Mercurial > libavcodec.hg
annotate mdct.c @ 11352:6e0af2cfdcfe libavcodec
Do MC and IDCT in coding (hilbert) order
This increases the slice size to 64 pixels, due to having to decode an
entire chroma superblock row per slice.
This can be up to 6% slower depending on clip and CPU, but is necessary
for future optimizations that gain significantly more than was lost.
author | conrad |
---|---|
date | Wed, 03 Mar 2010 23:27:40 +0000 |
parents | 3d011a01a6a0 |
children | 4b3da727d832 |
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 |
10827
3d011a01a6a0
Add support for hard-coded MDCT-related ff_sine_windows tables.
reimar
parents:
10204
diff
changeset
|
51 #include "mdct_tablegen.h" |
7094
b0820b8bd4dd
Add generic ff_sine_window_init function and implement in codecs appropriately
superdump
parents:
6498
diff
changeset
|
52 |
1106 | 53 /** |
54 * init MDCT or IMDCT computation. | |
781 | 55 */ |
10199 | 56 av_cold int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale) |
781 | 57 { |
58 int n, n4, i; | |
9658
67a20f0eb42c
Support for getting (i)MDCT output multiplied by a constant scaling factor.
serge
parents:
9480
diff
changeset
|
59 double alpha, theta; |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
60 int tstep; |
781 | 61 |
62 memset(s, 0, sizeof(*s)); | |
63 n = 1 << nbits; | |
10199 | 64 s->mdct_bits = nbits; |
65 s->mdct_size = n; | |
781 | 66 n4 = n >> 2; |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
67 s->permutation = FF_MDCT_PERM_NONE; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
68 |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
69 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
|
70 goto fail; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
71 |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
72 s->tcos = av_malloc(n/2 * sizeof(FFTSample)); |
781 | 73 if (!s->tcos) |
74 goto fail; | |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
75 |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
76 switch (s->permutation) { |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
77 case FF_MDCT_PERM_NONE: |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
78 s->tsin = s->tcos + n4; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
79 tstep = 1; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
80 break; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
81 case FF_MDCT_PERM_INTERLEAVE: |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
82 s->tsin = s->tcos + 1; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
83 tstep = 2; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
84 break; |
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
85 default: |
781 | 86 goto fail; |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
87 } |
781 | 88 |
9658
67a20f0eb42c
Support for getting (i)MDCT output multiplied by a constant scaling factor.
serge
parents:
9480
diff
changeset
|
89 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
|
90 scale = sqrt(fabs(scale)); |
781 | 91 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
|
92 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
|
93 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
|
94 s->tsin[i*tstep] = -sin(alpha) * scale; |
781 | 95 } |
96 return 0; | |
97 fail: | |
10204
db033d1fbf44
Allow arch-specific mdct code to request interleaving of cos/sin tables
mru
parents:
10199
diff
changeset
|
98 ff_mdct_end(s); |
781 | 99 return -1; |
100 } | |
101 | |
102 /* complex multiplication: p = a * b */ | |
103 #define CMUL(pre, pim, are, aim, bre, bim) \ | |
