1 |
2 |
dgisselq |
#include <stdio.h>
|
2 |
|
|
#include <stdlib.h>
|
3 |
|
|
#include <unistd.h>
|
4 |
|
|
#include <sys/stat.h>
|
5 |
|
|
#include <string.h>
|
6 |
|
|
#include <math.h>
|
7 |
|
|
#include <ctype.h>
|
8 |
|
|
#include <assert.h>
|
9 |
|
|
|
10 |
|
|
#define COREDIR "fft-core"
|
11 |
|
|
|
12 |
|
|
const char cpyleft[] =
|
13 |
|
|
"///////////////////////////////////////////////////////////////////////////\n"
|
14 |
|
|
"//\n"
|
15 |
|
|
"// Copyright (C) 2015, Gisselquist Technology, LLC\n"
|
16 |
|
|
"//\n"
|
17 |
|
|
"// This program is free software (firmware): you can redistribute it and/or\n"
|
18 |
|
|
"// modify it under the terms of the GNU General Public License as published\n"
|
19 |
|
|
"// by the Free Software Foundation, either version 3 of the License, or (at\n"
|
20 |
|
|
"// your option) any later version.\n"
|
21 |
|
|
"//\n"
|
22 |
|
|
"// This program is distributed in the hope that it will be useful, but WITHOUT\n"
|
23 |
|
|
"// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or\n"
|
24 |
|
|
"// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License\n"
|
25 |
|
|
"// for more details.\n"
|
26 |
|
|
"//\n"
|
27 |
|
|
"// You should have received a copy of the GNU General Public License along\n"
|
28 |
5 |
dgisselq |
"// with this program. (It's in the $(ROOT)/doc directory, run make with no\n"
|
29 |
|
|
"// target there if the PDF file isn\'t present.) If not, see\n"
|
30 |
|
|
"// <http://www.gnu.org/licenses/> for a copy.\n"
|
31 |
|
|
"//\n"
|
32 |
2 |
dgisselq |
"// License: GPL, v3, as defined and found on www.gnu.org,\n"
|
33 |
|
|
"// http://www.gnu.org/licenses/gpl.html\n"
|
34 |
|
|
"//\n"
|
35 |
|
|
"//\n"
|
36 |
|
|
"///////////////////////////////////////////////////////////////////////////\n";
|
37 |
|
|
const char prjname[] = "A Doubletime Pipelined FFT\n";
|
38 |
|
|
const char creator[] = "// Creator: Dan Gisselquist, Ph.D.\n"
|
39 |
|
|
"// Gisselquist Tecnology, LLC\n";
|
40 |
|
|
|
41 |
|
|
int lgval(int vl) {
|
42 |
|
|
int lg;
|
43 |
|
|
|
44 |
|
|
for(lg=1; (1<<lg) < vl; lg++)
|
45 |
|
|
;
|
46 |
|
|
return lg;
|
47 |
|
|
}
|
48 |
|
|
|
49 |
|
|
int nextlg(int vl) {
|
50 |
|
|
int r;
|
51 |
|
|
|
52 |
|
|
for(r=1; r<vl; r<<=1)
|
53 |
|
|
;
|
54 |
|
|
return r;
|
55 |
|
|
}
|
56 |
|
|
|
57 |
|
|
int lgdelay(int nbits, int xtra) {
|
58 |
|
|
int cbits = nbits + xtra;
|
59 |
|
|
int delay = nbits + 2;
|
60 |
|
|
if (nbits+1<cbits)
|
61 |
5 |
dgisselq |
delay = nbits+4;
|
62 |
2 |
dgisselq |
else
|
63 |
5 |
dgisselq |
delay = cbits+3;
|
64 |
2 |
dgisselq |
return lgval(delay);
|
65 |
|
|
}
|
66 |
|
|
|
67 |
|
|
void build_quarters(const char *fname) {
|
68 |
|
|
FILE *fp = fopen(fname, "w");
|
69 |
|
|
if (NULL == fp) {
|
70 |
|
|
fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
|
71 |
|
|
perror("O/S Err was:");
|
72 |
|
|
return;
|
73 |
|
|
}
|
74 |
|
|
|
75 |
|
|
fprintf(fp,
|
76 |
|
|
"///////////////////////////////////////////////////////////////////////////\n"
|
77 |
|
|
"//\n"
|
78 |
|
|
"// Filename: qtrstage.v\n"
|
79 |
|
|
"// \n"
|
80 |
|
|
"// Project: %s\n"
|
81 |
|
|
"//\n"
|
82 |
5 |
dgisselq |
"// Purpose: This file encapsulates the 4 point stage of a decimation in\n"
|
83 |
|
|
"// frequency FFT. This particular implementation is optimized\n"
|
84 |
|
|
"// so that all of the multiplies are accomplished by additions\n"
|
85 |
|
|
"// and multiplexers only.\n"
|
86 |
|
|
"//\n"
|
87 |
2 |
dgisselq |
"//\n%s"
|
88 |
|
|
"//\n",
|
89 |
|
|
prjname, creator);
|
90 |
|
|
fprintf(fp, "%s", cpyleft);
|
91 |
|
|
|
92 |
|
|
fprintf(fp,
|
93 |
|
|
"module\tqtrstage(i_clk, i_rst, i_ce, i_sync, i_data, o_data, o_sync);\n"
|
94 |
5 |
dgisselq |
"\tparameter IWIDTH=16, OWIDTH=IWIDTH+1;\n"
|
95 |
|
|
"\t// Parameters specific to the core that should be changed when this\n"
|
96 |
|
|
"\t// core is built ... Note that the minimum LGSPAN is 2. Smaller \n"
|
97 |
|
|
"\t// spans must use the fftdoubles stage.\n"
|
98 |
|
|
"\tparameter\tLGWIDTH=8, ODD=0, INVERSE=0,SHIFT=0;\n"
|
99 |
|
|
"\tinput\t i_clk, i_rst, i_ce, i_sync;\n"
|
100 |
|
|
"\tinput\t [(2*IWIDTH-1):0] i_data;\n"
|
101 |
|
|
"\toutput\treg [(2*OWIDTH-1):0] o_data;\n"
|
102 |
|
|
"\toutput\treg o_sync;\n"
|
103 |
|
|
"\t\n"
|
104 |
|
|
"\treg\t wait_for_sync;\n"
|
105 |
|
|
"\treg\t[2:0] pipeline;\n"
|
106 |
2 |
dgisselq |
"\n"
|
107 |
5 |
dgisselq |
"\treg\t[(IWIDTH):0] sum_r, sum_i, diff_r, diff_i;\n"
|
108 |
|
|
"\twire\t[(IWIDTH):0] n_diff_i;\n"
|
109 |
|
|
"\tassign n_diff_i = -diff_i;\n"
|
110 |
2 |
dgisselq |
"\n"
|
111 |
5 |
dgisselq |
"\treg\t[(2*OWIDTH-1):0] ob_a;\n"
|
112 |
|
|
"\twire\t[(2*OWIDTH-1):0] ob_b;\n"
|
113 |
|
|
"\treg\t[(OWIDTH-1):0] ob_b_r, ob_b_i;\n"
|
114 |
|
|
"\tassign ob_b = { ob_b_r, ob_b_i };\n"
|
115 |
2 |
dgisselq |
"\n"
|
116 |
5 |
dgisselq |
"\treg\t[(LGWIDTH-1):0] iaddr;\n"
|
117 |
|
|
"\treg\t[(2*IWIDTH-1):0] imem;\n"
|
118 |
2 |
dgisselq |
"\n"
|
119 |
5 |
dgisselq |
"\twire\tsigned\t[(IWIDTH-1):0]\timem_r, imem_i;\n"
|
120 |
|
|
"\tassign\timem_r = imem[(2*IWIDTH-1):(IWIDTH)];\n"
|
121 |
|
|
"\tassign\timem_i = imem[(IWIDTH-1):0];\n"
|
122 |
2 |
dgisselq |
"\n"
|
123 |
5 |
dgisselq |
"\twire\tsigned\t[(IWIDTH-1):0]\ti_data_r, i_data_i;\n"
|
124 |
|
|
"\tassign\ti_data_r = i_data[(2*IWIDTH-1):(IWIDTH)];\n"
|
125 |
|
|
"\tassign\ti_data_i = i_data[(IWIDTH-1):0];\n"
|
126 |
2 |
dgisselq |
"\n"
|
127 |
5 |
dgisselq |
"\treg [(2*OWIDTH-1):0] omem;\n"
|
128 |
2 |
dgisselq |
"\n"
|
129 |
5 |
dgisselq |
"\twire [(IWIDTH-1):0] rnd;\n"
|
130 |
|
|
"\tassign rnd = ((IWIDTH+1-OWIDTH-SHIFT)!=0) ? { {(IWIDTH-1){1'b0}}, (OWIDTH<IWIDTH+1)? 1'b1:1'b0 } : {{(IWIDTH){1'b0}}};\n"
|
131 |
2 |
dgisselq |
"\n"
|
132 |
5 |
dgisselq |
"\talways @(posedge i_clk)\n"
|
133 |
|
|
"\t\tif (i_rst)\n"
|
134 |
|
|
"\t\tbegin\n"
|
135 |
|
|
"\t\t\twait_for_sync <= 1'b1;\n"
|
136 |
|
|
"\t\t\tiaddr <= 0;\n"
|
137 |
|
|
"\t\t\tpipeline <= 3'b000;\n"
|
138 |
|
|
"\t\tend\n"
|
139 |
|
|
"\t\telse if ((i_ce)&&((~wait_for_sync)||(i_sync)))\n"
|
140 |
|
|
"\t\tbegin\n"
|
141 |
|
|
"\t\t\t// Always\n"
|
142 |
|
|
"\t\t\timem <= i_data;\n"
|
143 |
|
|
"\t\t\tiaddr <= iaddr + 1;\n"
|
144 |
|
|
"\t\t\twait_for_sync <= 1'b0;\n"
|
145 |
2 |
dgisselq |
"\n"
|
146 |
5 |
dgisselq |
"\t\t\t// In sequence, clock = 0\n"
|
147 |
|
|
"\t\t\tif (iaddr[0])\n"
|
148 |
|
|
"\t\t\tbegin\n"
|
149 |
|
|
"\t\t\t\tsum_r <= imem_r + i_data_r + rnd;\n"
|
150 |
|
|
"\t\t\t\tsum_i <= imem_i + i_data_i + rnd;\n"
|
151 |
|
|
"\t\t\t\tdiff_r <= imem_r - i_data_r + rnd;\n"
|
152 |
|
|
"\t\t\t\tdiff_i <= imem_i - i_data_i + rnd;\n"
|
153 |
2 |
dgisselq |
"\n"
|
154 |
5 |
dgisselq |
"\t\t\t\tpipeline[2:0] <= { pipeline[1:0], 1'b1 };\n"
|
155 |
|
|
"\t\t\tend else\n"
|
156 |
|
|
"\t\t\t\tpipeline[2:0] <= { pipeline[1:0], 1'b0 };\n"
|
157 |
2 |
dgisselq |
"\n"
|
158 |
5 |
dgisselq |
"\t\t\t// In sequence, clock = 1\n"
|
159 |
|
|
"\t\t\tif (pipeline[1])\n"
|
160 |
|
|
"\t\t\tbegin\n"
|
161 |
|
|
"\t\t\t\tob_a <= { sum_r[(IWIDTH):(IWIDTH+1-OWIDTH)],\n"
|
162 |
|
|
"\t\t\t\t\t\tsum_i[(IWIDTH):(IWIDTH+1-OWIDTH)] };\n"
|
163 |
|
|
"\t\t\t\t// on Even, W = e^{-j2pi 1/4 0} = 1\n"
|
164 |
|
|
"\t\t\t\tif (~ODD)\n"
|
165 |
|
|
"\t\t\t\tbegin\n"
|
166 |
2 |
dgisselq |
"\t\t\t\t\tob_b_r <= diff_r[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
|
167 |
|
|
"\t\t\t\t\tob_b_i <= diff_i[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
|
168 |
5 |
dgisselq |
"\t\t\t\tend else if (~INVERSE) begin\n"
|
169 |
2 |
dgisselq |
"\t\t\t\t\t// on Odd, W = e^{-j2pi 1/4} = -j\n"
|
170 |
|
|
"\t\t\t\t\tob_b_r <= diff_i[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
|
171 |
|
|
"\t\t\t\t\tob_b_i <= diff_r[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
|
172 |
5 |
dgisselq |
"\t\t\t\tend else begin\n"
|
173 |
2 |
dgisselq |
"\t\t\t\t\t// on Odd, W = e^{j2pi 1/4} = j\n"
|
174 |
|
|
"\t\t\t\t\tob_b_r <= n_diff_i[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
|
175 |
|
|
"\t\t\t\t\tob_b_i <= diff_r[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
|
176 |
5 |
dgisselq |
"\t\t\t\tend\n"
|
177 |
|
|
"\t\t\t\t// (wire) ob_b <= { ob_b_r, ob_b_i };\n"
|
178 |
|
|
"\t\t\tend\n"
|
179 |
|
|
"\t\t\t// In sequence, clock = 2\n"
|
180 |
|
|
"\t\t\tif (pipeline[2])\n"
|
181 |
|
|
"\t\t\tbegin\n"
|
182 |
|
|
"\t\t\t\tomem <= ob_b;\n"
|
183 |
|
|
"\t\t\t\to_data <= ob_a;\n"
|
184 |
|
|
"\t\t\tend else\n"
|
185 |
|
|
"\t\t\t\to_data <= omem;\n"
|
186 |
6 |
dgisselq |
"\t\t\t// Don\'t forget in the sync check that we are running\n"
|
187 |
|
|
"\t\t\t// at two clocks per sample. Thus we need to\n"
|
188 |
|
|
"\t\t\t// produce a sync every 2^(LGWIDTH-1) clocks.\n"
|
189 |
|
|
"\t\t\to_sync <= &(~iaddr[(LGWIDTH-2):3]) && (iaddr[2:0] == 3'b100);\n"
