Mercurial > libavcodec.hg
annotate wmadec.c @ 2045:9447bbd8a7e9 libavcodec
rewrite h261 loop filter
no malloc(64) memcpy free stuff
no filter1 A->B then filter2 A->B (yes not B->A)
no incorrect rouding after the 1d filter
author | michael |
---|---|
date | Sun, 30 May 2004 21:21:13 +0000 |
parents | 2b0fc6b25ab8 |
children | 770743eb4329 |
rev | line source |
---|---|
783 | 1 /* |
2 * WMA compatible decoder | |
3 * Copyright (c) 2002 The FFmpeg Project. | |
4 * | |
5 * This library is free software; you can redistribute it and/or | |
6 * modify it under the terms of the GNU Lesser General Public | |
7 * License as published by the Free Software Foundation; either | |
8 * version 2 of the License, or (at your option) any later version. | |
9 * | |
10 * This library is distributed in the hope that it will be useful, | |
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 * Lesser General Public License for more details. | |
14 * | |
15 * You should have received a copy of the GNU Lesser General Public | |
16 * License along with this library; if not, write to the Free Software | |
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
18 */ | |
1106 | 19 |
20 /** | |
21 * @file wmadec.c | |
22 * WMA compatible decoder. | |
1967
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
23 * This decoder handles Microsoft Windows Media Audio data, versions 1 & 2. |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
24 * WMA v1 is identified by audio format 0x160 in Microsoft media files |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
25 * (ASF/AVI/WAV). WMA v2 is identified by audio format 0x161. |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
26 * |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
27 * To use this decoder, a calling application must supply the extra data |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
28 * bytes provided with the WMA data. These are the extra, codec-specific |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
29 * bytes at the end of a WAVEFORMATEX data structure. Transmit these bytes |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
30 * to the decoder using the extradata[_size] fields in AVCodecContext. There |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
31 * should be 4 extra bytes for v1 data and 6 extra bytes for v2 data. |
1106 | 32 */ |
33 | |
783 | 34 #include "avcodec.h" |
35 #include "dsputil.h" | |
36 | |
37 /* size of blocks */ | |
38 #define BLOCK_MIN_BITS 7 | |
39 #define BLOCK_MAX_BITS 11 | |
40 #define BLOCK_MAX_SIZE (1 << BLOCK_MAX_BITS) | |
41 | |
42 #define BLOCK_NB_SIZES (BLOCK_MAX_BITS - BLOCK_MIN_BITS + 1) | |
43 | |
44 /* XXX: find exact max size */ | |
45 #define HIGH_BAND_MAX_SIZE 16 | |
46 | |
47 #define NB_LSP_COEFS 10 | |
48 | |
817 | 49 /* XXX: is it a suitable value ? */ |
783 | 50 #define MAX_CODED_SUPERFRAME_SIZE 4096 |
51 | |
52 #define MAX_CHANNELS 2 | |
53 | |
54 #define NOISE_TAB_SIZE 8192 | |
55 | |
56 #define LSP_POW_BITS 7 | |
57 | |
58 typedef struct WMADecodeContext { | |
59 GetBitContext gb; | |
60 int sample_rate; | |
61 int nb_channels; | |
62 int bit_rate; | |
63 int version; /* 1 = 0x160 (WMAV1), 2 = 0x161 (WMAV2) */ | |
64 int block_align; | |
65 int use_bit_reservoir; | |
66 int use_variable_block_len; | |
67 int use_exp_vlc; /* exponent coding: 0 = lsp, 1 = vlc + delta */ | |
