FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
opus_celt.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2012 Andrew D'Addesio
3  * Copyright (c) 2013-2014 Mozilla Corporation
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 
22 /**
23  * @file
24  * Opus CELT decoder
25  */
26 
27 #include <stdint.h>
28 
29 #include "libavutil/float_dsp.h"
30 #include "libavutil/libm.h"
31 
32 #include "imdct15.h"
33 #include "opus.h"
34 
35 enum CeltSpread {
40 };
41 
42 typedef struct CeltFrame {
45 
47 
48  /* buffer for mdct output + postfilter */
49  DECLARE_ALIGNED(32, float, buf)[2048];
50 
51  /* postfilter parameters */
53  float pf_gains_new[3];
54  int pf_period;
55  float pf_gains[3];
57  float pf_gains_old[3];
58 
59  float deemph_coeff;
60 } CeltFrame;
61 
62 struct CeltContext {
63  // constant values that do not change during context lifetime
68 
69  // values that have inter-frame effect and must be reset on flush
71  uint32_t seed;
72  int flushed;
73 
74  // values that only affect a single frame
76  int framebits;
77  int duration;
78 
79  /* number of iMDCT blocks in the frame */
80  int blocks;
81  /* size of each block */
82  int blocksize;
83 
84  int startband;
85  int endband;
87 
89 
93 
94  int remaining;
100 
102  DECLARE_ALIGNED(32, float, scratch)[22 * 8]; // MAX(celt_freq_range) * 1<<CELT_MAX_LOG_BLOCKS
103 };
104 
105 static const uint16_t celt_model_tapset[] = { 4, 2, 3, 4 };
106 
107 static const uint16_t celt_model_spread[] = { 32, 7, 9, 30, 32 };
108 
109 static const uint16_t celt_model_alloc_trim[] = {
110  128, 2, 4, 9, 19, 41, 87, 109, 119, 124, 126, 128
111 };
112 
113 static const uint16_t celt_model_energy_small[] = { 4, 2, 3, 4 };
114 
115 static const uint8_t celt_freq_bands[] = { /* in steps of 200Hz */
116  0, 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, 28, 34, 40, 48, 60, 78, 100
117 };
118 
119 static const uint8_t celt_freq_range[] = {
120  1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 4, 4, 4, 6, 6, 8, 12, 18, 22
121 };
122 
123 static const uint8_t celt_log_freq_range[] = {
124  0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 16, 16, 16, 21, 21, 24, 29, 34, 36
125 };
126 
127 static const int8_t celt_tf_select[4][2][2][2] = {
128  { { { 0, -1 }, { 0, -1 } }, { { 0, -1 }, { 0, -1 } } },
129  { { { 0, -1 }, { 0, -2 } }, { { 1, 0 }, { 1, -1 } } },
130  { { { 0, -2 }, { 0, -3 } }, { { 2, 0 }, { 1, -1 } } },
131  { { { 0, -2 }, { 0, -3 } }, { { 3, 0 }, { 1, -1 } } }
132 };
133 
134 static const float celt_mean_energy[] = {
135  6.437500f, 6.250000f, 5.750000f, 5.312500f, 5.062500f,
136  4.812500f, 4.500000f, 4.375000f, 4.875000f, 4.687500f,
137  4.562500f, 4.437500f, 4.875000f, 4.625000f, 4.312500f,
138  4.500000f, 4.375000f, 4.625000f, 4.750000f, 4.437500f,
139  3.750000f, 3.750000f, 3.750000f, 3.750000f, 3.750000f
140 };
141 
142 static const float celt_alpha_coef[] = {
143  29440.0f/32768.0f, 26112.0f/32768.0f, 21248.0f/32768.0f, 16384.0f/32768.0f
144 };
145 
146 static const float celt_beta_coef[] = { /* TODO: precompute 1 minus this if the code ends up neater */
147  30147.0f/32768.0f, 22282.0f/32768.0f, 12124.0f/32768.0f, 6554.0f/32768.0f
148 };
149 
150 static const uint8_t celt_coarse_energy_dist[4][2][42] = {
151  {
152  { // 120-sample inter
153  72, 127, 65, 129, 66, 128, 65, 128, 64, 128, 62, 128, 64, 128,
154  64, 128, 92, 78, 92, 79, 92, 78, 90, 79, 116, 41, 115, 40,
155  114, 40, 132, 26, 132, 26, 145, 17, 161, 12, 176, 10, 177, 11
156  }, { // 120-sample intra
157  24, 179, 48, 138, 54, 135, 54, 132, 53, 134, 56, 133, 55, 132,
158  55, 132, 61, 114, 70, 96, 74, 88, 75, 88, 87, 74, 89, 66,
159  91, 67, 100, 59, 108, 50, 120, 40, 122, 37, 97, 43, 78, 50
160  }
161  }, {
162  { // 240-sample inter
163  83, 78, 84, 81, 88, 75, 86, 74, 87, 71, 90, 73, 93, 74,
164  93, 74, 109, 40, 114, 36, 117, 34, 117, 34, 143, 17, 145, 18,
165  146, 19, 162, 12, 165, 10, 178, 7, 189, 6, 190, 8, 177, 9
166  }, { // 240-sample intra
167  23, 178, 54, 115, 63, 102, 66, 98, 69, 99, 74, 89, 71, 91,
168  73, 91, 78, 89, 86, 80, 92, 66, 93, 64, 102, 59, 103, 60,
169  104, 60, 117, 52, 123, 44, 138, 35, 133, 31, 97, 38, 77, 45
170  }
171  }, {
172  { // 480-sample inter
173  61, 90, 93, 60, 105, 42, 107, 41, 110, 45, 116, 38, 113, 38,
174  112, 38, 124, 26, 132, 27, 136, 19, 140, 20, 155, 14, 159, 16,
175  158, 18, 170, 13, 177, 10, 187, 8, 192, 6, 175, 9, 159, 10
176  }, { // 480-sample intra
177  21, 178, 59, 110, 71, 86, 75, 85, 84, 83, 91, 66, 88, 73,
178  87, 72, 92, 75, 98, 72, 105, 58, 107, 54, 115, 52, 114, 55,
179  112, 56, 129, 51, 132, 40, 150, 33, 140, 29, 98, 35, 77, 42
180  }
181  }, {
182  { // 960-sample inter
183  42, 121, 96, 66, 108, 43, 111, 40, 117, 44, 123, 32, 120, 36,
184  119, 33, 127, 33, 134, 34, 139, 21, 147, 23, 152, 20, 158, 25,
185  154, 26, 166, 21, 173, 16, 184, 13, 184, 10, 150, 13, 139, 15
186  }, { // 960-sample intra
187  22, 178, 63, 114, 74, 82, 84, 83, 92, 82, 103, 62, 96, 72,
188  96, 67, 101, 73, 107, 72, 113, 55, 118, 52, 125, 52, 118, 52,
189  117, 55, 135, 49, 137, 39, 157, 32, 145, 29, 97, 33, 77, 40
190  }
191  }
192 };
193 
194 static const uint8_t celt_static_alloc[11][21] = { /* 1/32 bit/sample */
195  { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
196  { 90, 80, 75, 69, 63, 56, 49, 40, 34, 29, 20, 18, 10, 0, 0, 0, 0, 0, 0, 0, 0 },
197  { 110, 100, 90, 84, 78, 71, 65, 58, 51, 45, 39, 32, 26, 20, 12, 0, 0, 0, 0, 0, 0 },
198  { 118, 110, 103, 93, 86, 80, 75, 70, 65, 59, 53, 47, 40, 31, 23, 15, 4, 0, 0, 0, 0 },
199  { 126, 119, 112, 104, 95, 89, 83, 78, 72, 66, 60, 54, 47, 39, 32, 25, 17, 12, 1, 0, 0 },
200  { 134, 127, 120, 114, 103, 97, 91, 85, 78, 72, 66, 60, 54, 47, 41, 35, 29, 23, 16, 10, 1 },
201  { 144, 137, 130, 124, 113, 107, 101, 95, 88, 82, 76, 70, 64, 57, 51, 45, 39, 33, 26, 15, 1 },
202  { 152, 145, 138, 132, 123, 117, 111, 105, 98, 92, 86, 80, 74, 67, 61, 55, 49, 43, 36, 20, 1 },
203  { 162, 155, 148, 142, 133, 127, 121, 115, 108, 102, 96, 90, 84, 77, 71, 65, 59, 53, 46, 30, 1 },
204  { 172, 165, 158, 152, 143, 137, 131, 125, 118, 112, 106, 100, 94, 87, 81, 75, 69, 63, 56, 45, 20 },
205  { 200, 200, 200, 200, 200, 200, 200, 200, 198, 193, 188, 183, 178, 173, 168, 163, 158, 153, 148, 129, 104 }
206 };
207 
208 static const uint8_t celt_static_caps[4][2][21] = {
209  { // 120-sample
210  {224, 224, 224, 224, 224, 224, 224, 224, 160, 160,
211  160, 160, 185, 185, 185, 178, 178, 168, 134, 61, 37},
212  {224, 224, 224, 224, 224, 224, 224, 224, 240, 240,
213  240, 240, 207, 207, 207, 198, 198, 183, 144, 66, 40},
214  }, { // 240-sample
215  {160, 160, 160, 160, 160, 160, 160, 160, 185, 185,
216  185, 185, 193, 193, 193, 183, 183, 172, 138, 64, 38},
217  {240, 240, 240, 240, 240, 240, 240, 240, 207, 207,
218  207, 207, 204, 204, 204, 193, 193, 180, 143, 66, 40},
219  }, { // 480-sample
220  {185, 185, 185, 185, 185, 185, 185, 185, 193, 193,
221  193, 193, 193, 193, 193, 183, 183, 172, 138, 65, 39},
222  {207, 207, 207, 207, 207, 207, 207, 207, 204, 204,
223  204, 204, 201, 201, 201, 188, 188, 176, 141, 66, 40},
224  }, { // 960-sample
225  {193, 193, 193, 193, 193, 193, 193, 193, 193, 193,
226  193, 193, 194, 194, 194, 184, 184, 173, 139, 65, 39},
227  {204, 204, 204, 204, 204, 204, 204, 204, 201, 201,
228  201, 201, 198, 198, 198, 187, 187, 175, 140, 66, 40}
229  }
230 };
231 
232 static const uint8_t celt_cache_bits[392] = {
233  40, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
234  7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
235  7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 40, 15, 23, 28,
236  31, 34, 36, 38, 39, 41, 42, 43, 44, 45, 46, 47, 47, 49, 50,
237  51, 52, 53, 54, 55, 55, 57, 58, 59, 60, 61, 62, 63, 63, 65,
238  66, 67, 68, 69, 70, 71, 71, 40, 20, 33, 41, 48, 53, 57, 61,
239  64, 66, 69, 71, 73, 75, 76, 78, 80, 82, 85, 87, 89, 91, 92,
240  94, 96, 98, 101, 103, 105, 107, 108, 110, 112, 114, 117, 119, 121, 123,
241  124, 126, 128, 40, 23, 39, 51, 60, 67, 73, 79, 83, 87, 91, 94,
242  97, 100, 102, 105, 107, 111, 115, 118, 121, 124, 126, 129, 131, 135, 139,
243  142, 145, 148, 150, 153, 155, 159, 163, 166, 169, 172, 174, 177, 179, 35,
244  28, 49, 65, 78, 89, 99, 107, 114, 120, 126, 132, 136, 141, 145, 149,
245  153, 159, 165, 171, 176, 180, 185, 189, 192, 199, 205, 211, 216, 220, 225,
246  229, 232, 239, 245, 251, 21, 33, 58, 79, 97, 112, 125, 137, 148, 157,
247  166, 174, 182, 189, 195, 201, 207, 217, 227, 235, 243, 251, 17, 35, 63,
248  86, 106, 123, 139, 152, 165, 177, 187, 197, 206, 214, 222, 230, 237, 250,
249  25, 31, 55, 75, 91, 105, 117, 128, 138, 146, 154, 161, 168, 174, 180,
250  185, 190, 200, 208, 215, 222, 229, 235, 240, 245, 255, 16, 36, 65, 89,
251  110, 128, 144, 159, 173, 185, 196, 207, 217, 226, 234, 242, 250, 11, 41,
252  74, 103, 128, 151, 172, 191, 209, 225, 241, 255, 9, 43, 79, 110, 138,
253  163, 186, 207, 227, 246, 12, 39, 71, 99, 123, 144, 164, 182, 198, 214,
254  228, 241, 253, 9, 44, 81, 113, 142, 168, 192, 214, 235, 255, 7, 49,
255  90, 127, 160, 191, 220, 247, 6, 51, 95, 134, 170, 203, 234, 7, 47,
256  87, 123, 155, 184, 212, 237, 6, 52, 97, 137, 174, 208, 240, 5, 57,
257  106, 151, 192, 231, 5, 59, 111, 158, 202, 243, 5, 55, 103, 147, 187,
258  224, 5, 60, 113, 161, 206, 248, 4, 65, 122, 175, 224, 4, 67, 127,
259  182, 234
260 };
261 
262 static const int16_t celt_cache_index[105] = {
263  -1, -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 41, 41, 41,
264  82, 82, 123, 164, 200, 222, 0, 0, 0, 0, 0, 0, 0, 0, 41,
265  41, 41, 41, 123, 123, 123, 164, 164, 240, 266, 283, 295, 41, 41, 41,
266  41, 41, 41, 41, 41, 123, 123, 123, 123, 240, 240, 240, 266, 266, 305,
267  318, 328, 336, 123, 123, 123, 123, 123, 123, 123, 123, 240, 240, 240, 240,
268  305, 305, 305, 318, 318, 343, 351, 358, 364, 240, 240, 240, 240, 240, 240,
269  240, 240, 305, 305, 305, 305, 343, 343, 343, 351, 351, 370, 376, 382, 387,
270 };
271 
272 static const uint8_t celt_log2_frac[] = {
273  0, 8, 13, 16, 19, 21, 23, 24, 26, 27, 28, 29, 30, 31, 32, 32, 33, 34, 34, 35, 36, 36, 37, 37
274 };
275 
276 static const uint8_t celt_bit_interleave[] = {
277  0, 1, 1, 1, 2, 3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3
278 };
279 
280 static const uint8_t celt_bit_deinterleave[] = {
281  0x00, 0x03, 0x0C, 0x0F, 0x30, 0x33, 0x3C, 0x3F,
282  0xC0, 0xC3, 0xCC, 0xCF, 0xF0, 0xF3, 0xFC, 0xFF
283 };
284 
285 static const uint8_t celt_hadamard_ordery[] = {
286  1, 0,
287  3, 0, 2, 1,
288  7, 0, 4, 3, 6, 1, 5, 2,
289  15, 0, 8, 7, 12, 3, 11, 4, 14, 1, 9, 6, 13, 2, 10, 5
290 };
291 
292 static const uint16_t celt_qn_exp2[] = {
293  16384, 17866, 19483, 21247, 23170, 25267, 27554, 30048
294 };
295 
296 static const uint32_t celt_pvq_u[1272] = {
297  /* N = 0, K = 0...176 */
298  1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
299  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
300  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
301  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
302  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
303  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
304  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
305  /* N = 1, K = 1...176 */
306  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
