6fd87167bd2bb2980274c84f1f41bbf44413139a
[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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19
20 /**
21 * @file fft.c
22 * FFT/IFFT transforms.
23 */
24
25 #include "dsputil.h"
26
27 /**
28 * The size of the FFT is 2^nbits. If inverse is TRUE, inverse FFT is
29 * done
30 */
31 int ff_fft_init(FFTContext *s, int nbits, int inverse)
32 {
33 int i, j, m, n;
34 float alpha, c1, s1, s2;
35
36 s->nbits = nbits;
37 n = 1 << nbits;
38
39 s->exptab = av_malloc((n / 2) * sizeof(FFTComplex));
40 if (!s->exptab)
41 goto fail;
42 s->revtab = av_malloc(n * sizeof(uint16_t));
43 if (!s->revtab)
44 goto fail;
45 s->inverse = inverse;
46
47 s2 = inverse ? 1.0 : -1.0;
48
49 for(i=0;i<(n/2);i++) {
50 alpha = 2 * M_PI * (float)i / (float)n;
51 c1 = cos(alpha);
52 s1 = sin(alpha) * s2;
53 s->exptab[i].re = c1;
54 s->exptab[i].im = s1;
55 }
56 s->fft_calc = ff_fft_calc_c;
57 s->imdct_calc = ff_imdct_calc;
58 s->exptab1 = NULL;
59
60 /* compute constant table for HAVE_SSE version */
61 #if (defined(HAVE_MMX) && (defined(HAVE_BUILTIN_VECTOR) || defined(HAVE_MM3DNOW))) \
62 || (defined(HAVE_ALTIVEC) && !defined(ALTIVEC_USE_REFERENCE_C_CODE))
63 {
64 int has_vectors = mm_support();
65
66 if (has_vectors) {
67 #if defined(HAVE_MMX)
68 if (has_vectors & MM_3DNOWEXT)
69 s->imdct_calc = ff_imdct_calc_3dn2;
70 #ifdef HAVE_MM3DNOW
71 if (has_vectors & MM_3DNOWEXT)
72 /* 3DNowEx for Athlon(XP) */
73 s->fft_calc = ff_fft_calc_3dn2;
74 else if (has_vectors & MM_3DNOW)
75 /* 3DNow! for K6-2/3 */
76 s->fft_calc = ff_fft_calc_3dn;
77 #endif
78 #ifdef HAVE_BUILTIN_VECTOR
79 if (has_vectors & MM_SSE2)
80 /* SSE for P4/K8 */
81 s->fft_calc = ff_fft_calc_sse;
82 else if ((has_vectors & MM_SSE) &&
83 s->fft_calc == ff_fft_calc_c)
84 /* SSE for P3 */
85 s->fft_calc = ff_fft_calc_sse;
86 #endif
87 #else /* HAVE_MMX */
88 if (has_vectors & MM_ALTIVEC)
89 s->fft_calc = ff_fft_calc_altivec;
90 #endif
91 }
92 if (s->fft_calc != ff_fft_calc_c) {
93 int np, nblocks, np2, l;
94 FFTComplex *q;
95
96 np = 1 << nbits;
97 nblocks = np >> 3;
98 np2 = np >> 1;
99 s->exptab1 = av_malloc(np * 2 * sizeof(FFTComplex));
100 if (!s->exptab1)
101 goto fail;
102 q = s->exptab1;
103 do {
104 for(l = 0; l < np2; l += 2 * nblocks) {
105 *q++ = s->exptab[l];
106 *q++ = s->exptab[l + nblocks];
107
108 q->re = -s->exptab[l].im;
109 q->im = s->exptab[l].re;
110 q++;
111 q->re = -s->exptab[l + nblocks].im;
112 q->im = s->exptab[l + nblocks].re;
113 q++;
114 }
115 nblocks = nblocks >> 1;
116 } while (nblocks != 0);
117 av_freep(&s->exptab);
118 }
119 }
120 #endif
121
122 /* compute bit reverse table */
123
124 for(i=0;i<n;i++) {
125 m=0;
126 for(j=0;j<nbits;j++) {
127 m |= ((i >> j) & 1) << (nbits-j-1);
128 }
129 s->revtab[i]=m;
130 }
131 return 0;
132 fail:
133 av_freep(&s->revtab);
134 av_freep(&s->exptab);
135 av_freep(&s->exptab1);
136 return -1;
137 }
138
139 /* butter fly op */
140 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
141 {\
142 FFTSample ax, ay, bx, by;\
143 bx=pre1;\
144 by=pim1;\
145 ax=qre1;\
146 ay=qim1;\
147 pre = (bx + ax);\
148 pim = (by + ay);\
149 qre = (bx - ax);\
150 qim = (by - ay);\
151 }
152
153 #define MUL16(a,b) ((a) * (b))
154
155 #define CMUL(pre, pim, are, aim, bre, bim) \
156 {\
157 pre = (MUL16(are, bre) - MUL16(aim, bim));\
158 pim = (MUL16(are, bim) + MUL16(bre, aim));\
159 }
160
161 /**
162 * Do a complex FFT with the parameters defined in ff_fft_init(). The
163 * input data must be permuted before with s->revtab table. No
164 * 1.0/sqrt(n) normalization is done.