68 int use_noise_coding; /* true if perceptual noise is added */ | |
69 int byte_offset_bits; | |
70 VLC exp_vlc; | |
71 int exponent_sizes[BLOCK_NB_SIZES]; | |
72 uint16_t exponent_bands[BLOCK_NB_SIZES][25]; | |
73 int high_band_start[BLOCK_NB_SIZES]; /* index of first coef in high band */ | |
74 int coefs_start; /* first coded coef */ | |
75 int coefs_end[BLOCK_NB_SIZES]; /* max number of coded coefficients */ | |
76 int exponent_high_sizes[BLOCK_NB_SIZES]; | |
77 int exponent_high_bands[BLOCK_NB_SIZES][HIGH_BAND_MAX_SIZE]; | |
78 VLC hgain_vlc; | |
79 | |
80 /* coded values in high bands */ | |
81 int high_band_coded[MAX_CHANNELS][HIGH_BAND_MAX_SIZE]; | |
82 int high_band_values[MAX_CHANNELS][HIGH_BAND_MAX_SIZE]; | |
83 | |
84 /* there are two possible tables for spectral coefficients */ | |
85 VLC coef_vlc[2]; | |
86 uint16_t *run_table[2]; | |
87 uint16_t *level_table[2]; | |
88 /* frame info */ | |
89 int frame_len; /* frame length in samples */ | |
90 int frame_len_bits; /* frame_len = 1 << frame_len_bits */ | |
91 int nb_block_sizes; /* number of block sizes */ | |
92 /* block info */ | |
93 int reset_block_lengths; | |
94 int block_len_bits; /* log2 of current block length */ | |
95 int next_block_len_bits; /* log2 of next block length */ | |
96 int prev_block_len_bits; /* log2 of prev block length */ | |
97 int block_len; /* block length in samples */ | |
98 int block_num; /* block number in current frame */ | |
99 int block_pos; /* current position in frame */ | |
100 uint8_t ms_stereo; /* true if mid/side stereo mode */ | |
101 uint8_t channel_coded[MAX_CHANNELS]; /* true if channel is coded */ | |
972 | 102 float exponents[MAX_CHANNELS][BLOCK_MAX_SIZE] __attribute__((aligned(16))); |
783 | 103 float max_exponent[MAX_CHANNELS]; |
104 int16_t coefs1[MAX_CHANNELS][BLOCK_MAX_SIZE]; | |
972 | 105 float coefs[MAX_CHANNELS][BLOCK_MAX_SIZE] __attribute__((aligned(16))); |
783 | 106 MDCTContext mdct_ctx[BLOCK_NB_SIZES]; |
1031
19de1445beb2
use av_malloc() functions - added av_strdup and av_realloc()
bellard
parents:
1025
diff
changeset
|
107 float *windows[BLOCK_NB_SIZES]; |
972 | 108 FFTSample mdct_tmp[BLOCK_MAX_SIZE] __attribute__((aligned(16))); /* temporary storage for imdct */ |
783 | 109 /* output buffer for one frame and the last for IMDCT windowing */ |
972 | 110 float frame_out[MAX_CHANNELS][BLOCK_MAX_SIZE * 2] __attribute__((aligned(16))); |
783 | 111 /* last frame info */ |
112 uint8_t last_superframe[MAX_CODED_SUPERFRAME_SIZE + 4]; /* padding added */ | |
113 int last_bitoffset; | |
114 int last_superframe_len; | |
115 float noise_table[NOISE_TAB_SIZE]; | |
116 int noise_index; | |
117 float noise_mult; /* XXX: suppress that and integrate it in the noise array */ | |
118 /* lsp_to_curve tables */ | |
119 float lsp_cos_table[BLOCK_MAX_SIZE]; | |
120 float lsp_pow_e_table[256]; | |
121 float lsp_pow_m_table1[(1 << LSP_POW_BITS)]; | |
122 float lsp_pow_m_table2[(1 << LSP_POW_BITS)]; | |
1343 | 123 |
124 #ifdef TRACE | |
125 int frame_count; | |
126 #endif | |
783 | 127 } WMADecodeContext; |
128 | |
129 typedef struct CoefVLCTable { | |
130 int n; /* total number of codes */ | |
131 const uint32_t *huffcodes; /* VLC bit values */ | |
132 const uint8_t *huffbits; /* VLC bit size */ | |
133 const uint16_t *levels; /* table to build run/level tables */ | |
