Statistics
| Branch: | Revision:

ffmpeg / libavcodec / fft.c @ 2912e87a

History | View | Annotate | Download (7.62 KB)

1
/*
2
 * FFT/IFFT transforms
3
 * Copyright (c) 2008 Loren Merritt
4
 * Copyright (c) 2002 Fabrice Bellard
5
 * Partly based on libdjbfft by D. J. Bernstein
6
 *
7
 * This file is part of Libav.
8
 *
9
 * Libav is free software; you can redistribute it and/or
10
 * modify it under the terms of the GNU Lesser General Public
11
 * License as published by the Free Software Foundation; either
12
 * version 2.1 of the License, or (at your option) any later version.
13
 *
14
 * Libav is distributed in the hope that it will be useful,
15
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17
 * Lesser General Public License for more details.
18
 *
19
 * You should have received a copy of the GNU Lesser General Public
20
 * License along with Libav; if not, write to the Free Software
21
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22
 */
23

    
24
/**
25
 * @file
26
 * FFT/IFFT transforms.
27
 */
28

    
29
#include <stdlib.h>
30
#include <string.h>
31
#include "libavutil/mathematics.h"
32
#include "fft.h"
33

    
34
/* cos(2*pi*x/n) for 0<=x<=n/4, followed by its reverse */
35
#if !CONFIG_HARDCODED_TABLES
36
COSTABLE(16);
37
COSTABLE(32);
38
COSTABLE(64);
39
COSTABLE(128);
40
COSTABLE(256);
41
COSTABLE(512);
42
COSTABLE(1024);
43
COSTABLE(2048);
44
COSTABLE(4096);
45
COSTABLE(8192);
46
COSTABLE(16384);
47
COSTABLE(32768);
48
COSTABLE(65536);
49
#endif
50
COSTABLE_CONST FFTSample * const ff_cos_tabs[] = {
51
    NULL, NULL, NULL, NULL,
52
    ff_cos_16, ff_cos_32, ff_cos_64, ff_cos_128, ff_cos_256, ff_cos_512, ff_cos_1024,
53
    ff_cos_2048, ff_cos_4096, ff_cos_8192, ff_cos_16384, ff_cos_32768, ff_cos_65536,
54
};
55

    
56
static void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
57
static void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
58

    
59
static int split_radix_permutation(int i, int n, int inverse)
60
{
61
    int m;
62
    if(n <= 2) return i&1;
63
    m = n >> 1;
64
    if(!(i&m))            return split_radix_permutation(i, m, inverse)*2;
65
    m >>= 1;
66
    if(inverse == !(i&m)) return split_radix_permutation(i, m, inverse)*4 + 1;
67
    else                  return split_radix_permutation(i, m, inverse)*4 - 1;
68
}
69

    
70
av_cold void ff_init_ff_cos_tabs(int index)
71
{
72
#if !CONFIG_HARDCODED_TABLES
73
    int i;
74
    int m = 1<<index;
75
    double freq = 2*M_PI/m;
76
    FFTSample *tab = ff_cos_tabs[index];
77
    for(i=0; i<=m/4; i++)
78
        tab[i] = cos(i*freq);
79
    for(i=1; i<m/4; i++)
80
        tab[m/2-i] = tab[i];
81
#endif
82
}
83

    
84
av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)
85
{
86
    int i, j, n;
87

    
88
    if (nbits < 2 || nbits > 16)
89
        goto fail;
90
    s->nbits = nbits;
91
    n = 1 << nbits;
92

    
93
    s->revtab = av_malloc(n * sizeof(uint16_t));
94
    if (!s->revtab)
95
        goto fail;
96
    s->tmp_buf = av_malloc(n * sizeof(FFTComplex));
97
    if (!s->tmp_buf)
98
        goto fail;
99
    s->inverse = inverse;
100
    s->fft_permutation = FF_FFT_PERM_DEFAULT;
101

    
102
    s->fft_permute = ff_fft_permute_c;
103
    s->fft_calc    = ff_fft_calc_c;
104
#if CONFIG_MDCT
105
    s->imdct_calc  = ff_imdct_calc_c;
106
    s->imdct_half  = ff_imdct_half_c;
107
    s->mdct_calc   = ff_mdct_calc_c;
108
#endif
109