|
190 |
5 |
dgisselq |
"\t\tend\n"
|
191 |
2 |
dgisselq |
"endmodule\n");
|
192 |
|
|
}
|
193 |
|
|
|
194 |
|
|
void build_dblstage(const char *fname) {
|
195 |
|
|
FILE *fp = fopen(fname, "w");
|
196 |
|
|
if (NULL == fp) {
|
197 |
|
|
fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
|
198 |
|
|
perror("O/S Err was:");
|
199 |
|
|
return;
|
200 |
|
|
}
|
201 |
|
|
|
202 |
|
|
fprintf(fp,
|
203 |
|
|
"///////////////////////////////////////////////////////////////////////////\n"
|
204 |
|
|
"//\n"
|
205 |
|
|
"// Filename: dblstage.v\n"
|
206 |
|
|
"//\n"
|
207 |
|
|
"// Project: %s\n"
|
208 |
|
|
"//\n"
|
209 |
|
|
"// Purpose: This is part of an FPGA implementation that will process\n"
|
210 |
5 |
dgisselq |
"// the final stage of a decimate-in-frequency FFT, running\n"
|
211 |
|
|
"// through the data at two samples per clock. If you notice\n"
|
212 |
|
|
"// from the derivation of an FFT, the only time both even and\n"
|
213 |
|
|
"// odd samples are used at the same time is in this stage.\n"
|
214 |
|
|
"// Therefore, other than this stage and these twiddles, all of\n"
|
215 |
|
|
"// the other stages can run two stages at a time at one sample\n"
|
216 |
|
|
"// per clock.\n"
|
217 |
2 |
dgisselq |
"//\n"
|
218 |
|
|
"// In this implementation, the output is valid one clock after\n"
|
219 |
|
|
"// the input is valid. The output also accumulates one bit\n"
|
220 |
|
|
"// above and beyond the number of bits in the input.\n"
|
221 |
|
|
"// \n"
|
222 |
|
|
"// i_clk A system clock\n"
|
223 |
6 |
dgisselq |
"// i_rst A synchronous reset\n"
|
224 |
2 |
dgisselq |
"// i_ce Circuit enable--nothing happens unless this line is high\n"
|
225 |
6 |
dgisselq |
"// i_sync A synchronization signal, high once per FFT at the start\n"
|
226 |
2 |
dgisselq |
"// i_left The first (even) complex sample input. The higher order\n"
|
227 |
|
|
"// bits contain the real portion, low order bits the\n"
|
228 |
|
|
"// imaginary portion, all in two\'s complement.\n"
|
229 |
|
|
"// i_right The next (odd) complex sample input, same format as\n"
|
230 |
|
|
"// i_left.\n"
|
231 |
|
|
"// o_left The first (even) complex output.\n"
|
232 |
|
|
"// o_right The next (odd) complex output.\n"
|
233 |
6 |
dgisselq |
"// o_sync Output synchronization signal.\n"
|
234 |
2 |
dgisselq |
"//\n%s"
|
235 |
|
|
"//\n", prjname, creator);
|
236 |
|
|
|
237 |
|
|
fprintf(fp, "%s", cpyleft);
|
238 |
|
|
fprintf(fp,
|
239 |
6 |
dgisselq |
"module dblstage(i_clk, i_rst, i_ce, i_sync, i_left, i_right, o_left, o_right, o_sync);\n"
|
240 |
5 |
dgisselq |
"\tparameter\tIWIDTH=16,OWIDTH=IWIDTH+1, SHIFT=0;\n"
|
241 |
6 |
dgisselq |
"\tinput\t\ti_clk, i_rst, i_ce, i_sync;\n"
|
242 |
5 |
dgisselq |
"\tinput\t\t[(2*IWIDTH-1):0]\ti_left, i_right;\n"
|
243 |
6 |
dgisselq |
"\toutput\twire\t[(2*OWIDTH-1):0]\to_left, o_right;\n"
|
244 |
|
|
"\toutput\treg\t\t\to_sync;\n"
|
245 |
2 |
dgisselq |
"\n"
|
246 |
5 |
dgisselq |
"\twire\tsigned\t[(IWIDTH-1):0]\ti_in_0r, i_in_0i, i_in_1r, i_in_1i;\n"
|
247 |
|
|
"\tassign\ti_in_0r = i_left[(2*IWIDTH-1):(IWIDTH)]; \n"
|
248 |
|
|
"\tassign\ti_in_0i = i_left[(IWIDTH-1):0]; \n"
|
249 |
|
|
"\tassign\ti_in_1r = i_right[(2*IWIDTH-1):(IWIDTH)]; \n"
|
250 |
|
|
"\tassign\ti_in_1i = i_right[(IWIDTH-1):0]; \n"
|
251 |
|
|
"\twire\t[(OWIDTH-1):0]\t\to_out_0r, o_out_0i,\n"
|
252 |
|
|
"\t\t\t\t\to_out_1r, o_out_1i;\n"
|
253 |
2 |
dgisselq |
"\n"
|
254 |
5 |
dgisselq |
"\t// Don't forget that we accumulate a bit by adding two values\n"
|
255 |
|
|
"\t// together. Therefore our intermediate value must have one more\n"
|
256 |
|
|
"\t// bit than the two originals.\n"
|
257 |
|
|
"\treg\t[IWIDTH:0]\tout_0r, out_0i, out_1r, out_1i;\n"
|
258 |
2 |
dgisselq |
"\n"
|
259 |
6 |
dgisselq |
"\treg\twait_for_sync;\n"
|
260 |
|
|
"\n"
|
261 |
5 |
dgisselq |
"\talways @(posedge i_clk)\n"
|
262 |
6 |
dgisselq |
"\t\tif (i_rst)\n"
|
263 |
|
|
"\t\t\twait_for_sync <= 1'b1;\n"
|
264 |
|
|
"\t\telse if ((i_ce)&&((~wait_for_sync)||(i_sync)))\n"
|
265 |
5 |
dgisselq |
"\t\tbegin\n"
|
266 |
6 |
dgisselq |
"\t\t\twait_for_sync <= 1'b0;\n"
|
267 |
|
|
"\t\t\t//\n"
|
268 |
5 |
dgisselq |
"\t\t\tout_0r <= i_in_0r + i_in_1r;\n"
|
269 |
|
|
"\t\t\tout_0i <= i_in_0i + i_in_1i;\n"
|
270 |
|
|
"\t\t\t//\n"
|
271 |
|
|
"\t\t\tout_1r <= i_in_0r - i_in_1r;\n"
|
272 |
|
|
"\t\t\tout_1i <= i_in_0i - i_in_1i;\n"
|
273 |
6 |
dgisselq |
"\t\t\t//\n"
|
274 |
|
|
"\t\t\to_sync <= i_sync;\n"
|
275 |
5 |
dgisselq |
"\t\tend\n"
|
276 |
2 |
dgisselq |
"\n"
|
277 |
5 |
dgisselq |
"\t// Now, if the master control program doesn't want to keep all of\n"
|
278 |
|
|
"\t// our bits, we can shift down to OWIDTH bits here.\n"
|
279 |
|
|
"\tassign\to_out_0r = out_0r[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
|
280 |
|
|
"\tassign\to_out_0i = out_0i[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
|
281 |
|
|
"\tassign\to_out_1r = out_1r[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
|
282 |
|
|
"\tassign\to_out_1i = out_1i[(IWIDTH-SHIFT):(IWIDTH+1-OWIDTH-SHIFT)];\n"
|
283 |
2 |
dgisselq |
"\n"
|
284 |
5 |
dgisselq |
"\tassign\to_left = { o_out_0r, o_out_0i };\n"
|
285 |
|
|
"\tassign\to_right = { o_out_1r, o_out_1i };\n"
|
286 |
2 |
dgisselq |
"\n"
|
287 |
|
|
"endmodule\n");
|
288 |
|
|
fclose(fp);
|
289 |
|
|
}
|
290 |
|
|
|
291 |
|
|
void build_multiply(const char *fname) {
|
292 |
|
|
FILE *fp = fopen(fname, "w");
|
293 |
|
|
if (NULL == fp) {
|
294 |
|
|
fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
|
295 |
|
|
perror("O/S Err was:");
|
296 |
|
|
return;
|
297 |
|
|
}
|
298 |
|
|
|
299 |
|
|
fprintf(fp,
|
300 |
|
|
"///////////////////////////////////////////////////////////////////////////\n"
|
301 |
|
|
"//\n"
|
302 |
|
|
"// Filename: shiftaddmpy.v\n"
|
303 |
|
|
"//\n"
|
304 |
|
|
"// Project: %s\n"
|
305 |
|
|
"//\n"
|
306 |
|
|
"// Purpose: A portable shift and add multiply.\n"
|
307 |
|
|
"//\n"
|
308 |
|
|
"// While both Xilinx and Altera will offer single clock \n"
|
309 |
|
|
"// multiplies, this simple approach will multiply two numbers\n"
|
310 |
|
|
"// on any architecture. The result maintains the full width\n"
|
311 |
|
|
"// of the multiply, there are no extra stuff bits, no rounding,\n"
|
312 |
|
|
"// no shifted bits, etc.\n"
|
313 |
|
|
"//\n"
|
314 |
|
|
"// Further, for those applications that can support it, this\n"
|
315 |
|
|
"// multiply is pipelined and will produce one answer per clock.\n"
|
316 |
|
|
"//\n"
|
317 |
|
|
"// For minimal processing delay, make the first parameter\n"
|
318 |
|
|
"// the one with the least bits, so that AWIDTH <= BWIDTH.\n"
|
319 |
|
|
"//\n"
|
320 |
|
|
"// The processing delay in this multiply is (AWIDTH+1) cycles.\n"
|
321 |
|
|
"// That is, if the data is present on the input at clock t=0,\n"
|
322 |
|
|
"// the result will be present on the output at time t=AWIDTH+1;\n"
|
323 |
|
|
"//\n"
|
324 |
|
|
"//\n%s"
|
325 |
|
|
"//\n", prjname, creator);
|
326 |
|
|
|
327 |
|
|
fprintf(fp, "%s", cpyleft);
|
328 |
|
|
fprintf(fp,
|
329 |
|
|
"module shiftaddmpy(i_clk, i_ce, i_a, i_b, o_r);\n"
|
330 |
|
|
"\tparameter\tAWIDTH=16,BWIDTH=AWIDTH;\n"
|
331 |
|
|
"\tinput\t\t\t\t\ti_clk, i_ce;\n"
|
332 |
|
|
"\tinput\t\t[(AWIDTH-1):0]\t\ti_a;\n"
|
333 |
|
|
"\tinput\t\t[(BWIDTH-1):0]\t\ti_b;\n"
|
334 |
|
|
"\toutput\treg\t[(AWIDTH+BWIDTH-1):0]\to_r;\n"
|
335 |
|
|
"\n"
|
336 |
|
|
"\treg\t[(AWIDTH-1):0]\tu_a;\n"
|
337 |
|
|
"\treg\t[(BWIDTH-1):0]\tu_b;\n"
|
338 |
|
|
"\treg\t\t\tsgn;\n"
|
339 |
|
|
"\n"
|
340 |
|
|
"\treg\t[(AWIDTH-2):0]\t\tr_a[0:(AWIDTH-1)];\n"
|
341 |
|
|
"\treg\t[(AWIDTH+BWIDTH-2):0]\tr_b[0:(AWIDTH-1)];\n"
|
342 |
|
|
"\treg\t\t\t\tr_s[0:(AWIDTH-1)];\n"
|
343 |
|
|
"\treg\t[(AWIDTH+BWIDTH-1):0]\tacc[0:(AWIDTH-1)];\n"
|
344 |
|
|
"\tgenvar k;\n"
|
345 |
|
|
"\n"
|
346 |
5 |
dgisselq |
"\t// If we were forced to stay within two\'s complement arithmetic,\n"
|
347 |
|
|
"\t// taking the absolute value here would require an additional bit.\n"
|
348 |
|
|
"\t// However, because our results are now unsigned, we can stay\n"
|
349 |
|
|
"\t// within the number of bits given (for now).\n"
|
350 |
2 |
dgisselq |
"\talways @(posedge i_clk)\n"
|
351 |
|
|
"\t\tif (i_ce)\n"
|
352 |
|
|
"\t\tbegin\n"
|
353 |
|
|
"\t\t\tu_a <= (i_a[AWIDTH-1])?(-i_a):(i_a);\n"
|
354 |
|
|
"\t\t\tu_b <= (i_b[BWIDTH-1])?(-i_b):(i_b);\n"
|
355 |
|
|
"\t\t\tsgn <= i_a[AWIDTH-1] ^ i_b[BWIDTH-1];\n"
|
356 |
|
|
"\t\tend\n"
|
357 |
|
|
"\n"
|
358 |
|
|
"\talways @(posedge i_clk)\n"
|
359 |
|
|
"\t\tif (i_ce)\n"
|
360 |
|
|
"\t\tbegin\n"
|
361 |
|
|
"\t\t\tacc[0] <= (u_a[0]) ? { {(AWIDTH){1'b0}}, u_b }\n"
|
362 |
|
|
"\t\t\t\t\t: {(AWIDTH+BWIDTH){1'b0}};\n"
|
363 |
|
|
"\t\t\tr_a[0] <= { u_a[(AWIDTH-1):1] };\n"
|
364 |
|