134 } CoefVLCTable; | |
135 | |
136 static void wma_lsp_to_curve_init(WMADecodeContext *s, int frame_len); | |
137 | |
138 #include "wmadata.h" | |
139 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
140 #ifdef TRACE |
783 | 141 static void dump_shorts(const char *name, const short *tab, int n) |
142 { | |
143 int i; | |
144 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
145 tprintf("%s[%d]:\n", name, n); |
783 | 146 for(i=0;i<n;i++) { |
147 if ((i & 7) == 0) | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
148 tprintf("%4d: ", i); |
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
149 tprintf(" %5d.0", tab[i]); |
783 | 150 if ((i & 7) == 7) |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
151 tprintf("\n"); |
783 | 152 } |
153 } | |
154 | |
155 static void dump_floats(const char *name, int prec, const float *tab, int n) | |
156 { | |
157 int i; | |
158 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
159 tprintf("%s[%d]:\n", name, n); |
783 | 160 for(i=0;i<n;i++) { |
161 if ((i & 7) == 0) | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
162 tprintf("%4d: ", i); |
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
163 tprintf(" %8.*f", prec, tab[i]); |
783 | 164 if ((i & 7) == 7) |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
165 tprintf("\n"); |
783 | 166 } |
167 if ((i & 7) != 0) | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
168 tprintf("\n"); |
783 | 169 } |
170 #endif | |
171 | |
172 /* XXX: use same run/length optimization as mpeg decoders */ | |
173 static void init_coef_vlc(VLC *vlc, | |
174 uint16_t **prun_table, uint16_t **plevel_table, | |
175 const CoefVLCTable *vlc_table) | |
176 { | |
177 int n = vlc_table->n; | |
178 const uint8_t *table_bits = vlc_table->huffbits; | |
179 const uint32_t *table_codes = vlc_table->huffcodes; | |
180 const uint16_t *levels_table = vlc_table->levels; | |
181 uint16_t *run_table, *level_table; | |
182 const uint16_t *p; | |
183 int i, l, j, level; | |
184 | |
185 init_vlc(vlc, 9, n, table_bits, 1, 1, table_codes, 4, 4); | |
186 | |
1031
19de1445beb2
use av_malloc() functions - added av_strdup and av_realloc()
bellard
parents:
1025
diff
changeset
|
187 run_table = av_malloc(n * sizeof(uint16_t)); |
19de1445beb2
use av_malloc() functions - added av_strdup and av_realloc()
bellard
parents:
1025
diff
changeset
|
188 level_table = av_malloc(n * sizeof(uint16_t)); |
783 | 189 p = levels_table; |
190 i = 2; | |
191 level = 1; | |
192 while (i < n) { | |
193 l = *p++; | |
194 for(j=0;j<l;j++) { | |
195 run_table[i] = j; | |
196 level_table[i] = level; | |
197 i++; | |
198 } | |
199 level++; | |
200 } | |
201 *prun_table = run_table; | |
202 *plevel_table = level_table; | |
203 } | |
204 | |
205 static int wma_decode_init(AVCodecContext * avctx) | |
206 { | |
207 WMADecodeContext *s = avctx->priv_data; | |
208 int i, flags1, flags2; | |
209 float *window; | |
210 uint8_t *extradata; | |
211 float bps1, high_freq, bps; | |
212 int sample_rate1; | |
213 int coef_vlc_table; | |
214 | |
215 s->sample_rate = avctx->sample_rate; | |
216 s->nb_channels = avctx->channels; | |
217 s->bit_rate = avctx->bit_rate; | |
218 s->block_align = avctx->block_align; | |
219 | |
808
e9bfaabcf07d
fixed nb_block_sizes detection - fixed codec_id test (avctx->codec_id does not need to be initialized)