307  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
308  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
309  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
310  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
311  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
312  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
313  /* N = 2, K = 2...176 */
314  3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31, 33, 35, 37, 39, 41,
315  43, 45, 47, 49, 51, 53, 55, 57, 59, 61, 63, 65, 67, 69, 71, 73, 75, 77, 79,
316  81, 83, 85, 87, 89, 91, 93, 95, 97, 99, 101, 103, 105, 107, 109, 111, 113,
317  115, 117, 119, 121, 123, 125, 127, 129, 131, 133, 135, 137, 139, 141, 143,
318  145, 147, 149, 151, 153, 155, 157, 159, 161, 163, 165, 167, 169, 171, 173,
319  175, 177, 179, 181, 183, 185, 187, 189, 191, 193, 195, 197, 199, 201, 203,
320  205, 207, 209, 211, 213, 215, 217, 219, 221, 223, 225, 227, 229, 231, 233,
321  235, 237, 239, 241, 243, 245, 247, 249, 251, 253, 255, 257, 259, 261, 263,
322  265, 267, 269, 271, 273, 275, 277, 279, 281, 283, 285, 287, 289, 291, 293,
323  295, 297, 299, 301, 303, 305, 307, 309, 311, 313, 315, 317, 319, 321, 323,
324  325, 327, 329, 331, 333, 335, 337, 339, 341, 343, 345, 347, 349, 351,
325  /* N = 3, K = 3...176 */
326  13, 25, 41, 61, 85, 113, 145, 181, 221, 265, 313, 365, 421, 481, 545, 613,
327  685, 761, 841, 925, 1013, 1105, 1201, 1301, 1405, 1513, 1625, 1741, 1861,
328  1985, 2113, 2245, 2381, 2521, 2665, 2813, 2965, 3121, 3281, 3445, 3613, 3785,
329  3961, 4141, 4325, 4513, 4705, 4901, 5101, 5305, 5513, 5725, 5941, 6161, 6385,
330  6613, 6845, 7081, 7321, 7565, 7813, 8065, 8321, 8581, 8845, 9113, 9385, 9661,
331  9941, 10225, 10513, 10805, 11101, 11401, 11705, 12013, 12325, 12641, 12961,
332  13285, 13613, 13945, 14281, 14621, 14965, 15313, 15665, 16021, 16381, 16745,
333  17113, 17485, 17861, 18241, 18625, 19013, 19405, 19801, 20201, 20605, 21013,
334  21425, 21841, 22261, 22685, 23113, 23545, 23981, 24421, 24865, 25313, 25765,
335  26221, 26681, 27145, 27613, 28085, 28561, 29041, 29525, 30013, 30505, 31001,
336  31501, 32005, 32513, 33025, 33541, 34061, 34585, 35113, 35645, 36181, 36721,
337  37265, 37813, 38365, 38921, 39481, 40045, 40613, 41185, 41761, 42341, 42925,
338  43513, 44105, 44701, 45301, 45905, 46513, 47125, 47741, 48361, 48985, 49613,
339  50245, 50881, 51521, 52165, 52813, 53465, 54121, 54781, 55445, 56113, 56785,
340  57461, 58141, 58825, 59513, 60205, 60901, 61601,
341  /* N = 4, K = 4...176 */
342  63, 129, 231, 377, 575, 833, 1159, 1561, 2047, 2625, 3303, 4089, 4991, 6017,
343  7175, 8473, 9919, 11521, 13287, 15225, 17343, 19649, 22151, 24857, 27775,
344  30913, 34279, 37881, 41727, 45825, 50183, 54809, 59711, 64897, 70375, 76153,
345  82239, 88641, 95367, 102425, 109823, 117569, 125671, 134137, 142975, 152193,
346  161799, 171801, 182207, 193025, 204263, 215929, 228031, 240577, 253575,
347  267033, 280959, 295361, 310247, 325625, 341503, 357889, 374791, 392217,
348  410175, 428673, 447719, 467321, 487487, 508225, 529543, 551449, 573951,
349  597057, 620775, 645113, 670079, 695681, 721927, 748825, 776383, 804609,
350  833511, 863097, 893375, 924353, 956039, 988441, 1021567, 1055425, 1090023,
351  1125369, 1161471, 1198337, 1235975, 1274393, 1313599, 1353601, 1394407,
352  1436025, 1478463, 1521729, 1565831, 1610777, 1656575, 1703233, 1750759,
353  1799161, 1848447, 1898625, 1949703, 2001689, 2054591, 2108417, 2163175,
354  2218873, 2275519, 2333121, 2391687, 2451225, 2511743, 2573249, 2635751,
355  2699257, 2763775, 2829313, 2895879, 2963481, 3032127, 3101825, 3172583,
356  3244409, 3317311, 3391297, 3466375, 3542553, 3619839, 3698241, 3777767,
357  3858425, 3940223, 4023169, 4107271, 4192537, 4278975, 4366593, 4455399,
358  4545401, 4636607, 4729025, 4822663, 4917529, 5013631, 5110977, 5209575,
359  5309433, 5410559, 5512961, 5616647, 5721625, 5827903, 5935489, 6044391,
360  6154617, 6266175, 6379073, 6493319, 6608921, 6725887, 6844225, 6963943,
361  7085049, 7207551,
362  /* N = 5, K = 5...176 */
363  321, 681, 1289, 2241, 3649, 5641, 8361, 11969, 16641, 22569, 29961, 39041,
364  50049, 63241, 78889, 97281, 118721, 143529, 172041, 204609, 241601, 283401,
365  330409, 383041, 441729, 506921, 579081, 658689, 746241, 842249, 947241,
366  1061761, 1186369, 1321641, 1468169, 1626561, 1797441, 1981449, 2179241,
367  2391489, 2618881, 2862121, 3121929, 3399041, 3694209, 4008201, 4341801,
368  4695809, 5071041, 5468329, 5888521, 6332481, 6801089, 7295241, 7815849,
369  8363841, 8940161, 9545769, 10181641, 10848769, 11548161, 12280841, 13047849,
370  13850241, 14689089, 15565481, 16480521, 17435329, 18431041, 19468809,
371  20549801, 21675201, 22846209, 24064041, 25329929, 26645121, 28010881,
372  29428489, 30899241, 32424449, 34005441, 35643561, 37340169, 39096641,
373  40914369, 42794761, 44739241, 46749249, 48826241, 50971689, 53187081,
374  55473921, 57833729, 60268041, 62778409, 65366401, 68033601, 70781609,
375  73612041, 76526529, 79526721, 82614281, 85790889, 89058241, 92418049,
376  95872041, 99421961, 103069569, 106816641, 110664969, 114616361, 118672641,
377  122835649, 127107241, 131489289, 135983681, 140592321, 145317129, 150160041,
378  155123009, 160208001, 165417001, 170752009, 176215041, 181808129, 187533321,
379  193392681, 199388289, 205522241, 211796649, 218213641, 224775361, 231483969,
380  238341641, 245350569, 252512961, 259831041, 267307049, 274943241, 282741889,
381  290705281, 298835721, 307135529, 315607041, 324252609, 333074601, 342075401,
382  351257409, 360623041, 370174729, 379914921, 389846081, 399970689, 410291241,
383  420810249, 431530241, 442453761, 453583369, 464921641, 476471169, 488234561,
384  500214441, 512413449, 524834241, 537479489, 550351881, 563454121, 576788929,
385  590359041, 604167209, 618216201, 632508801,
386  /* N = 6, K = 6...96 (technically V(109,5) fits in 32 bits, but that can't be
387  achieved by splitting an Opus band) */
388  1683, 3653, 7183, 13073, 22363, 36365, 56695, 85305, 124515, 177045, 246047,
389  335137, 448427, 590557, 766727, 982729, 1244979, 1560549, 1937199, 2383409,
390  2908411, 3522221, 4235671, 5060441, 6009091, 7095093, 8332863, 9737793,
391  11326283, 13115773, 15124775, 17372905, 19880915, 22670725, 25765455,
392  29189457, 32968347, 37129037, 41699767, 46710137, 52191139, 58175189,
393  64696159, 71789409, 79491819, 87841821, 96879431, 106646281, 117185651,
394  128542501, 140763503, 153897073, 167993403, 183104493, 199284183, 216588185,
395  235074115, 254801525, 275831935, 298228865, 322057867, 347386557, 374284647,
396  402823977, 433078547, 465124549, 499040399, 534906769, 572806619, 612825229,
397  655050231, 699571641, 746481891, 795875861, 847850911, 902506913, 959946283,
398  1020274013, 1083597703, 1150027593, 1219676595, 1292660325, 1369097135,
399  1449108145, 1532817275, 1620351277, 1711839767, 1807415257, 1907213187,
400  2011371957, 2120032959,
401  /* N = 7, K = 7...54 (technically V(60,6) fits in 32 bits, but that can't be
402  achieved by splitting an Opus band) */
403  8989, 19825, 40081, 75517, 134245, 227305, 369305, 579125, 880685, 1303777,
404  1884961, 2668525, 3707509, 5064793, 6814249, 9041957, 11847485, 15345233,
405  19665841, 24957661, 31388293, 39146185, 48442297, 59511829, 72616013,
406  88043969, 106114625, 127178701, 151620757, 179861305, 212358985, 249612805,
407  292164445, 340600625, 395555537, 457713341, 527810725, 606639529, 695049433,
408  793950709, 904317037, 1027188385, 1163673953, 1314955181, 1482288821,
409  1667010073, 1870535785, 2094367717,
410  /* N = 8, K = 8...37 (technically V(40,7) fits in 32 bits, but that can't be
411  achieved by splitting an Opus band) */
412  48639, 108545, 224143, 433905, 795455, 1392065, 2340495, 3800305, 5984767,
413  9173505, 13726991, 20103025, 28875327, 40754369, 56610575, 77500017,
414  104692735, 139703809, 184327311, 240673265, 311207743, 398796225, 506750351,
415  638878193, 799538175, 993696769, 1226990095, 1505789553, 1837271615,
416  2229491905,
417  /* N = 9, K = 9...28 (technically V(29,8) fits in 32 bits, but that can't be
418  achieved by splitting an Opus band) */
419  265729, 598417, 1256465, 2485825, 4673345, 8405905, 14546705, 24331777,
420  39490049, 62390545, 96220561, 145198913, 214828609, 312193553, 446304145,
421  628496897, 872893441, 1196924561, 1621925137, 2173806145,
422  /* N = 10, K = 10...24 */
423  1462563, 3317445, 7059735, 14218905, 27298155, 50250765, 89129247, 152951073,
424  254831667, 413442773, 654862247, 1014889769, 1541911931, 2300409629,
425  3375210671,
426  /* N = 11, K = 11...19 (technically V(20,10) fits in 32 bits, but that can't be
427  achieved by splitting an Opus band) */
428  8097453, 18474633, 39753273, 81270333, 158819253, 298199265, 540279585,
429  948062325, 1616336765,
430  /* N = 12, K = 12...18 */
431  45046719, 103274625, 224298231, 464387817, 921406335, 1759885185,
432  3248227095,
433  /* N = 13, K = 13...16 */
434  251595969, 579168825, 1267854873, 2653649025,
435  /* N = 14, K = 14 */
436  1409933619
437 };
438 
439 DECLARE_ALIGNED(32, static const float, celt_window)[120] = {
440  6.7286966e-05f, 0.00060551348f, 0.0016815970f, 0.0032947962f, 0.0054439943f,
441  0.0081276923f, 0.011344001f, 0.015090633f, 0.019364886f, 0.024163635f,
442  0.029483315f, 0.035319905f, 0.041668911f, 0.048525347f, 0.055883718f,
443  0.063737999f, 0.072081616f, 0.080907428f, 0.090207705f, 0.099974111f,
444  0.11019769f, 0.12086883f, 0.13197729f, 0.14351214f, 0.15546177f,
445  0.16781389f, 0.18055550f, 0.19367290f, 0.20715171f, 0.22097682f,
446  0.23513243f, 0.24960208f, 0.26436860f, 0.27941419f, 0.29472040f,
447  0.31026818f, 0.32603788f, 0.34200931f, 0.35816177f, 0.37447407f,
448  0.39092462f, 0.40749142f, 0.42415215f, 0.44088423f, 0.45766484f,
449  0.47447104f, 0.49127978f, 0.50806798f, 0.52481261f, 0.54149077f,
450  0.55807973f, 0.57455701f, 0.59090049f, 0.60708841f, 0.62309951f,
451  0.63891306f, 0.65450896f, 0.66986776f, 0.68497077f, 0.69980010f,
452  0.71433873f, 0.72857055f, 0.74248043f, 0.75605424f, 0.76927895f,
453  0.78214257f, 0.79463430f, 0.80674445f, 0.81846456f, 0.82978733f,
454  0.84070669f, 0.85121779f, 0.86131698f, 0.87100183f, 0.88027111f,
455  0.88912479f, 0.89756398f, 0.90559094f, 0.91320904f, 0.92042270f,
456  0.92723738f, 0.93365955f, 0.93969656f, 0.94535671f, 0.95064907f,
457  0.95558353f, 0.96017067f, 0.96442171f, 0.96834849f, 0.97196334f,
458  0.97527906f, 0.97830883f, 0.98106616f, 0.98356480f, 0.98581869f,
459  0.98784191f, 0.98964856f, 0.99125274f, 0.99266849f, 0.99390969f,
460  0.99499004f, 0.99592297f, 0.99672162f, 0.99739874f, 0.99796667f,
461  0.99843728f, 0.99882195f, 0.99913147f, 0.99937606f, 0.99956527f,
462  0.99970802f, 0.99981248f, 0.99988613f, 0.99993565f, 0.99996697f,
463  0.99998518f, 0.99999457f, 0.99999859f, 0.99999982f, 1.0000000f,
464 };
465 
466 /* square of the window, used for the postfilter */
467 const float ff_celt_window2[120] = {
468  4.5275357e-09f, 3.66647e-07f, 2.82777e-06f, 1.08557e-05f, 2.96371e-05f, 6.60594e-05f,
469  0.000128686f, 0.000227727f, 0.000374999f, 0.000583881f, 0.000869266f, 0.0012475f,
470  0.0017363f, 0.00235471f, 0.00312299f, 0.00406253f, 0.00519576f, 0.00654601f,