104 {\ | |
7545 | 105 FFTSample _are = (are);\ |
106 FFTSample _aim = (aim);\ | |
107 FFTSample _bre = (bre);\ | |
108 FFTSample _bim = (bim);\ | |
781 | 109 (pre) = _are * _bre - _aim * _bim;\ |
110 (pim) = _are * _bim + _aim * _bre;\ | |
111 } | |
112 | |
7544 | 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 */ | |
10199 | 119 void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input) |
781 | 120 { |
7544 | 121 int k, n8, n4, n2, n, j; |
10199 | 122 const uint16_t *revtab = s->revtab; |
781 | 123 const FFTSample *tcos = s->tcos; |
124 const FFTSample *tsin = s->tsin; | |
125 const FFTSample *in1, *in2; | |
7544 | 126 FFTComplex *z = (FFTComplex *)output; |
781 | 127 |
10199 | 128 n = 1 << s->mdct_bits; |
781 | 129 n2 = n >> 1; |
130 n4 = n >> 2; | |
7544 | 131 n8 = n >> 3; |
781 | 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 } | |
10199 | 142 ff_fft_calc(s, z); |
781 | 143 |
144 /* post rotation + reordering */ | |
7544 | 145 for(k = 0; k < n8; k++) { |
146 FFTSample r0, i0, r1, i1; | |
147 CMUL(r0, i1, z[n8-k-1].im, z[n8-k-1].re, tsin[n8-k-1], tcos[n8-k-1]); | |
148 CMUL(r1, i0, z[n8+k ].im, z[n8+k ].re, tsin[n8+k ], tcos[n8+k ]); | |
149 z[n8-k-1].re = r0; | |
150 z[n8-k-1].im = i0; | |
151 z[n8+k ].re = r1; | |
152 z[n8+k ].im = i1; | |
781 | 153 } |
7263 | 154 } |
155 | |
156 /** | |
157 * Compute inverse MDCT of size N = 2^nbits | |
158 * @param output N samples | |
159 * @param input N/2 samples | |
160 */ | |
10199 | 161 void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input) |
7263 | 162 { |
7544 | 163 int k; |
10199 | 164 int n = 1 << s->mdct_bits; |
7544 | 165 int n2 = n >> 1; |
166 int n4 = n >> 2; | |
781 | 167 |
7547 | 168 ff_imdct_half_c(s, output+n4, input); |
7263 | 169 |
7544 | 170 for(k = 0; k < n4; k++) { |
171 output[k] = -output[n2-k-1]; | |
172 output[n-k-1] = output[n2+k]; | |
7263 | 173 } |
174 } | |
175 | |
176 /** | |
781 | 177 * Compute MDCT of size N = 2^nbits |
178 * @param input N samples | |
179 * @param out N/2 samples | |
180 */ | |
10199 | 181 void ff_mdct_calc_c(FFTContext *s, FFTSample *out, const FFTSample *input) |
781 | 182 { |
183 int i, j, n, n8, n4, n2, n3; | |
7546 | 184 FFTSample re, im; |
10199 | 185 const uint16_t *revtab = s->revtab; |
781 | 186 const FFTSample *tcos = s->tcos; |
187 const FFTSample *tsin = s->tsin; | |
7544 | 188 FFTComplex *x = (FFTComplex *)out; |
781 | 189 |
10199 | 190 n = 1 << s->mdct_bits; |
781 | 191 n2 = n >> 1; |
192 n4 = n >> 2; | |
193 n8 = n >> 3; | |
194 n3 = 3 * n4; | |
195 | |
196 /* pre rotation */ | |
197 for(i=0;i<n8;i++) { | |
198 re = -input[2*i+3*n4] - input[n3-1-2*i]; | |
199 im = -input[n4+2*i] + input[n4-1-2*i]; | |
200 j = revtab[i]; | |
201 CMUL(x[j].re, x[j].im, re, im, -tcos[i], tsin[i]); | |
202 | |
203 re = input[2*i] - input[n2-1-2*i]; | |
204 im = -(input[n2+2*i] + input[n-1-2*i]); | |
205 j = revtab[n8 + i]; | |
206 CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]); | |
207 } | |
208 | |
10199 | 209 ff_fft_calc(s, x); |
2967 | 210 |
781 | 211 /* post rotation */ |
7544 | 212 for(i=0;i<n8;i++) { |
213 FFTSample r0, i0, r1, i1; | |
214 CMUL(i1, r0, x[n8-i-1].re, x[n8-i-1].im, -tsin[n8-i-1], -tcos[n8-i-1]); | |
215 CMUL(i0, r1, x[n8+i ].re, x[n8+i ].im, -tsin[n8+i ], -tcos[n8+i ]); | |
216 x[n8-i-1].re = r0; | |
217 x[n8-i-1].im = i0; | |
218 x[n8+i ].re = r1; | |
219 x[n8+i ].im = i1; | |
781 | 220 } |
221 } | |
222 | |
10199 | 223 av_cold void ff_mdct_end(FFTContext *s) |
781 | 224 { |
225 av_freep(&s->tcos); | |
10199 | 226 ff_fft_end(s); |
781 | 227 } |