\t\tassign fp_one_ic = f_rp2one_ic;\n"
|
616 |
|
|
"\t\tassign fp_one_id = f_rp2one_id;\n"
|
617 |
|
|
"\n"
|
618 |
|
|
"\t\tassign fp_two_ic = (!MPYDELAY[0])? f_rp2two_ic : f_rptwo_ic;\n"
|
619 |
|
|
"\t\tassign fp_two_id = (!MPYDELAY[0])? f_rp2two_id : f_rptwo_id;\n"
|
620 |
|
|
"\n"
|
621 |
|
|
"\t\tassign fp_three_ic= (MPYDELAY[0])? f_rpthree_ic : f_rp2three_ic;\n"
|
622 |
|
|
"\t\tassign fp_three_id= (MPYDELAY[0])? f_rpthree_id : f_rp2three_id;\n"
|
623 |
|
|
"`endif\n\n");
|
624 |
36 |
dgisselq |
|
625 |
37 |
dgisselq |
|
626 |
36 |
dgisselq |
/////////////////////////
|
627 |
|
|
///
|
628 |
|
|
/// Three clock per CE, so CE, no-ce, no-ce*, CE
|
629 |
|
|
///
|
630 |
|
|
fprintf(fp,
|
631 |
|
|
"\tend else if (CKPCE <= 3)\n\tbegin : CKPCE_THREE\n");
|
632 |
|
|
|
633 |
|
|
fprintf(fp,
|
634 |
|
|
"\t\t// Coefficient multiply inputs\n"
|
635 |
|
|
"\t\treg\t\t[3*(CWIDTH+1)-1:0]\tmpy_pipe_c;\n"
|
636 |
|
|
"\t\t// Data multiply inputs\n"
|
637 |
|
|
"\t\treg\t\t[3*(IWIDTH+2)-1:0]\tmpy_pipe_d;\n"
|
638 |
|
|
"\t\twire\tsigned [(CWIDTH):0] mpy_pipe_vc;\n"
|
639 |
|
|
"\t\twire\tsigned [(IWIDTH+1):0] mpy_pipe_vd;\n"
|
640 |
|
|
"\n"
|
641 |
|
|
"\t\tassign\tmpy_pipe_vc = mpy_pipe_c[3*(CWIDTH+1)-1:2*(CWIDTH+1)];\n"
|
642 |
|
|
"\t\tassign\tmpy_pipe_vd = mpy_pipe_d[3*(IWIDTH+2)-1:2*(IWIDTH+2)];\n"
|
643 |
|
|
"\n"
|
644 |
|
|
"\t\treg\t\t\tmpy_pipe_v;\n"
|
645 |
|
|
"\t\treg\t\t[2:0]\tce_phase;\n"
|
646 |
|
|
"\n"
|
647 |
|
|
"\t\treg\tsigned [ (CWIDTH+IWIDTH+3)-1:0] mpy_pipe_out;\n"
|
648 |
|
|
"\n");
|
649 |
37 |
dgisselq |
if (formal_property_flag) fprintf(fp,
|
650 |
|
|
"`ifdef FORMAL\n"
|
651 |
|
|
"\t\twire\t[CWIDTH:0] f_past_ic;\n"
|
652 |
|
|
"\t\twire\t[IWIDTH+1:0] f_past_id;\n"
|
653 |
|
|
"\n"
|
654 |
|
|
"\t\treg\t[CWIDTH:0] f_rpone_ic, f_rptwo_ic, f_rpthree_ic,\n"
|
655 |
|
|
"\t\t\t\t\tf_rp2one_ic, f_rp2two_ic, f_rp2three_ic,\n"
|
656 |
|
|
"\t\t\t\t\tf_rp3one_ic;\n"
|
657 |
|
|
"\t\treg\t[IWIDTH+1:0] f_rpone_id, f_rptwo_id, f_rpthree_id,\n"
|
658 |
|
|
"\t\t\t\t\tf_rp2one_id, f_rp2two_id, f_rp2three_id,\n"
|
659 |
|
|
"\t\t\t\t\tf_rp3one_id;\n"
|
660 |
|
|
"`endif\n"
|
661 |
|
|
"\n");
|
662 |
|
|
|
663 |
36 |
dgisselq |
fprintf(fp,
|
664 |
|
|
"\t\tinitial\tce_phase = 3'b011;\n"
|
665 |
|
|
"\t\talways @(posedge i_clk)\n"
|
666 |
|
|
"\t\tif (i_reset)\n"
|
667 |
|
|
"\t\t\tce_phase <= 3'b011;\n"
|
668 |
|
|
"\t\telse if (i_ce)\n"
|
669 |
|
|
"\t\t\tce_phase <= 3'b000;\n"
|
670 |
|
|
"\t\telse if (ce_phase != 3'b011)\n"
|
671 |
|
|
"\t\t\tce_phase <= ce_phase + 1'b1;\n"
|
672 |
|
|
"\n"
|
673 |
|
|
"\t\talways @(*)\n"
|
674 |
|
|
"\t\t\tmpy_pipe_v = (i_ce)||(ce_phase < 3'b010);\n"
|
675 |
|
|
"\n");
|
676 |
|
|
|
677 |
|
|
fprintf(fp,
|
678 |
|
|
"\t\talways @(posedge i_clk)\n"
|
679 |
37 |
dgisselq |
"\t\tif (ce_phase == 3\'b000)\n"
|
680 |
|
|
"\t\tbegin\n"
|
681 |
|
|
"\t\t\t// Second clock\n"
|
682 |
|
|
"\t\t\tmpy_pipe_c[3*(CWIDTH+1)-1:(CWIDTH+1)] <= {\n"
|
683 |
|
|
"\t\t\t\tir_coef_r[CWIDTH-1], ir_coef_r,\n"
|
684 |
|
|
"\t\t\t\tir_coef_i[CWIDTH-1], ir_coef_i };\n"
|
685 |
|
|
"\t\t\tmpy_pipe_c[CWIDTH:0] <= ir_coef_i + ir_coef_r;\n"
|
686 |
|
|
"\t\t\tmpy_pipe_d[3*(IWIDTH+2)-1:(IWIDTH+2)] <= {\n"
|
687 |
|
|
"\t\t\t\tr_dif_r[IWIDTH], r_dif_r,\n"
|
688 |
|
|
"\t\t\t\tr_dif_i[IWIDTH], r_dif_i };\n"
|
689 |
|
|
"\t\t\tmpy_pipe_d[(IWIDTH+2)-1:0] <= r_dif_r + r_dif_i;\n"
|
690 |
36 |
dgisselq |
"\n"
|
691 |
37 |
dgisselq |
"\t\tend else if (mpy_pipe_v)\n"
|
692 |
|
|
"\t\tbegin\n"
|
693 |
|
|
"\t\t\tmpy_pipe_c[3*(CWIDTH+1)-1:0] <= {\n"
|
694 |
|
|
"\t\t\t\tmpy_pipe_c[2*(CWIDTH+1)-1:0], {(CWIDTH+1){1\'b0}} };\n"
|
695 |
|
|
"\t\t\tmpy_pipe_d[3*(IWIDTH+2)-1:0] <= {\n"
|
696 |
|
|
"\t\t\t\tmpy_pipe_d[2*(IWIDTH+2)-1:0], {(IWIDTH+2){1\'b0}} };\n"
|
697 |
|
|
"\t\tend\n"
|
698 |
36 |
dgisselq |
"\n");
|
699 |
|
|
fprintf(fp,
|
700 |
|
|
"\t\tlongbimpy #(CWIDTH+1,IWIDTH+2) mpy(i_clk, mpy_pipe_v,\n"
|
701 |
37 |
dgisselq |
"\t\t\t\tmpy_pipe_vc, mpy_pipe_vd, mpy_pipe_out\n");
|
702 |
|
|
if (formal_property_flag) fprintf(fp,
|
703 |
|
|
"`ifdef FORMAL\n"
|
704 |
|
|
"\t\t\t\t, f_past_ic, f_past_id\n"
|
705 |
|
|
"`endif\n");
|
706 |
|
|
fprintf(fp,
|
707 |
|
|
"\t\t\t);\n"
|
708 |
36 |
dgisselq |
"\n");
|
709 |
|
|
|
710 |
|
|
fprintf(fp,
|
711 |
|
|
"\t\treg\tsigned\t[((IWIDTH+2)+(CWIDTH+1)-1):0]\n"
|
712 |
|
|
"\t\t\t\trp_one, rp_two, rp_three,\n"
|
713 |
|
|
"\t\t\t\trp2_one, rp2_two, rp2_three,\n"
|
714 |
|
|
"\t\t\t\trp3_one;\n"
|
715 |
|
|
"\n");
|
716 |
|
|
|
717 |
|
|
fprintf(fp,
|
718 |
|
|
"\t\talways @(posedge i_clk)\n"
|
719 |
|
|
"\t\tif (MPYREMAINDER == 0)\n"
|
720 |
|
|
"\t\tbegin\n\n"
|
721 |
|
|
"\t\t if (i_ce)\n"
|
722 |
37 |
dgisselq |
"\t\t begin\n"
|
723 |
|
|
"\t\t rp_two <= mpy_pipe_out;\n");
|
724 |
|
|
if (formal_property_flag) fprintf(fp,
|
725 |
|
|
"`ifdef FORMAL\n"
|
726 |
|
|
"\t\t f_rptwo_ic <= f_past_ic;\n"
|
727 |
|
|
"\t\t f_rptwo_id <= f_past_id;\n"
|
728 |
|
|
"`endif\n");
|
729 |
|
|
fprintf(fp,
|
730 |
|
|
"\t\t end else if (ce_phase == 3'b000)\n"
|
731 |
|
|
"\t\t begin\n"
|
732 |
|
|
"\t\t rp_three <= mpy_pipe_out;\n");
|
733 |
|
|
if (formal_property_flag) fprintf(fp,
|
734 |
|
|
"`ifdef FORMAL\n"
|
735 |
|
|
"\t\t f_rpthree_ic <= f_past_ic;\n"
|
736 |
|
|
"\t\t f_rpthree_id <= f_past_id;\n"
|
737 |
|
|
"`endif\n");
|
738 |
|
|
fprintf(fp,
|
739 |
|
|
"\t\t end else if (ce_phase == 3'b001)\n"
|
740 |
|
|
"\t\t begin\n"
|
741 |
|
|
"\t\t rp_one <= mpy_pipe_out;\n");
|
742 |
|
|
if (formal_property_flag) fprintf(fp,
|
743 |
|
|
"`ifdef FORMAL\n"
|
744 |
|
|
"\t\t f_rpone_ic <= f_past_ic;\n"
|
745 |
|
|
"\t\t f_rpone_id <= f_past_id;\n"
|
746 |
|
|
"`endif\n");
|
747 |
|
|
fprintf(fp,
|
748 |
|
|
"\t\t end\n"
|
749 |
36 |
dgisselq |
"\t\tend else if (MPYREMAINDER == 1)\n"
|
750 |
|
|
"\t\tbegin\n\n"
|
751 |
|
|
"\t\t if (i_ce)\n"
|
752 |
37 |
dgisselq |
"\t\t begin\n"
|
753 |
|
|
"\t\t rp_one <= mpy_pipe_out;\n");
|
754 |
|
|
if (formal_property_flag) fprintf(fp,
|
755 |
|
|
"`ifdef FORMAL\n"
|
756 |
|
|
"\t\t f_rpone_ic <= f_past_ic;\n"
|
757 |
|
|
"\t\t f_rpone_id <= f_past_id;\n"
|
758 |
|
|
"`endif\n");
|
759 |
|
|
fprintf(fp,
|
760 |
|
|
"\t\t end else if (ce_phase == 3'b000)\n"
|
761 |
|
|
"\t\t begin\n"
|
762 |
|
|
"\t\t rp_two <= mpy_pipe_out;\n");
|
763 |
|
|
if (formal_property_flag) fprintf(fp,
|
764 |
|
|
"`ifdef FORMAL\n"
|
765 |
|
|
"\t\t f_rptwo_ic <= f_past_ic;\n"
|
766 |
|
|
"\t\t f_rptwo_id <= f_past_id;\n"
|
767 |
|
|
"`endif\n");
|
768 |
|
|
fprintf(fp,
|
769 |
|
|
"\t\t end else if (ce_phase == 3'b001)\n"
|
770 |
|
|
"\t\t begin\n"
|
771 |
|
|
"\t\t rp_three <= mpy_pipe_out;\n");
|
772 |
|
|
if (formal_property_flag) fprintf(fp,
|
773 |
|
|
"`ifdef FORMAL\n"
|
774 |
|
|
"\t\t f_rpthree_ic <= f_past_ic;\n"
|
775 |
|
|
"\t\t f_rpthree_id <= f_past_id;\n"
|
776 |
|
|
"`endif\n");
|
777 |
|
|
fprintf(fp,
|
778 |
|
|
"\t\t end\n"
|
779 |
36 |
dgisselq |
"\t\tend else // if (MPYREMAINDER == 2)\n"
|
780 |
|
|
"\t\tbegin\n\n"
|
781 |
|
|
"\t\t if (i_ce)\n"
|
782 |
37 |
dgisselq |
"\t\t begin\n"
|
783 |
|
|
"\t\t rp_three <= mpy_pipe_out;\n");
|
784 |
|
|
if (formal_property_flag) fprintf(fp,
|
785 |
|
|
"`ifdef FORMAL\n"
|
786 |
|
|
"\t\t f_rpthree_ic <= f_past_ic;\n"
|
787 |
|
|
"\t\t f_rpthree_id <= f_past_id;\n"
|
788 |
|
|
"`endif\n");
|
789 |
|
|
fprintf(fp,
|
790 |
|
|
"\t\t end else if (ce_phase == 3'b000)\n"
|
791 |
|
|
"\t\t begin\n"
|
792 |
|
|
"\t\t rp_one <= mpy_pipe_out;\n");
|
793 |
|
|
if (formal_property_flag) fprintf(fp,
|
794 |
|
|
"`ifdef FORMAL\n"
|
795 |
|
|
"\t\t f_rpone_ic <= f_past_ic;\n"
|
796 |
|
|
"\t\t f_rpone_id <= f_past_id;\n"
|
797 |
|
|
"`endif\n");
|
798 |
|
|
fprintf(fp,
|
799 |
|
|
"\t\t end else if (ce_phase == 3'b001)\n"
|
800 |
|
|
"\t\t begin\n"
|
801 |
|
|
"\t\t rp_two <= mpy_pipe_out;\n");
|
802 |
|
|
if (formal_property_flag) fprintf(fp,
|
803 |
|
|
"`ifdef FORMAL\n"
|
804 |
|
|
"\t\t f_rptwo_ic <= f_past_ic;\n"
|
805 |
|
|
"\t\t f_rptwo_id <= f_past_id;\n"
|
806 |
|
|
"`endif\n");
|
807 |
|
|
fprintf(fp,
|
808 |
|
|
"\t\t end\n"
|
809 |
36 |
dgisselq |
"\t\tend\n\n");
|
810 |
|
|
|
811 |
|
|
fprintf(fp,
|
812 |
|
|
"\t\talways @(posedge i_clk)\n"
|
813 |
|
|
"\t\tif (i_ce)\n"
|
814 |
|
|
"\t\tbegin\n"
|
815 |
|
|
"\t\t\trp2_one <= rp_one;\n"
|
816 |
|
|
"\t\t\trp2_two <= rp_two;\n"
|
817 |
|
|