471  0.00813743f, 0.00999482f, 0.0121435f, 0.0146093f, 0.017418f, 0.0205957f, 0.0241684f,
472  0.0281615f, 0.0326003f, 0.0375092f, 0.0429118f, 0.0488308f, 0.0552873f, 0.0623012f,
473  0.0698908f, 0.0780723f, 0.0868601f, 0.0962664f, 0.106301f, 0.11697f, 0.12828f,
474  0.140231f, 0.152822f, 0.166049f, 0.179905f, 0.194379f, 0.209457f, 0.225123f, 0.241356f,
475  0.258133f, 0.275428f, 0.293212f, 0.311453f, 0.330116f, 0.349163f, 0.368556f, 0.388253f,
476  0.40821f, 0.428382f, 0.448723f, 0.469185f, 0.48972f, 0.51028f, 0.530815f, 0.551277f,
477  0.571618f, 0.59179f, 0.611747f, 0.631444f, 0.650837f, 0.669884f, 0.688547f, 0.706788f,
478  0.724572f, 0.741867f, 0.758644f, 0.774877f, 0.790543f, 0.805621f, 0.820095f, 0.833951f,
479  0.847178f, 0.859769f, 0.87172f, 0.88303f, 0.893699f, 0.903734f, 0.91314f, 0.921928f,
480  0.930109f, 0.937699f, 0.944713f, 0.951169f, 0.957088f, 0.962491f, 0.9674f, 0.971838f,
481  0.975832f, 0.979404f, 0.982582f, 0.985391f, 0.987857f, 0.990005f, 0.991863f, 0.993454f,
482  0.994804f, 0.995937f, 0.996877f, 0.997645f, 0.998264f, 0.998753f, 0.999131f, 0.999416f,
483  0.999625f, 0.999772f, 0.999871f, 0.999934f, 0.99997f, 0.999989f, 0.999997f, 0.99999964f, 1.0f,
484 };
485 
486 static const uint32_t * const celt_pvq_u_row[15] = {
487  celt_pvq_u + 0, celt_pvq_u + 176, celt_pvq_u + 351,
488  celt_pvq_u + 525, celt_pvq_u + 698, celt_pvq_u + 870,
489  celt_pvq_u + 1041, celt_pvq_u + 1131, celt_pvq_u + 1178,
490  celt_pvq_u + 1207, celt_pvq_u + 1226, celt_pvq_u + 1240,
491  celt_pvq_u + 1248, celt_pvq_u + 1254, celt_pvq_u + 1257
492 };
493 
494 static inline int16_t celt_cos(int16_t x)
495 {
496  x = (MUL16(x, x) + 4096) >> 13;
497  x = (32767-x) + ROUND_MUL16(x, (-7651 + ROUND_MUL16(x, (8277 + ROUND_MUL16(-626, x)))));
498  return 1+x;
499 }
500 
501 static inline int celt_log2tan(int isin, int icos)
502 {
503  int lc, ls;
504  lc = opus_ilog(icos);
505  ls = opus_ilog(isin);
506  icos <<= 15 - lc;
507  isin <<= 15 - ls;
508  return (ls << 11) - (lc << 11) +
509  ROUND_MUL16(isin, ROUND_MUL16(isin, -2597) + 7932) -
510  ROUND_MUL16(icos, ROUND_MUL16(icos, -2597) + 7932);
511 }
512 
513 static inline uint32_t celt_rng(CeltContext *s)
514 {
515  s->seed = 1664525 * s->seed + 1013904223;
516  return s->seed;
517 }
518 
520 {
521  int i, j;
522  float prev[2] = {0};
523  float alpha, beta;
524  const uint8_t *model;
525 
526  /* use the 2D z-transform to apply prediction in both */
527  /* the time domain (alpha) and the frequency domain (beta) */
528 
529  if (opus_rc_tell(rc)+3 <= s->framebits && opus_rc_p2model(rc, 3)) {
530  /* intra frame */
531  alpha = 0;
532  beta = 1.0f - 4915.0f/32768.0f;
533  model = celt_coarse_energy_dist[s->duration][1];
534  } else {
535  alpha = celt_alpha_coef[s->duration];
536  beta = 1.0f - celt_beta_coef[s->duration];
537  model = celt_coarse_energy_dist[s->duration][0];
538  }
539 
540  for (i = 0; i < CELT_MAX_BANDS; i++) {
541  for (j = 0; j < s->coded_channels; j++) {
542  CeltFrame *frame = &s->frame[j];
543  float value;
544  int available;
545 
546  if (i < s->startband || i >= s->endband) {
547  frame->energy[i] = 0.0;
548  continue;
549  }
550 
551  available = s->framebits - opus_rc_tell(rc);
552  if (available >= 15) {
553  /* decode using a Laplace distribution */
554  int k = FFMIN(i, 20) << 1;
555  value = opus_rc_laplace(rc, model[k] << 7, model[k+1] << 6);
556  } else if (available >= 2) {
558  value = (x>>1) ^ -(x&1);
559  } else if (available >= 1) {
560  value = -(float)opus_rc_p2model(rc, 1);
561  } else value = -1;
562 
563  frame->energy[i] = FFMAX(-9.0f, frame->energy[i]) * alpha + prev[j] + value;
564  prev[j] += beta * value;
565  }
566  }
567 }
568 
570 {
571  int i;
572  for (i = s->startband; i < s->endband; i++) {
573  int j;
574  if (!s->fine_bits[i])
575  continue;
576 
577  for (j = 0; j < s->coded_channels; j++) {
578  CeltFrame *frame = &s->frame[j];
579  int q2;
580  float offset;
581  q2 = opus_getrawbits(rc, s->fine_bits[i]);
582  offset = (q2 + 0.5f) * (1 << (14 - s->fine_bits[i])) / 16384.0f - 0.5f;
583  frame->energy[i] += offset;
584  }
585  }
586 }
587 
589  int bits_left)
590 {
591  int priority, i, j;
592 
593  for (priority = 0; priority < 2; priority++) {
594  for (i = s->startband; i < s->endband && bits_left >= s->coded_channels; i++) {
595  if (s->fine_priority[i] != priority || s->fine_bits[i] >= CELT_MAX_FINE_BITS)
596  continue;
597 
598  for (j = 0; j < s->coded_channels; j++) {
599  int q2;
600  float offset;
601  q2 = opus_getrawbits(rc, 1);
602  offset = (q2 - 0.5f) * (1 << (14 - s->fine_bits[i] - 1)) / 16384.0f;
603  s->frame[j].energy[i] += offset;
604  bits_left--;
605  }
606  }
607  }
608 }
609 
611  int transient)
612 {
613  int i, diff = 0, tf_select = 0, tf_changed = 0, tf_select_bit;
614  int consumed, bits = transient ? 2 : 4;
615 
616  consumed = opus_rc_tell(rc);
617  tf_select_bit = (s->duration != 0 && consumed+bits+1 <= s->framebits);
618 
619  for (i = s->startband; i < s->endband; i++) {
620  if (consumed+bits+tf_select_bit <= s->framebits) {
621  diff ^= opus_rc_p2model(rc, bits);
622  consumed = opus_rc_tell(rc);
623  tf_changed |= diff;
624  }
625  s->tf_change[i] = diff;
626  bits = transient ? 4 : 5;
627  }
628 
629  if (tf_select_bit && celt_tf_select[s->duration][transient][0][tf_changed] !=
630  celt_tf_select[s->duration][transient][1][tf_changed])
631  tf_select = opus_rc_p2model(rc, 1);
632 
633  for (i = s->startband; i < s->endband; i++) {
634  s->tf_change[i] = celt_tf_select[s->duration][transient][tf_select][s->tf_change[i]];
635  }
636 }
637 
639 {
640  // approx. maximum bit allocation for each band before boost/trim
641  int cap[CELT_MAX_BANDS];
642  int boost[CELT_MAX_BANDS];
643  int threshold[CELT_MAX_BANDS];
644  int bits1[CELT_MAX_BANDS];
645  int bits2[CELT_MAX_BANDS];
646  int trim_offset[CELT_MAX_BANDS];
647 
648  int skip_startband = s->startband;
649  int dynalloc = 6;
650  int alloctrim = 5;
651  int extrabits = 0;
652 
653  int skip_bit = 0;
654  int intensitystereo_bit = 0;
655  int dualstereo_bit = 0;
656 
657  int remaining, bandbits;
658  int low, high, total, done;
659  int totalbits;
660  int consumed;
661  int i, j;
662 
663  consumed = opus_rc_tell(rc);
664 
665  /* obtain spread flag */
667  if (consumed + 4 <= s->framebits)
669 
670  /* generate static allocation caps */
671  for (i = 0; i < CELT_MAX_BANDS; i++) {
672  cap[i] = (celt_static_caps[s->duration][s->coded_channels - 1][i] + 64)
673  * celt_freq_range[i] << (s->coded_channels - 1) << s->duration >> 2;
674  }
675 
676  /* obtain band boost */
677  totalbits = s->framebits << 3; // convert to 1/8 bits
678  consumed = opus_rc_tell_frac(rc);
679  for (i = s->startband; i < s->endband; i++) {
680  int quanta, band_dynalloc;
681 
682  boost[i] = 0;
683 
684  quanta = celt_freq_range[i] << (s->coded_channels - 1) << s->duration;
685  quanta = FFMIN(quanta << 3, FFMAX(6 << 3, quanta));
686  band_dynalloc = dynalloc;
687  while (consumed + (band_dynalloc<<3) < totalbits && boost[i] < cap[i]) {
688  int add = opus_rc_p2model(rc, band_dynalloc);
689  consumed = opus_rc_tell_frac(rc);
690  if (!add)
691  break;
692 
693  boost[i] += quanta;
694  totalbits -= quanta;
695  band_dynalloc = 1;
696  }
697  /* dynalloc is more likely to occur if it's already been used for earlier bands */
698  if (boost[i])
699  dynalloc = FFMAX(2, dynalloc - 1);
700  }
701 
702  /* obtain allocation trim */
703  if (consumed + (6 << 3) <= totalbits)
704  alloctrim = opus_rc_getsymbol(rc, celt_model_alloc_trim);
705 
706  /* anti-collapse bit reservation */
707  totalbits = (s->framebits << 3) - opus_rc_tell_frac(rc) - 1;
708  s->anticollapse_bit = 0;
709  if (s->blocks > 1 && s->duration >= 2 &&
710  totalbits >= ((s->duration + 2) << 3))
711  s->anticollapse_bit = 1 << 3;
712  totalbits -= s->anticollapse_bit;
713 
714  /* band skip bit reservation */
715  if (totalbits >= 1 << 3)
716  skip_bit = 1 << 3;
717  totalbits -= skip_bit;
718 
719  /* intensity/dual stereo bit reservation */
720  if (s->coded_channels == 2) {
721  intensitystereo_bit = celt_log2_frac[s->endband - s->startband];
722  if (intensitystereo_bit <= totalbits) {
723  totalbits -= intensitystereo_bit;
724  if (totalbits >= 1 << 3) {
725  dualstereo_bit = 1 << 3;
726  totalbits -= 1 << 3;
727  }
728  } else
729  intensitystereo_bit = 0;
730  }
731 
732  for (i = s->startband; i < s->endband; i++) {
733  int trim = alloctrim - 5 - s->duration;
734  int band = celt_freq_range[i] * (s->endband - i - 1);
735  int duration = s->duration + 3;
736  int scale = duration + s->coded_channels - 1;
737 
738  /* PVQ minimum allocation threshold, below this value the band is
739  * skipped */
740  threshold[i] = FFMAX(3 * celt_freq_range[i] << duration >> 4,
741  s->coded_channels << 3);
742 
743  trim_offset[i] = trim * (band << scale) >> 6;
744 
745  if (celt_freq_range[i] << s->duration == 1)
746  trim_offset[i] -= s->coded_channels << 3;
747  }
748 
749  /* bisection */
750  low = 1;
751  high = CELT_VECTORS - 1;
752  while (low <= high) {
753  int center = (low + high) >> 1;
754  done = total = 0;
755 
756  for (i = s->endband - 1; i >= s->startband; i--) {
757  bandbits = celt_freq_range[i] * celt_static_alloc[center][i]
758  << (s->coded_channels - 1) << s->duration >> 2;
759 
760  if (bandbits)
761  bandbits = FFMAX(0, bandbits + trim_offset[i]);
762  bandbits += boost[i];
763 
764  if (bandbits >= threshold[i] || done) {
765  done = 1;
766  total += FFMIN(bandbits, cap[i]);
767  } else if (bandbits >= s->coded_channels << 3)
768  total += s->coded_channels << 3;
769  }
770 
771  if (total > totalbits)
772  high = center - 1;
773  else
774  low = center + 1;
775  }
776  high = low--;
777 
778  for (i = s->startband; i < s->endband; i++) {
779  bits1[i] = celt_freq_range[i] * celt_static_alloc[low][i]
780  << (s->coded_channels - 1) << s->duration >> 2;
781  bits2[i] = high >= CELT_VECTORS ? cap[i] :
782  celt_freq_range[i] * celt_static_alloc[high][i]
783  << (s->coded_channels - 1) << s->duration >> 2;
784 
785  if (bits1[i])
786  bits1[i] = FFMAX(0, bits1[i] + trim_offset[i]);
787  if (bits2[i])
788  bits2[i] = FFMAX(0, bits2[i] + trim_offset[i]);
789  if (low)
790  bits1[i] += boost[i];
791  bits2[i] += boost[i];
792 
793  if (boost[i])
794  skip_startband = i;
795  bits2[i] = FFMAX(0, bits2[i] - bits1[i]);
796  }
797 
798  /* bisection */
799  low = 0;
800  high = 1 << CELT_ALLOC_STEPS;
801  for (i = 0; i < CELT_ALLOC_STEPS; i++) {
802  int center = (low + high) >> 1;
803  done = total = 0;
804 
805  for (j = s->endband - 1; j >= s->startband; j--) {
806  bandbits = bits1[j] + (center * bits2[j] >> CELT_ALLOC_STEPS);
807 
808  if (bandbits >= threshold[j] || done) {
809  done = 1;
810  total += FFMIN(bandbits, cap[j]);
811  } else if (bandbits >= s->coded_channels << 3)
812  total += s->coded_channels << 3;
813  }
814  if (total > totalbits)
815  high = center;
816  else
817  low = center;
818  }
819 
820  done = total = 0;
821  for (i = s->endband - 1; i >= s->startband; i--) {
822  bandbits = bits1[i] + (low * bits2[i] >> CELT_ALLOC_STEPS);
823 
824  if (bandbits >= threshold[i] || done)
825  done = 1;
826  else
827  bandbits = (bandbits >= s->coded_channels << 3) ?
