Mercurial > libavcodec.hg
annotate mdct.c @ 10199:38ab367d4231 libavcodec
Merge FFTContext and MDCTContext
author | mru |
---|---|
date | Sun, 20 Sep 2009 17:30:20 +0000 |
parents | 89cd870ca180 |
children | db033d1fbf44 |
rev | line source |
---|---|
781 | 1 /* |
2 * MDCT/IMDCT transforms | |
8629
04423b2f6e0b
cosmetics: Remove pointless period after copyright statement non-sentences.
diego
parents:
8567
diff
changeset
|
3 * Copyright (c) 2002 Fabrice Bellard |
781 | 4 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
5 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
6 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
7 * FFmpeg is free software; you can redistribute it and/or |
781 | 8 * modify it under the terms of the GNU Lesser General Public |
9 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
10 * version 2.1 of the License, or (at your option) any later version. |
781 | 11 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
12 * FFmpeg is distributed in the hope that it will be useful, |
781 | 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 | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
18 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
2967
diff
changeset
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
781 | 20 */ |
21 #include "dsputil.h" | |
22 | |
1106 | 23 /** |
8718
e9d9d946f213
Use full internal pathname in doxygen @file directives.
diego
parents:
8629
diff
changeset
|
24 * @file libavcodec/mdct.c |
1106 | 25 * MDCT/IMDCT transforms. |
26 */ | |
27 | |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
28 // Generate a Kaiser-Bessel Derived Window. |
6142
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
29 #define BESSEL_I0_ITER 50 // default: 50 iterations of Bessel I0 approximation |
8737
eeca2fc122f8
Add av_cold attributes to *_init and *_end functions.
alexc
parents:
8718
diff
changeset
|
30 av_cold void ff_kbd_window_init(float *window, float alpha, int n) |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
31 { |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
32 int i, j; |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
33 double sum = 0.0, bessel, tmp; |
6142
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
34 double local_window[n]; |
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
35 double alpha2 = (alpha * M_PI / n) * (alpha * M_PI / n); |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
36 |
6142
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
37 for (i = 0; i < n; i++) { |
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
38 tmp = i * (n - i) * alpha2; |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
39 bessel = 1.0; |
6142
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
40 for (j = BESSEL_I0_ITER; j > 0; j--) |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
41 bessel = bessel * tmp / (j * j) + 1; |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
42 sum += bessel; |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
43 local_window[i] = sum; |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
44 } |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
45 |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
46 sum++; |
6142
a35b838ab955
Add variable alpha and size of half window for Kaiser-Bessel Derived window
superdump
parents:
6139
diff
changeset
|
47 for (i = 0; i < n; i++) |
6139
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
48 window[i] = sqrt(local_window[i] / sum); |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
49 } |
5077d1562573
Make the Kaiser-Bessel window generator a common function
andoma
parents:
3947
diff
changeset
|
50 |
10174
89cd870ca180
Add two more sizes to ff_sine_windows[] and also pad it with NULLs so
vitor
parents:
10161
diff
changeset
|
51 DECLARE_ALIGNED(16, float, ff_sine_32 [ 32]); |
89cd870ca180
Add two more sizes to ff_sine_windows[] and also pad it with NULLs so
vitor
parents:
10161
diff
changeset
|
52 DECLARE_ALIGNED(16, float, ff_sine_64 [ 64]); |
7577
ed956c3c2cf3
The ff_sine_#[] should be aligned as they will commonly be used in dsputil
superdump
parents:
7573
diff
changeset
|
53 DECLARE_ALIGNED(16, float, ff_sine_128 [ 128]); |
ed956c3c2cf3
The ff_sine_#[] should be aligned as they will commonly be used in dsputil
superdump
parents:
7573
diff
changeset
|
54 DECLARE_ALIGNED(16, float, ff_sine_256 [ 256]); |
ed956c3c2cf3
The ff_sine_#[] should be aligned as they will commonly be used in dsputil
superdump
parents:
7573
diff
changeset
|
55 DECLARE_ALIGNED(16, float, ff_sine_512 [ 512]); |
ed956c3c2cf3
The ff_sine_#[] should be aligned as they will commonly be used in dsputil
superdump
parents:
7573
diff
changeset
|
56 DECLARE_ALIGNED(16, float, ff_sine_1024[1024]); |
ed956c3c2cf3
The ff_sine_#[] should be aligned as they will commonly be used in dsputil
superdump
parents:
7573
diff
changeset
|
57 DECLARE_ALIGNED(16, float, ff_sine_2048[2048]); |
8567 | 58 DECLARE_ALIGNED(16, float, ff_sine_4096[4096]); |
10174
89cd870ca180
Add two more sizes to ff_sine_windows[] and also pad it with NULLs so
vitor
parents:
10161
diff
changeset
|
59 float * const ff_sine_windows[] = { |
89cd870ca180
Add two more sizes to ff_sine_windows[] and also pad it with NULLs so
vitor
parents:
10161
diff
changeset
|
60 NULL, NULL, NULL, NULL, NULL, // unused |
89cd870ca180
Add two more sizes to ff_sine_windows[] and also pad it with NULLs so
vitor
parents:
10161
diff
changeset
|
61 ff_sine_32 , ff_sine_64 , |
8567 | 62 ff_sine_128, ff_sine_256, ff_sine_512, ff_sine_1024, ff_sine_2048, ff_sine_4096 |
7573
7802295cae6f
Add declarations for the sine tables used in wma.c (half window sizes: 128,
superdump
parents:
7547
diff
changeset
|
63 }; |
7802295cae6f
Add declarations for the sine tables used in wma.c (half window sizes: 128,
superdump
parents:
7547
diff
changeset
|
64 |
7094
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6498
diff
changeset
|
65 // Generate a sine window. |
8737
eeca2fc122f8
Add av_cold attributes to *_init and *_end functions.