    
110
    if (ARCH_ARM)     ff_fft_init_arm(s);
111
    if (HAVE_ALTIVEC) ff_fft_init_altivec(s);
112
    if (HAVE_MMX)     ff_fft_init_mmx(s);
113

    
114
    for(j=4; j<=nbits; j++) {
115
        ff_init_ff_cos_tabs(j);
116
    }
117
    for(i=0; i<n; i++) {
118
        int j = i;
119
        if (s->fft_permutation == FF_FFT_PERM_SWAP_LSBS)
120
            j = (j&~3) | ((j>>1)&1) | ((j<<1)&2);
121
        s->revtab[-split_radix_permutation(i, n, s->inverse) & (n-1)] = j;
122
    }
123

    
124
    return 0;
125
 fail:
126
    av_freep(&s->revtab);
127
    av_freep(&s->tmp_buf);
128
    return -1;
129
}
130

    
131
static void ff_fft_permute_c(FFTContext *s, FFTComplex *z)
132
{
133
    int j, np;
134
    const uint16_t *revtab = s->revtab;
135
    np = 1 << s->nbits;
136
    /* TODO: handle split-radix permute in a more optimal way, probably in-place */
137
    for(j=0;j<np;j++) s->tmp_buf[revtab[j]] = z[j];
138
    memcpy(z, s->tmp_buf, np * sizeof(FFTComplex));
139
}
140

    
141
av_cold void ff_fft_end(FFTContext *s)
142
{
143
    av_freep(&s->revtab);
144
    av_freep(&s->tmp_buf);
145
}
146

    
147
#define sqrthalf (float)M_SQRT1_2
148

    
149
#define BF(x,y,a,b) {\
150
    x = a - b;\
151
    y = a + b;\
152
}
153

    
154
#define BUTTERFLIES(a0,a1,a2,a3) {\
155
    BF(t3, t5, t5, t1);\
156
    BF(a2.re, a0.re, a0.re, t5);\
157
    BF(a3.im, a1.im, a1.im, t3);\
158
    BF(t4, t6, t2, t6);\
159
    BF(a3.re, a1.re, a1.re, t4);\
160
    BF(a2.im, a0.im, a0.im, t6);\
161
}
162

    
163
// force loading all the inputs before storing any.
164
// this is slightly slower for small data, but avoids store->load aliasing
165
// for addresses separated by large powers of 2.
166
#define BUTTERFLIES_BIG(a0,a1,a2,a3) {\
167
    FFTSample r0=a0.re, i0=a0.im, r1=a1.re, i1=a1.im;\
168
    BF(t3, t5, t5, t1);\
169
    BF(a2.re, a0.re, r0, t5);\
170
    BF(a3.im, a1.im, i1, t3);\
171
    BF(t4, t6, t2, t6);\
172
    BF(a3.re, a1.re, r1, t4);\
173
    BF(a2.im, a0.im, i0, t6);\
174
}
175

    
176
#define TRANSFORM(a0,a1,a2,a3,wre,wim) {\
177
    t1 = a2.re * wre + a2.im * wim;\
178
    t2 = a2.im * wre - a2.re * wim;\
179
    t5 = a3.re * wre - a3.im * wim;\
180
    t6 = a3.im * wre + a3.re * wim;\
181
    BUTTERFLIES(a0,a1,a2,a3)\
182
}
183

    
184
#define TRANSFORM_ZERO(a0,a1,a2,a3) {\
185
    t1 = a2.re;\
186
    t2 = a2.im;\
187
    t5 = a3.re;\
188
    t6 = a3.im;\
189
    BUTTERFLIES(a0,a1,a2,a3)\
190
}
191

    
192
/* z[0...8n-1], w[1...2n-1] */
193
#define PASS(name)\
194
static void name(FFTComplex *z, const FFTSample *wre, unsigned int n)\
195
{\
196
    FFTSample t1, t2, t3, t4, t5, t6;\
197
    int o1 = 2*n;\
198
    int o2 = 4*n;\
199
    int o3 = 6*n;\
200
    const FFTSample *wim = wre+o1;\
201
    n--;\
202
\
203
    TRANSFORM_ZERO(z[0],z[o1],z[o2],z[o3]);\
204
    TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
205
    do {\
206
        z += 2;\
207
        wre += 2;\
208
        wim -= 2;\
209
        TRANSFORM(z[0],z[o1],z[o2],z[o3],wre[0],wim[0]);\
210
        TRANSFORM(z[1],z[o1+1],z[o2+1],z[o3+1],wre[1],wim[-1]);\
211
    } while(--n);\
212
}
213

