daffffb936c06c99a0aa6de7df8a6e690ddc8062
[libav.git] / libavcodec / mdct.c
1 /*
2 * MDCT/IMDCT transforms
3 * Copyright (c) 2002 Fabrice Bellard
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21 #include "dsputil.h"
22
23 /**
24 * @file libavcodec/mdct.c
25 * MDCT/IMDCT transforms.
26 */
27
28 // Generate a Kaiser-Bessel Derived Window.
29 #define BESSEL_I0_ITER 50 // default: 50 iterations of Bessel I0 approximation
30 av_cold void ff_kbd_window_init(float *window, float alpha, int n)
31 {
32 int i, j;
33 double sum = 0.0, bessel, tmp;
34 double local_window[n];
35 double alpha2 = (alpha * M_PI / n) * (alpha * M_PI / n);
36
37 for (i = 0; i < n; i++) {
38 tmp = i * (n - i) * alpha2;
39 bessel = 1.0;
40 for (j = BESSEL_I0_ITER; j > 0; j--)
41 bessel = bessel * tmp / (j * j) + 1;
42 sum += bessel;
43 local_window[i] = sum;
44 }
45
46 sum++;
47 for (i = 0; i < n; i++)
48 window[i] = sqrt(local_window[i] / sum);
49 }
50
51 DECLARE_ALIGNED(16, float, ff_sine_128 [ 128]);
52 DECLARE_ALIGNED(16, float, ff_sine_256 [ 256]);
53 DECLARE_ALIGNED(16, float, ff_sine_512 [ 512]);
54 DECLARE_ALIGNED(16, float, ff_sine_1024[1024]);
55 DECLARE_ALIGNED(16, float, ff_sine_2048[2048]);
56 DECLARE_ALIGNED(16, float, ff_sine_4096[4096]);
57 float *ff_sine_windows[6] = {
58 ff_sine_128, ff_sine_256, ff_sine_512, ff_sine_1024, ff_sine_2048, ff_sine_4096
59 };
60
61 // Generate a sine window.
62 av_cold void ff_sine_window_init(float *window, int n) {
63 int i;
64 for(i = 0; i < n; i++)
65 window[i] = sinf((i + 0.5) * (M_PI / (2.0 * n)));
66 }
67
68 /**
69 * init MDCT or IMDCT computation.
70 */
71 av_cold int ff_mdct_init(MDCTContext *s, int nbits, int inverse)
72 {
73 int n, n4, i;
74 double alpha;
75
76 memset(s, 0, sizeof(*s));
77 n = 1 << nbits;
78 s->nbits = nbits;
79 s->n = n;
80 n4 = n >> 2;
81 s->tcos = av_malloc(n4 * sizeof(FFTSample));
82 if (!s->tcos)
83 goto fail;
84 s->tsin = av_malloc(n4 * sizeof(FFTSample));
85 if (!s->tsin)
86 goto fail;
87
88 for(i=0;i<n4;i++) {
89 alpha = 2 * M_PI * (i + 1.0 / 8.0) / n;
90 s->tcos[i] = -cos(alpha);
91 s->tsin[i] = -sin(alpha);
92 }
93 if (ff_fft_init(&s->fft, s->nbits - 2, inverse) < 0)
94 goto fail;
95 return 0;
96 fail:
97 av_freep(&s->tcos);
98 av_freep(&s->tsin);
99 return -1;
100 }
101
102 /* complex multiplication: p = a * b */
103 #define CMUL(pre, pim, are, aim, bre, bim) \
104 {\
105 FFTSample _are = (are);\
106 FFTSample _aim = (aim);\
107 FFTSample _bre = (bre);\
108 FFTSample _bim = (bim);\
109 (pre) = _are * _bre - _aim * _bim;\
110 (pim) = _are * _bim + _aim * _bre;\
111 }
112
113 /**
114 * Compute the middle half of the inverse MDCT of size N = 2^nbits,
115 * thus excluding the parts that can be derived by symmetry
116 * @param output N/2 samples
117 * @param input N/2 samples
118 */
119 void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