|
"\t\t\tr_b[0] <= { {(AWIDTH-1){1'b0}}, u_b };\n"
|
365 |
|
|
"\t\t\tr_s[0] <= sgn; // The final sign, needs to be preserved\n"
|
366 |
|
|
"\t\tend\n"
|
367 |
|
|
"\n"
|
368 |
|
|
"\tgenerate\n"
|
369 |
|
|
"\talways @(posedge i_clk)\n"
|
370 |
|
|
"\tif (i_ce)\n"
|
371 |
|
|
"\tbegin\n"
|
372 |
|
|
"\t\tfor(k=0; k<AWIDTH-1; k++)\n"
|
373 |
|
|
"\t\tbegin\n"
|
374 |
|
|
"\t\t\tacc[k+1] <= acc[k] + ((r_a[k][0]) ? {r_b[k],1'b0}:0);\n"
|
375 |
|
|
"\t\t\tr_a[k+1] <= { 1'b0, r_a[k][(AWIDTH-2):1] };\n"
|
376 |
|
|
"\t\t\tr_b[k+1] <= { r_b[k][(AWIDTH+BWIDTH-3):0], 1'b0};\n"
|
377 |
|
|
"\t\t\tr_s[k+1] <= r_s[k];\n"
|
378 |
|
|
"\t\tend\n"
|
379 |
|
|
"\tend\n"
|
380 |
|
|
"\tendgenerate\n"
|
381 |
|
|
"\n"
|
382 |
|
|
"\talways @(posedge i_clk)\n"
|
383 |
|
|
"\t\tif (i_ce)\n"
|
384 |
|
|
"\t\t\to_r <= (r_s[AWIDTH-1]) ? (-acc[AWIDTH-1]) : acc[AWIDTH-1];\n"
|
385 |
|
|
"\n"
|
386 |
|
|
"endmodule\n");
|
387 |
|
|
|
388 |
|
|
fclose(fp);
|
389 |
|
|
}
|
390 |
|
|
|
391 |
|
|
void build_dblreverse(const char *fname) {
|
392 |
|
|
FILE *fp = fopen(fname, "w");
|
393 |
|
|
if (NULL == fp) {
|
394 |
|
|
fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
|
395 |
|
|
perror("O/S Err was:");
|
396 |
|
|
return;
|
397 |
|
|
}
|
398 |
|
|
|
399 |
|
|
fprintf(fp,
|
400 |
|
|
"///////////////////////////////////////////////////////////////////////////\n"
|
401 |
|
|
"//\n"
|
402 |
|
|
"// Filename: dblreverse.v\n"
|
403 |
|
|
"//\n"
|
404 |
|
|
"// Project: %s\n"
|
405 |
|
|
"//\n"
|
406 |
|
|
"// Purpose: This module bitreverses a pipelined FFT input. Operation is\n"
|
407 |
|
|
"// expected as follows:\n"
|
408 |
|
|
"//\n"
|
409 |
|
|
"// i_clk A running clock at whatever system speed is offered.\n"
|
410 |
|
|
"// i_rst A synchronous reset signal, that resets all internals\n"
|
411 |
|
|
"// i_ce If this is one, one input is consumed and an output\n"
|
412 |
|
|
"// is produced.\n"
|
413 |
|
|
"// i_in_0, i_in_1\n"
|
414 |
|
|
"// Two inputs to be consumed, each of width WIDTH.\n"
|
415 |
|
|
"// o_out_0, o_out_1\n"
|
416 |
|
|
"// Two of the bitreversed outputs, also of the same\n"
|
417 |
|
|
"// width, WIDTH. Of course, there is a delay from the\n"
|
418 |
|
|
"// first input to the first output. For this purpose,\n"
|
419 |
|
|
"// o_sync is present.\n"
|
420 |
|
|
"// o_sync This will be a 1'b1 for the first value in any block.\n"
|
421 |
|
|
"// Following a reset, this will only become 1'b1 once\n"
|
422 |
|
|
"// the data has been loaded and is now valid. After that,\n"
|
423 |
|
|
"// all outputs will be valid.\n"
|
424 |
|
|
"//\n%s"
|
425 |
|
|
"//\n", prjname, creator);
|
426 |
|
|
fprintf(fp, "%s", cpyleft);
|
427 |
|
|
fprintf(fp,
|
428 |
|
|
"\n\n"
|
429 |
|
|
"//\n"
|
430 |
|
|
"// How do we do bit reversing at two smples per clock? Can we separate out\n"
|
431 |
|
|
"// our work into eight memory banks, writing two banks at once and reading\n"
|
432 |
|
|
"// another two banks in the same clock?\n"
|
433 |
|
|
"//\n"
|
434 |
|
|
"// mem[00xxx0] = s_0[n]\n"
|
435 |
|
|
"// mem[00xxx1] = s_1[n]\n"
|
436 |
|
|
"// o_0[n] = mem[10xxx0]\n"
|
437 |
|
|
"// o_1[n] = mem[11xxx0]\n"
|
438 |
|
|
"// ...\n"
|
439 |
|
|
"// mem[01xxx0] = s_0[m]\n"
|
440 |
|
|
"// mem[01xxx1] = s_1[m]\n"
|
441 |
|
|
"// o_0[m] = mem[10xxx1]\n"
|
442 |
|
|
"// o_1[m] = mem[11xxx1]\n"
|
443 |
|
|
"// ...\n"
|
444 |
|
|
"// mem[10xxx0] = s_0[n]\n"
|
445 |
|
|
"// mem[10xxx1] = s_1[n]\n"
|
446 |
|
|
"// o_0[n] = mem[00xxx0]\n"
|
447 |
|
|
"// o_1[n] = mem[01xxx0]\n"
|
448 |
|
|
"// ...\n"
|
449 |
|
|
"// mem[11xxx0] = s_0[m]\n"
|
450 |
|
|
"// mem[11xxx1] = s_1[m]\n"
|
451 |
|
|
"// o_0[m] = mem[00xxx1]\n"
|
452 |
|
|
"// o_1[m] = mem[01xxx1]\n"
|
453 |
|
|
"// ...\n"
|
454 |
|
|
"//\n"
|
455 |
5 |
dgisselq |
"// The answer is that, yes we can but: we need to use four memory banks\n"
|
456 |
|
|
"// to do it properly. These four banks are defined by the two bits\n"
|
457 |
|
|
"// that determine the top and bottom of the correct address. Larger\n"
|
458 |
|
|
"// FFT\'s would require more memories.\n"
|
459 |
|
|
"//\n"
|
460 |
2 |
dgisselq |
"//\n");
|
461 |
|
|
fprintf(fp,
|
462 |
|
|
"module dblreverse(i_clk, i_rst, i_ce, i_in_0, i_in_1,\n"
|
463 |
5 |
dgisselq |
"\t\to_out_0, o_out_1, o_sync);\n"
|
464 |
|
|
"\tparameter\t\t\tLGSIZE=4, WIDTH=24;\n"
|
465 |
|
|
"\tinput\t\t\t\ti_clk, i_rst, i_ce;\n"
|
466 |
|
|
"\tinput\t\t[(2*WIDTH-1):0]\ti_in_0, i_in_1;\n"
|
467 |
|
|
"\toutput\treg\t[(2*WIDTH-1):0]\to_out_0, o_out_1;\n"
|
468 |
|
|
"\toutput\treg\t\t\to_sync;\n"
|
469 |
2 |
dgisselq |
"\n"
|
470 |
5 |
dgisselq |
"\treg\tin_reset;\n"
|
471 |
|
|
"\treg\t[(LGSIZE):0]\tiaddr;\n"
|
472 |
|
|
"\treg\t[(2*WIDTH-1):0]\tmem_0e [0:((1<<(LGSIZE-1))-1)];\n"
|
473 |
|
|
"\treg\t[(2*WIDTH-1):0]\tmem_0o [0:((1<<(LGSIZE-1))-1)];\n"
|
474 |
|
|
"\treg\t[(2*WIDTH-1):0]\tmem_1e [0:((1<<(LGSIZE-1))-1)];\n"
|
475 |
|
|
"\treg\t[(2*WIDTH-1):0]\tmem_1o [0:((1<<(LGSIZE-1))-1)];\n"
|
476 |
2 |
dgisselq |
"\n"
|
477 |
5 |
dgisselq |
"\twire\t[(2*LGSIZE-1):0] braddr;\n"
|
478 |
|
|
"\tgenvar\tk;\n"
|
479 |
|
|
"\tgenerate for(k=0; k<LGSIZE; k++)\n"
|
480 |
|
|
"\t\tassign braddr[k] = iaddr[LGSIZE-1-k];\n"
|
481 |
|
|
"\tendgenerate\n"
|
482 |
2 |
dgisselq |
"\n"
|
483 |
5 |
dgisselq |
"\talways @(posedge i_clk)\n"
|
484 |
|
|
"\t\tif (i_rst)\n"
|
485 |
|
|
"\t\tbegin\n"
|
486 |
|
|
"\t\t\tiaddr <= 0;\n"
|
487 |
|
|
"\t\t\tin_reset <= 1'b1;\n"
|
488 |
|
|
"\t\tend else if (i_ce)\n"
|
489 |
|
|
"\t\tbegin\n"
|
490 |
|
|
"\t\t\tif (iaddr[(LGSIZE-1)])\n"
|
491 |
|
|
"\t\t\tbegin\n"
|
492 |
|
|
"\t\t\t\tmem_1e[{iaddr[LGSIZE],iaddr[(LGSIZE-2):1]}] <= i_in_0;\n"
|
493 |
|
|
"\t\t\t\tmem_1o[{iaddr[LGSIZE],iaddr[(LGSIZE-2):1]}] <= i_in_1;\n"
|
494 |
|
|
"\t\t\tend else begin\n"
|
495 |
|
|
"\t\t\t\tmem_0e[{iaddr[LGSIZE],iaddr[(LGSIZE-2):1]}] <= i_in_0;\n"
|
496 |
|
|
"\t\t\t\tmem_0o[{iaddr[LGSIZE],iaddr[(LGSIZE-2):1]}] <= i_in_1;\n"
|
497 |
|
|
"\t\t\tend\n"
|
498 |
|
|
"\t\t\tiaddr <= iaddr + 2;\n"
|
499 |
|
|
"\t\t\tif (&iaddr[(LGSIZE-1):1])\n"
|
500 |
|
|
"\t\t\t\tin_reset <= 1'b0;\n"
|
501 |
|
|
"\t\t\tif (in_reset)\n"
|
502 |
|
|
"\t\t\tbegin\n"
|
503 |
|
|
"\t\t\t\to_out_0 <= {(2*WIDTH){1'b0}};\n"
|
504 |
|
|
"\t\t\t\to_out_1 <= {(2*WIDTH){1'b0}};\n"
|
505 |
|
|
"\t\t\t\to_sync <= 1'b0;\n"
|
506 |
|
|
"\t\t\tend else\n"
|
507 |
|
|
"\t\t\tbegin\n"
|
508 |
|
|
"\t\t\t\tif (braddr[0])\n"
|
509 |
|
|
"\t\t\t\tbegin\n"
|
510 |
2 |
dgisselq |
"\t\t\t\t\to_out_0 <= mem_0o[{~iaddr[LGSIZE],braddr[(LGSIZE-2):1]}];\n"
|
511 |
|
|
"\t\t\t\t\to_out_1 <= mem_1o[{~iaddr[LGSIZE],braddr[(LGSIZE-2):1]}];\n"
|
512 |
5 |
dgisselq |
"\t\t\t\tend else begin\n"
|
513 |
2 |
dgisselq |
"\t\t\t\t\to_out_0 <= mem_0e[{~iaddr[LGSIZE],braddr[(LGSIZE-2):1]}];\n"
|
514 |
|
|
"\t\t\t\t\to_out_1 <= mem_1e[{~iaddr[LGSIZE],braddr[(LGSIZE-2):1]}];\n"
|
515 |
5 |
dgisselq |
"\t\t\t\tend\n"
|
516 |
|
|
"\t\t\t\to_sync <= ~(|iaddr[(LGSIZE-1):0]);\n"
|
517 |
|
|
"\t\t\tend\n"
|
518 |
|
|
"\t\tend\n"
|
519 |
2 |
dgisselq |
"\n"
|
520 |
|
|
"endmodule;\n");
|
521 |
|
|
|
522 |
|
|
fclose(fp);
|
523 |
|
|
}
|
524 |
|
|
|
525 |
|
|
void build_butterfly(const char *fname) {
|
526 |
|
|
FILE *fp = fopen(fname, "w");
|
527 |
|
|
if (NULL == fp) {
|
528 |
|
|
fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
|
529 |
|
|
perror("O/S Err was:");
|
530 |
|
|
return;
|
531 |
|
|
}
|
532 |
|
|
|
533 |
|
|
fprintf(fp,
|
534 |
|
|
"///////////////////////////////////////////////////////////////////////////\n"
|
535 |
|
|
"//\n"
|
536 |
|
|
"// Filename: butterfly.v\n"
|
537 |
|
|
"//\n"
|
538 |
|
|
"// Project: %s\n"
|
539 |
|
|
"//\n"
|
540 |
|
|
"// Purpose: This routine caculates a butterfly for a decimation\n"
|
541 |
|
|
"// in frequency version of an FFT. Specifically, given\n"
|
542 |
|
|
"// complex Left and Right values together with a \n"
|
543 |
|
|
"// coefficient, the output of this routine is given\n"
|
544 |
|
|
"// by:\n"
|
545 |
|
|
"//\n"
|
546 |
|
|
"// L' = L + R\n"
|
547 |
|
|
"// R' = (L - R)*C\n"
|
548 |
|
|
"//\n"
|
549 |
|
|
"// The rest of the junk below handles timing (mostly),\n"
|
550 |
|
|
"// to make certain that L' and R' reach the output at\n"
|
551 |
|
|
"// the same clock. Further, just to make certain\n"
|
552 |
|
|
"// that is the case, an 'aux' input exists. This\n"
|
553 |
|
|
"// aux value will come out of this routine synchronized\n"
|
554 |
|
|
"// to the values it came in with. (i.e., both L', R',\n"
|
555 |
|
|
"// and aux all have the same delay.) Hence, a caller\n"
|
556 |
|
|
"// of this routine may set aux on the first input with\n"
|
557 |
|
|
"// valid data, and then wait to see aux set on the output\n"
|
558 |
|
|
"// to know when to find the first output with valid data.\n"