828  s->coded_channels << 3 : 0;
829 
830  bandbits = FFMIN(bandbits, cap[i]);
831  s->pulses[i] = bandbits;
832  total += bandbits;
833  }
834 
835  /* band skipping */
836  for (s->codedbands = s->endband; ; s->codedbands--) {
837  int allocation;
838  j = s->codedbands - 1;
839 
840  if (j == skip_startband) {
841  /* all remaining bands are not skipped */
842  totalbits += skip_bit;
843  break;
844  }
845 
846  /* determine the number of bits available for coding "do not skip" markers */
847  remaining = totalbits - total;
848  bandbits = remaining / (celt_freq_bands[j+1] - celt_freq_bands[s->startband]);
849  remaining -= bandbits * (celt_freq_bands[j+1] - celt_freq_bands[s->startband]);
850  allocation = s->pulses[j] + bandbits * celt_freq_range[j]
851  + FFMAX(0, remaining - (celt_freq_bands[j] - celt_freq_bands[s->startband]));
852 
853  /* a "do not skip" marker is only coded if the allocation is
854  above the chosen threshold */
855  if (allocation >= FFMAX(threshold[j], (s->coded_channels + 1) <<3 )) {
856  if (opus_rc_p2model(rc, 1))
857  break;
858 
859  total += 1 << 3;
860  allocation -= 1 << 3;
861  }
862 
863  /* the band is skipped, so reclaim its bits */
864  total -= s->pulses[j];
865  if (intensitystereo_bit) {
866  total -= intensitystereo_bit;
867  intensitystereo_bit = celt_log2_frac[j - s->startband];
868  total += intensitystereo_bit;
869  }
870 
871  total += s->pulses[j] = (allocation >= s->coded_channels << 3) ?
872  s->coded_channels << 3 : 0;
873  }
874 
875  /* obtain stereo flags */
876  s->intensitystereo = 0;
877  s->dualstereo = 0;
878  if (intensitystereo_bit)
879  s->intensitystereo = s->startband +
880  opus_rc_unimodel(rc, s->codedbands + 1 - s->startband);
881  if (s->intensitystereo <= s->startband)
882  totalbits += dualstereo_bit; /* no intensity stereo means no dual stereo */
883  else if (dualstereo_bit)
884  s->dualstereo = opus_rc_p2model(rc, 1);
885 
886  /* supply the remaining bits in this frame to lower bands */
887  remaining = totalbits - total;
888  bandbits = remaining / (celt_freq_bands[s->codedbands] - celt_freq_bands[s->startband]);
889  remaining -= bandbits * (celt_freq_bands[s->codedbands] - celt_freq_bands[s->startband]);
890  for (i = s->startband; i < s->codedbands; i++) {
891  int bits = FFMIN(remaining, celt_freq_range[i]);
892 
893  s->pulses[i] += bits + bandbits * celt_freq_range[i];
894  remaining -= bits;
895  }
896 
897  for (i = s->startband; i < s->codedbands; i++) {
898  int N = celt_freq_range[i] << s->duration;
899  int prev_extra = extrabits;
900  s->pulses[i] += extrabits;
901 
902  if (N > 1) {
903  int dof; // degrees of freedom
904  int temp; // dof * channels * log(dof)
905  int offset; // fine energy quantization offset, i.e.
906  // extra bits assigned over the standard
907  // totalbits/dof
908  int fine_bits, max_bits;
909 
910  extrabits = FFMAX(0, s->pulses[i] - cap[i]);
911  s->pulses[i] -= extrabits;
912 
913  /* intensity stereo makes use of an extra degree of freedom */
914  dof = N * s->coded_channels
915  + (s->coded_channels == 2 && N > 2 && !s->dualstereo && i < s->intensitystereo);
916  temp = dof * (celt_log_freq_range[i] + (s->duration<<3));
917  offset = (temp >> 1) - dof * CELT_FINE_OFFSET;
918  if (N == 2) /* dof=2 is the only case that doesn't fit the model */
919  offset += dof<<1;
920 
921  /* grant an additional bias for the first and second pulses */
922  if (s->pulses[i] + offset < 2 * (dof << 3))
923  offset += temp >> 2;
924  else if (s->pulses[i] + offset < 3 * (dof << 3))
925  offset += temp >> 3;
926 
927  fine_bits = (s->pulses[i] + offset + (dof << 2)) / (dof << 3);
928  max_bits = FFMIN((s->pulses[i]>>3) >> (s->coded_channels - 1),
930 
931  max_bits = FFMAX(max_bits, 0);
932 
933  s->fine_bits[i] = av_clip(fine_bits, 0, max_bits);
934 
935  /* if fine_bits was rounded down or capped,
936  give priority for the final fine energy pass */
937  s->fine_priority[i] = (s->fine_bits[i] * (dof<<3) >= s->pulses[i] + offset);
938 
939  /* the remaining bits are assigned to PVQ */
940  s->pulses[i] -= s->fine_bits[i] << (s->coded_channels - 1) << 3;
941  } else {
942  /* all bits go to fine energy except for the sign bit */
943  extrabits = FFMAX(0, s->pulses[i] - (s->coded_channels << 3));
944  s->pulses[i] -= extrabits;
945  s->fine_bits[i] = 0;
946  s->fine_priority[i] = 1;
947  }
948 
949  /* hand back a limited number of extra fine energy bits to this band */
950  if (extrabits > 0) {
951  int fineextra = FFMIN(extrabits >> (s->coded_channels + 2),
952  CELT_MAX_FINE_BITS - s->fine_bits[i]);
953  s->fine_bits[i] += fineextra;
954 
955  fineextra <<= s->coded_channels + 2;
956  s->fine_priority[i] = (fineextra >= extrabits - prev_extra);
957  extrabits -= fineextra;
958  }
959  }
960  s->remaining = extrabits;
961 
962  /* skipped bands dedicate all of their bits for fine energy */
963  for (; i < s->endband; i++) {
964  s->fine_bits[i] = s->pulses[i] >> (s->coded_channels - 1) >> 3;
965  s->pulses[i] = 0;
966  s->fine_priority[i] = s->fine_bits[i] < 1;
967  }
968 }
969 
970 static inline int celt_bits2pulses(const uint8_t *cache, int bits)
971 {
972  // TODO: Find the size of cache and make it into an array in the parameters list
973  int i, low = 0, high;
974 
975  high = cache[0];
976  bits--;
977 
978  for (i = 0; i < 6; i++) {
979  int center = (low + high + 1) >> 1;
980  if (cache[center] >= bits)
981  high = center;
982  else
983  low = center;
984  }
985 
986  return (bits - (low == 0 ? -1 : cache[low]) <= cache[high] - bits) ? low : high;
987 }
988 
989 static inline int celt_pulses2bits(const uint8_t *cache, int pulses)
990 {
991  // TODO: Find the size of cache and make it into an array in the parameters list
992  return (pulses == 0) ? 0 : cache[pulses] + 1;
993 }
994 
995 static inline void celt_normalize_residual(const int * av_restrict iy, float * av_restrict X,
996  int N, float g)
997 {
998  int i;
999  for (i = 0; i < N; i++)
1000  X[i] = g * iy[i];
1001 }
1002 
1003 static void celt_exp_rotation1(float *X, unsigned int len, unsigned int stride,
1004  float c, float s)
1005 {
1006  float *Xptr;
1007  int i;
1008 
1009  Xptr = X;
1010  for (i = 0; i < len - stride; i++) {
1011  float x1, x2;
1012  x1 = Xptr[0];
1013  x2 = Xptr[stride];
1014  Xptr[stride] = c * x2 + s * x1;
1015  *Xptr++ = c * x1 - s * x2;
1016  }
1017 
1018  Xptr = &X[len - 2 * stride - 1];
1019  for (i = len - 2 * stride - 1; i >= 0; i--) {
1020  float x1, x2;
1021  x1 = Xptr[0];
1022  x2 = Xptr[stride];
1023  Xptr[stride] = c * x2 + s * x1;
1024  *Xptr-- = c * x1 - s * x2;
1025  }
1026 }
1027 
1028 static inline void celt_exp_rotation(float *X, unsigned int len,
1029  unsigned int stride, unsigned int K,
1030  enum CeltSpread spread)
1031 {
1032  unsigned int stride2 = 0;
1033  float c, s;
1034  float gain, theta;
1035  int i;
1036 
1037  if (2*K >= len || spread == CELT_SPREAD_NONE)
1038  return;
1039 
1040  gain = (float)len / (len + (20 - 5*spread) * K);
1041  theta = M_PI * gain * gain / 4;
1042 
1043  c = cos(theta);
1044  s = sin(theta);
1045 
1046  if (len >= stride << 3) {
1047  stride2 = 1;
1048  /* This is just a simple (equivalent) way of computing sqrt(len/stride) with rounding.
1049  It's basically incrementing long as (stride2+0.5)^2 < len/stride. */
1050  while ((stride2 * stride2 + stride2) * stride + (stride >> 2) < len)
1051  stride2++;
1052  }
1053 
1054  /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for
1055  extract_collapse_mask().*/
1056  len /= stride;
1057  for (i = 0; i < stride; i++) {
1058  if (stride2)
1059  celt_exp_rotation1(X + i * len, len, stride2, s, c);
1060  celt_exp_rotation1(X + i * len, len, 1, c, s);
1061  }
1062 }
1063 
1064 static inline unsigned int celt_extract_collapse_mask(const int *iy,
1065  unsigned int N,
1066  unsigned int B)
1067 {
1068  unsigned int collapse_mask;
1069  int N0;
1070  int i, j;
1071 
1072  if (B <= 1)
1073  return 1;
1074 
1075  /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for
1076  exp_rotation().*/
1077  N0 = N/B;
1078  collapse_mask = 0;
1079  for (i = 0; i < B; i++)
1080  for (j = 0; j < N0; j++)
1081  collapse_mask |= (iy[i*N0+j]!=0)<<i;
1082  return collapse_mask;
1083 }
1084 
1085 static inline void celt_renormalize_vector(float *X, int N, float gain)
1086 {
1087  int i;
1088  float g = 1e-15f;
1089  for (i = 0; i < N; i++)
1090  g += X[i] * X[i];
1091  g = gain / sqrtf(g);
1092 
1093  for (i = 0; i < N; i++)
1094  X[i] *= g;
1095 }
1096 
1097 static inline void celt_stereo_merge(float *X, float *Y, float mid, int N)
1098 {
1099  int i;
1100  float xp = 0, side = 0;
1101  float E[2];
1102  float mid2;
1103  float t, gain[2];
1104 
1105  /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */
1106  for (i = 0; i < N; i++) {
1107  xp += X[i] * Y[i];
1108  side += Y[i] * Y[i];
1109  }
1110 
1111  /* Compensating for the mid normalization */
1112  xp *= mid;
1113  mid2 = mid;
1114  E[0] = mid2 * mid2 + side - 2 * xp;
1115  E[1] = mid2 * mid2 + side + 2 * xp;
1116  if (E[0] < 6e-4f || E[1] < 6e-4f) {
1117  for (i = 0; i < N; i++)
1118  Y[i] = X[i];
1119  return;
1120  }
1121 
1122  t = E[0];
1123  gain[0] = 1.0f / sqrtf(t);
1124  t = E[1];
1125  gain[1] = 1.0f / sqrtf(t);
1126 
1127  for (i = 0; i < N; i++) {
1128  float value[2];
1129  /* Apply mid scaling (side is already scaled) */
1130  value[0] = mid * X[i];
1131  value[1] = Y[i];
1132  X[i] = gain[0] * (value[0] - value[1]);
1133  Y[i] = gain[1] * (value[0] + value[1]);
1134  }
1135 }
1136 
1137 static void celt_interleave_hadamard(float *tmp, float *X, int N0,
1138  int stride, int hadamard)
1139 {
1140  int i, j;
1141  int N = N0*stride;
1142 
1143  if (hadamard) {
1144  const uint8_t *ordery = celt_hadamard_ordery + stride - 2;
1145  for (i = 0; i < stride; i++)
1146  for (j = 0; j < N0; j++)
1147  tmp[j*stride+i] = X[ordery[i]*N0+j];
1148  } else {
1149  for (i = 0; i < stride; i++)
1150  for (j = 0; j < N0; j++)
1151  tmp[j*stride+i] = X[i*N0+j];
1152  }
1153 
1154  for (i = 0; i < N; i++)
1155  X[i] = tmp[i];
1156 }
1157 
1158 static void celt_deinterleave_hadamard(float *tmp, float *X, int N0,
1159  int stride, int hadamard)
1160 {
1161  int i, j;
1162  int N = N0*stride;
1163 
1164  if (hadamard) {
1165  const uint8_t *ordery = celt_hadamard_ordery + stride - 2;
1166  for (i = 0; i < stride; i++)
1167  for (j = 0; j < N0; j++)
1168  tmp[ordery[i]*N0+j] = X[j*stride+i];
1169  } else {
1170  for (i = 0; i < stride; i++)
1171  for (j = 0; j < N0; j++)
1172  tmp[i*N0+j] = X[j*stride+i];
1173  }
1174 
1175  for (i = 0; i < N; i++)
1176  X[i] = tmp[i];
1177 }
1178 
1179 static void celt_haar1(float *X, int N0, int stride)
1180 {
1181  int i, j;
