Mercurial > libavcodec.hg
annotate mdct.c @ 10553:86c7e3c6de00 libavcodec
Update libtheora wrapper to use the 1.0 API
author | conrad |
---|---|
date | Sun, 22 Nov 2009 21:08:37 +0000 |
parents | db033d1fbf44 |
children | 3d011a01a6a0 |
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; |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
79 int tstep; |
781 | 80 |
81 memset(s, 0, sizeof(*s)); | |
82 n = 1 << nbits; | |
10199 | 83 s->mdct_bits = nbits; |
84 s->mdct_size = n; | |
781 | 85 n4 = n >> 2; |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
86 s->permutation = FF_MDCT_PERM_NONE; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
87 |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
88 if (ff_fft_init(s, s->mdct_bits - 2, inverse) < 0) |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
89 goto fail; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
90 |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
91 s->tcos = av_malloc(n/2 * sizeof(FFTSample)); |
781 | 92 if (!s->tcos) |
93 goto fail; | |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
94 |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
95 switch (s->permutation) { |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
96 case FF_MDCT_PERM_NONE: |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
97 s->tsin = s->tcos + n4; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
98 tstep = 1; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
99 break; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
100 case FF_MDCT_PERM_INTERLEAVE: |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
101 s->tsin = s->tcos + 1; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
102 tstep = 2; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
103 break; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
104 default: |
781 | 105 goto fail; |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
106 } |
781 | 107 |
9658
67a20f0eb42c
Support for getting (i)MDCT output multiplied by a constant scaling factor.
serge
parents:
9480
diff
changeset
|
108 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
|
109 scale = sqrt(fabs(scale)); |
781 | 110 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
|
111 alpha = 2 * M_PI * (i + theta) / n; |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
112 s->tcos[i*tstep] = -cos(alpha) * scale; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
113 s->tsin[i*tstep] = -sin(alpha) * scale; |
781 | 114 } |
115 return 0; | |
116 fail: | |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
117 ff_mdct_end(s); |
781 | 118 return -1; |
119 } | |
120 | |
121 /* complex multiplication: p = a * b */ | |
122 #define CMUL(pre, pim, are, aim, bre, bim) \ | |
123 {\ | |
7545 | 124 FFTSample _are = (are);\ |
125 FFTSample _aim = (aim);\ | |
126 FFTSample _bre = (bre);\ | |
127 FFTSample _bim = (bim);\ | |
781 | 128 (pre) = _are * _bre - _aim * _bim;\ |
129 (pim) = _are * _bim + _aim * _bre;\ | |
130 } | |
131 | |
7544 | 132 /** |
133 * Compute the middle half of the inverse MDCT of size N = 2^nbits, | |
134 * thus excluding the parts that can be derived by symmetry | |
135 * @param output N/2 samples | |
136 * @param input N/2 samples | |
137 */ | |
10199 | 138 void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input) |
781 | 139 { |
7544 | 140 int k, n8, n4, n2, n, j; |
10199 | 141 const uint16_t *revtab = s->revtab; |
781 | 142 const FFTSample *tcos = s->tcos; |
143 const FFTSample *tsin = s->tsin; | |
144 const FFTSample *in1, *in2; | |
7544 | 145 FFTComplex *z = (FFTComplex *)output; |
781 | 146 |
10199 | 147 n = 1 << s->mdct_bits; |
781 | 148 n2 = n >> 1; |
149 n4 = n >> 2; | |
7544 | 150 n8 = n >> 3; |
781 | 151 |
152 /* pre rotation */ | |
153 in1 = input; | |
154 in2 = input + n2 - 1; | |
155 for(k = 0; k < n4; k++) { | |
156 j=revtab[k]; | |
157 CMUL(z[j].re, z[j].im, *in2, *in1, tcos[k], tsin[k]); | |
158 in1 += 2; | |
159 in2 -= 2; | |
160 } | |
10199 | 161 ff_fft_calc(s, z); |
781 | 162 |
163 /* post rotation + reordering */ | |
7544 | 164 for(k = 0; k < n8; k++) { |
165 FFTSample r0, i0, r1, i1; | |
166 CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]); | |
167 CMUL(r1, i0, z[n8+k ].im, z[n8+k ].re, tsin[n8+k ], tcos[n8+k ]); | |
168 z[n8-k-1].re = r0; | |
169 z[n8-k-1].im = i0; | |
170 z[n8+k ].re = r1; | |
171 z[n8+k ].im = i1; | |
781 | 172 } |
7263 | 173 } |
174 | |
175 /** | |
176 * Compute inverse MDCT of size N = 2^nbits | |
177 * @param output N samples | |
178 * @param input N/2 samples | |
179 */ | |
10199 | 180 void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input) |
7263 | 181 { |
7544 | 182 int k; |
10199 | 183 int n = 1 << s->mdct_bits; |
7544 | 184 int n2 = n >> 1; |
185 int n4 = n >> 2; | |
781 | 186 |
7547 | 187 ff_imdct_half_c(s, output+n4, input); |
7263 | 188 |
7544 | 189 for(k = 0; k < n4; k++) { |
190 output[k] = -output[n2-k-1]; | |
191 output[n-k-1] = output[n2+k]; | |
7263 | 192 } |
193 } | |
194 | |
195 /** | |
781 | 196 * Compute MDCT of size N = 2^nbits |
197 * @param input N samples | |
198 * @param out N/2 samples | |
199 */ | |
10199 | 200 void ff_mdct_calc_c(FFTContext *s, FFTSample *out, const FFTSample *input) |
781 | 201 { |
202 int i, j, n, n8, n4, n2, n3; | |
7546 | 203 FFTSample re, im; |
10199 | 204 const uint16_t *revtab = s->revtab; |
781 | 205 const FFTSample *tcos = s->tcos; |
206 const FFTSample *tsin = s->tsin; | |
7544 | 207 FFTComplex *x = (FFTComplex *)out; |
781 | 208 |
10199 | 209 n = 1 << s->mdct_bits; |
781 | 210 n2 = n >> 1; |
211 n4 = n >> 2; | |
212 n8 = n >> 3; | |
213 n3 = 3 * n4; | |
214 | |
215 /* pre rotation */ | |
216 for(i=0;i<n8;i++) { | |
217 re = -input[2*i+3*n4] - input[n3-1-2*i]; | |
218 im = -input[n4+2*i] + input[n4-1-2*i]; | |
219 j = revtab[i]; | |
220 CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]); | |
221 | |
222 re = input[2*i] - input[n2-1-2*i]; | |
223 im = -(input[n2+2*i] + input[n-1-2*i]); | |
224 j = revtab[n8 + i]; | |
225 CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]); | |
226 } | |
227 | |
10199 | 228 ff_fft_calc(s, x); |
2967 | 229 |
781 | 230 /* post rotation */ |
7544 | 231 for(i=0;i<n8;i++) { |
232 FFTSample r0, i0, r1, i1; | |
233 CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]); | |
234 CMUL(i0, r1, x[n8+i ].re, x[n8+i ].im, -tsin[n8+i ], -tcos[n8+i ]); | |
235 x[n8-i-1].re = r0; | |
236 x[n8-i-1].im = i0; | |
237 x[n8+i ].re = r1; | |
238 x[n8+i ].im = i1; | |
781 | 239 } |
240 } | |
241 | |
10199 | 242 av_cold void ff_mdct_end(FFTContext *s) |
781 | 243 { |
244 av_freep(&s->tcos); | |
10199 | 245 ff_fft_end(s); |
781 | 246 } |