|
559 |
|
|
"//\n"
|
560 |
|
|
"// All bits are preserved until the very last clock,\n"
|
561 |
|
|
"// where any more bits than OWIDTH will be quietly\n"
|
562 |
|
|
"// discarded.\n"
|
563 |
|
|
"//\n"
|
564 |
|
|
"// This design features no overflow checking.\n"
|
565 |
|
|
"// \n"
|
566 |
|
|
"// Notes:\n"
|
567 |
|
|
"// CORDIC:\n"
|
568 |
|
|
"// Much as we would like, we can't use a cordic here.\n"
|
569 |
|
|
"// The goal is to accomplish an FFT, as defined, and a\n"
|
570 |
|
|
"// CORDIC places a scale factor onto the data. Removing\n"
|
571 |
|
|
"// the scale factor would cost a two multiplies, which\n"
|
572 |
|
|
"// is precisely what we are trying to avoid.\n"
|
573 |
|
|
"//\n"
|
574 |
|
|
"//\n"
|
575 |
|
|
"// 3-MULTIPLIES:\n"
|
576 |
|
|
"// It should also be possible to do this with three \n"
|
577 |
|
|
"// multiplies and an extra two addition cycles. \n"
|
578 |
|
|
"//\n"
|
579 |
|
|
"// We want\n"
|
580 |
|
|
"// R+I = (a + jb) * (c + jd)\n"
|
581 |
|
|
"// R+I = (ac-bd) + j(ad+bc)\n"
|
582 |
|
|
"// We multiply\n"
|
583 |
|
|
"// P1 = ac\n"
|
584 |
|
|
"// P2 = bd\n"
|
585 |
|
|
"// P3 = (a+b)(c+d)\n"
|
586 |
|
|
"// Then \n"
|
587 |
|
|
"// R+I=(P1-P2)+j(P3-P2-P1)\n"
|
588 |
|
|
"//\n"
|
589 |
|
|
"// WIDTHS:\n"
|
590 |
|
|
"// On multiplying an X width number by an\n"
|
591 |
|
|
"// Y width number, X>Y, the result should be (X+Y)\n"
|
592 |
|
|
"// bits, right?\n"
|
593 |
|
|
"// -2^(X-1) <= a <= 2^(X-1) - 1\n"
|
594 |
|
|
"// -2^(Y-1) <= b <= 2^(Y-1) - 1\n"
|
595 |
|
|
"// (2^(Y-1)-1)*(-2^(X-1)) <= ab <= 2^(X-1)2^(Y-1)\n"
|
596 |
|
|
"// -2^(X+Y-2)+2^(X-1) <= ab <= 2^(X+Y-2) <= 2^(X+Y-1) - 1\n"
|
597 |
|
|
"// -2^(X+Y-1) <= ab <= 2^(X+Y-1)-1\n"
|
598 |
|
|
"// YUP! But just barely. Do this and you'll really want\n"
|
599 |
|
|
"// to drop a bit, although you will risk overflow in so\n"
|
600 |
|
|
"// doing.\n"
|
601 |
|
|
"//\n%s"
|
602 |
|
|
"//\n", prjname, creator);
|
603 |
|
|
fprintf(fp, "%s", cpyleft);
|
604 |
|
|
|
605 |
|
|
fprintf(fp,
|
606 |
6 |
dgisselq |
"module\tbutterfly(i_clk, i_rst, i_ce, i_coef, i_left, i_right, i_aux,\n"
|
607 |
5 |
dgisselq |
"\t\to_left, o_right, o_aux);\n"
|
608 |
|
|
"\t// Public changeable parameters ...\n"
|
609 |
|
|
"\tparameter IWIDTH=16,CWIDTH=IWIDTH+4,OWIDTH=IWIDTH+1;\n"
|
610 |
|
|
"\t// Parameters specific to the core that should not be changed.\n"
|
611 |
|
|
"\tparameter MPYDELAY=5'd20, // (IWIDTH+1 < CWIDTH)?(IWIDTH+4):(CWIDTH+3),\n"
|
612 |
|
|
"\t\t\tSHIFT=0, ROUND=0;\n"
|
613 |
|
|
"\t// The LGDELAY should be the base two log of the MPYDELAY. If\n"
|
614 |
|
|
"\t// this value is fractional, then round up to the nearest\n"
|
615 |
|
|
"\t// integer: LGDELAY=ceil(log(MPYDELAY)/log(2));\n"
|
616 |
|
|
"\tparameter\tLGDELAY=5;\n"
|
617 |
6 |
dgisselq |
"\tinput\t\ti_clk, i_rst, i_ce;\n"
|
618 |
5 |
dgisselq |
"\tinput\t\t[(2*CWIDTH-1):0] i_coef;\n"
|
619 |
|
|
"\tinput\t\t[(2*IWIDTH-1):0] i_left, i_right;\n"
|
620 |
|
|
"\tinput\t\ti_aux;\n"
|
621 |
|
|
"\toutput\twire [(2*OWIDTH-1):0] o_left, o_right;\n"
|
622 |
|
|
"\toutput\twire o_aux;\n"
|
623 |
2 |
dgisselq |
"\n"
|
624 |
5 |
dgisselq |
"\twire\t[(OWIDTH-1):0] o_left_r, o_left_i, o_right_r, o_right_i;\n"
|
625 |
2 |
dgisselq |
"\n"
|
626 |
5 |
dgisselq |
"\treg\t[(2*IWIDTH-1):0]\tr_left, r_right;\n"
|
627 |
|
|
"\treg\t\t\t\tr_aux, r_aux_2;\n"
|
628 |
|
|
"\treg\t[(2*CWIDTH-1):0]\tr_coef, r_coef_2;\n"
|
629 |
|
|
"\twire\tsigned\t[(CWIDTH-1):0]\tr_coef_r, r_coef_i;\n"
|
630 |
|
|
"\tassign\tr_coef_r = r_coef_2[ (2*CWIDTH-1):(CWIDTH)];\n"
|
631 |
|
|
"\tassign\tr_coef_i = r_coef_2[ ( CWIDTH-1):0];\n"
|
632 |
|
|
"\twire\tsigned\t[(IWIDTH-1):0]\tr_left_r, r_left_i, r_right_r, r_right_i;\n"
|
633 |
|
|
"\tassign\tr_left_r = r_left[ (2*IWIDTH-1):(IWIDTH)];\n"
|
634 |
|
|
"\tassign\tr_left_i = r_left[ (IWIDTH-1):0];\n"
|
635 |
|
|
"\tassign\tr_right_r = r_right[(2*IWIDTH-1):(IWIDTH)];\n"
|
636 |
|
|
"\tassign\tr_right_i = r_right[(IWIDTH-1):0];\n"
|
637 |
2 |
dgisselq |
"\n"
|
638 |
5 |
dgisselq |
"\treg\tsigned\t[(IWIDTH):0]\tr_sum_r, r_sum_i, r_dif_r, r_dif_i;\n"
|
639 |
2 |
dgisselq |
"\n"
|
640 |
5 |
dgisselq |
"\treg [(LGDELAY-1):0] fifo_addr;\n"
|
641 |
|
|
"\twire [(LGDELAY-1):0] fifo_read_addr;\n"
|
642 |
6 |
dgisselq |
"\tassign\tfifo_read_addr = fifo_addr - MPYDELAY;\n"
|
643 |
5 |
dgisselq |
"\treg [(2*IWIDTH+2):0] fifo_left [ 0:((1<<LGDELAY)-1)];\n"
|
644 |
6 |
dgisselq |
"\treg\t\t\t\tovalid;\n"
|
645 |
5 |
dgisselq |
"\n");
|
646 |
|
|
fprintf(fp,
|
647 |
|
|
"\t// Set up the input to the multiply\n"
|
648 |
2 |
dgisselq |
"\talways @(posedge i_clk)\n"
|
649 |
|
|
"\t\tif (i_ce)\n"
|
650 |
|
|
"\t\tbegin\n"
|
651 |
|
|
"\t\t\t// One clock just latches the inputs\n"
|
652 |
|
|
"\t\t\tr_left <= i_left; // No change in # of bits\n"
|
653 |
|
|
"\t\t\tr_right <= i_right;\n"
|
654 |
|
|
"\t\t\tr_aux <= i_aux;\n"
|
655 |
|
|
"\t\t\tr_coef <= i_coef;\n"
|
656 |
|
|
"\t\t\t// Next clock adds/subtracts\n"
|
657 |
|
|
"\t\t\tr_sum_r <= r_left_r + r_right_r; // Now IWIDTH+1 bits\n"
|
658 |
|
|
"\t\t\tr_sum_i <= r_left_i + r_right_i;\n"
|
659 |
|
|
"\t\t\tr_dif_r <= r_left_r - r_right_r;\n"
|
660 |
|
|
"\t\t\tr_dif_i <= r_left_i - r_right_i;\n"
|
661 |
|
|
"\t\t\t// Other inputs are simply delayed on second clock\n"
|
662 |
|
|
"\t\t\tr_aux_2 <= r_aux;\n"
|
663 |
|
|
"\t\t\tr_coef_2<= r_coef;\n"
|
664 |
|
|
"\t\tend\n"
|
665 |
5 |
dgisselq |
"\n");
|
666 |
|
|
fprintf(fp,
|
667 |
|
|
"\t// Don\'t forget to record the even side, since it doesn\'t need\n"
|
668 |
|
|
"\t// to be multiplied, but yet we still need the results in sync\n"
|
669 |
|
|
"\t// with the answer when it is ready.\n"
|
670 |
2 |
dgisselq |
"\talways @(posedge i_clk)\n"
|
671 |
6 |
dgisselq |
"\t\tif (i_rst)\n"
|
672 |
2 |
dgisselq |
"\t\tbegin\n"
|
673 |
6 |
dgisselq |
"\t\t\tfifo_addr <= 0;\n"
|
674 |
|
|
"\t\t\tovalid <= 1'b0;\n"
|
675 |
|
|
"\t\tend else if (i_ce)\n"
|
676 |
|
|
"\t\tbegin\n"
|
677 |
2 |
dgisselq |
"\t\t\t// Need to delay the sum side--nothing else happens\n"
|
678 |
|
|
"\t\t\t// to it, but it needs to stay synchronized with the\n"
|
679 |
|
|
"\t\t\t// right side.\n"
|
680 |
|
|
"\t\t\tfifo_left[fifo_addr] <= { r_aux_2, r_sum_r, r_sum_i };\n"
|
681 |
|
|
"\t\t\tfifo_addr <= fifo_addr + 1;\n"
|
682 |
6 |
dgisselq |
"\t\t\tovalid <= (ovalid) || (fifo_addr > MPYDELAY+1);\n"
|
683 |
2 |
dgisselq |
"\t\tend\n"
|
684 |
|
|
"\n"
|
685 |
5 |
dgisselq |
"\twire\tsigned\t[(CWIDTH-1):0] ir_coef_r, ir_coef_i;\n"
|
686 |
|
|
"\tassign\tir_coef_r = r_coef_2[(2*CWIDTH-1):CWIDTH];\n"
|
687 |
|
|
"\tassign\tir_coef_i = r_coef_2[(CWIDTH-1):0];\n"
|
688 |
|
|
"\twire\tsigned\t[((IWIDTH+2)+(CWIDTH+1)-1):0]\tp_one, p_two, p_three;\n"
|
689 |
2 |
dgisselq |
"\n"
|
690 |
5 |
dgisselq |
"\n");
|
691 |
|
|
fprintf(fp,
|
692 |
|
|
"\t// Multiply output is always a width of the sum of the widths of\n"
|
693 |
|
|
"\t// the two inputs. ALWAYS. This is independent of the number of\n"
|
694 |
|
|
"\t// bits in p_one, p_two, or p_three. These values needed to \n"
|
695 |
|
|
"\t// accumulate a bit (or two) each. However, this approach to a\n"
|
696 |
|
|
"\t// three multiply complex multiply cannot increase the total\n"
|
697 |
|
|
"\t// number of bits in our final output. We\'ll take care of\n"
|
698 |
|
|
"\t// dropping back down to the proper width, OWIDTH, in our routine\n"
|
699 |
|
|
"\t// below.\n"
|
700 |
2 |
dgisselq |
"\n"
|
701 |
5 |
dgisselq |
"\n");
|
702 |
|
|
fprintf(fp,
|
703 |
|
|
"\t// We accomplish here \"Karatsuba\" multiplication. That is,\n"
|
704 |
|
|
"\t// by doing three multiplies we accomplish the work of four.\n"
|
705 |
|
|
"\t// Let\'s prove to ourselves that this works ... We wish to\n"
|
706 |
|
|
"\t// multiply: (a+jb) * (c+jd), where a+jb is given by\n"
|
707 |
|
|
"\t//\ta + jb = r_dif_r + j r_dif_i, and\n"
|
708 |
|
|
"\t//\tc + jd = ir_coef_r + j ir_coef_i.\n"
|
709 |
|
|
"\t// We do this by calculating the intermediate products P1, P2,\n"
|
710 |
|
|
"\t// and P3 as\n"
|
711 |
|
|
"\t//\tP1 = ac\n"
|
712 |
|
|
"\t//\tP2 = bd\n"
|
713 |
|
|
"\t//\tP3 = (a + b) * (c + d)\n"
|
714 |
|
|
"\t// and then complete our final answer with\n"
|
715 |
|
|
"\t//\tac - bd = P1 - P2 (this checks)\n"
|
716 |
|
|
"\t//\tad + bc = P3 - P2 - P1\n"
|
717 |
|
|
"\t//\t = (ac + bc + ad + bd) - bd - ac\n"
|
718 |
|
|
"\t//\t = bc + ad (this checks)\n"
|
719 |
2 |
dgisselq |
"\n"
|
720 |
5 |
dgisselq |
"\n");
|
721 |
|
|
fprintf(fp,
|
722 |
|
|
"\t// This should really be based upon an IF, such as in\n"
|
723 |
|
|
"\t// if (IWIDTH < CWIDTH) then ...\n"
|
724 |
|
|
"\t// However, this is the only (other) way I know to do it.\n"
|
725 |
2 |
dgisselq |
"\tgenerate\n"
|
726 |
|
|
"\tif (CWIDTH < IWIDTH+1)\n"
|
727 |
|