165 */
166 void ff_fft_calc_c(FFTContext *s, FFTComplex *z)
167 {
168 int ln = s->nbits;
169 int j, np, np2;
170 int nblocks, nloops;
171 register FFTComplex *p, *q;
172 FFTComplex *exptab = s->exptab;
173 int l;
174 FFTSample tmp_re, tmp_im;
175
176 np = 1 << ln;
177
178 /* pass 0 */
179
180 p=&z[0];
181 j=(np >> 1);
182 do {
183 BF(p[0].re, p[0].im, p[1].re, p[1].im,
184 p[0].re, p[0].im, p[1].re, p[1].im);
185 p+=2;
186 } while (--j != 0);
187
188 /* pass 1 */
189
190
191 p=&z[0];
192 j=np >> 2;
193 if (s->inverse) {
194 do {
195 BF(p[0].re, p[0].im, p[2].re, p[2].im,
196 p[0].re, p[0].im, p[2].re, p[2].im);
197 BF(p[1].re, p[1].im, p[3].re, p[3].im,
198 p[1].re, p[1].im, -p[3].im, p[3].re);
199 p+=4;
200 } while (--j != 0);
201 } else {
202 do {
203 BF(p[0].re, p[0].im, p[2].re, p[2].im,
204 p[0].re, p[0].im, p[2].re, p[2].im);
205 BF(p[1].re, p[1].im, p[3].re, p[3].im,
206 p[1].re, p[1].im, p[3].im, -p[3].re);
207 p+=4;
208 } while (--j != 0);
209 }
210 /* pass 2 .. ln-1 */
211
212 nblocks = np >> 3;
213 nloops = 1 << 2;
214 np2 = np >> 1;
215 do {
216 p = z;
217 q = z + nloops;
218 for (j = 0; j < nblocks; ++j) {
219 BF(p->re, p->im, q->re, q->im,
220 p->re, p->im, q->re, q->im);
221
222 p++;
223 q++;
224 for(l = nblocks; l < np2; l += nblocks) {
225 CMUL(tmp_re, tmp_im, exptab[l].re, exptab[l].im, q->re, q->im);
226 BF(p->re, p->im, q->re, q->im,
227 p->re, p->im, tmp_re, tmp_im);
228 p++;
229 q++;
230 }
231
232 p += nloops;
233 q += nloops;
234 }
235 nblocks = nblocks >> 1;
236 nloops = nloops << 1;
237 } while (nblocks != 0);
238 }
239
240 /**
241 * Do the permutation needed BEFORE calling ff_fft_calc()
242 */
243 void ff_fft_permute(FFTContext *s, FFTComplex *z)
244 {
245 int j, k, np;
246 FFTComplex tmp;
247 const uint16_t *revtab = s->revtab;
248
249 /* reverse */
250 np = 1 << s->nbits;
251 for(j=0;j<np;j++) {
252 k = revtab[j];
253 if (k < j) {
254 tmp = z[k];
255 z[k] = z[j];
256 z[j] = tmp;
257 }
258 }
259 }
260
261 void ff_fft_end(FFTContext *s)
262 {
263 av_freep(&s->revtab);
264 av_freep(&s->exptab);
265 av_freep(&s->exptab1);
266 }
267