"\t\t\trp2_three <= (MPYREMAINDER == 2) ? mpy_pipe_out : rp_three;\n"
|
818 |
37 |
dgisselq |
"\t\t\trp3_one <= (MPYREMAINDER == 0) ? rp2_one : rp_one;\n");
|
819 |
|
|
|
820 |
|
|
if (formal_property_flag) fprintf(fp,
|
821 |
|
|
"`ifdef FORMAL\n"
|
822 |
|
|
"\t\t\tf_rp2one_ic <= f_rpone_ic;\n"
|
823 |
|
|
"\t\t\tf_rp2one_id <= f_rpone_id;\n"
|
824 |
|
|
"\n"
|
825 |
|
|
"\t\t\tf_rp2two_ic <= f_rptwo_ic;\n"
|
826 |
|
|
"\t\t\tf_rp2two_id <= f_rptwo_id;\n"
|
827 |
|
|
"\n"
|
828 |
|
|
"\t\t\tf_rp2three_ic <= (MPYREMAINDER==2) ? f_past_ic : f_rpthree_ic;\n"
|
829 |
|
|
"\t\t\tf_rp2three_id <= (MPYREMAINDER==2) ? f_past_id : f_rpthree_id;\n"
|
830 |
|
|
"\t\t\tf_rp3one_ic <= (MPYREMAINDER==0) ? f_rp2one_ic : f_rpone_ic;\n"
|
831 |
|
|
"\t\t\tf_rp3one_id <= (MPYREMAINDER==0) ? f_rp2one_id : f_rpone_id;\n"
|
832 |
|
|
"`endif\n");
|
833 |
|
|
|
834 |
|
|
|
835 |
36 |
dgisselq |
fprintf(fp,
|
836 |
37 |
dgisselq |
"\t\tend\n"
|
837 |
|
|
"\n"
|
838 |
36 |
dgisselq |
"\t\tassign\tp_one = rp3_one;\n"
|
839 |
|
|
"\t\tassign\tp_two = rp2_two;\n"
|
840 |
|
|
"\t\tassign\tp_three = rp2_three;\n"
|
841 |
|
|
"\n");
|
842 |
37 |
dgisselq |
if (formal_property_flag) fprintf(fp,
|
843 |
|
|
"`ifdef FORMAL\n"
|
844 |
|
|
"\t\tassign fp_one_ic = f_rp3one_ic;\n"
|
845 |
|
|
"\t\tassign fp_one_id = f_rp3one_id;\n"
|
846 |
|
|
"\n"
|
847 |
|
|
"\t\tassign fp_two_ic = f_rp2two_ic;\n"
|
848 |
|
|
"\t\tassign fp_two_id = f_rp2two_id;\n"
|
849 |
|
|
"\n"
|
850 |
|
|
"\t\tassign fp_three_ic = f_rp2three_ic;\n"
|
851 |
|
|
"\t\tassign fp_three_id = f_rp2three_id;\n"
|
852 |
|
|
"`endif\n"
|
853 |
|
|
"\n");
|
854 |
36 |
dgisselq |
|
855 |
|
|
fprintf(fp,
|
856 |
|
|
"\tend endgenerate\n");
|
857 |
|
|
|
858 |
|
|
fprintf(fp,
|
859 |
|
|
"\t// These values are held in memory and delayed during the\n"
|
860 |
|
|
"\t// multiply. Here, we recover them. During the multiply,\n"
|
861 |
|
|
"\t// values were multiplied by 2^(CWIDTH-2)*exp{-j*2*pi*...},\n"
|
862 |
|
|
"\t// therefore, the left_x values need to be right shifted by\n"
|
863 |
|
|
"\t// CWIDTH-2 as well. The additional bits come from a sign\n"
|
864 |
|
|
"\t// extension.\n"
|
865 |
|
|
"\twire\tsigned\t[(IWIDTH+CWIDTH):0] fifo_i, fifo_r;\n"
|
866 |
|
|
"\treg\t\t[(2*IWIDTH+1):0] fifo_read;\n"
|
867 |
37 |
dgisselq |
"\tassign\tfifo_r = { {2{fifo_read[2*(IWIDTH+1)-1]}},\n"
|
868 |
|
|
"\t\tfifo_read[(2*(IWIDTH+1)-1):(IWIDTH+1)], {(CWIDTH-2){1\'b0}} };\n"
|
869 |
|
|
"\tassign\tfifo_i = { {2{fifo_read[(IWIDTH+1)-1]}},\n"
|
870 |
|
|
"\t\tfifo_read[((IWIDTH+1)-1):0], {(CWIDTH-2){1\'b0}} };\n"
|
871 |
36 |
dgisselq |
"\n"
|
872 |
|
|
"\n"
|
873 |
|
|
"\treg\tsigned\t[(CWIDTH+IWIDTH+3-1):0] mpy_r, mpy_i;\n"
|
874 |
|
|
"\n");
|
875 |
|
|
fprintf(fp,
|
876 |
|
|
"\t// Let's do some rounding and remove unnecessary bits.\n"
|
877 |
|
|
"\t// We have (IWIDTH+CWIDTH+3) bits here, we need to drop down to\n"
|
878 |
|
|
"\t// OWIDTH, and SHIFT by SHIFT bits in the process. The trick is\n"
|
879 |
|
|
"\t// that we don\'t need (IWIDTH+CWIDTH+3) bits. We\'ve accumulated\n"
|
880 |
|
|
"\t// them, but the actual values will never fill all these bits.\n"
|
881 |
|
|
"\t// In particular, we only need:\n"
|
882 |
|
|
"\t//\t IWIDTH bits for the input\n"
|
883 |
|
|
"\t//\t +1 bit for the add/subtract\n"
|
884 |
|
|
"\t//\t+CWIDTH bits for the coefficient multiply\n"
|
885 |
|
|
"\t//\t +1 bit for the add/subtract in the complex multiply\n"
|
886 |
|
|
"\t//\t ------\n"
|
887 |
|
|
"\t//\t (IWIDTH+CWIDTH+2) bits at full precision.\n"
|
888 |
|
|
"\t//\n"
|
889 |
|
|
"\t// However, the coefficient multiply multiplied by a maximum value\n"
|
890 |
|
|
"\t// of 2^(CWIDTH-2). Thus, we only have\n"
|
891 |
|
|
"\t//\t IWIDTH bits for the input\n"
|
892 |
|
|
"\t//\t +1 bit for the add/subtract\n"
|
893 |
|
|
"\t//\t+CWIDTH-2 bits for the coefficient multiply\n"
|
894 |
|
|
"\t//\t +1 (optional) bit for the add/subtract in the cpx mpy.\n"
|
895 |
|
|
"\t//\t -------- ... multiply. (This last bit may be shifted out.)\n"
|
896 |
|
|
"\t//\t (IWIDTH+CWIDTH) valid output bits.\n"
|
897 |
|
|
"\t// Now, if the user wants to keep any extras of these (via OWIDTH),\n"
|
898 |
|
|
"\t// or if he wishes to arbitrarily shift some of these off (via\n"
|
899 |
|
|
"\t// SHIFT) we accomplish that here.\n"
|
900 |
|
|
"\n");
|
901 |
|
|
fprintf(fp,
|
902 |
|
|
"\twire\tsigned\t[(OWIDTH-1):0]\trnd_left_r, rnd_left_i, rnd_right_r, rnd_right_i;\n\n");
|
903 |
|
|
|
904 |
|
|
fprintf(fp,
|
905 |
|
|
"\twire\tsigned\t[(CWIDTH+IWIDTH+3-1):0]\tleft_sr, left_si;\n"
|
906 |
|
|
"\tassign left_sr = { {(2){fifo_r[(IWIDTH+CWIDTH)]}}, fifo_r };\n"
|
907 |
|
|
"\tassign left_si = { {(2){fifo_i[(IWIDTH+CWIDTH)]}}, fifo_i };\n\n");
|
908 |
|
|
|
909 |
|
|
fprintf(fp,
|
910 |
|
|
"\t%s #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_left_r(i_clk, i_ce,\n"
|
911 |
|
|
"\t\t\t\tleft_sr, rnd_left_r);\n\n",
|
912 |
|
|
rnd_string);
|
913 |
|
|
fprintf(fp,
|
914 |
|
|
"\t%s #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_left_i(i_clk, i_ce,\n"
|
915 |
|
|
"\t\t\t\tleft_si, rnd_left_i);\n\n",
|
916 |
|
|
rnd_string);
|
917 |
|
|
fprintf(fp,
|
918 |
|
|
"\t%s #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_right_r(i_clk, i_ce,\n"
|
919 |
|
|
"\t\t\t\tmpy_r, rnd_right_r);\n\n", rnd_string);
|
920 |
|
|
fprintf(fp,
|
921 |
|
|
"\t%s #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_right_i(i_clk, i_ce,\n"
|
922 |
|
|
"\t\t\t\tmpy_i, rnd_right_i);\n\n", rnd_string);
|
923 |
|
|
fprintf(fp,
|
924 |
|
|
"\talways @(posedge i_clk)\n"
|
925 |
37 |
dgisselq |
"\tif (i_ce)\n"
|
926 |
|
|
"\tbegin\n"
|
927 |
|
|
"\t\t// First clock, recover all values\n"
|
928 |
|
|
"\t\tfifo_read <= fifo_left[fifo_read_addr];\n"
|
929 |
|
|
"\t\t// These values are IWIDTH+CWIDTH+3 bits wide\n"
|
930 |
|
|
"\t\t// although they only need to be (IWIDTH+1)\n"
|
931 |
|
|
"\t\t// + (CWIDTH) bits wide. (We\'ve got two\n"
|
932 |
|
|
"\t\t// extra bits we need to get rid of.)\n"
|
933 |
|
|
"\t\tmpy_r <= p_one - p_two;\n"
|
934 |
|
|
"\t\tmpy_i <= p_three - p_one - p_two;\n"
|
935 |
|
|
"\tend\n"
|
936 |
36 |
dgisselq |
"\n");
|
937 |
|
|
|
938 |
|
|
fprintf(fp,
|
939 |
|
|
"\treg\t[(AUXLEN-1):0]\taux_pipeline;\n"
|
940 |
|
|
"\tinitial\taux_pipeline = 0;\n");
|
941 |
|
|
if (async_reset)
|
942 |
37 |
dgisselq |
fprintf(fp, "\talways @(posedge i_clk, negedge i_areset_n)\n\tif (!i_areset_n)\n");
|
943 |
36 |
dgisselq |
else
|
944 |
37 |
dgisselq |
fprintf(fp, "\talways @(posedge i_clk)\n\tif (i_reset)\n");
|
945 |
36 |
dgisselq |
fprintf(fp,
|
946 |
37 |
dgisselq |
"\t\taux_pipeline <= 0;\n"
|
947 |
|
|
"\telse if (i_ce)\n"
|
948 |
|
|
"\t\taux_pipeline <= { aux_pipeline[(AUXLEN-2):0], i_aux };\n"
|
949 |
36 |
dgisselq |
"\n");
|
950 |
|
|
fprintf(fp,
|
951 |
|
|
"\tinitial o_aux = 1\'b0;\n");
|
952 |
|
|
if (async_reset)
|
953 |
37 |
dgisselq |
fprintf(fp, "\talways @(posedge i_clk, negedge i_areset_n)\n\tif (!i_areset_n)\n");
|
954 |
36 |
dgisselq |
else
|
955 |
37 |
dgisselq |
fprintf(fp, "\talways @(posedge i_clk)\n\tif (i_reset)\n");
|
956 |
36 |
dgisselq |
fprintf(fp,
|
957 |
37 |
dgisselq |
"\t\to_aux <= 1\'b0;\n"
|
958 |
|
|
"\telse if (i_ce)\n"
|
959 |
|
|
"\tbegin\n"
|
960 |
|
|
"\t\t// Second clock, latch for final clock\n"
|
961 |
|
|
"\t\to_aux <= aux_pipeline[AUXLEN-1];\n"
|
962 |
|
|
"\tend\n"
|
963 |
36 |
dgisselq |
"\n");
|
964 |
|
|
|
965 |
|
|
fprintf(fp,
|
966 |
|
|
"\t// As a final step, we pack our outputs into two packed two\'s\n"
|
967 |
|
|
"\t// complement numbers per output word, so that each output word\n"
|
968 |
|
|
"\t// has (2*OWIDTH) bits in it, with the top half being the real\n"
|
969 |
|
|
"\t// portion and the bottom half being the imaginary portion.\n"
|
970 |
|
|
"\tassign o_left = { rnd_left_r, rnd_left_i };\n"
|
971 |
|
|
"\tassign o_right= { rnd_right_r,rnd_right_i};\n"
|
972 |
|
|
"\n");
|
973 |
|
|
|
974 |
37 |
dgisselq |
fprintf(fp,
|
975 |
|
|
"`ifdef FORMAL\n");
|
976 |
36 |
dgisselq |
if (formal_property_flag) {
|
977 |
|
|
fprintf(fp,
|
978 |
|
|
"\tinitial\tf_dlyaux[0] = 0;\n"
|
979 |
|
|
"\talways @(posedge i_clk)\n"
|
980 |
|
|
"\tif (i_reset)\n"
|
981 |
|
|
"\t\tf_dlyaux\t<= 0;\n"
|
982 |
|
|
"\telse if (i_ce)\n"
|
983 |
|
|
"\t\tf_dlyaux\t<= { f_dlyaux[F_DEPTH-2:0], i_aux };\n"
|
984 |
|
|
"\n"
|
985 |
|
|
"\talways @(posedge i_clk)\n"