|
"\tbegin\n"
|
728 |
|
|
"\t\t// We need to pad these first two multiplies by an extra\n"
|
729 |
5 |
dgisselq |
"\t\t// bit just to keep them aligned with the third,\n"
|
730 |
|
|
"\t\t// simpler, multiply.\n"
|
731 |
2 |
dgisselq |
"\t\tshiftaddmpy #(CWIDTH+1,IWIDTH+2) p1(i_clk, i_ce,\n"
|
732 |
|
|
"\t\t\t\t{ir_coef_r[CWIDTH-1],ir_coef_r},\n"
|
733 |
|
|
"\t\t\t\t{r_dif_r[IWIDTH],r_dif_r}, p_one);\n"
|
734 |
|
|
"\t\tshiftaddmpy #(CWIDTH+1,IWIDTH+2) p2(i_clk, i_ce,\n"
|
735 |
5 |
dgisselq |
"\t\t\t\t{ir_coef_i[CWIDTH-1],ir_coef_i},\n"
|
736 |
2 |
dgisselq |
"\t\t\t\t{r_dif_i[IWIDTH],r_dif_i}, p_two);\n"
|
737 |
|
|
"\t\tshiftaddmpy #(CWIDTH+1,IWIDTH+2) p3(i_clk, i_ce,\n"
|
738 |
5 |
dgisselq |
"\t\t\t\tir_coef_i+ir_coef_r,\n"
|
739 |
|
|
"\t\t\t\tr_dif_r + r_dif_i,\n"
|
740 |
|
|
"\t\t\t\tp_three);\n"
|
741 |
2 |
dgisselq |
"\tend else begin\n"
|
742 |
|
|
"\t\tshiftaddmpy #(IWIDTH+2,CWIDTH+1) p1a(i_clk, i_ce,\n"
|
743 |
|
|
"\t\t\t\t{r_dif_r[IWIDTH],r_dif_r},\n"
|
744 |
|
|
"\t\t\t\t{ir_coef_r[CWIDTH-1],ir_coef_r}, p_one);\n"
|
745 |
|
|
"\t\tshiftaddmpy #(IWIDTH+2,CWIDTH+1) p2a(i_clk, i_ce,\n"
|
746 |
|
|
"\t\t\t\t{r_dif_i[IWIDTH], r_dif_i},\n"
|
747 |
5 |
dgisselq |
"\t\t\t\t{ir_coef_i[CWIDTH-1],ir_coef_i}, p_two);\n"
|
748 |
2 |
dgisselq |
"\t\tshiftaddmpy #(IWIDTH+2,CWIDTH+1) p3a(i_clk, i_ce,\n"
|
749 |
|
|
"\t\t\t\tr_dif_r+r_dif_i,\n"
|
750 |
|
|
"\t\t\t\tir_coef_i+ir_coef_r,\n"
|
751 |
|
|
"\t\t\t\tp_three);\n"
|
752 |
|
|
"\tend\n"
|
753 |
|
|
"\tendgenerate\n"
|
754 |
5 |
dgisselq |
"\n");
|
755 |
|
|
fprintf(fp,
|
756 |
|
|
"\t// These values are held in memory and delayed during the\n"
|
757 |
|
|
"\t// multiply. Here, we recover them. During the multiply,\n"
|
758 |
|
|
"\t// values were multiplied by 2^(CWIDTH-2)*exp{-j*2*pi*...},\n"
|
759 |
|
|
"\t// therefore, the left_x values need to be right shifted by\n"
|
760 |
|
|
"\t// CWIDTH-2 as well. The additional bits come from a sign\n"
|
761 |
|
|
"\t// extension.\n"
|
762 |
2 |
dgisselq |
"\twire aux;\n"
|
763 |
5 |
dgisselq |
"\twire\tsigned\t[(IWIDTH+CWIDTH):0] fifo_i, fifo_r;\n"
|
764 |
|
|
"\treg\t\t[(2*IWIDTH+2):0] fifo_read;\n"
|
765 |
|
|
"\tassign\tfifo_r = { {2{fifo_read[2*(IWIDTH+1)-1]}}, fifo_read[(2*(IWIDTH+1)-1):(IWIDTH+1)], {(CWIDTH-2){1'b0}} };\n"
|
766 |
|
|
"\tassign\tfifo_i = { {2{fifo_read[(IWIDTH+1)-1]}}, fifo_read[((IWIDTH+1)-1):0], {(CWIDTH-2){1'b0}} };\n"
|
767 |
|
|
"\tassign\taux = fifo_read[2*IWIDTH+2];\n"
|
768 |
2 |
dgisselq |
"\n"
|
769 |
|
|
"\n"
|
770 |
5 |
dgisselq |
"\treg\tsigned\t[(CWIDTH+IWIDTH+3-1):0] b_left_r, b_left_i,\n"
|
771 |
|
|
"\t\t\t\t\t\tb_right_r, b_right_i;\n"
|
772 |
|
|
"\treg\tsigned\t[(CWIDTH+IWIDTH+3-1):0] mpy_r, mpy_i;\n"
|
773 |
|
|
"\treg\tsigned\t[(CWIDTH+IWIDTH+3-1):0] rnd;\n"
|
774 |
|
|
"\tgenerate\n"
|
775 |
|
|
"\tif ((~ROUND)||(CWIDTH+IWIDTH-OWIDTH-SHIFT<1))\n"
|
776 |
|
|
"\t\tassign rnd = ({(CWIDTH+IWIDTH+3){1'b0}});\n"
|
777 |
|
|
"\telse\n"
|
778 |
|
|
"\t\tassign rnd = ({ {(OWIDTH+3+SHIFT){1'b0}},1'b1,\n"
|
779 |
|
|
"\t\t\t\t{(CWIDTH+IWIDTH-OWIDTH-SHIFT-1){1'b0}} });\n"
|
780 |
|
|
"\tendgenerate\n"
|
781 |
|
|
"\n");
|
782 |
|
|
fprintf(fp,
|
783 |
2 |
dgisselq |
"\talways @(posedge i_clk)\n"
|
784 |
|
|
"\t\tif (i_ce)\n"
|
785 |
|
|
"\t\tbegin\n"
|
786 |
|
|
"\t\t\t// First clock, recover all values\n"
|
787 |
|
|
"\t\t\tfifo_read <= fifo_left[fifo_read_addr];\n"
|
788 |
|
|
"\t\t\t// These values are IWIDTH+CWIDTH+3 bits wide\n"
|
789 |
5 |
dgisselq |
"\t\t\t// although they only need to be (IWIDTH+1)\n"
|
790 |
|
|
"\t\t\t// + (CWIDTH) bits wide. (We\'ve got two\n"
|
791 |
|
|
"\t\t\t// extra bits we need to get rid of.)\n"
|
792 |
2 |
dgisselq |
"\t\t\tmpy_r <= p_one - p_two;\n"
|
793 |
|
|
"\t\t\tmpy_i <= p_three - p_one - p_two;\n"
|
794 |
|
|
"\n"
|
795 |
|
|
"\t\t\t// Second clock, round and latch for final clock\n"
|
796 |
|
|
"\t\t\tb_right_r <= mpy_r + rnd;\n"
|
797 |
|
|
"\t\t\tb_right_i <= mpy_i + rnd;\n"
|
798 |
5 |
dgisselq |
"\t\t\tb_left_r <= { {2{fifo_r[(IWIDTH+CWIDTH)]}},fifo_r } + rnd;\n"
|
799 |
|
|
"\t\t\tb_left_i <= { {2{fifo_i[(IWIDTH+CWIDTH)]}},fifo_i } + rnd;\n"
|
800 |
6 |
dgisselq |
"\t\t\to_aux <= aux & ovalid;\n"
|
801 |
2 |
dgisselq |
"\t\tend\n"
|
802 |
5 |
dgisselq |
"\n");
|
803 |
|
|
fprintf(fp,
|
804 |
2 |
dgisselq |
"\t// Final clock--clock and remove unnecessary bits.\n"
|
805 |
5 |
dgisselq |
"\t// We have (IWIDTH+CWIDTH+3) bits here, we need to drop down to\n"
|
806 |
|
|
"\t// OWIDTH, and SHIFT by SHIFT bits in the process. The trick is\n"
|
807 |
|
|
"\t// that we don\'t need (IWIDTH+CWIDTH+3) bits. We\'ve accumulated\n"
|
808 |
|
|
"\t// them, but the actual values will never fill all these bits.\n"
|
809 |
|
|
"\t// In particular, we only need:\n"
|
810 |
|
|
"\t//\t IWIDTH bits for the input\n"
|
811 |
|
|
"\t//\t +1 bit for the add/subtract\n"
|
812 |
|
|
"\t//\t+CWIDTH bits for the coefficient multiply\n"
|
813 |
|
|
"\t//\t +1 bit for the add/subtract in the complex multiply\n"
|
814 |
|
|
"\t//\t ------\n"
|
815 |
|
|
"\t//\t (IWIDTH+CWIDTH+2) bits at full precision.\n"
|
816 |
|
|
"\t//\n"
|
817 |
|
|
"\t// However, the coefficient multiply multiplied by a maximum value\n"
|
818 |
|
|
"\t// of 2^(CWIDTH-2). Thus, we only have\n"
|
819 |
|
|
"\t//\t IWIDTH bits for the input\n"
|
820 |
|
|
"\t//\t +1 bit for the add/subtract\n"
|
821 |
|
|
"\t//\t+CWIDTH-2 bits for the coefficient multiply\n"
|
822 |
|
|
"\t//\t +1 (optional) bit for the add/subtract in the cpx mpy.\n"
|
823 |
|
|
"\t//\t -------- ... multiply. (This last bit may be shifted out.)\n"
|
824 |
|
|
"\t//\t (IWIDTH+CWIDTH) valid output bits. \n"
|
825 |
|
|
"\t// Now, if the user wants to keep any extras of these (via OWIDTH),\n"
|
826 |
|
|
"\t// or if he wishes to arbitrarily shift some of these off (via\n"
|
827 |
|
|
"\t// SHIFT) we accomplish that here.\n"
|
828 |
|
|
"\tassign o_left_r = b_left_r[ (CWIDTH+IWIDTH-1-SHIFT-1):(CWIDTH+IWIDTH-OWIDTH-SHIFT-1)];\n"
|
829 |
|
|
"\tassign o_left_i = b_left_i[ (CWIDTH+IWIDTH-1-SHIFT-1):(CWIDTH+IWIDTH-OWIDTH-SHIFT-1)];\n"
|
830 |
|
|
"\tassign o_right_r = b_right_r[(CWIDTH+IWIDTH-1-SHIFT-1):(CWIDTH+IWIDTH-OWIDTH-SHIFT-1)];\n"
|
831 |
|
|
"\tassign o_right_i = b_right_i[(CWIDTH+IWIDTH-1-SHIFT-1):(CWIDTH+IWIDTH-OWIDTH-SHIFT-1)];\n"
|
832 |
2 |
dgisselq |
"\n"
|
833 |
5 |
dgisselq |
"\t// As a final step, we pack our outputs into two packed two\'s\n"
|
834 |
|
|
"\t// complement numbers per output word, so that each output word\n"
|
835 |
|
|
"\t// has (2*OWIDTH) bits in it, with the top half being the real\n"
|
836 |
|
|
"\t// portion and the bottom half being the imaginary portion.\n"
|
837 |
2 |
dgisselq |
"\tassign o_left = { o_left_r, o_left_i };\n"
|
838 |
|
|
"\tassign o_right= { o_right_r,o_right_i};\n"
|
839 |
|
|
"\n"
|
840 |
|
|
"endmodule\n");
|
841 |
|
|
fclose(fp);
|
842 |
|
|
}
|
843 |
|
|
|
844 |
|
|
void build_stage(const char *fname, int stage, bool odd, int nbits, bool inv, int xtra) {
|
845 |
|
|
FILE *fstage = fopen(fname, "w");
|
846 |
|
|
int cbits = nbits + xtra;
|
847 |
|
|
|
848 |
|
|
if ((cbits * 2) >= sizeof(long long)*8) {
|
849 |
|
|
fprintf(stderr, "ERROR: CMEM Coefficient precision requested overflows long long data type.\n");
|
850 |
|
|
exit(-1);
|
851 |
|
|
}
|
852 |
|
|
|
853 |
|
|
if (fstage == NULL) {
|
854 |
|
|
fprintf(stderr, "ERROR: Could not open %s for writing!\n", fname);
|
855 |
|
|
perror("O/S Err was:");
|
856 |
|
|
fprintf(stderr, "Attempting to continue, but this file will be missing.\n");
|
857 |
|
|
return;
|
858 |
|
|
}
|
859 |
|
|
|
860 |
|
|
fprintf(fstage,
|
861 |
|
|
"////////////////////////////////////////////////////////////////////////////\n"
|
862 |
|
|
"//\n"
|
863 |
|
|
"// Filename: %sfftstage_%c%d.v\n"
|
864 |
|
|
"//\n"
|
865 |
|
|
"// Project: %s\n"
|
866 |
|
|
"//\n"
|
867 |
|
|
"// Purpose: This file is (almost) a Verilog source file. It is meant to\n"
|
868 |
|
|
"// be used by a FFT core compiler to generate FFTs which may be\n"
|
869 |
|
|
"// used as part of an FFT core. Specifically, this file \n"
|
870 |
|
|
"// encapsulates the options of an FFT-stage. For any 2^N length\n"
|
871 |
|
|
"// FFT, there shall be (N-1) of these stages. \n"
|
872 |
|
|
"//\n%s"
|
873 |
|
|
"//\n",
|
874 |
|
|
(inv)?"i":"", (odd)?'o':'e', stage*2, prjname, creator);
|
875 |
|
|
fprintf(fstage, "%s", cpyleft);
|
876 |
|
|
fprintf(fstage, "module\t%sfftstage_%c%d(i_clk, i_rst, i_ce, i_sync, i_data, o_data, o_sync);\n",
|
877 |
|
|
(inv)?"i":"", (odd)?'o':'e', stage*2);
|
878 |
|
|
// These parameter values are useless at this point--they are to be
|
879 |
|
|
// replaced by the parameter values in the calling program. Only
|
880 |
|
|
// problem is, the CWIDTH needs to match exactly!
