| 1 |
36 |
dgisselq |
////////////////////////////////////////////////////////////////////////////////
|
| 2 |
|
|
//
|
| 3 |
|
|
// Filename: fftlib.cpp
|
| 4 |
|
|
//
|
| 5 |
|
|
// Project: A General Purpose Pipelined FFT Implementation
|
| 6 |
|
|
//
|
| 7 |
|
|
// Purpose:
|
| 8 |
|
|
//
|
| 9 |
|
|
// Creator: Dan Gisselquist, Ph.D.
|
| 10 |
|
|
// Gisselquist Technology, LLC
|
| 11 |
|
|
//
|
| 12 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
| 13 |
|
|
//
|
| 14 |
|
|
// Copyright (C) 2015-2018, Gisselquist Technology, LLC
|
| 15 |
|
|
//
|
| 16 |
|
|
// This program is free software (firmware): you can redistribute it and/or
|
| 17 |
|
|
// modify it under the terms of the GNU General Public License as published
|
| 18 |
|
|
// by the Free Software Foundation, either version 3 of the License, or (at
|
| 19 |
|
|
// your option) any later version.
|
| 20 |
|
|
//
|
| 21 |
|
|
// This program is distributed in the hope that it will be useful, but WITHOUT
|
| 22 |
|
|
// ANY WARRANTY; without even the implied warranty of MERCHANTIBILITY or
|
| 23 |
|
|
// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
| 24 |
|
|
// for more details.
|
| 25 |
|
|
//
|
| 26 |
|
|
// You should have received a copy of the GNU General Public License along
|
| 27 |
|
|
// with this program. (It's in the $(ROOT)/doc directory, run make with no
|
| 28 |
|
|
// target there if the PDF file isn't present.) If not, see
|
| 29 |
|
|
// <http://www.gnu.org/licenses/> for a copy.
|
| 30 |
|
|
//
|
| 31 |
|
|
// License: GPL, v3, as defined and found on www.gnu.org,
|
| 32 |
|
|
// http://www.gnu.org/licenses/gpl.html
|
| 33 |
|
|
//
|
| 34 |
|
|
//
|
| 35 |
|
|
////////////////////////////////////////////////////////////////////////////////
|
| 36 |
|
|
//
|
| 37 |
|
|
//
|
| 38 |
|
|
#define _CRT_SECURE_NO_WARNINGS // ms vs 2012 doesn't like fopen
|
| 39 |
|
|
#include <stdio.h>
|
| 40 |
|
|
#include <stdlib.h>
|
| 41 |
|
|
|
| 42 |
|
|
#ifdef _MSC_VER // added for ms vs compatibility
|
| 43 |
|
|
|
| 44 |
|
|
#include <io.h>
|
| 45 |
|
|
#include <direct.h>
|
| 46 |
|
|
#define _USE_MATH_DEFINES
|
| 47 |
|
|
|
| 48 |
|
|
#if _MSC_VER <= 1700
|
| 49 |
|
|
|
| 50 |
|
|
long long llround(double d) {
|
| 51 |
|
|
if (d<0) return -(long long)(-d+0.5);
|
| 52 |
|
|
else return (long long)(d+0.5); }