1182  N0 >>= 1;
1183  for (i = 0; i < stride; i++) {
1184  for (j = 0; j < N0; j++) {
1185  float x0 = X[stride * (2 * j + 0) + i];
1186  float x1 = X[stride * (2 * j + 1) + i];
1187  X[stride * (2 * j + 0) + i] = (x0 + x1) * M_SQRT1_2;
1188  X[stride * (2 * j + 1) + i] = (x0 - x1) * M_SQRT1_2;
1189  }
1190  }
1191 }
1192 
1193 static inline int celt_compute_qn(int N, int b, int offset, int pulse_cap,
1194  int dualstereo)
1195 {
1196  int qn, qb;
1197  int N2 = 2 * N - 1;
1198  if (dualstereo && N == 2)
1199  N2--;
1200 
1201  /* The upper limit ensures that in a stereo split with itheta==16384, we'll
1202  * always have enough bits left over to code at least one pulse in the
1203  * side; otherwise it would collapse, since it doesn't get folded. */
1204  qb = FFMIN3(b - pulse_cap - (4 << 3), (b + N2 * offset) / N2, 8 << 3);
1205  qn = (qb < (1 << 3 >> 1)) ? 1 : ((celt_qn_exp2[qb & 0x7] >> (14 - (qb >> 3))) + 1) >> 1 << 1;
1206  return qn;
1207 }
1208 
1209 // this code was adapted from libopus
1210 static inline uint64_t celt_cwrsi(unsigned int N, unsigned int K, unsigned int i, int *y)
1211 {
1212  uint64_t norm = 0;
1213  uint32_t p;
1214  int s, val;
1215  int k0;
1216 
1217  while (N > 2) {
1218  uint32_t q;
1219 
1220  /*Lots of pulses case:*/
1221  if (K >= N) {
1222  const uint32_t *row = celt_pvq_u_row[N];
1223 
1224  /* Are the pulses in this dimension negative? */
1225  p = row[K + 1];
1226  s = -(i >= p);
1227  i -= p & s;
1228 
1229  /*Count how many pulses were placed in this dimension.*/
1230  k0 = K;
1231  q = row[N];
1232  if (q > i) {
1233  K = N;
1234  do {
1235  p = celt_pvq_u_row[--K][N];
1236  } while (p > i);
1237  } else
1238  for (p = row[K]; p > i; p = row[K])
1239  K--;
1240 
1241  i -= p;
1242  val = (k0 - K + s) ^ s;
1243  norm += val * val;
1244  *y++ = val;
1245  } else { /*Lots of dimensions case:*/
1246  /*Are there any pulses in this dimension at all?*/
1247  p = celt_pvq_u_row[K ][N];
1248  q = celt_pvq_u_row[K + 1][N];
1249 
1250  if (p <= i && i < q) {
1251  i -= p;
1252  *y++ = 0;
1253  } else {
1254  /*Are the pulses in this dimension negative?*/
1255  s = -(i >= q);
1256  i -= q & s;
1257 
1258  /*Count how many pulses were placed in this dimension.*/
1259  k0 = K;
1260  do p = celt_pvq_u_row[--K][N];
1261  while (p > i);
1262 
1263  i -= p;
1264  val = (k0 - K + s) ^ s;
1265  norm += val * val;
1266  *y++ = val;
1267  }
1268  }
1269  N--;
1270  }
1271 
1272  /* N == 2 */
1273  p = 2 * K + 1;
1274  s = -(i >= p);
1275  i -= p & s;
1276  k0 = K;
1277  K = (i + 1) / 2;
1278 
1279  if (K)
1280  i -= 2 * K - 1;
1281 
1282  val = (k0 - K + s) ^ s;
1283  norm += val * val;
1284  *y++ = val;
1285 
1286  /* N==1 */
1287  s = -i;
1288  val = (K + s) ^ s;
1289  norm += val * val;
1290  *y = val;
1291 
1292  return norm;
1293 }
1294 
1295 static inline float celt_decode_pulses(OpusRangeCoder *rc, int *y, unsigned int N, unsigned int K)
1296 {
1297  unsigned int idx;
1298 #define CELT_PVQ_U(n, k) (celt_pvq_u_row[FFMIN(n, k)][FFMAX(n, k)])
1299 #define CELT_PVQ_V(n, k) (CELT_PVQ_U(n, k) + CELT_PVQ_U(n, (k) + 1))
1300  idx = opus_rc_unimodel(rc, CELT_PVQ_V(N, K));
1301  return celt_cwrsi(N, K, idx, y);
1302 }
1303 
1304 /** Decode pulse vector and combine the result with the pitch vector to produce
1305  the final normalised signal in the current band. */
1306 static inline unsigned int celt_alg_unquant(OpusRangeCoder *rc, float *X,
1307  unsigned int N, unsigned int K,
1308  enum CeltSpread spread,
1309  unsigned int blocks, float gain)
1310 {
1311  int y[176];
1312 
1313  gain /= sqrtf(celt_decode_pulses(rc, y, N, K));
1314  celt_normalize_residual(y, X, N, gain);
1315  celt_exp_rotation(X, N, blocks, K, spread);
1316  return celt_extract_collapse_mask(y, N, blocks);
1317 }
1318 
1319 static unsigned int celt_decode_band(CeltContext *s, OpusRangeCoder *rc,
1320  const int band, float *X, float *Y,
1321  int N, int b, unsigned int blocks,
1322  float *lowband, int duration,
1323  float *lowband_out, int level,
1324  float gain, float *lowband_scratch,
1325  int fill)
1326 {
1327  const uint8_t *cache;
1328  int dualstereo, split;
1329  int imid = 0, iside = 0;
1330  unsigned int N0 = N;
1331  int N_B;
1332  int N_B0;
1333  int B0 = blocks;
1334  int time_divide = 0;
1335  int recombine = 0;
1336  int inv = 0;
1337  float mid = 0, side = 0;
1338  int longblocks = (B0 == 1);
1339  unsigned int cm = 0;
1340 
1341  N_B0 = N_B = N / blocks;
1342  split = dualstereo = (Y != NULL);
1343 
1344  if (N == 1) {
1345  /* special case for one sample */
1346  int i;
1347  float *x = X;
1348  for (i = 0; i <= dualstereo; i++) {
1349  int sign = 0;
1350  if (s->remaining2 >= 1<<3) {
1351  sign = opus_getrawbits(rc, 1);
1352  s->remaining2 -= 1 << 3;
1353  b -= 1 << 3;
1354  }
1355  x[0] = sign ? -1.0f : 1.0f;
1356  x = Y;
1357  }
1358  if (lowband_out)
1359  lowband_out[0] = X[0];
1360  return 1;
1361  }
1362 
1363  if (!dualstereo && level == 0) {
1364  int tf_change = s->tf_change[band];
1365  int k;
1366  if (tf_change > 0)
1367  recombine = tf_change;
1368  /* Band recombining to increase frequency resolution */
1369 
1370  if (lowband &&
1371  (recombine || ((N_B & 1) == 0 && tf_change < 0) || B0 > 1)) {
1372  int j;
1373  for (j = 0; j < N; j++)
1374  lowband_scratch[j] = lowband[j];
1375  lowband = lowband_scratch;
1376  }
1377 
1378  for (k = 0; k < recombine; k++) {
1379  if (lowband)
1380  celt_haar1(lowband, N >> k, 1 << k);
1381  fill = celt_bit_interleave[fill & 0xF] | celt_bit_interleave[fill >> 4] << 2;
1382  }
1383  blocks >>= recombine;
1384  N_B <<= recombine;
1385 
1386  /* Increasing the time resolution */
1387  while ((N_B & 1) == 0 && tf_change < 0) {
1388  if (lowband)
1389  celt_haar1(lowband, N_B, blocks);
1390  fill |= fill << blocks;
1391  blocks <<= 1;
1392  N_B >>= 1;
1393  time_divide++;
1394  tf_change++;
1395  }
1396  B0 = blocks;
1397  N_B0 = N_B;
1398 
1399  /* Reorganize the samples in time order instead of frequency order */
1400  if (B0 > 1 && lowband)
1401  celt_deinterleave_hadamard(s->scratch, lowband, N_B >> recombine,
1402  B0 << recombine, longblocks);
1403  }
1404 
1405  /* If we need 1.5 more bit than we can produce, split the band in two. */
1406  cache = celt_cache_bits +
1407  celt_cache_index[(duration + 1) * CELT_MAX_BANDS + band];
1408  if (!dualstereo && duration >= 0 && b > cache[cache[0]] + 12 && N > 2) {
1409  N >>= 1;
1410  Y = X + N;
1411  split = 1;
1412  duration -= 1;
1413  if (blocks == 1)
1414  fill = (fill & 1) | (fill << 1);
1415  blocks = (blocks + 1) >> 1;
1416  }
1417 
1418  if (split) {
1419  int qn;
1420  int itheta = 0;
1421  int mbits, sbits, delta;
1422  int qalloc;
1423  int pulse_cap;
1424  int offset;
1425  int orig_fill;
1426  int tell;
1427 
1428  /* Decide on the resolution to give to the split parameter theta */
1429  pulse_cap = celt_log_freq_range[band] + duration * 8;
1430  offset = (pulse_cap >> 1) - (dualstereo && N == 2 ? CELT_QTHETA_OFFSET_TWOPHASE :
1432  qn = (dualstereo && band >= s->intensitystereo) ? 1 :
1433  celt_compute_qn(N, b, offset, pulse_cap, dualstereo);
1434  tell = opus_rc_tell_frac(rc);
1435  if (qn != 1) {
1436  /* Entropy coding of the angle. We use a uniform pdf for the
1437  time split, a step for stereo, and a triangular one for the rest. */
1438  if (dualstereo && N > 2)
1439  itheta = opus_rc_stepmodel(rc, qn/2);
1440  else if (dualstereo || B0 > 1)
1441  itheta = opus_rc_unimodel(rc, qn+1);
1442  else
1443  itheta = opus_rc_trimodel(rc, qn);
1444  itheta = itheta * 16384 / qn;
1445  /* NOTE: Renormalising X and Y *may* help fixed-point a bit at very high rate.
1446  Let's do that at higher complexity */
1447  } else if (dualstereo) {
1448  inv = (b > 2 << 3 && s->remaining2 > 2 << 3) ? opus_rc_p2model(rc, 2) : 0;
1449  itheta = 0;
1450  }
1451  qalloc = opus_rc_tell_frac(rc) - tell;
1452  b -= qalloc;
1453 
1454  orig_fill = fill;
1455  if (itheta == 0) {
1456  imid = 32767;
1457  iside = 0;
1458  fill = av_mod_uintp2(fill, blocks);
1459  delta = -16384;
1460  } else if (itheta == 16384) {
1461  imid = 0;
1462  iside = 32767;
1463  fill &= ((1 << blocks) - 1) << blocks;
1464  delta = 16384;
1465  } else {
1466  imid = celt_cos(itheta);
1467  iside = celt_cos(16384-itheta);
1468  /* This is the mid vs side allocation that minimizes squared error
1469  in that band. */
1470  delta = ROUND_MUL16((N - 1) << 7, celt_log2tan(iside, imid));
1471  }
1472 
1473  mid = imid / 32768.0f;
1474  side = iside / 32768.0f;
1475 
1476  /* This is a special case for N=2 that only works for stereo and takes
1477  advantage of the fact that mid and side are orthogonal to encode
1478  the side with just one bit. */
1479  if (N == 2 && dualstereo) {
1480  int c;
1481  int sign = 0;
1482  float tmp;
1483  float *x2, *y2;
1484  mbits = b;
1485  /* Only need one bit for the side */
1486  sbits = (itheta != 0 && itheta != 16384) ? 1 << 3 : 0;
1487  mbits -= sbits;
1488  c = (itheta > 8192);
1489  s->remaining2 -= qalloc+sbits;
1490 
1491  x2 = c ? Y : X;
1492  y2 = c ? X : Y;
1493  if (sbits)
1494  sign = opus_getrawbits(rc, 1);
1495  sign = 1 - 2 * sign;
1496  /* We use orig_fill here because we want to fold the side, but if
1497  itheta==16384, we'll have cleared the low bits of fill. */
1498  cm = celt_decode_band(s, rc, band, x2, NULL, N, mbits, blocks,
1499  lowband, duration, lowband_out, level, gain,
1500  lowband_scratch, orig_fill);
1501  /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse),
1502  and there's no need to worry about mixing with the other channel. */
1503  y2[0] = -sign * x2[1];
1504  y2[1] = sign * x2[0];
1505  X[0] *= mid;
1506  X[1] *= mid;
1507  Y[0] *= side;
1508  Y[1] *= side;
1509  tmp = X[0];
1510  X[0] = tmp - Y[0];
1511  Y[0] = tmp + Y[0];
1512  tmp = X[1];
1513  X[1] = tmp - Y[1];
1514  Y[1] = tmp + Y[1];
1515  } else {
1516  /* "Normal" split code */
1517  float *next_lowband2 = NULL;
1518  float *next_lowband_out1 = NULL;
1519  int next_level = 0;
1520  int rebalance;
1521 
1522  /* Give more bits to low-energy MDCTs than they would
1523  * otherwise deserve */
1524  if (B0 > 1 && !dualstereo && (itheta & 0x3fff)) {
1525  if (itheta > 8192)
1526  /* Rough approximation for pre-echo masking */
1527  delta -= delta >> (4 - duration);
1528  else
1529  /* Corresponds to a forward-masking slope of
1530  * 1.5 dB per 10 ms */
1531  delta = FFMIN(0, delta + (N << 3 >> (5 - duration)));
1532  }
1533  mbits = av_clip((b - delta) / 2, 0, b);
1534  sbits = b - mbits;
1535  s->remaining2 -= qalloc;
1536 
1537  if (lowband && !dualstereo)
1538  next_lowband2 = lowband + N; /* >32-bit split case */
1539 
1540  /* Only stereo needs to pass on lowband_out.