|
881 |
|
|
fprintf(fstage, "\tparameter\tIWIDTH=%d,CWIDTH=%d,OWIDTH=%d;\n",
|
882 |
|
|
nbits, cbits, nbits+1);
|
883 |
|
|
fprintf(fstage,
|
884 |
|
|
"\t// Parameters specific to the core that should be changed when this\n"
|
885 |
|
|
"\t// core is built ... Note that the minimum LGSPAN (the base two log\n"
|
886 |
|
|
"\t// of the span, or the base two log of the current FFT size) is 3.\n"
|
887 |
|
|
"\t// Smaller spans (i.e. the span of 2) must use the dblstage module.\n"
|
888 |
6 |
dgisselq |
"\tparameter\tLGWIDTH=11, LGSPAN=9, LGBDLY=5, BFLYSHIFT=0;\n");
|
889 |
2 |
dgisselq |
fprintf(fstage,
|
890 |
|
|
"\tinput i_clk, i_rst, i_ce, i_sync;\n"
|
891 |
|
|
"\tinput [(2*IWIDTH-1):0] i_data;\n"
|
892 |
|
|
"\toutput reg [(2*OWIDTH-1):0] o_data;\n"
|
893 |
|
|
"\toutput reg o_sync;\n"
|
894 |
|
|
"\n"
|
895 |
|
|
"\treg wait_for_sync;\n"
|
896 |
|
|
"\treg [(2*IWIDTH-1):0] ib_a, ib_b;\n"
|
897 |
|
|
"\treg [(2*CWIDTH-1):0] ib_c;\n"
|
898 |
|
|
"\treg ib_sync, b_ce;\n"
|
899 |
|
|
"\n"
|
900 |
|
|
"\treg b_started;\n"
|
901 |
|
|
"\twire ob_sync;\n"
|
902 |
|
|
"\twire [(2*OWIDTH-1):0] ob_a, ob_b;\n");
|
903 |
|
|
fprintf(fstage,
|
904 |
|
|
"\n"
|
905 |
|
|
"\t// %scmem is defined as an array of real and complex values,\n"
|
906 |
|
|
"\t// where the top CWIDTH bits are the real value and the bottom\n"
|
907 |
|
|
"\t// CWIDTH bits are the imaginary value.\n"
|
908 |
|
|
"\t//\n"
|
909 |
|
|
"\t// cmem[i] = { (2^(CWIDTH-2)) * cos(2*pi*i/(2^LGWIDTH)),\n"
|
910 |
|
|
"\t// (2^(CWIDTH-2)) * sin(2*pi*i/(2^LGWIDTH)) };\n"
|
911 |
|
|
"\t//\n"
|
912 |
|
|
"\treg [(2*CWIDTH-1):0] %scmem [0:((1<<LGSPAN)-1)];\n"
|
913 |
|
|
"\tinitial\t$readmemh(\"%scmem_%c%d.hex\",%scmem);\n\n",
|
914 |
|
|
(inv)?"i":"", (inv)?"i":"",
|
915 |
|
|
(inv)?"i":"", (odd)?'o':'e',stage<<1,
|
916 |
|
|
(inv)?"i":"");
|
917 |
|
|
{
|
918 |
|
|
FILE *cmem;
|
919 |
|
|
char memfile[128], *ptr;
|
920 |
|
|
|
921 |
|
|
strncpy(memfile, fname, 125);
|
922 |
|
|
if ((NULL != (ptr = strrchr(memfile, '/')))&&(ptr>memfile)) {
|
923 |
|
|
ptr++;
|
924 |
|
|
sprintf(ptr, "%scmem_%c%d.hex", (inv)?"i":"", (odd)?'o':'e', stage*2);
|
925 |
|
|
} else {
|
926 |
|
|
sprintf(memfile, "%s/%scmem_%c%d.hex",
|
927 |
|
|
COREDIR, (inv)?"i":"",
|
928 |
|
|
(odd)?'o':'e', stage*2);
|
929 |
|
|
}
|
930 |
|
|
// strcpy(&memfile[strlen(memfile)-2], ".hex");
|
931 |
|
|
cmem = fopen(memfile, "w");
|
932 |
|
|
// fprintf(cmem, "// CBITS = %d, inv = %s\n", cbits, (inv)?"true":"false");
|
933 |
|
|
for(int i=0; i<stage/2; i++) {
|
934 |
|
|
int k = 2*i+odd;
|
935 |
|
|
double W = ((inv)?1:-1)*2.0*M_PI*k/(double)stage;
|
936 |
|
|
double c, s;
|
937 |
|
|
long long ic, is, vl;
|
938 |
|
|
|
939 |
|
|
c = cos(W); s = sin(W);
|
940 |
|
|
ic = (long long)((double)((1ll<<(cbits-2)) * c + 0.5));
|
941 |
|
|
is = (long long)((double)((1ll<<(cbits-2)) * s + 0.5));
|
942 |
|
|
vl = (ic & (~(-1ll << (cbits))));
|
943 |
|
|
vl <<= (cbits);
|
944 |
|
|
vl |= (is & (~(-1ll << (cbits))));
|
945 |
|
|
fprintf(cmem, "%0*llx\n", ((cbits*2+3)/4), vl);
|
946 |
|
|
/*
|
947 |
|
|
fprintf(cmem, "%0*llx\t\t// %f+j%f -> %llx +j%llx\n",
|
948 |
|
|
((cbits*2+3)/4), vl, c, s,
|
949 |
|
|
ic & (~(-1ll<<(((cbits+3)/4)*4))),
|
950 |
|
|
is & (~(-1ll<<(((cbits+3)/4)*4))));
|
951 |
|
|
*/
|
952 |
|
|
} fclose(cmem);
|
953 |
|
|
}
|
954 |
|
|
|
955 |
|
|
fprintf(fstage,
|
956 |
6 |
dgisselq |
"\treg [(LGWIDTH-2):0] iaddr;\n"
|
957 |
2 |
dgisselq |
"\treg [(2*IWIDTH-1):0] imem [0:((1<<LGSPAN)-1)];\n"
|
958 |
|
|
"\n"
|
959 |
|
|
"\treg [(LGSPAN-1):0] oB;\n"
|
960 |
|
|
"\treg [(2*OWIDTH-1):0] omem [0:((1<<LGSPAN)-1)];\n"
|
961 |
|
|
"\n"
|
962 |
|
|
"\talways @(posedge i_clk)\n"
|
963 |
|
|
"\t\tif (i_rst)\n"
|
964 |
|
|
"\t\tbegin\n"
|
965 |
|
|
"\t\t\twait_for_sync <= 1'b1;\n"
|
966 |
|
|
"\t\t\tiaddr <= 0;\n"
|
967 |
|
|
"\t\t\toB <= 0;\n"
|
968 |
|
|
"\t\t\tb_ce <= 1'b0;\n"
|
969 |
|
|
"\t\tend\n"
|
970 |
|
|
"\t\telse if ((i_ce)&&((~wait_for_sync)||(i_sync)))\n"
|
971 |
|
|
"\t\tbegin\n"
|
972 |
|
|
"\t\t\t//\n"
|
973 |
|
|
"\t\t\t// First step: Record what we\'re not ready to use yet\n"
|
974 |
|
|
"\t\t\t//\n"
|
975 |
|
|
"\t\t\timem[iaddr[(LGSPAN-1):0]] <= i_data;\n"
|
976 |
|
|
"\t\t\tiaddr <= iaddr + 1;\n"
|
977 |
|
|
"\t\t\twait_for_sync <= 1'b0;\n"
|
978 |
|
|
"\n"
|
979 |
|
|
"\t\t\t//\n"
|
980 |
|
|
"\t\t\t// Now, we have all the inputs, so let\'s feed the\n"
|
981 |
|
|
"\t\t\t// butterfly\n"
|
982 |
|
|
"\t\t\t//\n"
|
983 |
6 |
dgisselq |
"\t\t\tif (iaddr[LGSPAN])\n"
|
984 |
2 |
dgisselq |
"\t\t\tbegin\n"
|
985 |
|
|
"\t\t\t\t// One input from memory, ...\n"
|
986 |
|
|
"\t\t\t\tib_a <= imem[iaddr[(LGSPAN-1):0]];\n"
|
987 |
|
|
"\t\t\t\t// One input clocked in from the top\n"
|
988 |
|
|
"\t\t\t\tib_b <= i_data;\n"
|
989 |
|
|
"\t\t\t\t// Set the sync to true on the very first\n"
|
990 |
|
|
"\t\t\t\t// valid input in, and hence on the very\n"
|
991 |
|
|
"\t\t\t\t// first valid data out per FFT.\n"
|
992 |
6 |
dgisselq |
"\t\t\t\tib_sync <= (iaddr==(1<<(LGSPAN)));\n"
|
993 |
2 |
dgisselq |
"\t\t\t\tib_c <= %scmem[iaddr[(LGSPAN-1):0]];\n"
|
994 |
|
|
"\t\t\t\tb_ce <= 1'b1;\n"
|
995 |
|
|
"\t\t\tend else\n"
|
996 |
|
|
"\t\t\t\tb_ce <= 1'b0;\n"
|
997 |
|
|
"\n"
|
998 |
|
|
"\t\t\t//\n"
|
999 |
|
|
"\t\t\t// Next step: recover the outputs from the butterfly\n"
|
1000 |
|
|
"\t\t\t//\n"
|
1001 |
|
|
"\t\t\tif ((ob_sync||b_started)&&(b_ce))\n"
|
1002 |
|
|
"\t\t\tbegin // A butterfly output is available\n"
|
1003 |
|
|
"\t\t\t\tb_started <= 1'b1;\n"
|
1004 |
|
|
"\t\t\t\tomem[oB] <= ob_b;\n"
|
1005 |
|
|
"\t\t\t\toB <= oB+1;\n"
|
1006 |
|
|
"\n"
|
1007 |
6 |
dgisselq |
"\t\t\t\to_sync <= (ob_sync);\n"
|
1008 |
2 |
dgisselq |
"\t\t\t\to_data <= ob_a;\n"
|
1009 |
|
|
"\t\t\tend else if (b_started)\n"
|
1010 |
|
|
"\t\t\tbegin // and keep outputting once you start--at a rate\n"
|
1011 |
|
|
"\t\t\t// of one guaranteed output per clock that has i_ce set.\n"
|
1012 |
|
|
"\t\t\t\to_data <= omem[oB];\n"
|
1013 |
|
|
"\t\t\t\toB <= oB + 1;\n"
|
1014 |
|
|
"\t\t\t\to_sync <= 1'b0;\n"
|
1015 |
|
|
"\t\t\tend else\n"
|
1016 |
|
|
"\t\t\t\to_sync <= 1'b0;\n"
|
1017 |
|
|
"\t\tend\n"
|
1018 |
5 |
dgisselq |
"\n", (inv)?"i":"");
|
1019 |
|
|
fprintf(fstage,
|
1020 |
2 |
dgisselq |
"\tbutterfly #(.IWIDTH(IWIDTH),.CWIDTH(CWIDTH),.OWIDTH(OWIDTH),\n"
|
1021 |
5 |
dgisselq |
"\t\t\t.MPYDELAY(%d\'d%d),.LGDELAY(LGBDLY),.SHIFT(BFLYSHIFT))\n"
|
1022 |
6 |
dgisselq |
"\t\tbfly(i_clk, i_rst, (b_ce&i_ce), ib_c,\n"
|
1023 |
2 |
dgisselq |
"\t\t\tib_a, ib_b, ib_sync, ob_a, ob_b, ob_sync);\n"
|
1024 |
|
|
"endmodule;\n",
|
1025 |
5 |
dgisselq |
lgdelay(nbits, xtra), (1<xtra)?(nbits+4):(nbits+xtra+3));
|
1026 |
2 |
dgisselq |
}
|
1027 |
|
|
|
1028 |
|
|
void usage(void) {
|
1029 |
|
|
fprintf(stderr,
|
1030 |
|
|
"USAGE:\tfftgen [-f <size>] [-d dir] [-c cbits] [-n nbits] [-m mxbits] [-s01]\n"
|
1031 |
|
|
// "\tfftgen -i\n"
|
1032 |
|
|
"\t-c <cbits>\tCauses all internal complex coefficients to be\n"
|
1033 |
|
|
"\t\tlonger than the corresponding data bits, to help avoid\n"
|
1034 |
|
|
"\t\tcoefficient truncation errors.\n"
|
1035 |
|
|
"\t-d <dir>\tPlaces all of the generated verilog files into <dir>.\n"
|
1036 |
|
|
"\t-f <size>\tSets the size of the FFT as the number of complex\n"
|
1037 |
|
|
"\t\tsamples input to the transform.\n"
|
1038 |
|
|
"\t-n <nbits>\tSets the number of bits in the twos complement input\n"
|
1039 |
|
|
"\t\tto the FFT routine.\n"
|
1040 |
|
|
"\t-m <mxbits>\tSets the maximum bit width that the FFT should ever\n"
|
1041 |
|
|
"\t\tproduce. Internal values greater than this value will be\n"
|
1042 |
|
|
"\t\ttruncated to this value.\n"
|
1043 |
|
|
"\t-s\tSkip the final bit reversal stage. This is useful in\n"
|
1044 |
|
|
"\t\talgorithms that need to apply a filter without needing to do\n"
|
1045 |
|
|
"\t\tbin shifting, as these algorithms can, with this option, just\n"
|
1046 |
|
|
"\t\tmultiply by a bit reversed correlation sequence and then\n"
|
1047 |
|
|
"\t\tinverse FFT the (still bit reversed) result.\n"
|
1048 |
|
|
"\t-S\tInclude the final bit reversal stage (default).\n"
|
1049 |
|
|
"\t-0\tA forward FFT (default), meaning that the coefficients are\n"
|
1050 |
|
|
"\t\tgiven by e^{-j 2 pi k/N n }.\n"
|
1051 |
|
|
"\t-1\tAn inverse FFT, meaning that the coefficients are\n"
|
1052 |
|
|
"\t\tgiven by e^{ j 2 pi k/N n }.\n");
|
1053 |
|
|
}
|
1054 |
|
|
|
1055 |
|
|
// Features still needed:
|
1056 |
|
|
// Interactivity.
|
1057 |
|
|
// Some number of maximum bits, beyond which we won't accumulate any more.
|
1058 |
|
|
// Obviously, the build_stage above.
|
1059 |
|
|
// Copying the files of interest into the fft-core directory, from
|
1060 |
|
|
// whatever directory this file is run out of.
