new generic FFT/MDCT code for audio codecs
[libav.git] / libavcodec / fft.c
1 /*
2 * FFT/IFFT transforms
3 * Copyright (c) 2002 Fabrice Bellard.
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19 #include "dsputil.h"
20
21 /**
22 * The size of the FFT is 2^nbits. If inverse is TRUE, inverse FFT is
23 * done
24 */
25 int fft_init(FFTContext *s, int nbits, int inverse)
26 {
27 int i, j, m, n;
28 float alpha, c1, s1, s2;
29
30 s->nbits = nbits;
31 n = 1 << nbits;
32
33 s->exptab = av_malloc((n / 2) * sizeof(FFTComplex));
34 if (!s->exptab)
35 goto fail;
36 s->revtab = av_malloc(n * sizeof(uint16_t));
37 if (!s->revtab)
38 goto fail;
39 s->inverse = inverse;
40
41 s2 = inverse ? 1.0 : -1.0;
42
43 for(i=0;i<(n/2);i++) {
44 alpha = 2 * M_PI * (float)i / (float)n;
45 c1 = cos(alpha);
46 s1 = sin(alpha) * s2;
47 s->exptab[i].re = c1;
48 s->exptab[i].im = s1;
49 }
50 s->fft_calc = fft_calc_c;
51 s->exptab1 = NULL;
52
53 /* compute constant table for HAVE_SSE version */
54 #if defined(HAVE_MMX) && 0
55 if (mm_flags & MM_SSE) {
56 int np, nblocks, np2, l;
57 FFTComplex *q;
58
59 np = 1 << nbits;
60 nblocks = np >> 3;
61 np2 = np >> 1;
62 s->exptab1 = av_malloc(np * 2 * sizeof(FFTComplex));
63 if (!s->exptab1)
64 goto fail;
65 q = s->exptab1;
66 do {
67 for(l = 0; l < np2; l += 2 * nblocks) {
68 *q++ = s->exptab[l];
69 *q++ = s->exptab[l + nblocks];
70
71 q->re = -s->exptab[l].im;
72 q->im = s->exptab[l].re;
73 q++;
74 q->re = -s->exptab[l + nblocks].im;
75 q->im = s->exptab[l + nblocks].re;
76 q++;
77 }
78 nblocks = nblocks >> 1;
79 } while (nblocks != 0);
80 av_freep(&s->exptab);
81 }
82 #endif
83
84 /* compute bit reverse table */
85
86 for(i=0;i<n;i++) {
87 m=0;
88 for(j=0;j<nbits;j++) {
89 m |= ((i >> j) & 1) << (nbits-j-1);
90 }
91 s->revtab[i]=m;
92 }
93 return 0;
94 fail:
95 av_freep(&s->revtab);
96 av_freep(&s->exptab);
97 av_freep(&s->exptab1);
98 return -1;
99 }
100
101 /* butter fly op */
102 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
103 {\
104 FFTSample ax, ay, bx, by;\
105 bx=pre1;\
106 by=pim1;\
107 ax=qre1;\
108 ay=qim1;\
109 pre = (bx + ax);\
110 pim = (by + ay);\
111 qre = (bx - ax);\
112 qim = (by - ay);\
113 }
114
115 #define MUL16(a,b) ((a) * (b))
116
117 #define CMUL(pre, pim, are, aim, bre, bim) \
118 {\
119 pre = (MUL16(are, bre) - MUL16(aim, bim));\
120 pim = (MUL16(are, bim) + MUL16(bre, aim));\
121 }
122
123 /**
124 * Do a complex FFT with the parameters defined in fft_init(). The
125 * input data must be permuted before with s->revtab table. No
126 * 1.0/sqrt(n) normalization is done.
127 */
128 void fft_calc_c(FFTContext *s, FFTComplex *z)
129 {
130 int ln = s->nbits;
131 int j, np, np2;
132 int nblocks, nloops;
133 register FFTComplex *p, *q;
134 FFTComplex *exptab = s->exptab;
135 int l;
136 FFTSample tmp_re, tmp_im;
137
138 np = 1 << ln;
139
140 /* pass 0 */
141
142 p=&z[0];
143 j=(np >> 1);
144 do {
145 BF(p[0].re, p[0].im, p[1].re, p[1].im,
146 p[0].re, p[0].im, p[1].re, p[1].im);
147 p+=2;
148 } while (--j != 0);
149
150 /* pass 1 */
151
152
153 p=&z[0];
154 j=np >> 2;
155 if (s->inverse) {
156 do {
157 BF(p[0].re, p[0].im, p[2].re, p[2].im,
158 p[0].re, p[0].im, p[2].re, p[2].im);
159 BF(p[1].re, p[1].im, p[3].re, p[3].im,
160 p[1].re, p[1].im, -p[3].im, p[3].re);
161 p+=4;
162 } while (--j != 0);
163 } else {
164 do {
165 BF(p[0].re, p[0].im, p[2].re, p[2].im,
166 p[0].re, p[0].im, p[2].re, p[2].im);
167 BF(p[1].re, p[1].im, p[3].re, p[3].im,
168 p[1].re, p[1].im, p[3].im, -p[3].re);
169 p+=4;
170 } while (--j != 0);
171 }
172 /* pass 2 .. ln-1 */
173
174 nblocks = np >> 3;
175 nloops = 1 << 2;
176 np2 = np >> 1;
177 do {
178 p = z;
179 q = z + nloops;
180 for (j = 0; j < nblocks; ++j) {
181 BF(p->re, p->im, q->re, q->im,
182 p->re, p->im, q->re, q->im);
183
184 p++;
185 q++;
186 for(l = nblocks; l < np2; l += nblocks) {
187 CMUL(tmp_re, tmp_im, exptab[l].re, exptab[l].im, q->re, q->im);
188 BF(p->re, p->im, q->re, q->im,
189 p->re, p->im, tmp_re, tmp_im);
190 p++;
191 q++;
192 }
193
194 p += nloops;
195 q += nloops;
196 }
197 nblocks = nblocks >> 1;
198 nloops = nloops << 1;
199 } while (nblocks != 0);
200 }
201
202 /**
203 * Do the permutation needed BEFORE calling fft_calc()
204 */
205 void fft_permute(FFTContext *s, FFTComplex *z)
206 {
207 int j, k, np;
208 FFTComplex tmp;
209 const uint16_t *revtab = s->revtab;
210
211 /* reverse */
212 np = 1 << s->nbits;
213 for(j=0;j<np;j++) {
214 k = revtab[j];
215 if (k < j) {
216 tmp = z[k];
217 z[k] = z[j];
218 z[j] = tmp;
219 }
220 }
221 }
222
223 void fft_end(FFTContext *s)
224 {
225 av_freep(&s->revtab);
226 av_freep(&s->exptab);
227 av_freep(&s->exptab1);
228 }
229