|
986 |
|
|
"\tif (i_ce)\n"
|
987 |
|
|
"\tbegin\n"
|
988 |
|
|
"\t f_dlyleft_r[0] <= i_left[ (2*IWIDTH-1):IWIDTH];\n"
|
989 |
|
|
"\t f_dlyleft_i[0] <= i_left[ ( IWIDTH-1):0];\n"
|
990 |
|
|
"\t f_dlyright_r[0] <= i_right[(2*IWIDTH-1):IWIDTH];\n"
|
991 |
|
|
"\t f_dlyright_i[0] <= i_right[( IWIDTH-1):0];\n"
|
992 |
|
|
"\t f_dlycoeff_r[0] <= i_coef[ (2*CWIDTH-1):CWIDTH];\n"
|
993 |
|
|
"\t f_dlycoeff_i[0] <= i_coef[ ( CWIDTH-1):0];\n"
|
994 |
|
|
"\tend\n"
|
995 |
|
|
"\n"
|
996 |
|
|
"\tgenvar k;\n"
|
997 |
|
|
"\tgenerate for(k=1; k<F_DEPTH; k=k+1)\n"
|
998 |
|
|
"\tbegin : F_PROPAGATE_DELAY_LINES\n"
|
999 |
|
|
"\n"
|
1000 |
|
|
"\n"
|
1001 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1002 |
|
|
"\t\tif (i_ce)\n"
|
1003 |
|
|
"\t\tbegin\n"
|
1004 |
|
|
"\t\t f_dlyleft_r[k] <= f_dlyleft_r[ k-1];\n"
|
1005 |
|
|
"\t\t f_dlyleft_i[k] <= f_dlyleft_i[ k-1];\n"
|
1006 |
|
|
"\t\t f_dlyright_r[k] <= f_dlyright_r[k-1];\n"
|
1007 |
|
|
"\t\t f_dlyright_i[k] <= f_dlyright_i[k-1];\n"
|
1008 |
|
|
"\t\t f_dlycoeff_r[k] <= f_dlycoeff_r[k-1];\n"
|
1009 |
|
|
"\t\t f_dlycoeff_i[k] <= f_dlycoeff_i[k-1];\n"
|
1010 |
|
|
"\t\tend\n"
|
1011 |
|
|
"\n"
|
1012 |
|
|
"\tend endgenerate\n"
|
1013 |
|
|
"\n"
|
1014 |
|
|
"`ifndef VERILATOR\n"
|
1015 |
37 |
dgisselq |
"\t//\n"
|
1016 |
|
|
"\t// Make some i_ce restraining assumptions. These are necessary\n"
|
1017 |
|
|
"\t// to get the design to pass induction.\n"
|
1018 |
|
|
"\t//\n"
|
1019 |
36 |
dgisselq |
"\tgenerate if (CKPCE <= 1)\n"
|
1020 |
|
|
"\tbegin\n"
|
1021 |
|
|
"\n"
|
1022 |
37 |
dgisselq |
"\t\t// No primary i_ce assumption. i_ce can be anything\n"
|
1023 |
|
|
"\t\t//\n"
|
1024 |
|
|
"\t\t// First induction i_ce assumption: No more than one\n"
|
1025 |
|
|
"\t\t// empty cycle between used cycles. Without this\n"
|
1026 |
|
|
"\t\t// assumption, or one like it, induction would never\n"
|
1027 |
|
|
"\t\t// complete.\n"
|
1028 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1029 |
|
|
"\t\tif ((!$past(i_ce)))\n"
|
1030 |
|
|
"\t\t\tassume(i_ce);\n"
|
1031 |
36 |
dgisselq |
"\n"
|
1032 |
37 |
dgisselq |
"\t\t// Second induction i_ce assumption: avoid skipping an\n"
|
1033 |
|
|
"\t\t// i_ce and thus stretching out the i_ce cycle two i_ce\n"
|
1034 |
|
|
"\t\t// cycles in a row. Without this assumption, induction\n"
|
1035 |
|
|
"\t\t// would still complete, it would just take longer\n"
|
1036 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1037 |
|
|
"\t\tif (($past(i_ce))&&(!$past(i_ce,2)))\n"
|
1038 |
|
|
"\t\t\tassume(i_ce);\n"
|
1039 |
|
|
"\n"
|
1040 |
36 |
dgisselq |
"\tend else if (CKPCE == 2)\n"
|
1041 |
|
|
"\tbegin : F_CKPCE_TWO\n"
|
1042 |
|
|
"\n"
|
1043 |
37 |
dgisselq |
"\t\t// Primary i_ce assumption: Every i_ce cycle is followed\n"
|
1044 |
|
|
"\t\t// by a non-i_ce cycle, so the multiplies can be\n"
|
1045 |
|
|
"\t\t// multiplexed\n"
|
1046 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1047 |
|
|
"\t\tif ($past(i_ce))\n"
|
1048 |
|
|
"\t\t\tassume(!i_ce);\n"
|
1049 |
|
|
|
1050 |
|
|
"\t\t// First induction assumption: Don't let this stretch\n"
|
1051 |
|
|
"\t\t// out too far. This is necessary to pass induction\n"
|
1052 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1053 |
|
|
"\t\tif ((!$past(i_ce))&&(!$past(i_ce,2)))\n"
|
1054 |
|
|
"\t\t\tassume(i_ce);\n"
|
1055 |
36 |
dgisselq |
"\n"
|
1056 |
37 |
dgisselq |
"\t\talways @(posedge i_clk)\n"
|
1057 |
|
|
"\t\tif ((!$past(i_ce))&&($past(i_ce,2))\n"
|
1058 |
|
|
"\t\t\t\t&&(!$past(i_ce,3))&&(!$past(i_ce,4)))\n"
|
1059 |
|
|
"\t\t\tassume(i_ce);\n"
|
1060 |
|
|
"\n"
|
1061 |
36 |
dgisselq |
"\tend else if (CKPCE == 3)\n"
|
1062 |
|
|
"\tbegin : F_CKPCE_THREE\n"
|
1063 |
|
|
"\n"
|
1064 |
37 |
dgisselq |
"\t\t// Primary i_ce assumption: Following any i_ce cycle,\n"
|
1065 |
|
|
"\t\t// there must be two clock cycles with i_ce de-asserted\n"
|
1066 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1067 |
|
|
"\t\tif (($past(i_ce))||($past(i_ce,2)))\n"
|
1068 |
|
|
"\t\t\tassume(!i_ce);\n"
|
1069 |
36 |
dgisselq |
"\n"
|
1070 |
37 |
dgisselq |
"\t\t// Induction assumption: Allow i_ce's every third or\n"
|
1071 |
|
|
"\t\t// fourth clock, but don't allow them to be separated\n"
|
1072 |
|
|
"\t\t// further than that\n"
|
1073 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1074 |
|
|
"\t\tif ((!$past(i_ce))&&(!$past(i_ce,2))&&(!$past(i_ce,3)))\n"
|
1075 |
|
|
"\t\t\tassume(i_ce);\n"
|
1076 |
|
|
"\n"
|
1077 |
|
|
"\t\t// Second induction assumption, to speed up the proof:\n"
|
1078 |
|
|
"\t\t// If it's the earliest possible opportunity for an\n"
|
1079 |
|
|
"\t\t// i_ce, and the last i_ce was late, don't let this one\n"
|
1080 |
|
|
"\t\t// be late as well.\n"
|
1081 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1082 |
|
|
"\t\tif ((!$past(i_ce))&&(!$past(i_ce,2))\n"
|
1083 |
|
|
"\t\t\t&&($past(i_ce,3))&&(!$past(i_ce,4))\n"
|
1084 |
|
|
"\t\t\t&&(!$past(i_ce,5))&&(!$past(i_ce,6)))\n"
|
1085 |
|
|
"\t\t\tassume(i_ce);\n"
|
1086 |
|
|
"\n"
|
1087 |
36 |
dgisselq |
"\tend endgenerate\n"
|
1088 |
|
|
"`endif\n"
|
1089 |
|
|
"\n"
|
1090 |
|
|
"\treg [F_LGDEPTH:0] f_startup_counter;\n"
|
1091 |
|
|
"\tinitial f_startup_counter = 0;\n"
|
1092 |
|
|
"\talways @(posedge i_clk)\n"
|
1093 |
|
|
"\tif (i_reset)\n"
|
1094 |
|
|
"\t f_startup_counter <= 0;\n"
|
1095 |
|
|
"\telse if ((i_ce)&&(!(&f_startup_counter)))\n"
|
1096 |
|
|
"\t f_startup_counter <= f_startup_counter + 1;\n"
|
1097 |
|
|
"\n"
|
1098 |
|
|
"\talways @(*)\n"
|
1099 |
|
|
"\tbegin\n"
|
1100 |
|
|
"\t f_sumr = f_dlyleft_r[F_D] + f_dlyright_r[F_D];\n"
|
1101 |
|
|
"\t f_sumi = f_dlyleft_i[F_D] + f_dlyright_i[F_D];\n"
|
1102 |
|
|
"\tend\n"
|
1103 |
|
|
"\n"
|
1104 |
|
|
"\tassign\tf_sumrx = { {(4){f_sumr[IWIDTH]}}, f_sumr, {(CWIDTH-2){1'b0}} };\n"
|
1105 |
|
|
"\tassign\tf_sumix = { {(4){f_sumi[IWIDTH]}}, f_sumi, {(CWIDTH-2){1'b0}} };\n"
|
1106 |
|
|
"\n"
|
1107 |
|
|
"\talways @(*)\n"
|
1108 |
|
|
"\tbegin\n"
|
1109 |
|
|
"\t f_difr = f_dlyleft_r[F_D] - f_dlyright_r[F_D];\n"
|
1110 |
|
|
"\t f_difi = f_dlyleft_i[F_D] - f_dlyright_i[F_D];\n"
|
1111 |
|
|
"\tend\n"
|
1112 |
|
|
"\n"
|
1113 |
|
|
"\tassign\tf_difrx = { {(CWIDTH+2){f_difr[IWIDTH]}}, f_difr };\n"
|
1114 |
|
|
"\tassign\tf_difix = { {(CWIDTH+2){f_difi[IWIDTH]}}, f_difi };\n"
|
1115 |
|
|
"\n"
|
1116 |
|
|
"\tassign\tf_widecoeff_r ={ {(IWIDTH+3){f_dlycoeff_r[F_D][CWIDTH-1]}},\n"
|
1117 |
|
|
"\t\t\t\t\t\tf_dlycoeff_r[F_D] };\n"
|
1118 |
|
|
"\tassign\tf_widecoeff_i ={ {(IWIDTH+3){f_dlycoeff_i[F_D][CWIDTH-1]}},\n"
|
1119 |
|
|
"\t\t\t\t\t\tf_dlycoeff_i[F_D] };\n"
|
1120 |
|
|
"\n"
|
1121 |
|
|
"\talways @(posedge i_clk)\n"
|
1122 |
|
|
"\tif (f_startup_counter > {1'b0, F_D})\n"
|
1123 |
|
|
"\tbegin\n"
|
1124 |
|
|
"\t assert(aux_pipeline == f_dlyaux);\n"
|
1125 |
|
|
"\t assert(left_sr == f_sumrx);\n"
|
1126 |
|
|
"\t assert(left_si == f_sumix);\n"
|
1127 |
|
|
"\t assert(aux_pipeline[AUXLEN-1] == f_dlyaux[F_D]);\n"
|
1128 |
|
|
"\n"
|
1129 |
|
|
"\t if ((f_difr == 0)&&(f_difi == 0))\n"
|
1130 |
|
|
"\t begin\n"
|
1131 |
|
|
"\t assert(mpy_r == 0);\n"
|
1132 |
|
|
"\t assert(mpy_i == 0);\n"
|
1133 |
|
|
"\t end else if ((f_dlycoeff_r[F_D] == 0)\n"
|
1134 |
|
|
"\t &&(f_dlycoeff_i[F_D] == 0))\n"
|
1135 |
|
|
"\t begin\n"
|
1136 |
|
|
"\t assert(mpy_r == 0);\n"
|
1137 |
|
|
"\t assert(mpy_i == 0);\n"
|
1138 |
|
|
"\t end\n"
|
1139 |
|
|
"\n"
|
1140 |
|
|
"\t if ((f_dlycoeff_r[F_D] == 1)&&(f_dlycoeff_i[F_D] == 0))\n"
|
1141 |
|
|
"\t begin\n"
|
1142 |
|
|
"\t assert(mpy_r == f_difrx);\n"
|
1143 |
|
|
"\t assert(mpy_i == f_difix);\n"
|
1144 |
|
|
"\t end\n"
|
1145 |
|
|
"\n"
|
1146 |
|
|
"\t if ((f_dlycoeff_r[F_D] == 0)&&(f_dlycoeff_i[F_D] == 1))\n"
|
1147 |
|
|
"\t begin\n"
|
1148 |
|
|
"\t assert(mpy_r == -f_difix);\n"
|
1149 |
|
|
"\t assert(mpy_i == f_difrx);\n"
|
1150 |
|
|
"\t end\n"
|
1151 |
|
|
"\n"
|
1152 |
|
|
"\t if ((f_difr == 1)&&(f_difi == 0))\n"
|
1153 |
|
|
"\t begin\n"
|
1154 |
|
|
"\t assert(mpy_r == f_widecoeff_r);\n"
|
1155 |
|
|
"\t assert(mpy_i == f_widecoeff_i);\n"
|
1156 |
|
|
"\t end\n"
|
1157 |
|
|
"\n"
|
1158 |
|
|
"\t if ((f_difr == 0)&&(f_difi == 1))\n"
|
1159 |
|
|
"\t begin\n"
|
1160 |
|
|
"\t assert(mpy_r == -f_widecoeff_i);\n"
|
1161 |
|
|
"\t assert(mpy_i == f_widecoeff_r);\n"
|
1162 |
|
|
"\t end\n"
|
1163 |
|
|
"\tend\n"