alexc
parents:
8718
diff
changeset
|
66 av_cold void ff_sine_window_init(float *window, int n) { |
7094
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6498
diff
changeset
|
67 int i; |
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6498
diff
changeset
|
68 for(i = 0; i < n; i++) |
7822 | 69 window[i] = sinf((i + 0.5) * (M_PI / (2.0 * n))); |
7094
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6498
diff
changeset
|
70 } |
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6498
diff
changeset
|
71 |
1106 | 72 /** |
73 * init MDCT or IMDCT computation. | |
781 | 74 */ |
10199 | 75 av_cold int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale) |
781 | 76 { |
77 int n, n4, i; | |
9658
67a20f0eb42c
Support for getting (i)MDCT output multiplied by a constant scaling factor.
serge
parents:
9480
diff
changeset
|
78 double alpha, theta; |
781 | 79 |
80 memset(s, 0, sizeof(*s)); | |
81 n = 1 << nbits; | |
10199 | 82 s->mdct_bits = nbits; |
83 s->mdct_size = n; | |
781 | 84 n4 = n >> 2; |
970 | 85 s->tcos = av_malloc(n4 * sizeof(FFTSample)); |
781 | 86 if (!s->tcos) |
87 goto fail; | |
970 | 88 s->tsin = av_malloc(n4 * sizeof(FFTSample)); |
781 | 89 if (!s->tsin) |
90 goto fail; | |
91 | |
9658
67a20f0eb42c
Support for getting (i)MDCT output multiplied by a constant scaling factor.
serge
parents:
9480
diff
changeset
|
92 theta = 1.0 / 8.0 + (scale < 0 ? n4 : 0); |
67a20f0eb42c
Support for getting (i)MDCT output multiplied by a constant scaling factor.
serge
parents:
9480
diff
changeset
|
93 scale = sqrt(fabs(scale)); |
781 | 94 for(i=0;i<n4;i++) { |
9658
67a20f0eb42c
Support for getting (i)MDCT output multiplied by a constant scaling factor.
serge
parents:
9480
diff
changeset
|
95 alpha = 2 * M_PI * (i + theta) / n; |
67a20f0eb42c
Support for getting (i)MDCT output multiplied by a constant scaling factor.
serge
parents:
9480
diff
changeset
|
96 s->tcos[i] = -cos(alpha) * scale; |
67a20f0eb42c
Support for getting (i)MDCT output multiplied by a constant scaling factor.