    
214
PASS(pass)
215
#undef BUTTERFLIES
216
#define BUTTERFLIES BUTTERFLIES_BIG
217
PASS(pass_big)
218

    
219
#define DECL_FFT(n,n2,n4)\
220
static void fft##n(FFTComplex *z)\
221
{\
222
    fft##n2(z);\
223
    fft##n4(z+n4*2);\
224
    fft##n4(z+n4*3);\
225
    pass(z,ff_cos_##n,n4/2);\
226
}
227

    
228
static void fft4(FFTComplex *z)
229
{
230
    FFTSample t1, t2, t3, t4, t5, t6, t7, t8;
231

    
232
    BF(t3, t1, z[0].re, z[1].re);
233
    BF(t8, t6, z[3].re, z[2].re);
234
    BF(z[2].re, z[0].re, t1, t6);
235
    BF(t4, t2, z[0].im, z[1].im);
236
    BF(t7, t5, z[2].im, z[3].im);
237
    BF(z[3].im, z[1].im, t4, t8);
238
    BF(z[3].re, z[1].re, t3, t7);
239
    BF(z[2].im, z[0].im, t2, t5);
240
}
241

    
242
static void fft8(FFTComplex *z)
243
{
244
    FFTSample t1, t2, t3, t4, t5, t6, t7, t8;
245

    
246
    fft4(z);
247

    
248
    BF(t1, z[5].re, z[4].re, -z[5].re);
249
    BF(t2, z[5].im, z[4].im, -z[5].im);
250
    BF(t3, z[7].re, z[6].re, -z[7].re);
251
    BF(t4, z[7].im, z[6].im, -z[7].im);
252
    BF(t8, t1, t3, t1);
253
    BF(t7, t2, t2, t4);
254
    BF(z[4].re, z[0].re, z[0].re, t1);
255
    BF(z[4].im, z[0].im, z[0].im, t2);
256
    BF(z[6].re, z[2].re, z[2].re, t7);
257
    BF(z[6].im, z[2].im, z[2].im, t8);
258

    
259
    TRANSFORM(z[1],z[3],z[5],z[7],sqrthalf,sqrthalf);
260
}
261

    
262
#if !CONFIG_SMALL
263
static void fft16(FFTComplex *z)
264
{
265
    FFTSample t1, t2, t3, t4, t5, t6;
266

    
267
    fft8(z);
268
    fft4(z+8);
269
    fft4(z+12);
270

    
271
    TRANSFORM_ZERO(z[0],z[4],z[8],z[12]);
272
    TRANSFORM(z[2],z[6],z[10],z[14],sqrthalf,sqrthalf);
273
    TRANSFORM(z[1],z[5],z[9],z[13],ff_cos_16[1],ff_cos_16[3]);
274
    TRANSFORM(z[3],z[7],z[11],z[15],ff_cos_16[3],ff_cos_16[1]);
275
}
276
#else
277
DECL_FFT(16,8,4)
278
#endif
279
DECL_FFT(32,16,8)
280
DECL_FFT(64,32,16)
281
DECL_FFT(128,64,32)
282
DECL_FFT(256,128,64)
283
DECL_FFT(512,256,128)
284
#if !CONFIG_SMALL
285
#define pass pass_big
286
#endif
287
DECL_FFT(1024,512,256)
288
DECL_FFT(2048,1024,512)
289
DECL_FFT(4096,2048,1024)
290
DECL_FFT(8192,4096,2048)
291
DECL_FFT(16384,8192,4096)
292
DECL_FFT(32768,16384,8192)
293
DECL_FFT(65536,32768,16384)
294

    
295
static void (* const fft_dispatch[])(FFTComplex*) = {
296
    fft4, fft8, fft16, fft32, fft64, fft128, fft256, fft512, fft1024,
297
    fft2048, fft4096, fft8192, fft16384, fft32768, fft65536,
298
};
299

    
300
static void ff_fft_calc_c(FFTContext *s, FFTComplex *z)
301
{
302
    fft_dispatch[s->nbits-2](z);
303
}
304