|
1061 |
|
|
int main(int argc, char **argv) {
|
1062 |
|
|
int fftsize = -1, lgsize = -1;
|
1063 |
|
|
int nbitsin = 16, xtracbits = 4;
|
1064 |
|
|
int nbitsout, maxbitsout = -1;
|
1065 |
|
|
bool bitreverse = true, inverse=false, interactive = false,
|
1066 |
|
|
verbose_flag = false;
|
1067 |
|
|
FILE *vmain;
|
1068 |
|
|
char fname[128], coredir[1024] = "fft-core";
|
1069 |
|
|
|
1070 |
|
|
|
1071 |
|
|
if (argc <= 1)
|
1072 |
|
|
usage();
|
1073 |
|
|
|
1074 |
|
|
for(int argn=1; argn<argc; argn++) {
|
1075 |
|
|
if ('-' == argv[argn][0]) {
|
1076 |
|
|
for(int j=1; (argv[argn][j])&&(j<100); j++) {
|
1077 |
|
|
switch(argv[argn][j]) {
|
1078 |
|
|
case '0':
|
1079 |
|
|
inverse = false;
|
1080 |
|
|
break;
|
1081 |
|
|
case '1':
|
1082 |
|
|
inverse = true;
|
1083 |
|
|
break;
|
1084 |
|
|
case 'c':
|
1085 |
|
|
if (argn+1 >= argc) {
|
1086 |
|
|
printf("No extra number of coefficient bits given\n");
|
1087 |
|
|
usage(); exit(-1);
|
1088 |
|
|
}
|
1089 |
|
|
xtracbits = atoi(argv[++argn]);
|
1090 |
|
|
j+= 200;
|
1091 |
|
|
break;
|
1092 |
|
|
case 'd':
|
1093 |
|
|
if (argn+1 >= argc) {
|
1094 |
|
|
printf("No extra number of coefficient bits given\n");
|
1095 |
|
|
usage(); exit(-1);
|
1096 |
|
|
}
|
1097 |
|
|
strcpy(coredir, argv[++argn]);
|
1098 |
|
|
j += 200;
|
1099 |
|
|
break;
|
1100 |
|
|
case 'f':
|
1101 |
|
|
if (argn+1 >= argc) {
|
1102 |
|
|
printf("No FFT Size given\n");
|
1103 |
|
|
usage(); exit(-1);
|
1104 |
|
|
}
|
1105 |
|
|
fftsize = atoi(argv[++argn]);
|
1106 |
|
|
{ int sln = strlen(argv[argn]);
|
1107 |
|
|
if (!isdigit(argv[argn][sln-1])){
|
1108 |
|
|
switch(argv[argn][sln-1]) {
|
1109 |
|
|
case 'k': case 'K':
|
1110 |
|
|
fftsize <<= 10;
|
1111 |
|
|
break;
|
1112 |
|
|
case 'm': case 'M':
|
1113 |
|
|
fftsize <<= 20;
|
1114 |
|
|
break;
|
1115 |
|
|
case 'g': case 'G':
|
1116 |
|
|
fftsize <<= 30;
|
1117 |
|
|
break;
|
1118 |
|
|
default:
|
1119 |
|
|
printf("Unknown FFT size, %s\n", argv[argn]);
|
1120 |
|
|
exit(-1);
|
1121 |
|
|
}
|
1122 |
|
|
}}
|
1123 |
|
|
j += 200;
|
1124 |
|
|
break;
|
1125 |
|
|
case 'h':
|
1126 |
|
|
usage();
|
1127 |
|
|
exit(0);
|
1128 |
|
|
break;
|
1129 |
|
|
case 'i':
|
1130 |
|
|
interactive = true;
|
1131 |
|
|
break;
|
1132 |
|
|
case 'm':
|
1133 |
|
|
if (argn+1 >= argc) {
|
1134 |
|
|
printf("No maximum output bit value given\n");
|
1135 |
|
|
exit(-1);
|
1136 |
|
|
}
|
1137 |
|
|
maxbitsout = atoi(argv[++argn]);
|
1138 |
|
|
j += 200;
|
1139 |
|
|
break;
|
1140 |
|
|
case 'n':
|
1141 |
|
|
if (argn+1 >= argc) {
|
1142 |
|
|
printf("No input bit size given\n");
|
1143 |
|
|
exit(-1);
|
1144 |
|
|
}
|
1145 |
|
|
nbitsin = atoi(argv[++argn]);
|
1146 |
|
|
j += 200;
|
1147 |
|
|
break;
|
1148 |
|
|
case 'S':
|
1149 |
|
|
bitreverse = true;
|
1150 |
|
|
break;
|
1151 |
|
|
case 's':
|
1152 |
|
|
bitreverse = false;
|
1153 |
|
|
break;
|
1154 |
|
|
case 'v':
|
1155 |
|
|
verbose_flag = true;
|
1156 |
|
|
break;
|
1157 |
|
|
default:
|
1158 |
|
|
printf("Unknown argument, -%c\n", argv[argn][j]);
|
1159 |
|
|
usage();
|
1160 |
|
|
exit(-1);
|
1161 |
|
|
}
|
1162 |
|
|
}
|
1163 |
|
|
} else {
|
1164 |
|
|
printf("Unrecognized argument, %s\n", argv[argn]);
|
1165 |
|
|
usage();
|
1166 |
|
|
exit(-1);
|
1167 |
|
|
}
|
1168 |
|
|
}
|
1169 |
|
|
|
1170 |
|
|
if ((lgsize < 0)&&(fftsize > 1)) {
|
1171 |
|
|
for(lgsize=1; (1<<lgsize) < fftsize; lgsize++)
|
1172 |
|
|
;
|
1173 |
|
|
}
|
1174 |
|
|
|
1175 |
|
|
if ((fftsize <= 0)||(nbitsin < 1)||(nbitsin>48)) {
|
1176 |
|
|
printf("INVALID PARAMETERS!!!!\n");
|
1177 |
|
|
exit(-1);
|
1178 |
|
|
}
|
1179 |
|
|
|
1180 |
|
|
|
1181 |
|
|
if (nextlg(fftsize) != fftsize) {
|
1182 |
|
|
fprintf(stderr, "ERR: FFTSize (%d) *must* be a power of two\n",
|
1183 |
|
|
fftsize);
|
1184 |
|
|
exit(-1);
|
1185 |
|
|
} else if (fftsize < 2) {
|
1186 |
|
|
fprintf(stderr, "ERR: Minimum FFTSize is 2, not %d\n",
|
1187 |
|
|
fftsize);
|
1188 |
|
|
if (fftsize == 1) {
|
1189 |
|
|
fprintf(stderr, "You do realize that a 1 point FFT makes very little sense\n");
|
1190 |
|
|
fprintf(stderr, "in an FFT operation that handles two samples per clock?\n");
|
1191 |
|
|
fprintf(stderr, "If you really need to do an FFT of this size, the output\n");
|
1192 |
|
|
fprintf(stderr, "can be connected straight to the input.\n");
|
1193 |
|
|
} else {
|
1194 |
|
|
fprintf(stderr, "Indeed, a size of %d doesn\'t make much sense to me at all.\n", fftsize);
|
1195 |
|
|
fprintf(stderr, "Is such an operation even defined?\n");
|
1196 |
|
|
}
|
1197 |
|
|
exit(-1);
|
1198 |
|
|
}
|
1199 |
|
|
|
1200 |
|
|
// Calculate how many output bits we'll have, and what the log
|
1201 |
|
|
// based two size of our FFT is.
|
1202 |
|
|
{
|
1203 |
|
|
int tmp_size = fftsize;
|
1204 |
|
|
|
1205 |
|
|
// The first stage always accumulates one bit, regardless
|
1206 |
|
|
// of whether you need to or not.
|
1207 |
|
|
nbitsout = nbitsin + 1;
|
1208 |
|
|
tmp_size >>= 1;
|
1209 |
|
|
|
1210 |
|
|
while(tmp_size > 4) {
|
1211 |
|
|
nbitsout += 1;
|
1212 |
|
|
tmp_size >>= 2;
|
1213 |
|
|
}
|
1214 |
|
|
|
1215 |
|
|
if (tmp_size > 1)
|
1216 |
|
|
nbitsout ++;
|
1217 |
|
|
|
1218 |
|
|
if (fftsize <= 2)
|
1219 |
|
|
bitreverse = false;
|
1220 |
|
|
} if ((maxbitsout > 0)&&(nbitsout > maxbitsout))
|
1221 |
|
|
nbitsout = maxbitsout;
|
1222 |
|
|
|
1223 |
|
|
|
1224 |
|
|
{
|
1225 |
|
|
struct stat sbuf;
|
1226 |
|
|
if (lstat(coredir, &sbuf)==0) {
|
1227 |
|
|
if (!S_ISDIR(sbuf.st_mode)) {
|
1228 |
|
|
fprintf(stderr, "\'%s\' already exists, and is not a directory!\n", coredir);
|
1229 |
|
|
fprintf(stderr, "I will stop now, lest I overwrite something you care about.\n");
|
1230 |
|
|
fprintf(stderr, "To try again, please remove this file.\n");
|
1231 |
|
|
exit(-1);
|
1232 |
|
|
}
|
1233 |
|
|
} else
|
1234 |
|
|
mkdir(coredir, 0755);
|
1235 |
|
|
if (access(coredir, X_OK|W_OK) != 0) {
|
1236 |
|
|
fprintf(stderr, "I have no access to the directory \'%s\'.\n", coredir);
|
1237 |
|
|
exit(-1);
|
1238 |
|
|
}
|
1239 |
|
|
}
|
1240 |
|
|
|
1241 |
|
|
sprintf(fname, "%s/%sfftmain.v", coredir, (inverse)?"i":"");
|
1242 |
|
|
vmain = fopen(fname, "w");
|
1243 |
|
|
if (NULL == vmain) {
|
1244 |
|
|
fprintf(stderr, "Could not open \'%s\' for writing\n", fname);
|
1245 |
|
|
perror("Err from O/S:");
|
1246 |
|
|
exit(-1);
|
1247 |
|
|
}
|
1248 |
|
|
|
1249 |
|
|
fprintf(vmain, "/////////////////////////////////////////////////////////////////////////////\n");
|
1250 |
|
|
fprintf(vmain, "//\n");
|
1251 |
|
|
fprintf(vmain, "// Filename: %sfftmain.v\n", (inverse)?"i":"");
|
1252 |
|
|
fprintf(vmain, "//\n");
|
1253 |
|
|
fprintf(vmain, "// Project: %s\n", prjname);
|
1254 |
|
|
fprintf(vmain, "//\n");
|
1255 |
|
|
fprintf(vmain, "// Purpose: This is the main module in the Doubletime FPGA FFT project.\n");
|
1256 |
|
|
fprintf(vmain, "// As such, all other modules are subordinate to this one.\n");
|
1257 |
|
|
fprintf(vmain, "// (I have been reading too much legalese this week ...)\n");
|
1258 |
|
|
fprintf(vmain, "// This module accomplish a fixed size Complex FFT on %d data\n", fftsize);
|
1259 |
|
|
fprintf(vmain, "// points. The FFT is fully pipelined, and accepts as inputs\n");
|
1260 |
|
|
fprintf(vmain, "// two complex two\'s complement samples per clock.\n");
|
1261 |
|
|
fprintf(vmain, "//\n");
|
1262 |
|
|
fprintf(vmain, "// Parameters:\n");
|
1263 |
|
|
fprintf(vmain, "// i_clk\tThe clock. All operations are synchronous with this clock.\n");
|
1264 |
|
|
fprintf(vmain, "//\ti_rst\tSynchronous reset, active high. Setting this line will\n");
|
1265 |
|
|
fprintf(vmain, "//\t\t\tforce the reset of all of the internals to this routine.\n");
|
1266 |
|
|
fprintf(vmain, "//\t\t\tFurther, following a reset, the o_sync line will go\n");
|
1267 |
|
|
fprintf(vmain, "//\t\t\thigh the same time the first output sample is valid.\n");
|
1268 |
|
|
fprintf(vmain, "// i_ce\tA clock enable line. If this line is set, this module\n");
|
1269 |
|
|
fprintf(vmain, "//\t\t\twill accept two complex values as inputs, and produce\n");
|
1270 |
|
|
fprintf(vmain, "//\t\t\ttwo (possibly empty) complex values as outputs.\n");
|
1271 |
|
|
fprintf(vmain, "//\t\ti_left\tThe first of two complex input samples. This value\n");
|
1272 |
|
|
fprintf(vmain, "//\t\t\tis split into two two\'s complement numbers, of \n");
|
1273 |
|
|
fprintf(vmain, "//\t\t\t%d bits each, with the real portion in the high\n", nbitsin);
|
1274 |
|
|
fprintf(vmain, "//\t\t\torder bits, and the imaginary portion taking the\n");
|
1275 |
|
|
fprintf(vmain, "//\t\t\tbottom %d bits.\n", nbitsin);
|
1276 |
|
|
fprintf(vmain, "//\t\ti_right\tThis is the same thing as i_left, only this is the\n");
|
1277 |
|
|
fprintf(vmain, "//\t\t\tsecond of two such samples. Hence, i_left would\n");
|
1278 |
|
|
fprintf(vmain, "//\t\t\tcontain input sample zero, i_right would contain\n");
|
1279 |
|
|
fprintf(vmain, "//\t\t\tsample one. On the next clock i_left would contain\n");
|
1280 |
|
|
fprintf(vmain, "//\t\t\tinput sample two, i_right number three and so forth.\n");
|
1281 |
|
|
fprintf(vmain, "//\t\to_left\tThe first of two output samples, of the same\n");
|
1282 |
|
|
fprintf(vmain, "//\t\t\tformat as i_left, only having %d bits for each of\n", nbitsout);
|
1283 |
|
|
fprintf(vmain, "//\t\t\tthe real and imaginary components, leading to %d\n", nbitsout*2);
|
1284 |
|
|
fprintf(vmain, "//\t\t\tbits total.\n");
|
1285 |
|
|
fprintf(vmain, "//\t\to_right\tThe second of two output samples produced each clock.\n");
|
1286 |
|
|
fprintf(vmain, "//\t\t\tThis has the same format as o_left.\n");
|
1287 |
|
|
fprintf(vmain, "//\t\to_sync\tA one bit output indicating the first valid sample\n");
|
1288 |
|
|
fprintf(vmain, "//\t\t\tproduced by this FFT following a reset. Ever after,\n");
|
1289 |
|
|