bellard
parents:
797
diff
changeset
|
220 if (avctx->codec->id == CODEC_ID_WMAV1) { |
783 | 221 s->version = 1; |
222 } else { | |
223 s->version = 2; | |
224 } | |
225 | |
226 /* extract flag infos */ | |
227 flags1 = 0; | |
228 flags2 = 0; | |
229 extradata = avctx->extradata; | |
230 if (s->version == 1 && avctx->extradata_size >= 4) { | |
231 flags1 = extradata[0] | (extradata[1] << 8); | |
232 flags2 = extradata[2] | (extradata[3] << 8); | |
233 } else if (s->version == 2 && avctx->extradata_size >= 6) { | |
234 flags1 = extradata[0] | (extradata[1] << 8) | | |
235 (extradata[2] << 16) | (extradata[3] << 24); | |
236 flags2 = extradata[4] | (extradata[5] << 8); | |
237 } | |
238 s->use_exp_vlc = flags2 & 0x0001; | |
239 s->use_bit_reservoir = flags2 & 0x0002; | |
240 s->use_variable_block_len = flags2 & 0x0004; | |
241 | |
242 /* compute MDCT block size */ | |
243 if (s->sample_rate <= 16000) { | |
244 s->frame_len_bits = 9; | |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
245 } else if (s->sample_rate <= 22050 || |
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
246 (s->sample_rate <= 32000 && s->version == 1)) { |
783 | 247 s->frame_len_bits = 10; |
248 } else { | |
249 s->frame_len_bits = 11; | |
250 } | |
251 s->frame_len = 1 << s->frame_len_bits; | |
252 if (s->use_variable_block_len) { | |
808
e9bfaabcf07d
fixed nb_block_sizes detection - fixed codec_id test (avctx->codec_id does not need to be initialized)
bellard
parents:
797
diff
changeset
|
253 int nb_max, nb; |
e9bfaabcf07d
fixed nb_block_sizes detection - fixed codec_id test (avctx->codec_id does not need to be initialized)
bellard
parents:
797
diff
changeset
|
254 nb = ((flags2 >> 3) & 3) + 1; |
e9bfaabcf07d
fixed nb_block_sizes detection - fixed codec_id test (avctx->codec_id does not need to be initialized)
bellard
parents:
797
diff
changeset
|
255 if ((s->bit_rate / s->nb_channels) >= 32000) |
e9bfaabcf07d
fixed nb_block_sizes detection - fixed codec_id test (avctx->codec_id does not need to be initialized)
bellard
parents:
797
diff
changeset
|
256 nb += 2; |
e9bfaabcf07d
fixed nb_block_sizes detection - fixed codec_id test (avctx->codec_id does not need to be initialized)
bellard
parents:
797
diff
changeset
|
257 nb_max = s->frame_len_bits - BLOCK_MIN_BITS; |
e9bfaabcf07d
fixed nb_block_sizes detection - fixed codec_id test (avctx->codec_id does not need to be initialized)
bellard
parents:
797
diff
changeset
|
258 if (nb > nb_max) |
e9bfaabcf07d
fixed nb_block_sizes detection - fixed codec_id test (avctx->codec_id does not need to be initialized)
bellard
parents:
797
diff
changeset
|
259 nb = nb_max; |
e9bfaabcf07d
fixed nb_block_sizes detection - fixed codec_id test (avctx->codec_id does not need to be initialized)
bellard
parents:
797
diff
changeset
|
260 s->nb_block_sizes = nb + 1; |
783 | 261 } else { |
262 s->nb_block_sizes = 1; | |
263 } | |
264 | |
265 /* init rate dependant parameters */ | |
266 s->use_noise_coding = 1; | |
267 high_freq = s->sample_rate * 0.5; | |
268 | |
269 /* if version 2, then the rates are normalized */ | |
270 sample_rate1 = s->sample_rate; | |
271 if (s->version == 2) { | |
272 if (sample_rate1 >= 44100) | |
273 sample_rate1 = 44100; | |
274 else if (sample_rate1 >= 22050) | |
275 sample_rate1 = 22050; | |