120 {
121 int k, n8, n4, n2, n, j;
122 const uint16_t *revtab = s->fft.revtab;
123 const FFTSample *tcos = s->tcos;
124 const FFTSample *tsin = s->tsin;
125 const FFTSample *in1, *in2;
126 FFTComplex *z = (FFTComplex *)output;
127
128 n = 1 << s->nbits;
129 n2 = n >> 1;
130 n4 = n >> 2;
131 n8 = n >> 3;
132
133 /* pre rotation */
134 in1 = input;
135 in2 = input + n2 - 1;
136 for(k = 0; k < n4; k++) {
137 j=revtab[k];
138 CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]);
139 in1 += 2;
140 in2 -= 2;
141 }
142 ff_fft_calc(&s->fft, z);
143
144 /* post rotation + reordering */
145 output += n4;
146 for(k = 0; k < n8; k++) {
147 FFTSample r0, i0, r1, i1;
148 CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]);
149 CMUL(r1, i0, z[n8+k ].im, z[n8+k ].re, tsin[n8+k ], tcos[n8+k ]);
150 z[n8-k-1].re = r0;
151 z[n8-k-1].im = i0;
152 z[n8+k ].re = r1;
153 z[n8+k ].im = i1;
154 }
155 }
156
157 /**
158 * Compute inverse MDCT of size N = 2^nbits
159 * @param output N samples
160 * @param input N/2 samples
161 * @param tmp N/2 samples
162 */
163 void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
164 {
165 int k;
166 int n = 1 << s->nbits;
167 int n2 = n >> 1;
168 int n4 = n >> 2;
169
170 ff_imdct_half_c(s, output+n4, input);
171
172 for(k = 0; k < n4; k++) {
173 output[k] = -output[n2-k-1];
174 output[n-k-1] = output[n2+k];
175 }
176 }
177
178 /**
179 * Compute MDCT of size N = 2^nbits
180 * @param input N samples
181 * @param out N/2 samples
182 * @param tmp temporary storage of N/2 samples
183 */
184 void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input)
185 {
186 int i, j, n, n8, n4, n2, n3;
187 FFTSample re, im;
188 const uint16_t *revtab = s->fft.revtab;
189 const FFTSample *tcos = s->tcos;
190 const FFTSample *tsin = s->tsin;
191 FFTComplex *x = (FFTComplex *)out;
192
193 n = 1 << s->nbits;
194 n2 = n >> 1;
195 n4 = n >> 2;
196 n8 = n >> 3;
197 n3 = 3 * n4;
198
199 /* pre rotation */
200 for(i=0;i<n8;i++) {
201 re = -input[2*i+3*n4] - input[n3-1-2*i];
202 im = -input[n4+2*i] + input[n4-1-2*i];
203 j = revtab[i];
204 CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]);
205
206 re = input[2*i] - input[n2-1-2*i];
207 im = -(input[n2+2*i] + input[n-1-2*i]);
208 j = revtab[n8 + i];
209 CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]);
210 }
211
212 ff_fft_calc(&s->fft, x);
213
214 /* post rotation */
215 for(i=0;i<n8;i++) {
216 FFTSample r0, i0, r1, i1;
217 CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]);
218 CMUL(i0, r1, x[n8+i ].re, x[n8+i ].im, -tsin[n8+i ], -tcos[n8+i ]);
219 x[n8-i-1].re = r0;
220 x[n8-i-1].im = i0;
221 x[n8+i ].re = r1;
222 x[n8+i ].im = i1;
223 }
224 }
225
226 av_cold void ff_mdct_end(MDCTContext *s)
227 {
228 av_freep(&s->tcos);
229 av_freep(&s->tsin);
230 ff_fft_end(&s->fft);
231 }