|
1164 |
|
|
"\n");
|
1165 |
|
|
|
1166 |
|
|
fprintf(fp,
|
1167 |
|
|
"\t// Let's see if we can improve our performance at all by\n"
|
1168 |
|
|
"\t// moving our test one clock earlier. If nothing else, it should\n"
|
1169 |
|
|
"\t// help induction finish one (or more) clocks ealier than\n"
|
1170 |
|
|
"\t// otherwise\n"
|
1171 |
|
|
"\n\n"
|
1172 |
|
|
"\talways @(*)\n"
|
1173 |
|
|
"\tbegin\n"
|
1174 |
|
|
"\t\tf_predifr = f_dlyleft_r[F_D-1] - f_dlyright_r[F_D-1];\n"
|
1175 |
|
|
"\t\tf_predifi = f_dlyleft_i[F_D-1] - f_dlyright_i[F_D-1];\n"
|
1176 |
|
|
"\tend\n"
|
1177 |
|
|
"\n"
|
1178 |
|
|
"\tassign f_predifrx = { {(CWIDTH+2){f_predifr[IWIDTH]}}, f_predifr };\n"
|
1179 |
|
|
"\tassign f_predifix = { {(CWIDTH+2){f_predifi[IWIDTH]}}, f_predifi };\n"
|
1180 |
|
|
"\n"
|
1181 |
|
|
"\talways @(*)\n"
|
1182 |
|
|
"\tbegin\n"
|
1183 |
|
|
"\t\tf_sumcoef = f_dlycoeff_r[F_D-1] + f_dlycoeff_i[F_D-1];\n"
|
1184 |
|
|
"\t\tf_sumdiff = f_predifr + f_predifi;\n"
|
1185 |
|
|
"\tend\n"
|
1186 |
|
|
"\n"
|
1187 |
|
|
"\t// Induction helpers\n"
|
1188 |
|
|
"\talways @(posedge i_clk)\n"
|
1189 |
|
|
"\tif (f_startup_counter >= { 1'b0, F_D })\n"
|
1190 |
|
|
"\tbegin\n"
|
1191 |
|
|
"\t\tif (f_dlycoeff_r[F_D-1] == 0)\n"
|
1192 |
|
|
"\t\t\tassert(p_one == 0);\n"
|
1193 |
|
|
"\t\tif (f_dlycoeff_i[F_D-1] == 0)\n"
|
1194 |
|
|
"\t\t\tassert(p_two == 0);\n"
|
1195 |
|
|
"\n"
|
1196 |
|
|
"\t\tif (f_dlycoeff_r[F_D-1] == 1)\n"
|
1197 |
|
|
"\t\t\tassert(p_one == f_predifrx);\n"
|
1198 |
|
|
"\t\tif (f_dlycoeff_i[F_D-1] == 1)\n"
|
1199 |
|
|
"\t\t\tassert(p_two == f_predifix);\n"
|
1200 |
|
|
"\n"
|
1201 |
|
|
"\t\tif (f_predifr == 0)\n"
|
1202 |
|
|
"\t\t\tassert(p_one == 0);\n"
|
1203 |
|
|
"\t\tif (f_predifi == 0)\n"
|
1204 |
|
|
"\t\t\tassert(p_two == 0);\n"
|
1205 |
|
|
"\n"
|
1206 |
|
|
"\t\t// verilator lint_off WIDTH\n"
|
1207 |
|
|
"\t\tif (f_predifr == 1)\n"
|
1208 |
|
|
"\t\t\tassert(p_one == f_dlycoeff_r[F_D-1]);\n"
|
1209 |
|
|
"\t\tif (f_predifi == 1)\n"
|
1210 |
|
|
"\t\t\tassert(p_two == f_dlycoeff_i[F_D-1]);\n"
|
1211 |
|
|
"\t\t// verilator lint_on WIDTH\n"
|
1212 |
|
|
"\n"
|
1213 |
|
|
"\t\tif (f_sumcoef == 0)\n"
|
1214 |
|
|
"\t\t\tassert(p_three == 0);\n"
|
1215 |
|
|
"\t\tif (f_sumdiff == 0)\n"
|
1216 |
|
|
"\t\t\tassert(p_three == 0);\n"
|
1217 |
|
|
"\t\t// verilator lint_off WIDTH\n"
|
1218 |
|
|
"\t\tif (f_sumcoef == 1)\n"
|
1219 |
|
|
"\t\t\tassert(p_three == f_sumdiff);\n"
|
1220 |
|
|
"\t\tif (f_sumdiff == 1)\n"
|
1221 |
|
|
"\t\t\tassert(p_three == f_sumcoef);\n"
|
1222 |
|
|
"\t\t// verilator lint_on WIDTH\n"
|
1223 |
|
|
"`ifdef VERILATOR\n"
|
1224 |
37 |
dgisselq |
"\t\t// Check that the multiplies match--but *ONLY* if using\n"
|
1225 |
|
|
"\t\t// Verilator, and not if using formal proper\n"
|
1226 |
36 |
dgisselq |
"\t\tassert(p_one == f_predifr * f_dlycoeff_r[F_D-1]);\n"
|
1227 |
|
|
"\t\tassert(p_two == f_predifi * f_dlycoeff_i[F_D-1]);\n"
|
1228 |
|
|
"\t\tassert(p_three == f_sumdiff * f_sumcoef);\n"
|
1229 |
|
|
"`endif // VERILATOR\n"
|
1230 |
|
|
"\tend\n\n");
|
1231 |
|
|
|
1232 |
|
|
fprintf(fp,
|
1233 |
37 |
dgisselq |
"\t// The following logic formally insists that our version of the\n"
|
1234 |
|
|
"\t// inputs to the multiply matches what the (multiclock) multiply\n"
|
1235 |
|
|
"\t// thinks its inputs were. While this may seem redundant, the\n"
|
1236 |
|
|
"\t// proof will not complete in any reasonable amount of time\n"
|
1237 |
|
|
"\t// without these assertions.\n"
|
1238 |
|
|
"\n"
|
1239 |
|
|
"\tassign\tf_p3c_in = f_dlycoeff_i[F_D-1] + f_dlycoeff_r[F_D-1];\n"
|
1240 |
|
|
"\tassign\tf_p3d_in = f_predifi + f_predifr;\n"
|
1241 |
|
|
"\n"
|
1242 |
|
|
"\talways @(*)\n"
|
1243 |
|
|
"\tif (f_startup_counter >= { 1'b0, F_D })\n"
|
1244 |
|
|
"\tbegin\n"
|
1245 |
|
|
"\t\tassert(fp_one_ic == { f_dlycoeff_r[F_D-1][CWIDTH-1],\n"
|
1246 |
|
|
"\t\t\t\tf_dlycoeff_r[F_D-1][CWIDTH-1:0] });\n"
|
1247 |
|
|
"\t\tassert(fp_two_ic == { f_dlycoeff_i[F_D-1][CWIDTH-1],\n"
|
1248 |
|
|
"\t\t\t\tf_dlycoeff_i[F_D-1][CWIDTH-1:0] });\n"
|
1249 |
|
|
"\t\tassert(fp_one_id == { f_predifr[IWIDTH], f_predifr });\n"
|
1250 |
|
|
"\t\tassert(fp_two_id == { f_predifi[IWIDTH], f_predifi });\n"
|
1251 |
|
|
"\t\tassert(fp_three_ic == f_p3c_in);\n"
|
1252 |
|
|
"\t\tassert(fp_three_id == f_p3d_in);\n"
|
1253 |
|
|
"\tend\n"
|
1254 |
|
|
"\n");
|
1255 |
|
|
|
1256 |
|
|
|
1257 |
|
|
fprintf(fp,
|
1258 |
36 |
dgisselq |
"\t// F_CHECK will be set externally by the solver, so that we can\n"
|
1259 |
|
|
"\t// double check that the solver is actually testing what we think\n"
|
1260 |
|
|
"\t// it is testing. We'll set it here to MPYREMAINDER, which will\n"
|
1261 |
|
|
"\t// essentially eliminate the check--unless overridden by the\n"
|
1262 |
|
|
"\t// solver.\n"
|
1263 |
|
|
"\tparameter F_CHECK = MPYREMAINDER;\n"
|
1264 |
|
|
"\tinitial assert(MPYREMAINDER == F_CHECK);\n\n");
|
1265 |
|
|
|
1266 |
37 |
dgisselq |
} else {
|
1267 |
|
|
fprintf(fp, "// Set the formal_property_flag to enable formal\n"
|
1268 |
|
|
"// property generation\n");
|
1269 |
|
|
}
|
1270 |
36 |
dgisselq |
fprintf(fp,
|
1271 |
|
|
"`endif // FORMAL\n");
|
1272 |
|
|
|
1273 |
|
|
fprintf(fp,
|
1274 |
|
|
"endmodule\n");
|
1275 |
|
|
fclose(fp);
|
1276 |
|
|
}
|
1277 |
|
|
|
1278 |
|
|
void build_hwbfly(const char *fname, int xtracbits, ROUND_T rounding,
|
1279 |
|
|
int ckpce, const bool async_reset) {
|
1280 |
|
|
FILE *fp = fopen(fname, "w");
|
1281 |
|
|
if (NULL == fp) {
|
1282 |
|
|
fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
|
1283 |
|
|
perror("O/S Err was:");
|
1284 |
|
|
return;
|
1285 |
|
|
}
|
1286 |
|
|
|
1287 |
|
|
const char *rnd_string;
|
1288 |
|
|
if (rounding == RND_TRUNCATE)
|
1289 |
|
|
rnd_string = "truncate";
|
1290 |
|
|
else if (rounding == RND_FROMZERO)
|
1291 |
|
|
rnd_string = "roundfromzero";
|
1292 |
|
|
else if (rounding == RND_HALFUP)
|
1293 |
|
|
rnd_string = "roundhalfup";
|
1294 |
|
|
else
|
1295 |
|
|
rnd_string = "convround";
|
1296 |
|
|
|
1297 |
|
|
std::string resetw("i_reset");
|
1298 |
|
|
if (async_reset)
|
1299 |
|
|
resetw = std::string("i_areset_n");
|
1300 |
|
|
|
1301 |
|
|
|
1302 |
|
|
fprintf(fp,
|
1303 |
|
|
SLASHLINE
|
1304 |
|
|
"//\n"
|
1305 |
|
|
"// Filename:\thwbfly.v\n"
|
1306 |
|
|
"//\n"
|
1307 |
|
|
"// Project:\t%s\n"
|
1308 |
|
|
"//\n"
|
1309 |
|
|
"// Purpose:\tThis routine is identical to the butterfly.v routine found\n"
|
1310 |
|
|
"// in 'butterfly.v', save only that it uses the verilog\n"
|
1311 |
|
|
"// operator '*' in hopes that the synthesizer would be able to optimize\n"
|
1312 |
|
|
"// it with hardware resources.\n"
|
1313 |
|
|
"//\n"
|
1314 |
|
|
"// It is understood that a hardware multiply can complete its operation in\n"
|
1315 |
|
|
"// a single clock.\n"
|
1316 |
|
|
"//\n"
|
1317 |
|
|
"// Operation:\n"
|
1318 |
|
|
"//\n"
|
1319 |
|
|
"// Given two inputs, A (i_left) and B (i_right), and a complex\n"
|
1320 |
|
|
"// coefficient C (i_coeff), return two outputs, O1 and O2, where:\n"
|
1321 |
|
|
"//\n"
|
1322 |
|
|
"// O1 = A + B, and\n"
|
1323 |
|
|
"// O2 = (A - B)*C\n"
|
1324 |
|
|
"//\n"
|
1325 |
|
|
"// This operation is commonly known as a Decimation in Frequency (DIF)\n"
|
1326 |
|
|
"// Radix-2 Butterfly.\n"
|
1327 |
|
|
"// O1 and O2 are rounded before being returned in (o_left) and o_right\n"
|
1328 |
|
|
"// to OWIDTH bits. If SHIFT is one, an extra bit is dropped from these\n"
|
1329 |
|
|
"// values during the rounding process.\n"
|
1330 |
|
|
"//\n"
|
1331 |
|
|
"// Further, since these outputs will take some number of clocks to\n"
|
1332 |
|
|
"// calculate, we'll pipe a value (i_aux) through the system and return\n"
|
1333 |
|
|
"// it with the results (o_aux), so you can synchronize to the outgoing\n"
|
1334 |
|
|
"// output stream.\n"
|
1335 |
|
|
"//\n"
|
1336 |
|
|
"//\n%s"
|
1337 |
|
|
"//\n", prjname, creator);
|
1338 |
|
|
fprintf(fp, "%s", cpyleft);
|
1339 |
|
|
fprintf(fp, "//\n//\n`default_nettype\tnone\n//\n");
|
1340 |
|
|
fprintf(fp,
|
1341 |
|
|
"module hwbfly(i_clk, %s, i_ce, i_coef, i_left, i_right, i_aux,\n"
|
1342 |
|
|
"\t\to_left, o_right, o_aux);\n"
|
1343 |
|
|
"\t// Public changeable parameters ...\n"
|
1344 |
|
|
"\t// - IWIDTH, number of bits in each component of the input\n"
|
1345 |
|
|