1541  * Otherwise, it's handled at the end */
1542  if (dualstereo)
1543  next_lowband_out1 = lowband_out;
1544  else
1545  next_level = level + 1;
1546 
1547  rebalance = s->remaining2;
1548  if (mbits >= sbits) {
1549  /* In stereo mode, we do not apply a scaling to the mid
1550  * because we need the normalized mid for folding later */
1551  cm = celt_decode_band(s, rc, band, X, NULL, N, mbits, blocks,
1552  lowband, duration, next_lowband_out1,
1553  next_level, dualstereo ? 1.0f : (gain * mid),
1554  lowband_scratch, fill);
1555 
1556  rebalance = mbits - (rebalance - s->remaining2);
1557  if (rebalance > 3 << 3 && itheta != 0)
1558  sbits += rebalance - (3 << 3);
1559 
1560  /* For a stereo split, the high bits of fill are always zero,
1561  * so no folding will be done to the side. */
1562  cm |= celt_decode_band(s, rc, band, Y, NULL, N, sbits, blocks,
1563  next_lowband2, duration, NULL,
1564  next_level, gain * side, NULL,
1565  fill >> blocks) << ((B0 >> 1) & (dualstereo - 1));
1566  } else {
1567  /* For a stereo split, the high bits of fill are always zero,
1568  * so no folding will be done to the side. */
1569  cm = celt_decode_band(s, rc, band, Y, NULL, N, sbits, blocks,
1570  next_lowband2, duration, NULL,
1571  next_level, gain * side, NULL,
1572  fill >> blocks) << ((B0 >> 1) & (dualstereo - 1));
1573 
1574  rebalance = sbits - (rebalance - s->remaining2);
1575  if (rebalance > 3 << 3 && itheta != 16384)
1576  mbits += rebalance - (3 << 3);
1577 
1578  /* In stereo mode, we do not apply a scaling to the mid because
1579  * we need the normalized mid for folding later */
1580  cm |= celt_decode_band(s, rc, band, X, NULL, N, mbits, blocks,
1581  lowband, duration, next_lowband_out1,
1582  next_level, dualstereo ? 1.0f : (gain * mid),
1583  lowband_scratch, fill);
1584  }
1585  }
1586  } else {
1587  /* This is the basic no-split case */
1588  unsigned int q = celt_bits2pulses(cache, b);
1589  unsigned int curr_bits = celt_pulses2bits(cache, q);
1590  s->remaining2 -= curr_bits;
1591 
1592  /* Ensures we can never bust the budget */
1593  while (s->remaining2 < 0 && q > 0) {
1594  s->remaining2 += curr_bits;
1595  curr_bits = celt_pulses2bits(cache, --q);
1596  s->remaining2 -= curr_bits;
1597  }
1598 
1599  if (q != 0) {
1600  /* Finally do the actual quantization */
1601  cm = celt_alg_unquant(rc, X, N, (q < 8) ? q : (8 + (q & 7)) << ((q >> 3) - 1),
1602  s->spread, blocks, gain);
1603  } else {
1604  /* If there's no pulse, fill the band anyway */
1605  int j;
1606  unsigned int cm_mask = (1 << blocks) - 1;
1607  fill &= cm_mask;
1608  if (!fill) {
1609  for (j = 0; j < N; j++)
1610  X[j] = 0.0f;
1611  } else {
1612  if (!lowband) {
1613  /* Noise */
1614  for (j = 0; j < N; j++)
1615  X[j] = (((int32_t)celt_rng(s)) >> 20);
1616  cm = cm_mask;
1617  } else {
1618  /* Folded spectrum */
1619  for (j = 0; j < N; j++) {
1620  /* About 48 dB below the "normal" folding level */
1621  X[j] = lowband[j] + (((celt_rng(s)) & 0x8000) ? 1.0f / 256 : -1.0f / 256);
1622  }
1623  cm = fill;
1624  }
1625  celt_renormalize_vector(X, N, gain);
1626  }
1627  }
1628  }
1629 
1630  /* This code is used by the decoder and by the resynthesis-enabled encoder */
1631  if (dualstereo) {
1632  int j;
1633  if (N != 2)
1634  celt_stereo_merge(X, Y, mid, N);
1635  if (inv) {
1636  for (j = 0; j < N; j++)
1637  Y[j] *= -1;
1638  }
1639  } else if (level == 0) {
1640  int k;
1641 
1642  /* Undo the sample reorganization going from time order to frequency order */
1643  if (B0 > 1)
1644  celt_interleave_hadamard(s->scratch, X, N_B>>recombine,
1645  B0<<recombine, longblocks);
1646 
1647  /* Undo time-freq changes that we did earlier */
1648  N_B = N_B0;
1649  blocks = B0;
1650  for (k = 0; k < time_divide; k++) {
1651  blocks >>= 1;
1652  N_B <<= 1;
1653  cm |= cm >> blocks;
1654  celt_haar1(X, N_B, blocks);
1655  }
1656 
1657  for (k = 0; k < recombine; k++) {
1658  cm = celt_bit_deinterleave[cm];
1659  celt_haar1(X, N0>>k, 1<<k);
1660  }
1661  blocks <<= recombine;
1662 
1663  /* Scale output for later folding */
1664  if (lowband_out) {
1665  int j;
1666  float n = sqrtf(N0);
1667  for (j = 0; j < N0; j++)
1668  lowband_out[j] = n * X[j];
1669  }
1670  cm = av_mod_uintp2(cm, blocks);
1671  }
1672  return cm;
1673 }
1674 
1676 {
1677  int i, j;
1678 
1679  for (i = s->startband; i < s->endband; i++) {
1680  float *dst = data + (celt_freq_bands[i] << s->duration);
1681  float norm = exp2(frame->energy[i] + celt_mean_energy[i]);
1682 
1683  for (j = 0; j < celt_freq_range[i] << s->duration; j++)
1684  dst[j] *= norm;
1685  }
1686 }
1687 
1689 {
1690  const int T0 = frame->pf_period_old;
1691  const int T1 = frame->pf_period;
1692 
1693  float g00, g01, g02;
1694  float g10, g11, g12;
1695 
1696  float x0, x1, x2, x3, x4;
1697 
1698  int i;
1699 
1700  if (frame->pf_gains[0] == 0.0 &&
1701  frame->pf_gains_old[0] == 0.0)
1702  return;
1703 
1704  g00 = frame->pf_gains_old[0];
1705  g01 = frame->pf_gains_old[1];
1706  g02 = frame->pf_gains_old[2];
1707  g10 = frame->pf_gains[0];
1708  g11 = frame->pf_gains[1];
1709  g12 = frame->pf_gains[2];
1710 
1711  x1 = data[-T1 + 1];
1712  x2 = data[-T1];
1713  x3 = data[-T1 - 1];
1714  x4 = data[-T1 - 2];
1715 
1716  for (i = 0; i < CELT_OVERLAP; i++) {
1717  float w = ff_celt_window2[i];
1718  x0 = data[i - T1 + 2];
1719 
1720  data[i] += (1.0 - w) * g00 * data[i - T0] +
1721  (1.0 - w) * g01 * (data[i - T0 - 1] + data[i - T0 + 1]) +
1722  (1.0 - w) * g02 * (data[i - T0 - 2] + data[i - T0 + 2]) +
1723  w * g10 * x2 +
1724  w * g11 * (x1 + x3) +
1725  w * g12 * (x0 + x4);
1726  x4 = x3;
1727  x3 = x2;
1728  x2 = x1;
1729  x1 = x0;
1730  }
1731 }
1732 
1734  float *data, int len)
1735 {
1736  const int T = frame->pf_period;
1737  float g0, g1, g2;
1738  float x0, x1, x2, x3, x4;
1739  int i;
1740 
1741  if (frame->pf_gains[0] == 0.0 || len <= 0)
1742  return;
1743 
1744  g0 = frame->pf_gains[0];
1745  g1 = frame->pf_gains[1];
1746  g2 = frame->pf_gains[2];
1747 
1748  x4 = data[-T - 2];
1749  x3 = data[-T - 1];
1750  x2 = data[-T];
1751  x1 = data[-T + 1];
1752 
1753  for (i = 0; i < len; i++) {
1754  x0 = data[i - T + 2];
1755  data[i] += g0 * x2 +
1756  g1 * (x1 + x3) +
1757  g2 * (x0 + x4);
1758  x4 = x3;
1759  x3 = x2;
1760  x2 = x1;
1761  x1 = x0;
1762  }
1763 }
1764 
1766 {
1767  int len = s->blocksize * s->blocks;
1768 
1769  celt_postfilter_apply_transition(frame, frame->buf + 1024);
1770 
1771  frame->pf_period_old = frame->pf_period;
1772  memcpy(frame->pf_gains_old, frame->pf_gains, sizeof(frame->pf_gains));
1773 
1774  frame->pf_period = frame->pf_period_new;
1775  memcpy(frame->pf_gains, frame->pf_gains_new, sizeof(frame->pf_gains));
1776 
1777  if (len > CELT_OVERLAP) {
1778  celt_postfilter_apply_transition(frame, frame->buf + 1024 + CELT_OVERLAP);
1779  celt_postfilter_apply(frame, frame->buf + 1024 + 2 * CELT_OVERLAP,
1780  len - 2 * CELT_OVERLAP);
1781 
1782  frame->pf_period_old = frame->pf_period;
1783  memcpy(frame->pf_gains_old, frame->pf_gains, sizeof(frame->pf_gains));
1784  }
1785 
1786  memmove(frame->buf, frame->buf + len, (1024 + CELT_OVERLAP / 2) * sizeof(float));
1787 }
1788 
1789 static int parse_postfilter(CeltContext *s, OpusRangeCoder *rc, int consumed)
1790 {
1791  static const float postfilter_taps[3][3] = {
1792  { 0.3066406250f, 0.2170410156f, 0.1296386719f },
1793  { 0.4638671875f, 0.2680664062f, 0.0 },
1794  { 0.7998046875f, 0.1000976562f, 0.0 }
1795  };
1796  int i;
1797 
1798  memset(s->frame[0].pf_gains_new, 0, sizeof(s->frame[0].pf_gains_new));
1799  memset(s->frame[1].pf_gains_new, 0, sizeof(s->frame[1].pf_gains_new));
1800 
1801  if (s->startband == 0 && consumed + 16 <= s->framebits) {
1802  int has_postfilter = opus_rc_p2model(rc, 1);
1803  if (has_postfilter) {
1804  float gain;
1805  int tapset, octave, period;
1806 
1807  octave = opus_rc_unimodel(rc, 6);
1808  period = (16 << octave) + opus_getrawbits(rc, 4 + octave) - 1;
1809  gain = 0.09375f * (opus_getrawbits(rc, 3) + 1);
1810  tapset = (opus_rc_tell(rc) + 2 <= s->framebits) ?