276 else if (sample_rate1 >= 16000) | |
277 sample_rate1 = 16000; | |
278 else if (sample_rate1 >= 11025) | |
279 sample_rate1 = 11025; | |
280 else if (sample_rate1 >= 8000) | |
281 sample_rate1 = 8000; | |
282 } | |
283 | |
284 bps = (float)s->bit_rate / (float)(s->nb_channels * s->sample_rate); | |
285 s->byte_offset_bits = av_log2((int)(bps * s->frame_len / 8.0)) + 2; | |
286 | |
287 /* compute high frequency value and choose if noise coding should | |
288 be activated */ | |
289 bps1 = bps; | |
290 if (s->nb_channels == 2) | |
291 bps1 = bps * 1.6; | |
292 if (sample_rate1 == 44100) { | |
293 if (bps1 >= 0.61) | |
294 s->use_noise_coding = 0; | |
295 else | |
296 high_freq = high_freq * 0.4; | |
297 } else if (sample_rate1 == 22050) { | |
298 if (bps1 >= 1.16) | |
299 s->use_noise_coding = 0; | |
300 else if (bps1 >= 0.72) | |
301 high_freq = high_freq * 0.7; | |
302 else | |
303 high_freq = high_freq * 0.6; | |
304 } else if (sample_rate1 == 16000) { | |
305 if (bps > 0.5) | |
306 high_freq = high_freq * 0.5; | |
307 else | |
308 high_freq = high_freq * 0.3; | |
309 } else if (sample_rate1 == 11025) { | |
310 high_freq = high_freq * 0.7; | |
311 } else if (sample_rate1 == 8000) { | |
312 if (bps <= 0.625) { | |
313 high_freq = high_freq * 0.5; | |
314 } else if (bps > 0.75) { | |
315 s->use_noise_coding = 0; | |
316 } else { | |
317 high_freq = high_freq * 0.65; | |
318 } | |
319 } else { | |
320 if (bps >= 0.8) { | |
321 high_freq = high_freq * 0.75; | |
322 } else if (bps >= 0.6) { | |
323 high_freq = high_freq * 0.6; | |
324 } else { | |
325 high_freq = high_freq * 0.5; | |
326 } | |
327 } | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
328 dprintf("flags1=0x%x flags2=0x%x\n", flags1, flags2); |
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
329 dprintf("version=%d channels=%d sample_rate=%d bitrate=%d block_align=%d\n", |
783 | 330 s->version, s->nb_channels, s->sample_rate, s->bit_rate, |
331 s->block_align); | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
332 dprintf("bps=%f bps1=%f high_freq=%f bitoffset=%d\n", |
783 | 333 bps, bps1, high_freq, s->byte_offset_bits); |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
334 dprintf("use_noise_coding=%d use_exp_vlc=%d nb_block_sizes=%d\n", |
808
e9bfaabcf07d
fixed nb_block_sizes detection - fixed codec_id test (avctx->codec_id does not need to be initialized)
bellard
parents:
797
diff
changeset
|
335 s->use_noise_coding, s->use_exp_vlc, s->nb_block_sizes); |
783 | 336 |
337 /* compute the scale factor band sizes for each MDCT block size */ | |
338 { | |
339 int a, b, pos, lpos, k, block_len, i, j, n; | |
340 const uint8_t *table; | |
341 | |
342 if (s->version == 1) { | |
343 s->coefs_start = 3; | |
344 } else { | |
345 s->coefs_start = 0; | |
346 } | |
347 for(k = 0; k < s->nb_block_sizes; k++) { | |
348 block_len = s->frame_len >> k; | |
349 | |
350 if (s->version == 1) { | |
351 lpos = 0; | |
352 for(i=0;i<25;i++) { | |
353 a = wma_critical_freqs[i]; | |
354 b = s->sample_rate; | |
355 pos = ((block_len * 2 * a) + (b >> 1)) / b; | |
356 if (pos > block_len) | |
357 pos = block_len; | |
358 s->exponent_bands[0][i] = pos - lpos; | |
359 if (pos >= block_len) { | |
360 i++; | |
361 break; | |
362 } | |
363 lpos = pos; | |
364 } | |
365 s->exponent_sizes[0] = i; | |
366 } else { | |