"\t// - CWIDTH, number of bits in each component of the twiddle factor\n"
|
1346 |
|
|
"\t// - OWIDTH, number of bits in each component of the output\n"
|
1347 |
|
|
"\tparameter IWIDTH=16,CWIDTH=IWIDTH+%d,OWIDTH=IWIDTH+1;\n"
|
1348 |
|
|
"\t// Drop an additional bit on the output?\n"
|
1349 |
|
|
"\tparameter\t\tSHIFT=0;\n"
|
1350 |
|
|
"\t// The number of clocks per clock enable, 1, 2, or 3.\n"
|
1351 |
|
|
"\tparameter\t[1:0]\tCKPCE=%d;\n\t//\n", resetw.c_str(), xtracbits,
|
1352 |
|
|
ckpce);
|
1353 |
|
|
|
1354 |
|
|
fprintf(fp,
|
1355 |
37 |
dgisselq |
"\tinput\twire\ti_clk, %s, i_ce;\n"
|
1356 |
|
|
"\tinput\twire\t[(2*CWIDTH-1):0]\ti_coef;\n"
|
1357 |
|
|
"\tinput\twire\t[(2*IWIDTH-1):0]\ti_left, i_right;\n"
|
1358 |
|
|
"\tinput\twire\ti_aux;\n"
|
1359 |
36 |
dgisselq |
"\toutput\twire\t[(2*OWIDTH-1):0]\to_left, o_right;\n"
|
1360 |
|
|
"\toutput\treg\to_aux;\n\n"
|
1361 |
|
|
"\n", resetw.c_str());
|
1362 |
|
|
|
1363 |
|
|
fprintf(fp,
|
1364 |
|
|
"\treg\t[(2*IWIDTH-1):0] r_left, r_right;\n"
|
1365 |
|
|
"\treg\t r_aux, r_aux_2;\n"
|
1366 |
|
|
"\treg\t[(2*CWIDTH-1):0] r_coef;\n"
|
1367 |
|
|
"\twire signed [(IWIDTH-1):0] r_left_r, r_left_i, r_right_r, r_right_i;\n"
|
1368 |
|
|
"\tassign\tr_left_r = r_left[ (2*IWIDTH-1):(IWIDTH)];\n"
|
1369 |
|
|
"\tassign\tr_left_i = r_left[ (IWIDTH-1):0];\n"
|
1370 |
|
|
"\tassign\tr_right_r = r_right[(2*IWIDTH-1):(IWIDTH)];\n"
|
1371 |
|
|
"\tassign\tr_right_i = r_right[(IWIDTH-1):0];\n"
|
1372 |
|
|
"\treg signed [(CWIDTH-1):0] ir_coef_r, ir_coef_i;\n"
|
1373 |
|
|
"\n"
|
1374 |
|
|
"\treg signed [(IWIDTH):0] r_sum_r, r_sum_i, r_dif_r, r_dif_i;\n"
|
1375 |
|
|
"\n"
|
1376 |
|
|
"\treg [(2*IWIDTH+2):0] leftv, leftvv;\n"
|
1377 |
|
|
"\n"
|
1378 |
|
|
"\t// Set up the input to the multiply\n"
|
1379 |
|
|
"\tinitial r_aux = 1\'b0;\n"
|
1380 |
|
|
"\tinitial r_aux_2 = 1\'b0;\n");
|
1381 |
|
|
if (async_reset)
|
1382 |
|
|
fprintf(fp, "\talways @(posedge i_clk, negedge i_areset_n)\n\t\tif (!i_areset_n)\n");
|
1383 |
|
|
else
|
1384 |
|
|
fprintf(fp, "\talways @(posedge i_clk)\n\t\tif (i_reset)\n");
|
1385 |
|
|
fprintf(fp,
|
1386 |
|
|
"\t\tbegin\n"
|
1387 |
|
|
"\t\t\tr_aux <= 1\'b0;\n"
|
1388 |
|
|
"\t\t\tr_aux_2 <= 1\'b0;\n"
|
1389 |
|
|
"\t\tend else if (i_ce)\n"
|
1390 |
|
|
"\t\tbegin\n"
|
1391 |
|
|
"\t\t\t// One clock just latches the inputs\n"
|
1392 |
|
|
"\t\t\tr_aux <= i_aux;\n"
|
1393 |
|
|
"\t\t\t// Next clock adds/subtracts\n"
|
1394 |
|
|
"\t\t\t// Other inputs are simply delayed on second clock\n"
|
1395 |
|
|
"\t\t\tr_aux_2 <= r_aux;\n"
|
1396 |
|
|
"\t\tend\n"
|
1397 |
|
|
"\talways @(posedge i_clk)\n"
|
1398 |
|
|
"\t\tif (i_ce)\n"
|
1399 |
|
|
"\t\tbegin\n"
|
1400 |
|
|
"\t\t\t// One clock just latches the inputs\n"
|
1401 |
|
|
"\t\t\tr_left <= i_left; // No change in # of bits\n"
|
1402 |
|
|
"\t\t\tr_right <= i_right;\n"
|
1403 |
|
|
"\t\t\tr_coef <= i_coef;\n"
|
1404 |
|
|
"\t\t\t// Next clock adds/subtracts\n"
|
1405 |
|
|
"\t\t\tr_sum_r <= r_left_r + r_right_r; // Now IWIDTH+1 bits\n"
|
1406 |
|
|
"\t\t\tr_sum_i <= r_left_i + r_right_i;\n"
|
1407 |
|
|
"\t\t\tr_dif_r <= r_left_r - r_right_r;\n"
|
1408 |
|
|
"\t\t\tr_dif_i <= r_left_i - r_right_i;\n"
|
1409 |
|
|
"\t\t\t// Other inputs are simply delayed on second clock\n"
|
1410 |
|
|
"\t\t\tir_coef_r <= r_coef[(2*CWIDTH-1):CWIDTH];\n"
|
1411 |
|
|
"\t\t\tir_coef_i <= r_coef[(CWIDTH-1):0];\n"
|
1412 |
|
|
"\t\tend\n"
|
1413 |
|
|
"\n\n");
|
1414 |
|
|
fprintf(fp,
|
1415 |
|
|
"\t// See comments in the butterfly.v source file for a discussion of\n"
|
1416 |
|
|
"\t// these operations and the appropriate bit widths.\n\n");
|
1417 |
|
|
fprintf(fp,
|
1418 |
|
|
"\twire\tsigned [((IWIDTH+1)+(CWIDTH)-1):0] p_one, p_two;\n"
|
1419 |
|
|
"\twire\tsigned [((IWIDTH+2)+(CWIDTH+1)-1):0] p_three;\n"
|
1420 |
|
|
"\n"
|
1421 |
|
|
"\tinitial leftv = 0;\n"
|
1422 |
|
|
"\tinitial leftvv = 0;\n");
|
1423 |
|
|
if (async_reset)
|
1424 |
|
|
fprintf(fp, "\talways @(posedge i_clk, negedge i_areset_n)\n\t\tif (!i_areset_n)\n");
|
1425 |
|
|
else
|
1426 |
|
|
fprintf(fp, "\talways @(posedge i_clk)\n\t\tif (i_reset)\n");
|
1427 |
|
|
fprintf(fp,
|
1428 |
|
|
"\t\tbegin\n"
|
1429 |
|
|
"\t\t\tleftv <= 0;\n"
|
1430 |
|
|
"\t\t\tleftvv <= 0;\n"
|
1431 |
|
|
"\t\tend else if (i_ce)\n"
|
1432 |
|
|
"\t\tbegin\n"
|
1433 |
|
|
"\t\t\t// Second clock, pipeline = 1\n"
|
1434 |
|
|
"\t\t\tleftv <= { r_aux_2, r_sum_r, r_sum_i };\n"
|
1435 |
|
|
"\n"
|
1436 |
|
|
"\t\t\t// Third clock, pipeline = 3\n"
|
1437 |
|
|
"\t\t\t// As desired, each of these lines infers a DSP48\n"
|
1438 |
|
|
"\t\t\tleftvv <= leftv;\n"
|
1439 |
|
|
"\t\tend\n"
|
1440 |
|
|
"\n");
|
1441 |
|
|
|
1442 |
|
|
// Nominally, we should handle code for 1, 2, or 3 clocks per CE, with
|
1443 |
|
|
// one clock per CE meaning CE could be constant. The code below
|
1444 |
|
|
// instead handles 1 or 3 clocks per CE, leaving the two clocks per
|
1445 |
|
|
// CE optimization(s) unfulfilled.
|
1446 |
|
|
|
1447 |
|
|
// fprintf(fp,
|
1448 |
|
|
//"\tend else if (CKPCI == 2'b01)\n\tbegin\n");
|
1449 |
|
|
|
1450 |
|
|
///////////////////////////////////////////
|
1451 |
|
|
///
|
1452 |
|
|
/// One clock per CE, so CE, CE, CE, CE, CE is possible
|
1453 |
|
|
///
|
1454 |
|
|
fprintf(fp,
|
1455 |
|
|
"\tgenerate if (CKPCE <= 1)\n\tbegin : CKPCE_ONE\n");
|
1456 |
|
|
|
1457 |
|
|
fprintf(fp,
|
1458 |
|
|
"\t\t// Coefficient multiply inputs\n"
|
1459 |
|
|
"\t\treg\tsigned [(CWIDTH-1):0] p1c_in, p2c_in;\n"
|
1460 |
|
|
"\t\t// Data multiply inputs\n"
|
1461 |
|
|
"\t\treg\tsigned [(IWIDTH):0] p1d_in, p2d_in;\n"
|
1462 |
|
|
"\t\t// Product 3, coefficient input\n"
|
1463 |
|
|
"\t\treg\tsigned [(CWIDTH):0] p3c_in;\n"
|
1464 |
|
|
"\t\t// Product 3, data input\n"
|
1465 |
|
|
"\t\treg\tsigned [(IWIDTH+1):0] p3d_in;\n"
|
1466 |
|
|
"\n");
|
1467 |
|
|
fprintf(fp,
|
1468 |
|
|
"\t\treg\tsigned [((IWIDTH+1)+(CWIDTH)-1):0] rp_one, rp_two;\n"
|
1469 |
|
|
"\t\treg\tsigned [((IWIDTH+2)+(CWIDTH+1)-1):0] rp_three;\n"
|
1470 |
|
|
"\n");
|
1471 |
|
|
|
1472 |
|
|
fprintf(fp,
|
1473 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1474 |
|
|
"\t\tif (i_ce)\n"
|
1475 |
|
|
"\t\tbegin\n"
|
1476 |
|
|
"\t\t\t// Second clock, pipeline = 1\n"
|
1477 |
|
|
"\t\t\tp1c_in <= ir_coef_r;\n"
|
1478 |
|
|
"\t\t\tp2c_in <= ir_coef_i;\n"
|
1479 |
|
|
"\t\t\tp1d_in <= r_dif_r;\n"
|
1480 |
|
|
"\t\t\tp2d_in <= r_dif_i;\n"
|
1481 |
|
|
"\t\t\tp3c_in <= ir_coef_i + ir_coef_r;\n"
|
1482 |
|
|
"\t\t\tp3d_in <= r_dif_r + r_dif_i;\n"
|
1483 |
|
|
"\t\tend\n\n");
|
1484 |
|
|
|
1485 |
|
|
if (formal_property_flag)
|
1486 |
|
|
fprintf(fp,
|
1487 |
|
|
"`ifndef FORMAL\n");
|
1488 |
|
|
|
1489 |
|
|
fprintf(fp,
|
1490 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1491 |
|
|
"\t\tif (i_ce)\n"
|
1492 |
|
|
"\t\tbegin\n"
|
1493 |
|
|
"\t\t\t// Third clock, pipeline = 3\n"
|
1494 |
|
|
"\t\t\t// As desired, each of these lines infers a DSP48\n"
|
1495 |
|
|
"\t\t\trp_one <= p1c_in * p1d_in;\n"
|
1496 |
|
|
"\t\t\trp_two <= p2c_in * p2d_in;\n"
|
1497 |
|
|
"\t\t\trp_three <= p3c_in * p3d_in;\n"
|
1498 |
|
|
"\t\tend\n");
|
1499 |
|
|
|
1500 |
|
|
if (formal_property_flag)
|
1501 |
|
|
fprintf(fp,
|
1502 |
|
|
"`else\n"
|
1503 |
|
|
"\t\twire signed [((IWIDTH+1)+(CWIDTH)-1):0] pre_rp_one, pre_rp_two;\n"
|
1504 |
|
|
"\t\twire signed [((IWIDTH+2)+(CWIDTH+1)-1):0] pre_rp_three;\n"
|
1505 |
|
|
"\n"
|
1506 |
|
|
"\t\tabs_mpy #(CWIDTH,IWIDTH+1,1'b1)\n"
|
1507 |
|
|
"\t\t onei(p1c_in, p1d_in, pre_rp_one);\n"
|
1508 |
|
|
"\t\tabs_mpy #(CWIDTH,IWIDTH+1,1'b1)\n"
|
1509 |
|
|
"\t\t twoi(p2c_in, p2d_in, pre_rp_two);\n"
|
1510 |
|
|
"\t\tabs_mpy #(CWIDTH+1,IWIDTH+2,1'b1)\n"
|
1511 |
|
|
"\t\t threei(p3c_in, p3d_in, pre_rp_three);\n"
|
1512 |
|
|
"\n"
|
1513 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1514 |
|
|
"\t\tif (i_ce)\n"
|
1515 |
|
|
"\t\tbegin\n"
|
1516 |
|
|
"\t\t rp_one = pre_rp_one;\n"
|
1517 |
|
|
"\t\t rp_two = pre_rp_two;\n"
|
1518 |
|
|
"\t\t rp_three = pre_rp_three;\n"
|
1519 |
|
|
"\t\tend\n"
|
1520 |
|
|
"`endif // FORMAL\n");
|
1521 |
|
|
|
1522 |
|
|
fprintf(fp,"\n"
|
1523 |
|
|
"\t\tassign\tp_one = rp_one;\n"
|
1524 |
|
|
"\t\tassign\tp_two = rp_two;\n"
|
1525 |
|
|
"\t\tassign\tp_three = rp_three;\n"
|
1526 |
|
|
"\n");
|
1527 |
|
|
|
1528 |
|
|
///////////////////////////////////////////
|
1529 |
|
|