|
| 53 |
|
|
|
| 54 |
|
|
#endif
|
| 55 |
|
|
|
| 56 |
|
|
#else
|
| 57 |
|
|
// And for G++/Linux environment
|
| 58 |
|
|
|
| 59 |
|
|
#include <unistd.h> // Defines the R_OK/W_OK/etc. macros
|
| 60 |
|
|
#endif
|
| 61 |
|
|
|
| 62 |
|
|
#include <string.h>
|
| 63 |
|
|
#include <string>
|
| 64 |
|
|
#include <math.h>
|
| 65 |
|
|
// #include <ctype.h>
|
| 66 |
|
|
#include <assert.h>
|
| 67 |
|
|
|
| 68 |
|
|
#include "fftlib.h"
|
| 69 |
|
|
|
| 70 |
|
|
|
| 71 |
|
|
int lgval(int vl) {
|
| 72 |
|
|
int lg;
|
| 73 |
|
|
|
| 74 |
|
|
for(lg=1; (1<<lg) < vl; lg++)
|
| 75 |
|
|
;
|
| 76 |
|
|
return lg;
|
| 77 |
|
|
}
|
| 78 |
|
|
|
| 79 |
|
|
int nextlg(int vl) {
|
| 80 |
|
|
int r;
|
| 81 |
|
|
|
| 82 |
|
|
for(r=1; r<vl; r<<=1)
|
| 83 |
|
|
;
|
| 84 |
|
|
return r;
|
| 85 |
|
|
}
|
| 86 |
|
|
|
| 87 |
|
|
int bflydelay(int nbits, int xtra) {
|
| 88 |
|
|
int cbits = nbits + xtra;
|
| 89 |
|
|
int delay;
|
| 90 |
|
|
|
| 91 |
|
|
if (USE_OLD_MULTIPLY) {
|
| 92 |
|
|
if (nbits+1<cbits)
|
| 93 |
|
|
delay = nbits+4;
|
| 94 |
|
|
else
|
| 95 |
|
|
delay = cbits+3;
|
| 96 |
|
|
} else {
|
| 97 |
|
|
int na=nbits+2, nb=cbits+1;
|
| 98 |
|
|
if (nb<na) {
|
| 99 |
|
|
int tmp = nb;
|
| 100 |
|
|
nb = na; na = tmp;
|
| 101 |
|
|
} delay = ((na)/2+(na&1)+2);
|
| 102 |
|
|
}
|
| 103 |
|
|
return delay;
|
| 104 |
|
|
}
|
| 105 |
|
|
|
| 106 |
|
|
int lgdelay(int nbits, int xtra) {
|
| 107 |
|
|
// The butterfly code needs to compare a valid address, of this
|
| 108 |
|
|
// many bits, with an address two greater. This guarantees we
|
| 109 |
|
|
// have enough bits for that comparison. We'll also end up with
|
| 110 |
|
|
// more storage space to look for these values, but without a
|
| 111 |
|
|
// redesign that's just what we'll deal with.
|
| 112 |
|
|
return lgval(bflydelay(nbits, xtra)+3);
|
| 113 |
|
|
}
|
| 114 |
|
|
|
| 115 |
|
|
void gen_coeffs(FILE *cmem, int stage, int cbits,
|
| 116 |
|
|
int nwide, int offset, bool inv) {
|
| 117 |
|
|
//
|
| 118 |
|
|
// For an FFT stage of 2^n elements, we need 2^(n-1) butterfly
|
| 119 |
|
|
// coefficients, sometimes called twiddle factors. Stage captures the
|
| 120 |
|
|
// width of the FFT at this point. If thiss is a 2x at a time FFT,
|
| 121 |
|
|
// nwide will be equal to 2, and offset will be one or two.