367 /* hardcoded tables */ | |
368 table = NULL; | |
369 a = s->frame_len_bits - BLOCK_MIN_BITS - k; | |
370 if (a < 3) { | |
371 if (s->sample_rate >= 44100) | |
372 table = exponent_band_44100[a]; | |
373 else if (s->sample_rate >= 32000) | |
374 table = exponent_band_32000[a]; | |
375 else if (s->sample_rate >= 22050) | |
376 table = exponent_band_22050[a]; | |
377 } | |
378 if (table) { | |
379 n = *table++; | |
380 for(i=0;i<n;i++) | |
381 s->exponent_bands[k][i] = table[i]; | |
382 s->exponent_sizes[k] = n; | |
383 } else { | |
384 j = 0; | |
385 lpos = 0; | |
386 for(i=0;i<25;i++) { | |
387 a = wma_critical_freqs[i]; | |
388 b = s->sample_rate; | |
389 pos = ((block_len * 2 * a) + (b << 1)) / (4 * b); | |
390 pos <<= 2; | |
391 if (pos > block_len) | |
392 pos = block_len; | |
393 if (pos > lpos) | |
394 s->exponent_bands[k][j++] = pos - lpos; | |
395 if (pos >= block_len) | |
396 break; | |
397 lpos = pos; | |
398 } | |
399 s->exponent_sizes[k] = j; | |
400 } | |
401 } | |
402 | |
403 /* max number of coefs */ | |
404 s->coefs_end[k] = (s->frame_len - ((s->frame_len * 9) / 100)) >> k; | |
405 /* high freq computation */ | |
406 s->high_band_start[k] = (int)((block_len * 2 * high_freq) / | |
407 s->sample_rate + 0.5); | |
408 n = s->exponent_sizes[k]; | |
409 j = 0; | |
410 pos = 0; | |
411 for(i=0;i<n;i++) { | |
412 int start, end; | |
413 start = pos; | |
414 pos += s->exponent_bands[k][i]; | |
415 end = pos; | |
416 if (start < s->high_band_start[k]) | |
417 start = s->high_band_start[k]; | |
418 if (end > s->coefs_end[k]) | |
419 end = s->coefs_end[k]; | |
420 if (end > start) | |
421 s->exponent_high_bands[k][j++] = end - start; | |
422 } | |
423 s->exponent_high_sizes[k] = j; | |
424 #if 0 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
425 tprintf("%5d: coefs_end=%d high_band_start=%d nb_high_bands=%d: ", |
783 | 426 s->frame_len >> k, |
427 s->coefs_end[k], | |
428 s->high_band_start[k], | |
429 s->exponent_high_sizes[k]); | |
430 for(j=0;j<s->exponent_high_sizes[k];j++) | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
431 tprintf(" %d", s->exponent_high_bands[k][j]); |
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
432 tprintf("\n"); |
783 | 433 #endif |
434 } | |
435 } | |
436 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
437 #ifdef TRACE |
783 | 438 { |
439 int i, j; | |
440 for(i = 0; i < s->nb_block_sizes; i++) { | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
441 tprintf("%5d: n=%2d:", |
783 | 442 s->frame_len >> i, |
443 s->exponent_sizes[i]); | |
444 for(j=0;j<s->exponent_sizes[i];j++) | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
445 tprintf(" %d", s->exponent_bands[i][j]); |
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
446 tprintf("\n"); |
783 | 447 } |
448 } | |
449 #endif | |
450 | |
451 /* init MDCT */ | |
452 for(i = 0; i < s->nb_block_sizes; i++) | |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
453 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1); |
783 | 454 |
455 /* init MDCT windows : simple sinus window */ | |
456 for(i = 0; i < s->nb_block_sizes; i++) { | |
457 int n, j; | |
458 float alpha; | |
459 n = 1 << (s->frame_len_bits - i); | |
460 window = av_malloc(sizeof(float) * n); | |
461 alpha = M_PI / (2.0 * n); | |
462 for(j=0;j<n;j++) { | |
463 window[n - j - 1] = sin((j + 0.5) * alpha); | |