///
|
1530 |
|
|
/// Two clocks per CE, so CE, no-ce, CE, no-ce, etc
|
1531 |
|
|
///
|
1532 |
|
|
fprintf(fp,
|
1533 |
|
|
"\tend else if (CKPCE <= 2)\n"
|
1534 |
|
|
"\tbegin : CKPCE_TWO\n"
|
1535 |
|
|
"\t\t// Coefficient multiply inputs\n"
|
1536 |
|
|
"\t\treg [2*(CWIDTH)-1:0] mpy_pipe_c;\n"
|
1537 |
|
|
"\t\t// Data multiply inputs\n"
|
1538 |
|
|
"\t\treg [2*(IWIDTH+1)-1:0] mpy_pipe_d;\n"
|
1539 |
|
|
"\t\twire signed [(CWIDTH-1):0] mpy_pipe_vc;\n"
|
1540 |
|
|
"\t\twire signed [(IWIDTH):0] mpy_pipe_vd;\n"
|
1541 |
|
|
"\t\t//\n"
|
1542 |
|
|
"\t\treg signed [(CWIDTH+1)-1:0] mpy_cof_sum;\n"
|
1543 |
|
|
"\t\treg signed [(IWIDTH+2)-1:0] mpy_dif_sum;\n"
|
1544 |
|
|
"\n"
|
1545 |
|
|
"\t\tassign mpy_pipe_vc = mpy_pipe_c[2*(CWIDTH)-1:CWIDTH];\n"
|
1546 |
|
|
"\t\tassign mpy_pipe_vd = mpy_pipe_d[2*(IWIDTH+1)-1:IWIDTH+1];\n"
|
1547 |
|
|
"\n"
|
1548 |
|
|
"\t\treg mpy_pipe_v;\n"
|
1549 |
|
|
"\t\treg ce_phase;\n"
|
1550 |
|
|
"\n"
|
1551 |
|
|
"\t\treg signed [(CWIDTH+IWIDTH+1)-1:0] mpy_pipe_out;\n"
|
1552 |
|
|
"\t\treg signed [IWIDTH+CWIDTH+3-1:0] longmpy;\n"
|
1553 |
|
|
"\n"
|
1554 |
|
|
"\n"
|
1555 |
|
|
"\t\tinitial ce_phase = 1'b1;\n"
|
1556 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1557 |
|
|
"\t\tif (i_reset)\n"
|
1558 |
|
|
"\t\t\tce_phase <= 1'b1;\n"
|
1559 |
|
|
"\t\telse if (i_ce)\n"
|
1560 |
|
|
"\t\t\tce_phase <= 1'b0;\n"
|
1561 |
|
|
"\t\telse\n"
|
1562 |
|
|
"\t\t\tce_phase <= 1'b1;\n"
|
1563 |
|
|
"\n"
|
1564 |
|
|
"\t\talways @(*)\n"
|
1565 |
|
|
"\t\t\tmpy_pipe_v = (i_ce)||(!ce_phase);\n"
|
1566 |
|
|
"\n"
|
1567 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1568 |
|
|
"\t\tif (!ce_phase)\n"
|
1569 |
|
|
"\t\tbegin\n"
|
1570 |
|
|
"\t\t\t// Pre-clock\n"
|
1571 |
|
|
"\t\t\tmpy_pipe_c[2*CWIDTH-1:0] <=\n"
|
1572 |
|
|
"\t\t\t\t\t{ ir_coef_r, ir_coef_i };\n"
|
1573 |
|
|
"\t\t\tmpy_pipe_d[2*(IWIDTH+1)-1:0] <=\n"
|
1574 |
|
|
"\t\t\t\t\t{ r_dif_r, r_dif_i };\n"
|
1575 |
|
|
"\n"
|
1576 |
|
|
"\t\t\tmpy_cof_sum <= ir_coef_i + ir_coef_r;\n"
|
1577 |
|
|
"\t\t\tmpy_dif_sum <= r_dif_r + r_dif_i;\n"
|
1578 |
|
|
"\n"
|
1579 |
|
|
"\t\tend else if (i_ce)\n"
|
1580 |
|
|
"\t\tbegin\n"
|
1581 |
|
|
"\t\t\t// First clock\n"
|
1582 |
|
|
"\t\t\tmpy_pipe_c[2*(CWIDTH)-1:0] <= {\n"
|
1583 |
|
|
"\t\t\t\tmpy_pipe_c[(CWIDTH)-1:0], {(CWIDTH){1'b0}} };\n"
|
1584 |
|
|
"\t\t\tmpy_pipe_d[2*(IWIDTH+1)-1:0] <= {\n"
|
1585 |
|
|
"\t\t\t\tmpy_pipe_d[(IWIDTH+1)-1:0], {(IWIDTH+1){1'b0}} };\n"
|
1586 |
|
|
"\t\tend\n\n");
|
1587 |
|
|
|
1588 |
|
|
if (formal_property_flag)
|
1589 |
|
|
fprintf(fp, "`ifndef FORMAL\n");
|
1590 |
|
|
|
1591 |
|
|
fprintf(fp,
|
1592 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1593 |
|
|
"\t\tif (i_ce) // First clock\n"
|
1594 |
|
|
"\t\t\tlongmpy <= mpy_cof_sum * mpy_dif_sum;\n"
|
1595 |
|
|
"\n"
|
1596 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1597 |
|
|
"\t\tif (mpy_pipe_v)\n"
|
1598 |
|
|
"\t\t\tmpy_pipe_out <= mpy_pipe_vc * mpy_pipe_vd;\n");
|
1599 |
|
|
|
1600 |
|
|
if (formal_property_flag)
|
1601 |
|
|
fprintf(fp, "`else\n"
|
1602 |
|
|
"\t\twire signed [IWIDTH+CWIDTH+3-1:0] pre_longmpy;\n"
|
1603 |
|
|
"\t\twire signed [(CWIDTH+IWIDTH+1)-1:0] pre_mpy_pipe_out;\n"
|
1604 |
|
|
"\n"
|
1605 |
|
|
"\t\tabs_mpy #(CWIDTH+1,IWIDTH+2,1)\n"
|
1606 |
|
|
"\t\t longmpyi(mpy_cof_sum, mpy_dif_sum, pre_longmpy);\n"
|
1607 |
|
|
"\n"
|
1608 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1609 |
|
|
"\t\tif (i_ce)\n"
|
1610 |
|
|
"\t\t longmpy <= pre_longmpy;\n"
|
1611 |
|
|
"\n"
|
1612 |
|
|
"\n"
|
1613 |
|
|
"\t\tabs_mpy #(CWIDTH,IWIDTH+1,1)\n"
|
1614 |
|
|
"\t\t mpy_pipe_outi(mpy_pipe_vc, mpy_pipe_vd, pre_mpy_pipe_out);\n"
|
1615 |
|
|
"\n"
|
1616 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1617 |
|
|
"\t\tif (mpy_pipe_v)\n"
|
1618 |
|
|
"\t\t mpy_pipe_out <= pre_mpy_pipe_out;\n"
|
1619 |
|
|
"`endif\n");
|
1620 |
|
|
|
1621 |
|
|
fprintf(fp,"\n"
|
1622 |
|
|
"\t\treg\tsigned\t[((IWIDTH+1)+(CWIDTH)-1):0] rp_one,\n"
|
1623 |
|
|
"\t\t\t\t\t\t\trp2_one, rp_two;\n"
|
1624 |
|
|
"\t\treg\tsigned\t[((IWIDTH+2)+(CWIDTH+1)-1):0] rp_three;\n"
|
1625 |
|
|
"\n"
|
1626 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1627 |
|
|
"\t\tif (!ce_phase) // 1.5 clock\n"
|
1628 |
|
|
"\t\t\trp_one <= mpy_pipe_out;\n"
|
1629 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1630 |
|
|
"\t\tif (i_ce) // two clocks\n"
|
1631 |
|
|
"\t\t\trp_two <= mpy_pipe_out;\n"
|
1632 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1633 |
|
|
"\t\tif (i_ce) // Second clock\n"
|
1634 |
|
|
"\t\t\trp_three<= longmpy;\n"
|
1635 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1636 |
|
|
"\t\tif (i_ce)\n"
|
1637 |
|
|
"\t\t\trp2_one<= rp_one;\n"
|
1638 |
|
|
"\n"
|
1639 |
|
|
"\t\tassign p_one = rp2_one;\n"
|
1640 |
|
|
"\t\tassign p_two = rp_two;\n"
|
1641 |
|
|
"\t\tassign p_three= rp_three;\n"
|
1642 |
|
|
"\n");
|
1643 |
|
|
|
1644 |
|
|
/////////////////////////
|
1645 |
|
|
///
|
1646 |
|
|
/// Three clock per CE, so CE, no-ce, no-ce*, CE
|
1647 |
|
|
///
|
1648 |
|
|
fprintf(fp,
|
1649 |
|
|
"\tend else if (CKPCE <= 2'b11)\n\tbegin : CKPCE_THREE\n");
|
1650 |
|
|
|
1651 |
|
|
fprintf(fp,
|
1652 |
|
|
"\t\t// Coefficient multiply inputs\n"
|
1653 |
|
|
"\t\treg\t\t[3*(CWIDTH+1)-1:0]\tmpy_pipe_c;\n"
|
1654 |
|
|
"\t\t// Data multiply inputs\n"
|
1655 |
|
|
"\t\treg\t\t[3*(IWIDTH+2)-1:0]\tmpy_pipe_d;\n"
|
1656 |
|
|
"\t\twire\tsigned [(CWIDTH):0] mpy_pipe_vc;\n"
|
1657 |
|
|
"\t\twire\tsigned [(IWIDTH+1):0] mpy_pipe_vd;\n"
|
1658 |
|
|
"\n"
|
1659 |
|
|
"\t\tassign\tmpy_pipe_vc = mpy_pipe_c[3*(CWIDTH+1)-1:2*(CWIDTH+1)];\n"
|
1660 |
|
|
"\t\tassign\tmpy_pipe_vd = mpy_pipe_d[3*(IWIDTH+2)-1:2*(IWIDTH+2)];\n"
|
1661 |
|
|
"\n"
|
1662 |
|
|
"\t\treg\t\t\tmpy_pipe_v;\n"
|
1663 |
|
|
"\t\treg\t\t[2:0]\tce_phase;\n"
|
1664 |
|
|
"\n"
|
1665 |
|
|
"\t\treg\tsigned [ (CWIDTH+IWIDTH+3)-1:0] mpy_pipe_out;\n"
|
1666 |
|
|
"\n");
|
1667 |
|
|
fprintf(fp,
|
1668 |
|
|
"\t\tinitial\tce_phase = 3'b011;\n"
|
1669 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1670 |
|
|
"\t\tif (i_reset)\n"
|
1671 |
|
|
"\t\t\tce_phase <= 3'b011;\n"
|
1672 |
|
|
"\t\telse if (i_ce)\n"
|
1673 |
|
|
"\t\t\tce_phase <= 3'b000;\n"
|
1674 |
|
|
"\t\telse if (ce_phase != 3'b011)\n"
|
1675 |
|
|
"\t\t\tce_phase <= ce_phase + 1'b1;\n"
|
1676 |
|
|
"\n"
|
1677 |
|
|
"\t\talways @(*)\n"
|
1678 |
|
|
"\t\t\tmpy_pipe_v = (i_ce)||(ce_phase < 3'b010);\n"
|
1679 |
|
|
"\n");
|
1680 |
|
|
|
1681 |
|
|
fprintf(fp,
|
1682 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1683 |
|
|
"\t\t\tif (ce_phase == 3\'b000)\n"
|
1684 |
|
|
"\t\t\tbegin\n"
|
1685 |
|
|
"\t\t\t\t// Second clock\n"
|
1686 |
|
|
"\t\t\t\tmpy_pipe_c[3*(CWIDTH+1)-1:(CWIDTH+1)] <= {\n"
|
1687 |
|
|
"\t\t\t\t\tir_coef_r[CWIDTH-1], ir_coef_r,\n"
|
1688 |
|
|
"\t\t\t\t\tir_coef_i[CWIDTH-1], ir_coef_i };\n"
|
1689 |
|
|
"\t\t\t\tmpy_pipe_c[CWIDTH:0] <= ir_coef_i + ir_coef_r;\n"
|
1690 |
|
|
"\t\t\t\tmpy_pipe_d[3*(IWIDTH+2)-1:(IWIDTH+2)] <= {\n"
|
1691 |
|
|
"\t\t\t\t\tr_dif_r[IWIDTH], r_dif_r,\n"
|
1692 |
|
|
"\t\t\t\t\tr_dif_i[IWIDTH], r_dif_i };\n"
|
1693 |
|
|
"\t\t\t\tmpy_pipe_d[(IWIDTH+2)-1:0] <= r_dif_r + r_dif_i;\n"
|
1694 |
|
|
"\n"
|
1695 |
|
|
"\t\t\tend else if (mpy_pipe_v)\n"
|
1696 |
|
|
"\t\t\tbegin\n"
|
1697 |
|
|
"\t\t\t\tmpy_pipe_c[3*(CWIDTH+1)-1:0] <= {\n"
|
1698 |
|
|
"\t\t\t\t\tmpy_pipe_c[2*(CWIDTH+1)-1:0], {(CWIDTH+1){1\'b0}} };\n"
|
1699 |
|
|
"\t\t\t\tmpy_pipe_d[3*(IWIDTH+2)-1:0] <= {\n"
|
1700 |
|
|
"\t\t\t\t\tmpy_pipe_d[2*(IWIDTH+2)-1:0], {(IWIDTH+2){1\'b0}} };\n"
|
1701 |
|
|
"\t\t\tend\n\n");
|
1702 |
|
|
|
1703 |
|
|
if (formal_property_flag)
|
1704 |
|
|
fprintf(fp, "`ifndef\tFORMAL\n");
|
1705 |
|
|
|
1706 |
|
|
fprintf(fp,
|
1707 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1708 |
|
|
"\t\t\tif (mpy_pipe_v)\n"
|
1709 |
|
|
"\t\t\t\tmpy_pipe_out <= mpy_pipe_vc * mpy_pipe_vd;\n"
|
1710 |
|
|
"\n");
|
1711 |
|
|
|
1712 |
|
|
if (formal_property_flag)
|
1713 |
|
|
fprintf(fp,
|
1714 |
|
|
"`else\t// FORMAL\n"
|
1715 |
|
|
"\t\twire signed [ (CWIDTH+IWIDTH+3)-1:0] pre_mpy_pipe_out;\n"
|
1716 |
|
|
"\n"
|
1717 |
|
|
"\t\tabs_mpy #(CWIDTH+1,IWIDTH+2,1)\n"
|
1718 |
|
|
"\t\t mpy_pipe_outi(mpy_pipe_vc, mpy_pipe_vd, pre_mpy_pipe_out);\n"
|
1719 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1720 |
|
|
"\t\t if (mpy_pipe_v)\n"