|
| 122 |
|
|
//
|
| 123 |
|
|
assert(nwide > 0);
|
| 124 |
|
|
assert(offset < nwide);
|
| 125 |
|
|
assert(stage / nwide > 1);
|
| 126 |
|
|
assert(stage % nwide == 0);
|
| 127 |
|
|
printf("GEN-COEFFS(): stage =%4d, bits =%2d, nwide = %d, offset = %d, nverse = %d\n", stage, cbits, nwide, offset, inv);
|
| 128 |
|
|
int ncoeffs = stage/nwide/2;
|
| 129 |
|
|
for(int i=0; i<ncoeffs; i++) {
|
| 130 |
|
|
int k = nwide*i+offset;
|
| 131 |
|
|
double W = ((inv)?1:-1)*2.0*M_PI*k/(double)(stage);
|
| 132 |
|
|
double c, s;
|
| 133 |
|
|
long long ic, is, vl;
|
| 134 |
|
|
|
| 135 |
|
|
c = cos(W); s = sin(W);
|
| 136 |
|
|
ic = (long long)llround((1ll<<(cbits-2)) * c);
|
| 137 |
|
|
is = (long long)llround((1ll<<(cbits-2)) * s);
|
| 138 |
|
|
vl = (ic & (~(-1ll << (cbits))));
|
| 139 |
|
|
vl <<= (cbits);
|
| 140 |
|
|
vl |= (is & (~(-1ll << (cbits))));
|
| 141 |
|
|
fprintf(cmem, "%0*llx\n", ((cbits*2+3)/4), vl);
|
| 142 |
|
|
//
|
| 143 |
|
|
} fclose(cmem);
|
| 144 |
|
|
}
|
| 145 |
|
|
|
| 146 |
|
|
std::string gen_coeff_fname(const char *coredir,
|
| 147 |
|
|
int stage, int nwide, int offset, bool inv) {
|
| 148 |
|
|
std::string result;
|
| 149 |
|
|
char *memfile;
|
| 150 |
|
|
|
| 151 |
|
|
assert((nwide == 1)||(nwide == 2));
|
| 152 |
|
|
|
| 153 |
|
|
memfile = new char[strlen(coredir)+3+10+strlen(".hex")+64];
|
| 154 |
|
|
if (nwide == 2) {
|
| 155 |
|
|
if (coredir[0] == '\0') {
|
| 156 |
|
|
sprintf(memfile, "%scmem_%c%d.hex",
|
| 157 |
|
|
(inv)?"i":"", (offset==1)?'o':'e', stage*nwide);
|
| 158 |
|
|
} else {
|
| 159 |
|
|
sprintf(memfile, "%s/%scmem_%c%d.hex",
|
| 160 |
|
|
coredir, (inv)?"i":"",
|
| 161 |
|
|
(offset==1)?'o':'e', stage*nwide);
|
| 162 |
|
|
}
|
| 163 |
|
|
} else if (coredir[0] == '\0') // if (nwide == 1)
|
| 164 |
|
|
sprintf(memfile, "%scmem_%d.hex",
|
| 165 |
|
|
(inv)?"i":"", stage);
|
| 166 |
|
|
else
|
| 167 |
|
|
sprintf(memfile, "%s/%scmem_%d.hex",
|
| 168 |
|
|
coredir, (inv)?"i":"", stage);
|
| 169 |
|
|
|
| 170 |
|
|
result = std::string(memfile);
|
| 171 |
|
|
delete[] memfile;
|
| 172 |
|
|
return result;
|
| 173 |
|
|
}
|
| 174 |
|
|
|
| 175 |
|
|
FILE *gen_coeff_open(const char *fname) {
|
| 176 |
|
|
FILE *cmem;
|
| 177 |
|
|
|
| 178 |
|
|
cmem = fopen(fname, "w");
|
| 179 |
|
|
if (NULL == cmem) {
|
| 180 |
|
|
fprintf(stderr, "Could not open FFT coefficient file "
|
| 181 |
|
|
"\'%s\' for writing\n", fname);
|
| 182 |
|
|
perror("Err from O/S:");
|
| 183 |
|
|
exit(EXIT_FAILURE);
|
| 184 |
|
|
}
|
| 185 |
|
|
|
| 186 |
|
|
return cmem;
|
| 187 |
|
|
}
|
| 188 |
|
|
|
| 189 |
|
|
void gen_coeff_file(const char *coredir, const char *fname,
|
| 190 |
|
|
int stage, int cbits, int nwide, int offset, bool inv) {
|
| 191 |
|
|
std::string fstr;
|
| 192 |
|
|
FILE *cmem;
|
| 193 |
|
|
|
| 194 |
|
|
fstr= gen_coeff_fname(coredir, stage, nwide, offset, inv);
|
| 195 |
|
|
cmem = gen_coeff_open(fstr.c_str());
|
| 196 |
|
|
gen_coeffs(cmem, stage, cbits, nwide, offset, inv);
|
| 197 |
|
|
}
|