464 } | |
465 s->windows[i] = window; | |
466 } | |
467 | |
468 s->reset_block_lengths = 1; | |
469 | |
470 if (s->use_noise_coding) { | |
471 | |
472 /* init the noise generator */ | |
473 if (s->use_exp_vlc) | |
474 s->noise_mult = 0.02; | |
475 else | |
476 s->noise_mult = 0.04; | |
477 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
478 #ifdef TRACE |
783 | 479 for(i=0;i<NOISE_TAB_SIZE;i++) |
480 s->noise_table[i] = 1.0 * s->noise_mult; | |
481 #else | |
482 { | |
483 unsigned int seed; | |
484 float norm; | |
485 seed = 1; | |
486 norm = (1.0 / (float)(1LL << 31)) * sqrt(3) * s->noise_mult; | |
487 for(i=0;i<NOISE_TAB_SIZE;i++) { | |
488 seed = seed * 314159 + 1; | |
489 s->noise_table[i] = (float)((int)seed) * norm; | |
490 } | |
491 } | |
492 #endif | |
493 init_vlc(&s->hgain_vlc, 9, sizeof(hgain_huffbits), | |
494 hgain_huffbits, 1, 1, | |
495 hgain_huffcodes, 2, 2); | |
496 } | |
497 | |
498 if (s->use_exp_vlc) { | |
499 init_vlc(&s->exp_vlc, 9, sizeof(scale_huffbits), | |
500 scale_huffbits, 1, 1, | |
501 scale_huffcodes, 4, 4); | |
502 } else { | |
503 wma_lsp_to_curve_init(s, s->frame_len); | |
504 } | |
505 | |
506 /* choose the VLC tables for the coefficients */ | |
507 coef_vlc_table = 2; | |
508 if (s->sample_rate >= 32000) { | |
509 if (bps1 < 0.72) | |
510 coef_vlc_table = 0; | |
511 else if (bps1 < 1.16) | |
512 coef_vlc_table = 1; | |
513 } | |
514 | |
515 init_coef_vlc(&s->coef_vlc[0], &s->run_table[0], &s->level_table[0], | |
516 &coef_vlcs[coef_vlc_table * 2]); | |
517 init_coef_vlc(&s->coef_vlc[1], &s->run_table[1], &s->level_table[1], | |
518 &coef_vlcs[coef_vlc_table * 2 + 1]); | |
519 return 0; | |
520 } | |
521 | |
522 /* interpolate values for a bigger or smaller block. The block must | |
523 have multiple sizes */ | |
524 static void interpolate_array(float *scale, int old_size, int new_size) | |
525 { | |
526 int i, j, jincr, k; | |
527 float v; | |
528 | |
529 if (new_size > old_size) { | |
530 jincr = new_size / old_size; | |
531 j = new_size; | |
532 for(i = old_size - 1; i >=0; i--) { | |
533 v = scale[i]; | |
534 k = jincr; | |
535 do { | |
536 scale[--j] = v; | |
537 } while (--k); | |
538 } | |
539 } else if (new_size < old_size) { | |
540 j = 0; | |
541 jincr = old_size / new_size; | |
542 for(i = 0; i < new_size; i++) { | |
543 scale[i] = scale[j]; | |
544 j += jincr; | |
545 } | |
546 } | |
547 } | |
548 | |
549 /* compute x^-0.25 with an exponent and mantissa table. We use linear | |
550 interpolation to reduce the mantissa table size at a small speed | |
551 expense (linear interpolation approximately doubles the number of | |
552 bits of precision). */ | |
553 static inline float pow_m1_4(WMADecodeContext *s, float x) | |
554 { | |
555 union { | |
556 float f; | |
557 unsigned int v; | |
558 } u, t; | |
559 unsigned int e, m; | |
560 float a, b; | |
561 | |
562 u.f = x; | |
563 e = u.v >> 23; | |
564 m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1); | |
565 /* build interpolation scale: 1 <= t < 2. */ | |
566 t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23); | |
567 a = s->lsp_pow_m_table1[m]; | |
568 b = s->lsp_pow_m_table2[m]; | |
569 return s->lsp_pow_e_table[e] * (a + b * t.f); | |
570 } | |
571 | |
572 static void wma_lsp_to_curve_init(WMADecodeContext *s, int frame_len) | |
573 { | |