1812 
1813  for (i = 0; i < 2; i++) {
1814  CeltFrame *frame = &s->frame[i];
1815 
1816  frame->pf_period_new = FFMAX(period, CELT_POSTFILTER_MINPERIOD);
1817  frame->pf_gains_new[0] = gain * postfilter_taps[tapset][0];
1818  frame->pf_gains_new[1] = gain * postfilter_taps[tapset][1];
1819  frame->pf_gains_new[2] = gain * postfilter_taps[tapset][2];
1820  }
1821  }
1822 
1823  consumed = opus_rc_tell(rc);
1824  }
1825 
1826  return consumed;
1827 }
1828 
1830 {
1831  int i, j, k;
1832 
1833  for (i = s->startband; i < s->endband; i++) {
1834  int renormalize = 0;
1835  float *xptr;
1836  float prev[2];
1837  float Ediff, r;
1838  float thresh, sqrt_1;
1839  int depth;
1840 
1841  /* depth in 1/8 bits */
1842  depth = (1 + s->pulses[i]) / (celt_freq_range[i] << s->duration);
1843  thresh = exp2f(-1.0 - 0.125f * depth);
1844  sqrt_1 = 1.0f / sqrtf(celt_freq_range[i] << s->duration);
1845 
1846  xptr = X + (celt_freq_bands[i] << s->duration);
1847 
1848  prev[0] = frame->prev_energy[0][i];
1849  prev[1] = frame->prev_energy[1][i];
1850  if (s->coded_channels == 1) {
1851  CeltFrame *frame1 = &s->frame[1];
1852 
1853  prev[0] = FFMAX(prev[0], frame1->prev_energy[0][i]);
1854  prev[1] = FFMAX(prev[1], frame1->prev_energy[1][i]);
1855  }
1856  Ediff = frame->energy[i] - FFMIN(prev[0], prev[1]);
1857  Ediff = FFMAX(0, Ediff);
1858 
1859  /* r needs to be multiplied by 2 or 2*sqrt(2) depending on LM because
1860  short blocks don't have the same energy as long */
1861  r = exp2(1 - Ediff);
1862  if (s->duration == 3)
1863  r *= M_SQRT2;
1864  r = FFMIN(thresh, r) * sqrt_1;
1865  for (k = 0; k < 1 << s->duration; k++) {
1866  /* Detect collapse */
1867  if (!(frame->collapse_masks[i] & 1 << k)) {
1868  /* Fill with noise */
1869  for (j = 0; j < celt_freq_range[i]; j++)
1870  xptr[(j << s->duration) + k] = (celt_rng(s) & 0x8000) ? r : -r;
1871  renormalize = 1;
1872  }
1873  }
1874 
1875  /* We just added some energy, so we need to renormalize */
1876  if (renormalize)
1877  celt_renormalize_vector(xptr, celt_freq_range[i] << s->duration, 1.0f);
1878  }
1879 }
1880 
1882 {
1883  float lowband_scratch[8 * 22];
1884  float norm[2 * 8 * 100];
1885 
1886  int totalbits = (s->framebits << 3) - s->anticollapse_bit;
1887 
1888  int update_lowband = 1;
1889  int lowband_offset = 0;
1890 
1891  int i, j;
1892 
1893  memset(s->coeffs, 0, sizeof(s->coeffs));
1894 
1895  for (i = s->startband; i < s->endband; i++) {
1896  int band_offset = celt_freq_bands[i] << s->duration;
1897  int band_size = celt_freq_range[i] << s->duration;
1898  float *X = s->coeffs[0] + band_offset;
1899  float *Y = (s->coded_channels == 2) ? s->coeffs[1] + band_offset : NULL;
1900 
1901  int consumed = opus_rc_tell_frac(rc);
1902  float *norm2 = norm + 8 * 100;
1903  int effective_lowband = -1;
1904  unsigned int cm[2];
1905  int b;
1906 
1907  /* Compute how many bits we want to allocate to this band */
1908  if (i != s->startband)
1909  s->remaining -= consumed;
1910  s->remaining2 = totalbits - consumed - 1;
1911  if (i <= s->codedbands - 1) {
1912  int curr_balance = s->remaining / FFMIN(3, s->codedbands-i);
1913  b = av_clip_uintp2(FFMIN(s->remaining2 + 1, s->pulses[i] + curr_balance), 14);
1914  } else
1915  b = 0;
1916 
1918  (update_lowband || lowband_offset == 0))
1919  lowband_offset = i;
1920 
1921  /* Get a conservative estimate of the collapse_mask's for the bands we're
1922  going to be folding from. */
1923  if (lowband_offset != 0 && (s->spread != CELT_SPREAD_AGGRESSIVE ||
1924  s->blocks > 1 || s->tf_change[i] < 0)) {
1925  int foldstart, foldend;
1926 
1927  /* This ensures we never repeat spectral content within one band */
1928  effective_lowband = FFMAX(celt_freq_bands[s->startband],
1929  celt_freq_bands[lowband_offset] - celt_freq_range[i]);
1930  foldstart = lowband_offset;
1931  while (celt_freq_bands[--foldstart] > effective_lowband);
1932  foldend = lowband_offset - 1;
1933  while (celt_freq_bands[++foldend] < effective_lowband + celt_freq_range[i]);
1934 
1935  cm[0] = cm[1] = 0;
1936  for (j = foldstart; j < foldend; j++) {
1937  cm[0] |= s->frame[0].collapse_masks[j];
1938  cm[1] |= s->frame[s->coded_channels - 1].collapse_masks[j];
1939  }
1940  } else
1941  /* Otherwise, we'll be using the LCG to fold, so all blocks will (almost
1942  always) be non-zero.*/
1943  cm[0] = cm[1] = (1 << s->blocks) - 1;
1944 
1945  if (s->dualstereo && i == s->intensitystereo) {
1946  /* Switch off dual stereo to do intensity */
1947  s->dualstereo = 0;
1948  for (j = celt_freq_bands[s->startband] << s->duration; j < band_offset; j++)
1949  norm[j] = (norm[j] + norm2[j]) / 2;
1950  }
1951 
1952  if (s->dualstereo) {
1953  cm[0] = celt_decode_band(s, rc, i, X, NULL, band_size, b / 2, s->blocks,
1954  effective_lowband != -1 ? norm + (effective_lowband << s->duration) : NULL, s->duration,
1955  norm + band_offset, 0, 1.0f, lowband_scratch, cm[0]);
1956 
1957  cm[1] = celt_decode_band(s, rc, i, Y, NULL, band_size, b/2, s->blocks,
1958  effective_lowband != -1 ? norm2 + (effective_lowband << s->duration) : NULL, s->duration,
1959  norm2 + band_offset, 0, 1.0f, lowband_scratch, cm[1]);
1960  } else {
1961  cm[0] = celt_decode_band(s, rc, i, X, Y, band_size, b, s->blocks,
1962  effective_lowband != -1 ? norm + (effective_lowband << s->duration) : NULL, s->duration,
1963  norm + band_offset, 0, 1.0f, lowband_scratch, cm[0]|cm[1]);
1964 
1965  cm[1] = cm[0];
1966  }
1967 
1968  s->frame[0].collapse_masks[i] = (uint8_t)cm[0];
1969  s->frame[s->coded_channels - 1].collapse_masks[i] = (uint8_t)cm[1];
1970  s->remaining += s->pulses[i] + consumed;
1971 
1972  /* Update the folding position only as long as we have 1 bit/sample depth */
1973  update_lowband = (b > band_size << 3);
1974  }
1975 }
1976 
1978  float **output, int coded_channels, int frame_size,
1979  int startband, int endband)
1980 {
1981  int i, j;
1982 
1983  int consumed; // bits of entropy consumed thus far for this frame
1984  int silence = 0;
1985  int transient = 0;
1986  int anticollapse = 0;
1987  IMDCT15Context *imdct;
1988  float imdct_scale = 1.0;
1989 
1990  if (coded_channels != 1 && coded_channels != 2) {
1991  av_log(s->avctx, AV_LOG_ERROR, "Invalid number of coded channels: %d\n",
1992  coded_channels);
1993  return AVERROR_INVALIDDATA;
1994  }
1995  if (startband < 0 || startband > endband || endband > CELT_MAX_BANDS) {
1996  av_log(s->avctx, AV_LOG_ERROR, "Invalid start/end band: %d %d\n",
1997  startband, endband);
1998  return AVERROR_INVALIDDATA;
1999  }
2000 
2001  s->flushed = 0;
2002  s->coded_channels = coded_channels;
2003  s->startband = startband;
2004  s->endband = endband;
2005  s->framebits = rc->rb.bytes * 8;
2006 
2007  s->duration = av_log2(frame_size / CELT_SHORT_BLOCKSIZE);
2008  if (s->duration > CELT_MAX_LOG_BLOCKS ||
2009  frame_size != CELT_SHORT_BLOCKSIZE * (1 << s->duration)) {
2010  av_log(s->avctx, AV_LOG_ERROR, "Invalid CELT frame size: %d\n",
2011  frame_size);
2012  return AVERROR_INVALIDDATA;
2013  }
2014 
2015  if (!s->output_channels)
2016  s->output_channels = coded_channels;
2017 
2018  memset(s->frame[0].collapse_masks, 0, sizeof(s->frame[0].collapse_masks));
2019  memset(s->frame[1].collapse_masks, 0, sizeof(s->frame[1].collapse_masks));
2020 
2021  consumed = opus_rc_tell(rc);
2022 
2023  /* obtain silence flag */
2024  if (consumed >= s->framebits)
2025  silence = 1;
2026  else if (consumed == 1)
2027  silence = opus_rc_p2model(rc, 15);
2028 
2029 
2030  if (silence) {
2031  consumed = s->framebits;
2032  rc->total_read_bits += s->framebits - opus_rc_tell(rc);
2033  }
2034 
2035  /* obtain post-filter options */
2036  consumed = parse_postfilter(s, rc, consumed);
2037 
2038  /* obtain transient flag */
2039  if (s->duration != 0 && consumed+3 <= s->framebits)
2040  transient = opus_rc_p2model(rc, 3);
2041 
2042  s->blocks = transient ? 1 << s->duration : 1;
2043  s->blocksize = frame_size / s->blocks;
2044 
2045  imdct = s->imdct[transient ? 0 : s->duration];
2046 
2047  if (coded_channels == 1) {
2048  for (i = 0; i < CELT_MAX_BANDS; i++)
2049  s->frame[0].energy[i] = FFMAX(s->frame[0].energy[i], s->frame[1].energy[i]);
2050  }
2051 
2053  celt_decode_tf_changes (s, rc, transient);
2054  celt_decode_allocation (s, rc);
2055  celt_decode_fine_energy (s, rc);
2056  celt_decode_bands (s, rc);
2057 
2058  if (s->anticollapse_bit)
2059  anticollapse = opus_getrawbits(rc, 1);
2060 
2062 
2063  /* apply anti-collapse processing and denormalization to
2064  * each coded channel */
2065  for (i = 0; i < s->coded_channels; i++) {
2066  CeltFrame *frame = &s->frame[i];
2067 
2068  if (anticollapse)
2069  process_anticollapse(s, frame, s->coeffs[i]);
2070 
2071  celt_denormalize(s, frame, s->coeffs[i]);
2072  }
2073 
2074  /* stereo -> mono downmix */
2075  if (s->output_channels < s->coded_channels) {
2076  s->dsp->vector_fmac_scalar(s->coeffs[0], s->coeffs[1], 1.0, FFALIGN(frame_size, 16));
2077  imdct_scale = 0.5;
2078  } else if (s->output_channels > s->coded_channels)
2079  memcpy(s->coeffs[1], s->coeffs[0], frame_size * sizeof(float));
2080 
2081  if (silence) {
2082  for (i = 0; i < 2; i++) {
2083  CeltFrame *frame = &s->frame[i];
2084 
2085  for (j = 0; j < FF_ARRAY_ELEMS(frame->energy); j++)
2086  frame->energy[j] = CELT_ENERGY_SILENCE;
2087  }
2088  memset(s->coeffs, 0, sizeof(s->coeffs));
2089  }
2090 
2091  /* transform and output for each output channel */
2092  for (i = 0; i < s->output_channels; i++) {
2093  CeltFrame *frame = &s->frame[i];
2094  float m = frame->deemph_coeff;
2095 
2096  /* iMDCT and overlap-add */
2097  for (j = 0; j < s->blocks; j++) {
2098  float *dst = frame->buf + 1024 + j * s->blocksize;
2099 
2100  imdct->imdct_half(imdct, dst + CELT_OVERLAP / 2, s->coeffs[i] + j,
2101  s->blocks, imdct_scale);
2102  s->dsp->vector_fmul_window(dst, dst, dst + CELT_OVERLAP / 2,
2103  celt_window, CELT_OVERLAP / 2);
2104  }
2105 
2106  /* postfilter */
2107  celt_postfilter(s, frame);
2108 
2109  /* deemphasis and output scaling */
2110  for (j = 0; j < frame_size; j++) {
2111  float tmp = frame->buf[1024 - frame_size + j] + m;
2112  m = tmp * CELT_DEEMPH_COEFF;
2113  output[i][j] = tmp / 32768.;
2114  }
2115  frame->deemph_coeff = m;
2116  }
2117 
2118  if (coded_channels == 1)
2119  memcpy(s->frame[1].energy, s->frame[0].energy, sizeof(s->frame[0].energy));
2120 
2121  for (i = 0; i < 2; i++ ) {
2122  CeltFrame *frame = &s->frame[i];
2123 
2124  if (!transient) {
2125  memcpy(frame->prev_energy[1], frame->prev_energy[0], sizeof(frame->prev_energy[0]));
2126  memcpy(frame->prev_energy[0], frame->energy, sizeof(frame->prev_energy[0]));
2127  } else {
2128  for (j = 0; j < CELT_MAX_BANDS; j++)
2129  frame->prev_energy[0][j] = FFMIN(frame->prev_energy[0][j], frame->energy[j]);
2130  }
2131 
2132  for (j = 0; j < s->startband; j++) {
2133  frame->prev_energy[0][j] = CELT_ENERGY_SILENCE;
2134  frame->energy[j] = 0.0;
2135  }
2136  for (j = s->endband; j < CELT_MAX_BANDS; j++) {
2137  frame->prev_energy[0][j] = CELT_ENERGY_SILENCE;
2138  frame->energy[j] = 0.0;
2139  }
2140  }
2141 
2142  s->seed = rc->range;
2143 
2144  return 0;
2145 }
2146 
2148 {
2149  int i, j;
2150 
2151  if (s->flushed)
2152  return;
2153 
2154  for (i = 0; i < 2; i++) {
2155  CeltFrame *frame = &s->frame[i];
2156 
2157  for (j = 0; j < CELT_MAX_BANDS; j++)
2158  frame->prev_energy[0][j] = frame->prev_energy[1][j] = CELT_ENERGY_SILENCE;
2159 
2160  memset(frame->energy, 0, sizeof(frame->energy));
2161  memset(frame->buf, 0, sizeof(frame->buf));
2162 
2163  memset(frame->pf_gains, 0, sizeof(frame->pf_gains));
2164  memset(frame->pf_gains_old, 0, sizeof(frame->pf_gains_old));
2165  memset(frame->pf_gains_new, 0, sizeof(frame->pf_gains_new));
2166 
2167  frame->deemph_coeff = 0.0;
2168  }
2169  s->seed = 0;
2170 
2171  s->flushed = 1;
2172 }
2173 
2175 {
2176  CeltContext *s = *ps;
2177  int i;
2178 
2179  if (!s)
2180  return;
2181 
2182  for (i = 0; i < FF_ARRAY_ELEMS(s->imdct); i++)
2183  ff_imdct15_uninit(&s->imdct[i]);
2184 
2185  av_freep(&s->dsp);
2186  av_freep(ps);
2187 }
2188 
2189 int ff_celt_init(AVCodecContext *avctx, CeltContext **ps, int output_channels)
2190 {
2191  CeltContext *s;
2192  int i, ret;
2193 
2194  if (output_channels != 1 && output_channels != 2) {
2195  av_log(avctx, AV_LOG_ERROR, "Invalid number of output channels: %d\n",
2196  output_channels);
2197  return AVERROR(EINVAL);
2198  }
2199 
2200  s = av_mallocz(sizeof(*s));
2201  if (!s)
2202  return AVERROR(ENOMEM);
2203 
2204  s->avctx = avctx;
2205  s->output_channels = output_channels;
2206 
2207  for (i = 0; i < FF_ARRAY_ELEMS(s->imdct); i++) {
2208  ret = ff_imdct15_init(&s->imdct[i], i + 3);
2209  if (ret < 0)
2210  goto fail;
2211  }
2212 
2214  if (!s->dsp) {
2215  ret = AVERROR(ENOMEM);
2216  goto fail;
2217  }
2218 
2219  ff_celt_flush(s);
2220 
2221  *ps = s;
2222 
2223  return 0;
2224 fail:
2225  ff_celt_free(&s);
2226  return ret;
2227 }
#define NULL
Definition: coverity.c:32
const char const char void * val
Definition: avisynth_c.h:634
const char * s
Definition: avisynth_c.h:631
#define AVERROR_INVALIDDATA
Invalid data found when processing input.
Definition: error.h:59
static void process_anticollapse(CeltContext *s, CeltFrame *frame, float *X)
Definition: opus_celt.c:1829
ptrdiff_t const GLvoid * data
Definition: opengl_enc.c:101
static int celt_pulses2bits(const uint8_t *cache, int pulses)
Definition: opus_celt.c:989
static void celt_haar1(float *X, int N0, int stride)
Definition: opus_celt.c:1179
#define CELT_ALLOC_STEPS
Definition: opus.h:47
static av_always_inline unsigned int opus_rc_getsymbol(OpusRangeCoder *rc, const uint16_t *cdf)
Definition: opus.h:215
static const uint8_t celt_static_alloc[11][21]
Definition: opus_celt.c:194
AVCodecContext * avctx
Definition: opus_celt.c:64
float prev_energy[2][CELT_MAX_BANDS]
Definition: opus_celt.c:44
else temp
Definition: vf_mcdeint.c:259
static const float celt_alpha_coef[]
Definition: opus_celt.c:142
const char * g
Definition: vf_curves.c:108
static float celt_decode_pulses(OpusRangeCoder *rc, int *y, unsigned int N, unsigned int K)
Definition: opus_celt.c:1295
float energy[CELT_MAX_BANDS]
Definition: opus_celt.c:43
#define M_SQRT1_2
Definition: mathematics.h:52
static const int8_t pulses[4]
Number of non-zero pulses in the MP-MLQ excitation.