|
1721 |
|
|
"\t\t mpy_pipe_out <= pre_mpy_pipe_out;\n"
|
1722 |
|
|
"`endif\t// FORMAL\n\n");
|
1723 |
|
|
|
1724 |
|
|
|
1725 |
|
|
fprintf(fp,
|
1726 |
|
|
"\t\treg\tsigned\t[((IWIDTH+1)+(CWIDTH)-1):0]\trp_one, rp_two,\n"
|
1727 |
|
|
"\t\t\t\t\t\trp2_one, rp2_two;\n"
|
1728 |
|
|
"\t\treg\tsigned\t[((IWIDTH+2)+(CWIDTH+1)-1):0]\trp_three, rp2_three;\n"
|
1729 |
|
|
|
1730 |
|
|
"\n");
|
1731 |
|
|
|
1732 |
|
|
fprintf(fp,
|
1733 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1734 |
|
|
"\t\tif(i_ce)\n"
|
1735 |
|
|
"\t\t\trp_one <= mpy_pipe_out[(CWIDTH+IWIDTH):0];\n"
|
1736 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1737 |
|
|
"\t\tif(ce_phase == 3'b000)\n"
|
1738 |
|
|
"\t\t\trp_two <= mpy_pipe_out[(CWIDTH+IWIDTH):0];\n"
|
1739 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1740 |
|
|
"\t\tif(ce_phase == 3'b001)\n"
|
1741 |
|
|
"\t\t\trp_three <= mpy_pipe_out;\n"
|
1742 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1743 |
|
|
"\t\tif (i_ce)\n"
|
1744 |
|
|
"\t\tbegin\n"
|
1745 |
|
|
"\t\t\trp2_one<= rp_one;\n"
|
1746 |
|
|
"\t\t\trp2_two<= rp_two;\n"
|
1747 |
|
|
"\t\t\trp2_three<= rp_three;\n"
|
1748 |
|
|
"\t\tend\n");
|
1749 |
|
|
fprintf(fp,
|
1750 |
|
|
"\t\tassign p_one\t= rp2_one;\n"
|
1751 |
|
|
"\t\tassign p_two\t= rp2_two;\n"
|
1752 |
|
|
"\t\tassign\tp_three\t= rp2_three;\n"
|
1753 |
|
|
"\n");
|
1754 |
|
|
|
1755 |
|
|
fprintf(fp,
|
1756 |
|
|
"\tend endgenerate\n");
|
1757 |
|
|
|
1758 |
|
|
fprintf(fp,
|
1759 |
|
|
"\twire\tsigned [((IWIDTH+2)+(CWIDTH+1)-1):0] w_one, w_two;\n"
|
1760 |
|
|
"\tassign\tw_one = { {(2){p_one[((IWIDTH+1)+(CWIDTH)-1)]}}, p_one };\n"
|
1761 |
|
|
"\tassign\tw_two = { {(2){p_two[((IWIDTH+1)+(CWIDTH)-1)]}}, p_two };\n"
|
1762 |
|
|
"\n");
|
1763 |
|
|
|
1764 |
|
|
fprintf(fp,
|
1765 |
|
|
"\t// These values are held in memory and delayed during the\n"
|
1766 |
|
|
"\t// multiply. Here, we recover them. During the multiply,\n"
|
1767 |
|
|
"\t// values were multiplied by 2^(CWIDTH-2)*exp{-j*2*pi*...},\n"
|
1768 |
|
|
"\t// therefore, the left_x values need to be right shifted by\n"
|
1769 |
|
|
"\t// CWIDTH-2 as well. The additional bits come from a sign\n"
|
1770 |
|
|
"\t// extension.\n"
|
1771 |
|
|
"\twire\taux_s;\n"
|
1772 |
|
|
"\twire\tsigned\t[(IWIDTH+CWIDTH):0] left_si, left_sr;\n"
|
1773 |
|
|
"\treg\t\t[(2*IWIDTH+2):0] left_saved;\n"
|
1774 |
|
|
"\tassign\tleft_sr = { {2{left_saved[2*(IWIDTH+1)-1]}}, left_saved[(2*(IWIDTH+1)-1):(IWIDTH+1)], {(CWIDTH-2){1\'b0}} };\n"
|
1775 |
|
|
"\tassign\tleft_si = { {2{left_saved[(IWIDTH+1)-1]}}, left_saved[((IWIDTH+1)-1):0], {(CWIDTH-2){1\'b0}} };\n"
|
1776 |
|
|
"\tassign\taux_s = left_saved[2*IWIDTH+2];\n"
|
1777 |
|
|
"\n"
|
1778 |
|
|
"\t(* use_dsp48=\"no\" *)\n"
|
1779 |
|
|
"\treg signed [(CWIDTH+IWIDTH+3-1):0] mpy_r, mpy_i;\n"
|
1780 |
|
|
"\n");
|
1781 |
|
|
|
1782 |
|
|
fprintf(fp,
|
1783 |
|
|
"\tinitial left_saved = 0;\n"
|
1784 |
|
|
"\tinitial o_aux = 1\'b0;\n");
|
1785 |
|
|
if (async_reset)
|
1786 |
|
|
fprintf(fp, "\talways @(posedge i_clk, negedge i_areset_n)\n\t\tif (!i_areset_n)\n");
|
1787 |
|
|
else
|
1788 |
|
|
fprintf(fp, "\talways @(posedge i_clk)\n\t\tif (i_reset)\n");
|
1789 |
|
|
fprintf(fp,
|
1790 |
|
|
"\t\tbegin\n"
|
1791 |
|
|
"\t\t\tleft_saved <= 0;\n"
|
1792 |
|
|
"\t\t\to_aux <= 1\'b0;\n"
|
1793 |
|
|
"\t\tend else if (i_ce)\n"
|
1794 |
|
|
"\t\tbegin\n"
|
1795 |
|
|
"\t\t\t// First clock, recover all values\n"
|
1796 |
|
|
"\t\t\tleft_saved <= leftvv;\n"
|
1797 |
|
|
"\n"
|
1798 |
|
|
"\t\t\t// Second clock, round and latch for final clock\n"
|
1799 |
|
|
"\t\t\to_aux <= aux_s;\n"
|
1800 |
|
|
"\t\tend\n"
|
1801 |
|
|
"\talways @(posedge i_clk)\n"
|
1802 |
|
|
"\t\tif (i_ce)\n"
|
1803 |
|
|
"\t\tbegin\n"
|
1804 |
|
|
"\t\t\t// These values are IWIDTH+CWIDTH+3 bits wide\n"
|
1805 |
|
|
"\t\t\t// although they only need to be (IWIDTH+1)\n"
|
1806 |
|
|
"\t\t\t// + (CWIDTH) bits wide. (We've got two\n"
|
1807 |
|
|
"\t\t\t// extra bits we need to get rid of.)\n"
|
1808 |
|
|
"\n"
|
1809 |
|
|
"\t\t\t// These two lines also infer DSP48\'s.\n"
|
1810 |
|
|
"\t\t\t// To keep from using extra DSP48 resources,\n"
|
1811 |
|
|
"\t\t\t// they are prevented from using DSP48\'s\n"
|
1812 |
|
|
"\t\t\t// by the (* use_dsp48 ... *) comment above.\n"
|
1813 |
|
|
"\t\t\tmpy_r <= w_one - w_two;\n"
|
1814 |
|
|
"\t\t\tmpy_i <= p_three - w_one - w_two;\n"
|
1815 |
|
|
"\t\tend\n"
|
1816 |
|
|
"\n");
|
1817 |
|
|
|
1818 |
|
|
fprintf(fp,
|
1819 |
|
|
"\t// Round the results\n"
|
1820 |
|
|
"\twire\tsigned\t[(OWIDTH-1):0]\trnd_left_r, rnd_left_i, rnd_right_r, rnd_right_i;\n\n");
|
1821 |
|
|
fprintf(fp,
|
1822 |
|
|
"\t%s #(CWIDTH+IWIDTH+1,OWIDTH,SHIFT+2) do_rnd_left_r(i_clk, i_ce,\n"
|
1823 |
|
|
"\t\t\t\tleft_sr, rnd_left_r);\n\n",
|
1824 |
|
|
rnd_string);
|
1825 |
|
|
fprintf(fp,
|
1826 |
|
|
"\t%s #(CWIDTH+IWIDTH+1,OWIDTH,SHIFT+2) do_rnd_left_i(i_clk, i_ce,\n"
|
1827 |
|
|
"\t\t\t\tleft_si, rnd_left_i);\n\n",
|
1828 |
|
|
rnd_string);
|
1829 |
|
|
fprintf(fp,
|
1830 |
|
|
"\t%s #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_right_r(i_clk, i_ce,\n"
|
1831 |
|
|
"\t\t\t\tmpy_r, rnd_right_r);\n\n", rnd_string);
|
1832 |
|
|
fprintf(fp,
|
1833 |
|
|
"\t%s #(CWIDTH+IWIDTH+3,OWIDTH,SHIFT+4) do_rnd_right_i(i_clk, i_ce,\n"
|
1834 |
|
|
"\t\t\t\tmpy_i, rnd_right_i);\n\n", rnd_string);
|
1835 |
|
|
|
1836 |
|
|
|
1837 |
|
|
fprintf(fp,
|
1838 |
|
|
"\t// As a final step, we pack our outputs into two packed two's\n"
|
1839 |
|
|
"\t// complement numbers per output word, so that each output word\n"
|
1840 |
|
|
"\t// has (2*OWIDTH) bits in it, with the top half being the real\n"
|
1841 |
|
|
"\t// portion and the bottom half being the imaginary portion.\n"
|
1842 |
|
|
"\tassign\to_left = { rnd_left_r, rnd_left_i };\n"
|
1843 |
|
|
"\tassign\to_right= { rnd_right_r,rnd_right_i};\n"
|
1844 |
|
|
"\n");
|
1845 |
|
|
|
1846 |
|
|
if (formal_property_flag) {
|
1847 |
|
|
fprintf(fp,
|
1848 |
|
|
"`ifdef FORMAL\n"
|
1849 |
|
|
"\tlocalparam F_LGDEPTH = 3;\n"
|
1850 |
|
|
"\tlocalparam F_DEPTH = 5;\n"
|
1851 |
|
|
"\tlocalparam [F_LGDEPTH-1:0] F_D = F_DEPTH-1;\n"
|
1852 |
|
|
"\n"
|
1853 |
|
|
"\treg signed [IWIDTH-1:0] f_dlyleft_r [0:F_DEPTH-1];\n"
|
1854 |
|
|
"\treg signed [IWIDTH-1:0] f_dlyleft_i [0:F_DEPTH-1];\n"
|
1855 |
|
|
"\treg signed [IWIDTH-1:0] f_dlyright_r [0:F_DEPTH-1];\n"
|
1856 |
|
|
"\treg signed [IWIDTH-1:0] f_dlyright_i [0:F_DEPTH-1];\n"
|
1857 |
|
|
"\treg signed [CWIDTH-1:0] f_dlycoeff_r [0:F_DEPTH-1];\n"
|
1858 |
|
|
"\treg signed [CWIDTH-1:0] f_dlycoeff_i [0:F_DEPTH-1];\n"
|
1859 |
|
|
"\treg signed [F_DEPTH-1:0] f_dlyaux;\n"
|
1860 |
|
|
"\n"
|
1861 |
|
|
"\talways @(posedge i_clk)\n"
|
1862 |
|
|
"\tif (i_reset)\n"
|
1863 |
|
|
"\t\tf_dlyaux <= 0;\n"
|
1864 |
|
|
"\telse if (i_ce)\n"
|
1865 |
|
|
"\t\tf_dlyaux <= { f_dlyaux[F_DEPTH-2:0], i_aux };\n"
|
1866 |
|
|
"\n"
|
1867 |
|
|
"\talways @(posedge i_clk)\n"
|
1868 |
|
|
"\tif (i_ce)\n"
|
1869 |
|
|
"\tbegin\n"
|
1870 |
|
|
"\t\tf_dlyleft_r[0] <= i_left[ (2*IWIDTH-1):IWIDTH];\n"
|
1871 |
|
|
"\t\tf_dlyleft_i[0] <= i_left[ ( IWIDTH-1):0];\n"
|
1872 |
|
|
"\t\tf_dlyright_r[0] <= i_right[(2*IWIDTH-1):IWIDTH];\n"
|
1873 |
|
|
"\t\tf_dlyright_i[0] <= i_right[( IWIDTH-1):0];\n"
|
1874 |
|
|
"\t\tf_dlycoeff_r[0] <= i_coef[ (2*CWIDTH-1):CWIDTH];\n"
|
1875 |
|
|
"\t\tf_dlycoeff_i[0] <= i_coef[ ( CWIDTH-1):0];\n"
|
1876 |
|
|
"\tend\n"
|
1877 |
|
|
"\n"
|
1878 |
|
|
"\tgenvar k;\n"
|
1879 |
|
|
"\tgenerate for(k=1; k<F_DEPTH; k=k+1)\n"
|
1880 |
|
|
"\n"
|
1881 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1882 |
|
|
"\t\tif (i_ce)\n"
|
1883 |
|
|
"\t\tbegin\n"
|
1884 |
|
|
"\t\t\tf_dlyleft_r[k] <= f_dlyleft_r[ k-1];\n"
|
1885 |
|
|
"\t\t\tf_dlyleft_i[k] <= f_dlyleft_i[ k-1];\n"
|
1886 |
|
|
"\t\t\tf_dlyright_r[k] <= f_dlyright_r[k-1];\n"
|
1887 |
|
|
"\t\t\tf_dlyright_i[k] <= f_dlyright_i[k-1];\n"
|
1888 |
|
|
"\t\t\tf_dlycoeff_r[k] <= f_dlycoeff_r[k-1];\n"
|
1889 |
|
|
"\t\t\tf_dlycoeff_i[k] <= f_dlycoeff_i[k-1];\n"
|
1890 |
|
|
"\t\tend\n"
|
1891 |
|
|
"\n"
|
1892 |
|
|
"\tendgenerate\n"
|
1893 |
|
|
"\n"
|
1894 |
|
|
"`ifdef VERILATOR"
|
1895 |
|
|
/*
|
1896 |
|
|
"\tgenerate if (CKPCE <= 1)\n"
|
1897 |
|
|
"\tbegin\n"
|
1898 |
|
|
"\n"
|
1899 |
|
|
"\t\t// i_ce is allowed to be anything in this mode\n"
|
1900 |
|
|