serge
parents:
9480
diff
changeset
|
97 s->tsin[i] = -sin(alpha) * scale; |
781 | 98 } |
10199 | 99 if (ff_fft_init(s, s->mdct_bits - 2, inverse) < 0) |
781 | 100 goto fail; |
101 return 0; | |
102 fail: | |
103 av_freep(&s->tcos); | |
104 av_freep(&s->tsin); | |
105 return -1; | |
106 } | |
107 | |
108 /* complex multiplication: p = a * b */ | |
109 #define CMUL(pre, pim, are, aim, bre, bim) \ | |
110 {\ | |
7545 | 111 FFTSample _are = (are);\ |
112 FFTSample _aim = (aim);\ | |
113 FFTSample _bre = (bre);\ | |
114 FFTSample _bim = (bim);\ | |
781 | 115 (pre) = _are * _bre - _aim * _bim;\ |
116 (pim) = _are * _bim + _aim * _bre;\ | |
117 } | |
118 | |
7544 | 119 /** |
120 * Compute the middle half of the inverse MDCT of size N = 2^nbits, | |
121 * thus excluding the parts that can be derived by symmetry | |
122 * @param output N/2 samples | |
123 * @param input N/2 samples | |
124 */ | |
10199 | 125 void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input) |
781 | 126 { |
7544 | 127 int k, n8, n4, n2, n, j; |
10199 | 128 const uint16_t *revtab = s->revtab; |
781 | 129 const FFTSample *tcos = s->tcos; |
130 const FFTSample *tsin = s->tsin; | |
131 const FFTSample *in1, *in2; | |
7544 | 132 FFTComplex *z = (FFTComplex *)output; |
781 | 133 |
10199 | 134 n = 1 << s->mdct_bits; |
781 | 135 n2 = n >> 1; |
136 n4 = n >> 2; | |
7544 | 137 n8 = n >> 3; |
781 | 138 |
139 /* pre rotation */ | |
140 in1 = input; | |
141 in2 = input + n2 - 1; | |
142 for(k = 0; k < n4; k++) { | |
143 j=revtab[k]; | |
144 CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]); | |
145 in1 += 2; | |
146 in2 -= 2; | |
147 } | |
10199 | 148 ff_fft_calc(s, z); |
781 | 149 |
150 /* post rotation + reordering */ | |
7544 | 151 for(k = 0; k < n8; k++) { |
152 FFTSample r0, i0, r1, i1; | |
153 CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]); | |
154 CMUL(r1, i0, z[n8+k ].im, z[n8+k ].re, tsin[n8+k ], tcos[n8+k ]); | |
155 z[n8-k-1].re = r0; | |
156 z[n8-k-1].im = i0; | |
157 z[n8+k ].re = r1; | |
158 z[n8+k ].im = i1; | |
781 | 159 } |
7263 | 160 } |
161 | |
162 /** | |
163 * Compute inverse MDCT of size N = 2^nbits | |
164 * @param output N samples | |
165 * @param input N/2 samples | |
166 */ | |
10199 | 167 void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input) |
7263 | 168 { |
7544 | 169 int k; |
10199 | 170 int n = 1 << s->mdct_bits; |
7544 | 171 int n2 = n >> 1; |
172 int n4 = n >> 2; | |
781 | 173 |
7547 | 174 ff_imdct_half_c(s, output+n4, input); |
7263 | 175 |
7544 | 176 for(k = 0; k < n4; k++) { |
177 output[k] = -output[n2-k-1]; | |
178 output[n-k-1] = output[n2+k]; | |
7263 | 179 } |
180 } | |
181 | |
182 /** | |
781 | 183 * Compute MDCT of size N = 2^nbits |
184 * @param input N samples | |
185 * @param out N/2 samples | |
186 */ | |
10199 | 187 void ff_mdct_calc_c(FFTContext *s, FFTSample *out, const FFTSample *input) |
781 | 188 { |
189 int i, j, n, n8, n4, n2, n3; | |
7546 | 190 FFTSample re, im; |
10199 | 191 const uint16_t *revtab = s->revtab; |
781 | 192 const FFTSample *tcos = s->tcos; |
193 const FFTSample *tsin = s->tsin; | |
7544 | 194 FFTComplex *x = (FFTComplex *)out; |
781 | 195 |
10199 | 196 n = 1 << s->mdct_bits; |
781 | 197 n2 = n >> 1; |
198 n4 = n >> 2; | |
199 n8 = n >> 3; | |
200 n3 = 3 * n4; | |
201 | |
202 /* pre rotation */ | |
203 for(i=0;i<n8;i++) { | |
204 re = -input[2*i+3*n4] - input[n3-1-2*i]; | |
205 im = -input[n4+2*i] + input[n4-1-2*i]; | |
206 j = revtab[i]; | |
207 CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]); | |
208 | |
209 re = input[2*i] - input[n2-1-2*i]; | |
210 im = -(input[n2+2*i] + input[n-1-2*i]); | |
211 j = revtab[n8 + i]; | |
212 CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]); | |
213 } | |
214 | |
10199 | 215 ff_fft_calc(s, x); |
2967 | 216 |
781 | 217 /* post rotation */ |
7544 | 218 for(i=0;i<n8;i++) { |
219 FFTSample r0, i0, r1, i1; | |
220 CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]); | |
221 CMUL(i0, r1, x[n8+i ].re, x[n8+i ].im, -tsin[n8+i ], -tcos[n8+i ]); | |
222 x[n8-i-1].re = r0; | |
223 x[n8-i-1].im = i0; | |
224 x[n8+i ].re = r1; | |
225 x[n8+i ].im = i1; | |
781 | 226 } |
227 } | |
228 | |
10199 | 229 av_cold void ff_mdct_end(FFTContext *s) |
781 | 230 { |
231 av_freep(&s->tcos); | |
232 av_freep(&s->tsin); | |
10199 | 233 ff_fft_end(s); |
781 | 234 } |