Definition: g723_1.h:720
const char * b
Definition: vf_curves.c:109
#define CELT_POSTFILTER_MINPERIOD
Definition: opus.h:54
int av_log2(unsigned v)
Definition: intmath.c:26
#define DECLARE_ALIGNED(n, t, v)
Definition: mem.h:53
RawBitsContext rb
Definition: opus.h:96
static av_always_inline unsigned int opus_rc_p2model(OpusRangeCoder *rc, unsigned int bits)
Definition: opus.h:234
static const uint8_t celt_coarse_energy_dist[4][2][42]
Definition: opus_celt.c:150
int framebits
Definition: opus_celt.c:76
int pulses[CELT_MAX_BANDS]
Definition: opus_celt.c:98
void ff_celt_free(CeltContext **ps)
Definition: opus_celt.c:2174
int ff_celt_decode_frame(CeltContext *s, OpusRangeCoder *rc, float **output, int coded_channels, int frame_size, int startband, int endband)
Definition: opus_celt.c:1977
#define MUL16(a, b)
Definition: fft-test.c:50
int pf_period
Definition: opus_celt.c:54
static const uint16_t celt_model_energy_small[]
Definition: opus_celt.c:113
static uint64_t celt_cwrsi(unsigned int N, unsigned int K, unsigned int i, int *y)
Definition: opus_celt.c:1210
void(* vector_fmac_scalar)(float *dst, const float *src, float mul, int len)
Multiply a vector of floats by a scalar float and add to destination vector.
Definition: float_dsp.h:54
static void celt_decode_fine_energy(CeltContext *s, OpusRangeCoder *rc)
Definition: opus_celt.c:569
void(* vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len)
Overlap/add with window function.
Definition: float_dsp.h:103
uint8_t bits
Definition: crc.c:295
uint8_t
#define CELT_OVERLAP
Definition: opus.h:42
enum CeltSpread spread
Definition: opus_celt.c:92
float delta
#define Y
Definition: vf_boxblur.c:76
int flushed
Definition: opus_celt.c:72
unsigned int total_read_bits
Definition: opus.h:99
float scratch[22 *8]
Definition: opus_celt.c:102
int fine_bits[CELT_MAX_BANDS]
Definition: opus_celt.c:96
static void celt_denormalize(CeltContext *s, CeltFrame *frame, float *data)
Definition: opus_celt.c:1675
#define opus_ilog(i)
Definition: opus.h:62
#define CELT_DEEMPH_COEFF
Definition: opus.h:53
static void celt_normalize_residual(const int *av_restrict iy, float *av_restrict X, int N, float g)
Definition: opus_celt.c:995
static const uint16_t celt_model_alloc_trim[]
Definition: opus_celt.c:109
#define N
Definition: vf_pp7.c:73
#define CELT_MAX_LOG_BLOCKS
Definition: opus.h:43
static AVFrame * frame
float buf[2048]
Definition: opus_celt.c:49
static const uint8_t celt_log_freq_range[]
Definition: opus_celt.c:123
#define FFMIN3(a, b, c)
Definition: common.h:97
static const uint8_t bits2[81]
Definition: aactab.c:130
#define N2
Definition: vf_pp7.c:69
#define FFALIGN(x, a)
Definition: macros.h:48
static unsigned int celt_alg_unquant(OpusRangeCoder *rc, float *X, unsigned int N, unsigned int K, enum CeltSpread spread, unsigned int blocks, float gain)
Decode pulse vector and combine the result with the pitch vector to produce the final normalised sign...
Definition: opus_celt.c:1306
#define av_log(a,...)
#define cm
Definition: dvbsubdec.c:36
unsigned m
Definition: audioconvert.c:187
static void celt_postfilter(CeltContext *s, CeltFrame *frame)
Definition: opus_celt.c:1765
static const uint8_t celt_bit_interleave[]
Definition: opus_celt.c:276
static unsigned int celt_decode_band(CeltContext *s, OpusRangeCoder *rc, const int band, float *X, float *Y, int N, int b, unsigned int blocks, float *lowband, int duration, float *lowband_out, int level, float gain, float *lowband_scratch, int fill)
Definition: opus_celt.c:1319
CeltFrame frame[2]
Definition: opus_celt.c:70
static const uint8_t celt_freq_bands[]
Definition: opus_celt.c:115
static double alpha(void *priv, double x, double y)
Definition: vf_geq.c:99
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:176
static const int8_t celt_tf_select[4][2][2][2]
Definition: opus_celt.c:127
static const uint8_t celt_static_caps[4][2][21]
Definition: opus_celt.c:208
#define AVERROR(e)
Definition: error.h:43
int pf_period_new
Definition: opus_celt.c:52
static void celt_exp_rotation1(float *X, unsigned int len, unsigned int stride, float c, float s)
Definition: opus_celt.c:1003
void ff_celt_flush(CeltContext *s)
Definition: opus_celt.c:2147
const char * r
Definition: vf_curves.c:107
static void celt_stereo_merge(float *X, float *Y, float mid, int N)
Definition: opus_celt.c:1097
static const uint8_t celt_freq_range[]
Definition: opus_celt.c:119
int fine_priority[CELT_MAX_BANDS]
Definition: opus_celt.c:97
static void celt_renormalize_vector(float *X, int N, float gain)
Definition: opus_celt.c:1085
int flags
AV_CODEC_FLAG_*.
Definition: avcodec.h:1627
av_cold void ff_imdct15_uninit(IMDCT15Context **ps)
Free an iMDCT.
Definition: imdct15.c:69
int codedbands
Definition: opus_celt.c:86
static void celt_decode_allocation(CeltContext *s, OpusRangeCoder *rc)
Definition: opus_celt.c:638
static void celt_decode_tf_changes(CeltContext *s, OpusRangeCoder *rc, int transient)
Definition: opus_celt.c:610
static av_always_inline unsigned int opus_rc_tell_frac(const OpusRangeCoder *rc)
Definition: opus.h:260
static const uint8_t celt_log2_frac[]
Definition: opus_celt.c:272
static const float celt_mean_energy[]
Definition: opus_celt.c:134
static const uint8_t offset[127][2]
Definition: vf_spp.c:92
#define FFMAX(a, b)
Definition: common.h:94
static int celt_compute_qn(int N, int b, int offset, int pulse_cap, int dualstereo)
Definition: opus_celt.c:1193
#define fail()
Definition: checkasm.h:80
static void celt_postfilter_apply(CeltFrame *frame, float *data, int len)
Definition: opus_celt.c:1733
int depth
Definition: v4l.c:62
static const uint16_t celt_model_tapset[]
Definition: opus_celt.c:105
static void celt_decode_final_energy(CeltContext *s, OpusRangeCoder *rc, int bits_left)
Definition: opus_celt.c:588
static char * split(char *message, char delim)
Definition: af_channelmap.c:81
float coeffs[2][CELT_MAX_FRAME_SIZE]
Definition: opus_celt.c:101
int coded_channels
Definition: opus_celt.c:75
int pf_period_old
Definition: opus_celt.c:56
static const uint8_t celt_bit_deinterleave[]
Definition: opus_celt.c:280
#define T(x)
Definition: vp56_arith.h:29
#define E
Definition: avdct.c:32
#define AV_CODEC_FLAG_BITEXACT
Use only bitexact stuff (except (I)DCT).
Definition: avcodec.h:787
float pf_gains_old[3]
Definition: opus_celt.c:57
static const int16_t celt_cache_index[105]
Definition: opus_celt.c:262
#define CELT_ENERGY_SILENCE
Definition: opus.h:55
#define FFMIN(a, b)
Definition: common.h:96
#define CELT_SHORT_BLOCKSIZE
Definition: opus.h:41
static const uint32_t *const celt_pvq_u_row[15]
Definition: opus_celt.c:486
static const float celt_beta_coef[]
Definition: opus_celt.c:146
GLsizei GLboolean const GLfloat * value
Definition: opengl_enc.c:109
#define N0
Definition: vf_pp7.c:67
int32_t
int64_t duration
Definition: movenc-test.c:63
int n
Definition: avisynth_c.h:547
#define CELT_QTHETA_OFFSET
Definition: opus.h:51
static int parse_postfilter(CeltContext *s, OpusRangeCoder *rc, int consumed)
Definition: opus_celt.c:1789
#define FF_ARRAY_ELEMS(a)
int ff_celt_init(AVCodecContext *avctx, CeltContext **ps, int output_channels)
Definition: opus_celt.c:2189
Definition: vf_geq.c:46
AVFloatDSPContext * dsp
Definition: opus_celt.c:66
unsigned int bytes
Definition: opus.h:89
static int celt_bits2pulses(const uint8_t *cache, int bits)
Definition: opus_celt.c:970
#define exp2f(x)
Definition: libm.h:293
int frame_size
Definition: mxfenc.c:1821
static void celt_decode_bands(CeltContext *s, OpusRangeCoder *rc)
Definition: opus_celt.c:1881
static unsigned int celt_extract_collapse_mask(const int *iy, unsigned int N, unsigned int B)
Definition: opus_celt.c:1064
uint8_t collapse_masks[CELT_MAX_BANDS]
Definition: opus_celt.c:46
uint32_t seed
Definition: opus_celt.c:71
static av_always_inline unsigned int opus_rc_stepmodel(OpusRangeCoder *rc, int k0)
Definition: opus.h:361
#define CELT_MAX_BANDS
Definition: opus.h:45
static int celt_log2tan(int isin, int icos)
Definition: opus_celt.c:501
int remaining2
Definition: opus_celt.c:95
main external API structure.
Definition: avcodec.h:1532
static const uint32_t celt_pvq_u[1272]
Definition: opus_celt.c:296
Replacements for frequently missing libm functions.
CeltSpread
Definition: opus_celt.c:35
static uint32_t celt_rng(CeltContext *s)
Definition: opus_celt.c:513
void(* imdct_half)(struct IMDCT15Context *s, float *dst, const float *src, ptrdiff_t src_stride, float scale)
Calculate the middle half of the iMDCT.
Definition: imdct15.h:40
#define CELT_VECTORS
Definition: opus.h:46
int startband
Definition: opus_celt.c:84
av_cold AVFloatDSPContext * avpriv_float_dsp_alloc(int bit_exact)
Allocate a float DSP context.
Definition: float_dsp.c:119
av_cold int ff_imdct15_init(IMDCT15Context **ps, int N)
Init an iMDCT of the length 2 * 15 * (2^N)
Definition: imdct15.c:90
int output_channels
Definition: opus_celt.c:67
static const uint16_t celt_qn_exp2[]
Definition: opus_celt.c:292
float pf_gains_new[3]
Definition: opus_celt.c:53
uint8_t level
Definition: svq3.c:150
static const uint16_t celt_model_spread[]
Definition: opus_celt.c:107
int blocks
Definition: opus_celt.c:80
#define M_SQRT2
Definition: mathematics.h:55
int remaining
Definition: opus_celt.c:94
unsigned int range
Definition: opus.h:97
GLint GLenum GLboolean GLsizei stride
Definition: opengl_enc.c:105
int dualstereo
Definition: opus_celt.c:91
static av_always_inline unsigned int opus_rc_trimodel(OpusRangeCoder *rc, int qn)
Definition: opus.h:376
static av_always_inline unsigned int opus_getrawbits(OpusRangeCoder *rc, unsigned int count)
CELT: read 1-25 raw bits at the end of the frame, backwards byte-wise.
Definition: opus.h:282
#define ROUND_MUL16(a, b)
Definition: opus.h:61
if(ret< 0)
Definition: vf_mcdeint.c:282
#define exp2(x)
Definition: libm.h:288
int tf_change[CELT_MAX_BANDS]
Definition: opus_celt.c:99
static void celt_deinterleave_hadamard(float *tmp, float *X, int N0, int stride, int hadamard)
Definition: opus_celt.c:1158
static double c[64]
static av_always_inline unsigned int opus_rc_unimodel(OpusRangeCoder *rc, unsigned int size)
CELT: read a uniform distribution.
Definition: opus.h:303
static const uint8_t celt_cache_bits[392]
Definition: opus_celt.c:232
IMDCT15Context * imdct[4]
Definition: opus_celt.c:65
const float ff_celt_window2[120]
Definition: opus_celt.c:467
int intensitystereo
Definition: opus_celt.c:90
static av_always_inline int diff(const uint32_t a, const uint32_t b)
float pf_gains[3]
Definition: opus_celt.c:55
int len
static const uint8_t celt_hadamard_ordery[]
Definition: opus_celt.c:285
int anticollapse_bit
Definition: opus_celt.c:88
static av_always_inline unsigned int opus_rc_tell(const OpusRangeCoder *rc)
CELT: estimate bits of entropy that have thus far been consumed for the current CELT frame...
Definition: opus.h:255
static void celt_decode_coarse_energy(CeltContext *s, OpusRangeCoder *rc)
Definition: opus_celt.c:519
int duration
Definition: opus_celt.c:77
#define CELT_MAX_FRAME_SIZE
Definition: opus.h:44
static void celt_exp_rotation(float *X, unsigned int len, unsigned int stride, unsigned int K, enum CeltSpread spread)
Definition: opus_celt.c:1028
#define av_freep(p)
#define M_PI
Definition: mathematics.h:46
static void celt_postfilter_apply_transition(CeltFrame *frame, float *data)
Definition: opus_celt.c:1688
#define stride
static av_always_inline int opus_rc_laplace(OpusRangeCoder *rc, unsigned int symbol, int decay)
Definition: opus.h:322
static int16_t celt_cos(int16_t x)
Definition: opus_celt.c:494
#define CELT_QTHETA_OFFSET_TWOPHASE
Definition: opus.h:52
int endband
Definition: opus_celt.c:85
static const float celt_window[120]
Definition: opus_celt.c:439
static void celt_interleave_hadamard(float *tmp, float *X, int N0, int stride, int hadamard)
Definition: opus_celt.c:1137
#define CELT_MAX_FINE_BITS
Definition: opus.h:49
void * av_mallocz(size_t size)
Allocate a block of size bytes with alignment suitable for all memory accesses (including vectors if ...
Definition: mem.c:252
#define B0
Definition: faandct.c:40
int blocksize
Definition: opus_celt.c:82
static const uint8_t bits1[81]
Definition: aactab.c:107
float deemph_coeff
Definition: opus_celt.c:59
#define CELT_FINE_OFFSET
Definition: opus.h:48
#define CELT_PVQ_V(n, k)