"\n"
|
1901 |
|
|
"\tend else if (CKPCE == 2)\n"
|
1902 |
|
|
"\tbegin : F_CKPCE_TWO\n"
|
1903 |
|
|
"\n"
|
1904 |
|
|
"\t\tassert property (@(posedge i_clk)\n"
|
1905 |
|
|
"\t\t i_ce |=> !i_ce);\n"
|
1906 |
|
|
"\n"
|
1907 |
|
|
"\tend else if (CKPCE == 3)\n"
|
1908 |
|
|
"\tbegin : F_CKPCE_THREE\n"
|
1909 |
|
|
"\n"
|
1910 |
|
|
"\t\tassert property (@(posedge i_clk)\n"
|
1911 |
|
|
"\t\t i_ce |=> !i_ce ##1 !i_ce);\n"
|
1912 |
|
|
"\n"
|
1913 |
|
|
"\tend endgenerate\n"
|
1914 |
|
|
*/
|
1915 |
|
|
"\n"
|
1916 |
|
|
"`else\n"
|
1917 |
|
|
"\talways @(posedge i_clk)\n"
|
1918 |
|
|
"\tif ((!$past(i_ce))&&(!$past(i_ce,2))&&(!$past(i_ce,3))\n"
|
1919 |
|
|
"\t\t\t&&(!$past(i_ce,4)))\n"
|
1920 |
|
|
"\t\tassume(i_ce);\n"
|
1921 |
|
|
"\n"
|
1922 |
|
|
"\tgenerate if (CKPCE <= 1)\n"
|
1923 |
|
|
"\tbegin\n"
|
1924 |
|
|
"\n"
|
1925 |
|
|
"\t\t// i_ce is allowed to be anything in this mode\n"
|
1926 |
|
|
"\n"
|
1927 |
|
|
"\tend else if (CKPCE == 2)\n"
|
1928 |
|
|
"\tbegin : F_CKPCE_TWO\n"
|
1929 |
|
|
"\n"
|
1930 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1931 |
|
|
"\t\t if ($past(i_ce))\n"
|
1932 |
|
|
"\t\t assume(!i_ce);\n"
|
1933 |
|
|
"\n"
|
1934 |
|
|
"\tend else if (CKPCE == 3)\n"
|
1935 |
|
|
"\tbegin : F_CKPCE_THREE\n"
|
1936 |
|
|
"\n"
|
1937 |
|
|
"\t\talways @(posedge i_clk)\n"
|
1938 |
|
|
"\t\t if (($past(i_ce))||($past(i_ce,2)))\n"
|
1939 |
|
|
"\t\t assume(!i_ce);\n"
|
1940 |
|
|
"\n"
|
1941 |
|
|
"\tend endgenerate\n"
|
1942 |
|
|
"`endif"
|
1943 |
|
|
"\n"
|
1944 |
|
|
"\treg [F_LGDEPTH-1:0] f_startup_counter;\n"
|
1945 |
|
|
"\tinitial f_startup_counter = 0;\n"
|
1946 |
|
|
"\talways @(posedge i_clk)\n"
|
1947 |
|
|
"\tif (i_reset)\n"
|
1948 |
|
|
"\t\tf_startup_counter <= 0;\n"
|
1949 |
|
|
"\telse if ((i_ce)&&(!(&f_startup_counter)))\n"
|
1950 |
|
|
"\t\tf_startup_counter <= f_startup_counter + 1;\n"
|
1951 |
|
|
"\n"
|
1952 |
|
|
"\twire signed [IWIDTH:0] f_sumr, f_sumi;\n"
|
1953 |
|
|
"\talways @(*)\n"
|
1954 |
|
|
"\tbegin\n"
|
1955 |
|
|
"\t\tf_sumr = f_dlyleft_r[F_D] + f_dlyright_r[F_D];\n"
|
1956 |
|
|
"\t\tf_sumi = f_dlyleft_i[F_D] + f_dlyright_i[F_D];\n"
|
1957 |
|
|
"\tend\n"
|
1958 |
|
|
"\n"
|
1959 |
|
|
"\twire signed [IWIDTH+CWIDTH:0] f_sumrx, f_sumix;\n"
|
1960 |
|
|
"\tassign f_sumrx = { {(2){f_sumr[IWIDTH]}}, f_sumr, {(CWIDTH-2){1'b0}} };\n"
|
1961 |
|
|
"\tassign f_sumix = { {(2){f_sumi[IWIDTH]}}, f_sumi, {(CWIDTH-2){1'b0}} };\n"
|
1962 |
|
|
"\n"
|
1963 |
|
|
"\twire signed [IWIDTH:0] f_difr, f_difi;\n"
|
1964 |
|
|
"\talways @(*)\n"
|
1965 |
|
|
"\tbegin\n"
|
1966 |
|
|
"\t\tf_difr = f_dlyleft_r[F_D] - f_dlyright_r[F_D];\n"
|
1967 |
|
|
"\t\tf_difi = f_dlyleft_i[F_D] - f_dlyright_i[F_D];\n"
|
1968 |
|
|
"\tend\n"
|
1969 |
|
|
"\n"
|
1970 |
|
|
"\twire signed [IWIDTH+CWIDTH+3-1:0] f_difrx, f_difix;\n"
|
1971 |
|
|
"\tassign f_difrx = { {(CWIDTH+2){f_difr[IWIDTH]}}, f_difr };\n"
|
1972 |
|
|
"\tassign f_difix = { {(CWIDTH+2){f_difi[IWIDTH]}}, f_difi };\n"
|
1973 |
|
|
"\n"
|
1974 |
|
|
"\twire signed [IWIDTH+CWIDTH+3-1:0] f_widecoeff_r, f_widecoeff_i;\n"
|
1975 |
|
|
"\tassign f_widecoeff_r = {{(IWIDTH+3){f_dlycoeff_r[F_D][CWIDTH-1]}},\n"
|
1976 |
|
|
"\t f_dlycoeff_r[F_D] };\n"
|
1977 |
|
|
"\tassign f_widecoeff_i = {{(IWIDTH+3){f_dlycoeff_i[F_D][CWIDTH-1]}},\n"
|
1978 |
|
|
"\t f_dlycoeff_i[F_D] };\n"
|
1979 |
|
|
"\n"
|
1980 |
|
|
"\talways @(posedge i_clk)\n"
|
1981 |
|
|
"\tif (f_startup_counter > F_D)\n"
|
1982 |
|
|
"\tbegin\n"
|
1983 |
|
|
"\t\tassert(left_sr == f_sumrx);\n"
|
1984 |
|
|
"\t\tassert(left_si == f_sumix);\n"
|
1985 |
|
|
"\t\tassert(aux_s == f_dlyaux[F_D]);\n"
|
1986 |
|
|
"\n"
|
1987 |
|
|
"\t\tif ((f_difr == 0)&&(f_difi == 0))\n"
|
1988 |
|
|
"\t\tbegin\n"
|
1989 |
|
|
"\t\t assert(mpy_r == 0);\n"
|
1990 |
|
|
"\t\t assert(mpy_i == 0);\n"
|
1991 |
|
|
"\t\tend else if ((f_dlycoeff_r[F_D] == 0)\n"
|
1992 |
|
|
"\t\t &&(f_dlycoeff_i[F_D] == 0))\n"
|
1993 |
|
|
"\t\tbegin\n"
|
1994 |
|
|
"\t assert(mpy_r == 0);\n"
|
1995 |
|
|
"\t\t assert(mpy_i == 0);\n"
|
1996 |
|
|
"\t\tend\n"
|
1997 |
|
|
"\n"
|
1998 |
|
|
"\t\tif ((f_dlycoeff_r[F_D] == 1)&&(f_dlycoeff_i[F_D] == 0))\n"
|
1999 |
|
|
"\t\tbegin\n"
|
2000 |
|
|
"\t\t assert(mpy_r == f_difrx);\n"
|
2001 |
|
|
"\t\t assert(mpy_i == f_difix);\n"
|
2002 |
|
|
"\t\tend\n"
|
2003 |
|
|
"\n"
|
2004 |
|
|
"\t\tif ((f_dlycoeff_r[F_D] == 0)&&(f_dlycoeff_i[F_D] == 1))\n"
|
2005 |
|
|
"\t\tbegin\n"
|
2006 |
|
|
"\t\t assert(mpy_r == -f_difix);\n"
|
2007 |
|
|
"\t\t assert(mpy_i == f_difrx);\n"
|
2008 |
|
|
"\t\tend\n"
|
2009 |
|
|
"\n"
|
2010 |
|
|
"\t\tif ((f_difr == 1)&&(f_difi == 0))\n"
|
2011 |
|
|
"\t\tbegin\n"
|
2012 |
|
|
"\t\t assert(mpy_r == f_widecoeff_r);\n"
|
2013 |
|
|
"\t\t assert(mpy_i == f_widecoeff_i);\n"
|
2014 |
|
|
"\t\tend\n"
|
2015 |
|
|
"\n"
|
2016 |
|
|
"\t\tif ((f_difr == 0)&&(f_difi == 1))\n"
|
2017 |
|
|
"\t\tbegin\n"
|
2018 |
|
|
"\t\t assert(mpy_r == -f_widecoeff_i);\n"
|
2019 |
|
|
"\t\t assert(mpy_i == f_widecoeff_r);\n"
|
2020 |
|
|
"\t\tend\n"
|
2021 |
|
|
"\tend\n"
|
2022 |
|
|
"\n");
|
2023 |
|
|
|
2024 |
|
|
fprintf(fp,
|
2025 |
|
|
"\t// Let's see if we can improve our performance at all by\n"
|
2026 |
|
|
"\t// moving our test one clock earlier. If nothing else, it should\n"
|
2027 |
|
|
"\t// help induction finish one (or more) clocks ealier than\n"
|
2028 |
|
|
"\t// otherwise\n"
|
2029 |
|
|
"\n\n"
|
2030 |
|
|
"\twire signed [IWIDTH:0] f_predifr, f_predifi;\n"
|
2031 |
|
|
"\talways @(*)\n"
|
2032 |
|
|
"\tbegin\n"
|
2033 |
|
|
"\t\tf_predifr = f_dlyleft_r[F_D-1] - f_dlyright_r[F_D-1];\n"
|
2034 |
|
|
"\t\tf_predifi = f_dlyleft_i[F_D-1] - f_dlyright_i[F_D-1];\n"
|
2035 |
|
|
"\tend\n"
|
2036 |
|
|
"\n"
|
2037 |
|
|
"\twire signed [IWIDTH+CWIDTH+1-1:0] f_predifrx, f_predifix;\n"
|
2038 |
|
|
"\tassign f_predifrx = { {(CWIDTH){f_predifr[IWIDTH]}}, f_predifr };\n"
|
2039 |
|
|
"\tassign f_predifix = { {(CWIDTH){f_predifi[IWIDTH]}}, f_predifi };\n"
|
2040 |
|
|
"\n"
|
2041 |
|
|
"\twire signed [CWIDTH:0] f_sumcoef;\n"
|
2042 |
|
|
"\twire signed [IWIDTH+1:0] f_sumdiff;\n"
|
2043 |
|
|
"\talways @(*)\n"
|
2044 |
|
|
"\tbegin\n"
|
2045 |
|
|
"\t\tf_sumcoef = f_dlycoeff_r[F_D-1] + f_dlycoeff_i[F_D-1];\n"
|
2046 |
|
|
"\t\tf_sumdiff = f_predifr + f_predifi;\n"
|
2047 |
|
|
"\tend\n"
|
2048 |
|
|
"\n"
|
2049 |
|
|
"\t// Induction helpers\n"
|
2050 |
|
|
"\talways @(posedge i_clk)\n"
|
2051 |
|
|
"\tif (f_startup_counter >= F_D)\n"
|
2052 |
|
|
"\tbegin\n"
|
2053 |
|
|
"\t\tif (f_dlycoeff_r[F_D-1] == 0)\n"
|
2054 |
|
|
"\t\t\tassert(p_one == 0);\n"
|
2055 |
|
|
"\t\tif (f_dlycoeff_i[F_D-1] == 0)\n"
|
2056 |
|
|
"\t\t\tassert(p_two == 0);\n"
|
2057 |
|
|
"\n"
|
2058 |
|
|
"\t\tif (f_dlycoeff_r[F_D-1] == 1)\n"
|
2059 |
|
|
"\t\t\tassert(p_one == f_predifrx);\n"
|
2060 |
|
|
"\t\tif (f_dlycoeff_i[F_D-1] == 1)\n"
|
2061 |
|
|
"\t\t\tassert(p_two == f_predifix);\n"
|
2062 |
|
|
"\n"
|
2063 |
|
|
"\t\tif (f_predifr == 0)\n"
|
2064 |
|
|
"\t\t\tassert(p_one == 0);\n"
|
2065 |
|
|
"\t\tif (f_predifi == 0)\n"
|
2066 |
|
|
"\t\t\tassert(p_two == 0);\n"
|
2067 |
|
|
"\n"
|
2068 |
|
|
"\t\t// verilator lint_off WIDTH\n"
|
2069 |
|
|
"\t\tif (f_predifr == 1)\n"
|
2070 |
|
|
"\t\t\tassert(p_one == f_dlycoeff_r[F_D-1]);\n"
|
2071 |
|
|
"\t\tif (f_predifi == 1)\n"
|
2072 |
|
|
"\t\t\tassert(p_two == f_dlycoeff_i[F_D-1]);\n"
|
2073 |
|
|
"\t\t// verilator lint_on WIDTH\n"
|
2074 |
|
|
"\n"
|
2075 |
|
|
"\t\tif (f_sumcoef == 0)\n"
|
2076 |
|
|
"\t\t\tassert(p_three == 0);\n"
|
2077 |
|
|
"\t\tif (f_sumdiff == 0)\n"
|
2078 |
|
|
"\t\t\tassert(p_three == 0);\n"
|
2079 |
|
|
"\t\t// verilator lint_off WIDTH\n"
|
2080 |
|
|
"\t\tif (f_sumcoef == 1)\n"
|
2081 |
|
|
"\t\t\tassert(p_three == f_sumdiff);\n"
|
2082 |
|
|
"\t\tif (f_sumdiff == 1)\n"
|
2083 |
|
|
"\t\t\tassert(p_three == f_sumcoef);\n"
|
2084 |
|
|
"\t\t// verilator lint_on WIDTH\n"
|
2085 |
|
|
"`ifdef VERILATOR\n"
|
2086 |
|
|
"\t\tassert(p_one == f_predifr * f_dlycoeff_r[F_D-1]);\n"
|
2087 |
|
|
"\t\tassert(p_two == f_predifi * f_dlycoeff_i[F_D-1]);\n"
|
2088 |
|
|
"\t\tassert(p_three == f_sumdiff * f_sumcoef);\n"
|
2089 |
|
|
"`endif // VERILATOR\n"
|
2090 |
|
|
"\tend\n\n"
|
2091 |
|
|
"`endif // FORMAL\n");
|
2092 |
|
|
}
|
2093 |
|
|
|
2094 |
|
|
fprintf(fp,
|
2095 |
|
|
"endmodule\n");
|
2096 |
|
|
|
2097 |
|
|
fclose(fp);
|
2098 |
|
|
}
|