fprintf(vmain, "//\t\t\tthis will indicate the first sample of an FFT frame.\n");
|
1290 |
|
|
fprintf(vmain, "//\n");
|
1291 |
|
|
fprintf(vmain, "%s", creator);
|
1292 |
|
|
fprintf(vmain, "//\n");
|
1293 |
|
|
fprintf(vmain, "%s", cpyleft);
|
1294 |
|
|
|
1295 |
|
|
|
1296 |
|
|
fprintf(vmain, "//\n");
|
1297 |
|
|
fprintf(vmain, "//\n");
|
1298 |
|
|
fprintf(vmain, "module %sfftmain(i_clk, i_rst, i_ce,\n", (inverse)?"i":"");
|
1299 |
|
|
fprintf(vmain, "\t\ti_left, i_right,\n");
|
1300 |
|
|
fprintf(vmain, "\t\to_left, o_right, o_sync);\n");
|
1301 |
|
|
fprintf(vmain, "\tparameter\tIWIDTH=%d, OWIDTH=%d, LGWIDTH=%d;\n", nbitsin, nbitsout, lgsize);
|
1302 |
|
|
assert(lgsize > 0);
|
1303 |
|
|
fprintf(vmain, "\tinput\t\ti_clk, i_rst, i_ce;\n");
|
1304 |
|
|
fprintf(vmain, "\tinput\t\t[(2*IWIDTH-1):0]\ti_left, i_right;\n");
|
1305 |
|
|
fprintf(vmain, "\toutput\treg\t[(2*OWIDTH-1):0]\to_left, o_right;\n");
|
1306 |
|
|
fprintf(vmain, "\toutput\treg\t\t\to_sync;\n");
|
1307 |
|
|
fprintf(vmain, "\n\n");
|
1308 |
|
|
|
1309 |
|
|
fprintf(vmain, "\t// Outputs of the FFT, ready for bit reversal.\n");
|
1310 |
|
|
fprintf(vmain, "\twire\t[(2*OWIDTH-1):0]\tbr_left, br_right;\n");
|
1311 |
|
|
fprintf(vmain, "\n\n");
|
1312 |
|
|
|
1313 |
|
|
int tmp_size = fftsize, lgtmp = lgsize;
|
1314 |
|
|
if (fftsize == 2) {
|
1315 |
|
|
if (bitreverse) {
|
1316 |
|
|
fprintf(vmain, "\treg\tbr_start;\n");
|
1317 |
|
|
fprintf(vmain, "\talways @(posedge i_clk)\n");
|
1318 |
|
|
fprintf(vmain, "\t\tif (i_rst)\n");
|
1319 |
|
|
fprintf(vmain, "\t\t\tbr_start <= 1'b0;\n");
|
1320 |
|
|
fprintf(vmain, "\t\telse if (i_ce)\n");
|
1321 |
|
|
fprintf(vmain, "\t\t\tbr_start <= 1'b1;\n");
|
1322 |
|
|
}
|
1323 |
|
|
fprintf(vmain, "\n\n");
|
1324 |
6 |
dgisselq |
fprintf(vmain, "\tdblstage\t#(IWIDTH)\tstage_2(i_clk, i_rst, i_ce,\n");
|
1325 |
|
|
fprintf(vmain, "\t\t\t(~i_rst), i_left, i_right, br_left, br_right);\n");
|
1326 |
2 |
dgisselq |
fprintf(vmain, "\n\n");
|
1327 |
|
|
} else {
|
1328 |
|
|
int nbits = nbitsin, dropbit=0;
|
1329 |
|
|
// Always do a first stage
|
1330 |
|
|
fprintf(vmain, "\n\n");
|
1331 |
|
|
fprintf(vmain, "\twire\t\tw_s%d, w_os%d;\n", fftsize, fftsize);
|
1332 |
|
|
fprintf(vmain, "\twire\t[(2*IWIDTH+1):0]\tw_e%d, w_o%d;\n", fftsize, fftsize);
|
1333 |
|
|
fprintf(vmain, "\t%sfftstage_e%d\t#(IWIDTH,IWIDTH+%d,IWIDTH+1,%d,%d,%d,0)\tstage_e%d(i_clk, i_rst, i_ce,\n",
|
1334 |
|
|
(inverse)?"i":"", fftsize,
|
1335 |
|
|
xtracbits,
|
1336 |
|
|
lgsize, lgtmp-2, lgdelay(nbits,xtracbits),
|
1337 |
|
|
fftsize);
|
1338 |
|
|
fprintf(vmain, "\t\t\t(~i_rst), i_left, w_e%d, w_s%d);\n", fftsize, fftsize);
|
1339 |
|
|
fprintf(vmain, "\t%sfftstage_o%d\t#(IWIDTH,IWIDTH+%d,IWIDTH+1,%d,%d,%d,0)\tstage_o%d(i_clk, i_rst, i_ce,\n",
|
1340 |
|
|
(inverse)?"i":"", fftsize,
|
1341 |
|
|
xtracbits,
|
1342 |
|
|
lgsize, lgtmp-2, lgdelay(nbits,xtracbits),
|
1343 |
|
|
fftsize);
|
1344 |
|
|
fprintf(vmain, "\t\t\t(~i_rst), i_left, w_o%d, w_os%d);\n", fftsize, fftsize);
|
1345 |
|
|
fprintf(vmain, "\n\n");
|
1346 |
|
|
|
1347 |
|
|
sprintf(fname, "%s/%sfftstage_e%d.v", coredir, (inverse)?"i":"", fftsize);
|
1348 |
|
|
build_stage(fname, fftsize/2, 0, nbits, inverse, xtracbits); // Even stage
|
1349 |
|
|
sprintf(fname, "%s/%sfftstage_o%d.v", coredir, (inverse)?"i":"", fftsize);
|
1350 |
|
|
build_stage(fname, fftsize/2, 1, nbits, inverse, xtracbits); // Odd stage
|
1351 |
|
|
|
1352 |
|
|
nbits += 1; // New number of input bits
|
1353 |
|
|
tmp_size >>= 1; lgtmp--;
|
1354 |
|
|
dropbit = 0;
|
1355 |
|
|
fprintf(vmain, "\n\n");
|
1356 |
|
|
while(tmp_size >= 8) {
|
1357 |
|
|
int obits = nbits+((dropbit)?0:1);
|
1358 |
|
|
|
1359 |
|
|
if ((maxbitsout > 0)&&(obits > maxbitsout))
|
1360 |
|
|
obits = maxbitsout;
|
1361 |
|
|
|
1362 |
|
|
fprintf(vmain, "\twire\t\tw_s%d, w_os%d;\n", tmp_size, tmp_size);
|
1363 |
|
|
fprintf(vmain, "\twire\t[%d:0]\tw_e%d, w_o%d;\n", 2*obits-1, tmp_size, tmp_size);
|
1364 |
|
|
fprintf(vmain, "\t%sfftstage_e%d\t#(%d,%d,%d,%d,%d,%d,%d)\tstage_e%d(i_clk, i_rst, i_ce,\n",
|
1365 |
|
|
(inverse)?"i":"", tmp_size,
|
1366 |
|
|
nbits, nbits+xtracbits, obits,
|
1367 |
|
|
lgsize, lgtmp-2, lgdelay(nbits,xtracbits), dropbit,
|
1368 |
|
|
tmp_size);
|
1369 |
|
|
fprintf(vmain, "\t\t\t\t\t\tw_s%d, w_e%d, w_e%d, w_s%d);\n", tmp_size<<1, tmp_size<<1, tmp_size, tmp_size);
|
1370 |
|
|
fprintf(vmain, "\t%sfftstage_o%d\t#(%d,%d,%d,%d,%d,%d,%d)\tstage_o%d(i_clk, i_rst, i_ce,\n",
|
1371 |
|
|
(inverse)?"i":"", tmp_size,
|
1372 |
|
|
nbits, nbits+xtracbits, obits,
|
1373 |
|
|
lgsize, lgtmp-2, lgdelay(nbits,xtracbits), dropbit,
|
1374 |
|
|
tmp_size);
|
1375 |
|
|
fprintf(vmain, "\t\t\t\t\t\tw_s%d, w_o%d, w_o%d, w_os%d);\n", tmp_size<<1, tmp_size<<1, tmp_size, tmp_size);
|
1376 |
|
|
fprintf(vmain, "\n\n");
|
1377 |
|
|
|
1378 |
|
|
sprintf(fname, "%s/%sfftstage_e%d.v", coredir, (inverse)?"i":"", tmp_size);
|
1379 |
|
|
build_stage(fname, tmp_size/2, 0, nbits, inverse, xtracbits); // Even stage
|
1380 |
|
|
sprintf(fname, "%s/%sfftstage_o%d.v", coredir, (inverse)?"i":"", tmp_size);
|
1381 |
|
|
build_stage(fname, tmp_size/2, 1, nbits, inverse, xtracbits); // Odd stage
|
1382 |
|
|
|
1383 |
|
|
|
1384 |
|
|
dropbit ^= 1;
|
1385 |
|
|
nbits = obits;
|
1386 |
|
|
tmp_size >>= 1; lgtmp--;
|
1387 |
|
|
}
|
1388 |
|
|
|
1389 |
|
|
if (tmp_size == 4) {
|
1390 |
|
|
int obits = nbits+((dropbit)?0:1);
|
1391 |
|
|
|
1392 |
|
|
if ((maxbitsout > 0)&&(obits > maxbitsout))
|
1393 |
|
|
obits = maxbitsout;
|
1394 |
|
|
|
1395 |
|
|
fprintf(vmain, "\twire\t\tw_s4, w_os4;\n");
|
1396 |
|
|
fprintf(vmain, "\twire\t[%d:0]\tw_e4, w_o4;\n", 2*obits-1);
|
1397 |
|
|
fprintf(vmain, "\tqtrstage\t#(%d,%d,%d,0,%d,%d)\tstage_e4(i_clk, i_rst, i_ce,\n",
|
1398 |
|
|
nbits, obits, lgsize, (inverse)?1:0, dropbit);
|
1399 |
6 |
dgisselq |
fprintf(vmain, "\t\t\t\t\t\tw_s8, w_e8, w_e4, w_s4);\n");
|
1400 |
2 |
dgisselq |
fprintf(vmain, "\tqtrstage\t#(%d,%d,%d,1,%d,%d)\tstage_o4(i_clk, i_rst, i_ce,\n",
|
1401 |
|
|
nbits, obits, lgsize, (inverse)?1:0, dropbit);
|
1402 |
6 |
dgisselq |
fprintf(vmain, "\t\t\t\t\t\tw_s8, w_o8, w_o4, w_os4);\n");
|
1403 |
2 |
dgisselq |
dropbit ^= 1;
|
1404 |
|
|
nbits = obits;
|
1405 |
|
|
tmp_size >>= 1; lgtmp--;
|
1406 |
|
|
}
|
1407 |
|
|
|
1408 |
|
|
{
|
1409 |
|
|
int obits = nbits+((dropbit)?0:1);
|
1410 |
|
|
if (obits > nbitsout)
|
1411 |
|
|
obits = nbitsout;
|
1412 |
|
|
if ((maxbitsout>0)&&(obits > maxbitsout))
|
1413 |
|
|
obits = maxbitsout;
|
1414 |
|
|
fprintf(vmain, "\twire\t\tw_s2;\n");
|
1415 |
|
|
fprintf(vmain, "\twire\t[%d:0]\tw_e2, w_o2;\n", 2*obits-1);
|
1416 |
6 |
dgisselq |
fprintf(vmain, "\tdblstage\t#(%d,%d,%d)\tstage_2(i_clk, i_rst, i_ce,\n", nbits, obits,dropbit);
|
1417 |
|
|
fprintf(vmain, "\t\t\t\t\tw_s4, w_e4, w_o4, w_e2, w_o2, w_s2);\n");
|
1418 |
2 |
dgisselq |
|
1419 |
|
|
fprintf(vmain, "\n\n");
|
1420 |
|
|
nbits = obits;
|
1421 |
|
|
}
|
1422 |
|
|
|
1423 |
|
|
fprintf(vmain, "\t// Prepare for a (potential) bit-reverse stage.\n");
|
1424 |
|
|
fprintf(vmain, "\tassign\tbr_left = w_e2;\n");
|
1425 |
|
|
fprintf(vmain, "\tassign\tbr_right = w_o2;\n");
|
1426 |
|
|
fprintf(vmain, "\n");
|
1427 |
|
|
if (bitreverse) {
|
1428 |
|
|
fprintf(vmain, "\twire\tbr_start;\n");
|
1429 |
|
|
fprintf(vmain, "\treg\tr_br_started;\n");
|
1430 |
|
|
fprintf(vmain, "\t// A delay of one clock here is perfect, as it matches the delay in\n");
|
1431 |
|
|
fprintf(vmain, "\t// our dblstage.\n");
|
1432 |
|
|
fprintf(vmain, "\talways @(posedge i_clk)\n");
|
1433 |
|
|
fprintf(vmain, "\t\tif (i_rst)\n");
|
1434 |
|
|
fprintf(vmain, "\t\t\tr_br_started <= 1'b0;\n");
|
1435 |
|
|
fprintf(vmain, "\t\telse\n");
|
1436 |
|
|
fprintf(vmain, "\t\t\tr_br_started <= r_br_started || w_s4;\n");
|
1437 |
|
|
fprintf(vmain, "\tassign\tbr_start = r_br_started;\n");
|
1438 |
|
|
}
|
1439 |
|
|
}
|
1440 |
|
|
|
1441 |
|
|
fprintf(vmain, "\n");
|
1442 |
|
|
fprintf(vmain, "\t// Now for the bit-reversal stage.\n");
|
1443 |
|
|
fprintf(vmain, "\twire\tbr_sync;\n");
|
1444 |
|
|
fprintf(vmain, "\twire\t[(2*OWIDTH-1):0]\tbr_o_left, br_o_right;\n");
|
1445 |
|
|
if (bitreverse) {
|
1446 |
|
|
fprintf(vmain, "\tdblreverse\t#(%d,%d)\trevstage(i_clk, i_rst,\n", lgsize, nbitsout);
|
1447 |
|
|
fprintf(vmain, "\t\t\t(i_ce & br_start), br_left, br_right,\n");
|
1448 |
|
|
fprintf(vmain, "\t\t\tbr_o_left, br_o_right, br_sync);\n");
|
1449 |
|
|
} else {
|
1450 |
|
|
fprintf(vmain, "\tassign\tbr_o_left = br_left;\n");
|
1451 |
|
|
fprintf(vmain, "\tassign\tbr_o_right = br_right;\n");
|
1452 |
|
|
fprintf(vmain, "\tassign\tbr_sync = w_s2;\n");
|
1453 |
|
|
}
|
1454 |
|
|
|
1455 |
|
|
fprintf(vmain, "\n\n");
|
1456 |
|
|
fprintf(vmain, "\t// Last clock: Register our outputs, we\'re done.\n");
|
1457 |
|
|
fprintf(vmain, "\talways @(posedge i_clk)\n");
|
1458 |
|
|
fprintf(vmain, "\t\tbegin\n");
|
1459 |
|
|
fprintf(vmain, "\t\t\to_left <= br_o_left;\n");
|
1460 |
|
|
fprintf(vmain, "\t\t\to_right <= br_o_right;\n");
|
1461 |
|
|
fprintf(vmain, "\t\t\to_sync <= br_sync;\n");
|
1462 |
|
|
fprintf(vmain, "\t\tend\n");
|
1463 |
|
|
fprintf(vmain, "\n\n");
|
1464 |
|
|
fprintf(vmain, "endmodule\n");
|
1465 |
|
|
fclose(vmain);
|
1466 |
|
|
|
1467 |
|
|
sprintf(fname, "%s/butterfly.v", coredir);
|
1468 |
|
|
build_butterfly(fname);
|
1469 |
|
|
|
1470 |
|
|
sprintf(fname, "%s/shiftaddmpy.v", coredir);
|
1471 |
|
|
build_multiply(fname);
|
1472 |
|
|
|
1473 |
|
|
sprintf(fname, "%s/qtrstage.v", coredir);
|
1474 |
|
|
build_quarters(fname);
|
1475 |
|
|
|
1476 |
|
|
sprintf(fname, "%s/dblstage.v", coredir);
|
1477 |
|
|
build_dblstage(fname);
|
1478 |
|
|
|
1479 |
|
|
if (bitreverse) {
|
1480 |
|
|
sprintf(fname, "%s/dblreverse.v", coredir);
|
1481 |
|
|
build_dblreverse(fname);
|
1482 |
|
|
}
|
1483 |
|
|
}
|
1484 |
|
|
|