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