574 float wdel, a, b; | |
575 int i, e, m; | |
576 | |
577 wdel = M_PI / frame_len; | |
578 for(i=0;i<frame_len;i++) | |
579 s->lsp_cos_table[i] = 2.0f * cos(wdel * i); | |
580 | |
581 /* tables for x^-0.25 computation */ | |
582 for(i=0;i<256;i++) { | |
583 e = i - 126; | |
584 s->lsp_pow_e_table[i] = pow(2.0, e * -0.25); | |
585 } | |
586 | |
587 /* NOTE: these two tables are needed to avoid two operations in | |
588 pow_m1_4 */ | |
589 b = 1.0; | |
590 for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) { | |
591 m = (1 << LSP_POW_BITS) + i; | |
592 a = (float)m * (0.5 / (1 << LSP_POW_BITS)); | |
593 a = pow(a, -0.25); | |
594 s->lsp_pow_m_table1[i] = 2 * a - b; | |
595 s->lsp_pow_m_table2[i] = b - a; | |
596 b = a; | |
597 } | |
598 #if 0 | |
599 for(i=1;i<20;i++) { | |
600 float v, r1, r2; | |
601 v = 5.0 / i; | |
602 r1 = pow_m1_4(s, v); | |
603 r2 = pow(v,-0.25); | |
604 printf("%f^-0.25=%f e=%f\n", v, r1, r2 - r1); | |
605 } | |
606 #endif | |
607 } | |
608 | |
609 /* NOTE: We use the same code as Vorbis here */ | |
610 /* XXX: optimize it further with SSE/3Dnow */ | |
611 static void wma_lsp_to_curve(WMADecodeContext *s, | |
612 float *out, float *val_max_ptr, | |
613 int n, float *lsp) | |
614 { | |
615 int i, j; | |
616 float p, q, w, v, val_max; | |
617 | |
618 val_max = 0; | |
619 for(i=0;i<n;i++) { | |
620 p = 0.5f; | |
621 q = 0.5f; | |
622 w = s->lsp_cos_table[i]; | |
623 for(j=1;j<NB_LSP_COEFS;j+=2){ | |
624 q *= w - lsp[j - 1]; | |
625 p *= w - lsp[j]; | |
626 } | |
627 p *= p * (2.0f - w); | |
628 q *= q * (2.0f + w); | |
629 v = p + q; | |
630 v = pow_m1_4(s, v); | |
631 if (v > val_max) | |
632 val_max = v; | |
633 out[i] = v; | |
634 } | |
635 *val_max_ptr = val_max; | |
636 } | |
637 | |
638 /* decode exponents coded with LSP coefficients (same idea as Vorbis) */ | |
639 static void decode_exp_lsp(WMADecodeContext *s, int ch) | |
640 { | |
641 float lsp_coefs[NB_LSP_COEFS]; | |
642 int val, i; | |
643 | |
644 for(i = 0; i < NB_LSP_COEFS; i++) { | |
645 if (i == 0 || i >= 8) | |
646 val = get_bits(&s->gb, 3); | |
647 else | |
648 val = get_bits(&s->gb, 4); | |
649 lsp_coefs[i] = lsp_codebook[i][val]; | |
650 } | |
651 | |
652 wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch], | |
653 s->block_len, lsp_coefs); | |
654 } | |
655 | |
656 /* decode exponents coded with VLC codes */ | |
657 static int decode_exp_vlc(WMADecodeContext *s, int ch) | |
658 { | |
659 int last_exp, n, code; | |
660 const uint16_t *ptr, *band_ptr; | |
661 float v, *q, max_scale, *q_end; | |
662 | |
663 band_ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits]; | |
664 ptr = band_ptr; | |
665 q = s->exponents[ch]; | |
666 q_end = q + s->block_len; | |
667 max_scale = 0; | |
668 if (s->version == 1) { | |
669 last_exp = get_bits(&s->gb, 5) + 10; | |
670 /* XXX: use a table */ | |
671 v = pow(10, last_exp * (1.0 / 16.0)); | |
672 max_scale = v; | |
673 n = *ptr++; | |
674 do { | |
675 *q++ = v; | |
676 } while (--n); | |
677 } | |
678 last_exp = 36; | |
679 while (q < q_end) { | |
680 code = get_vlc(&s->gb, &s->exp_vlc); | |
681 if (code < 0) | |
682 return -1; | |
683 /* NOTE: this offset is the same as MPEG4 AAC ! */ | |
684 last_exp += code - 60; | |
685 /* XXX: use a table */ | |
686 v = pow(10, last_exp * (1.0 / 16.0)); | |
687 if (v > max_scale) | |
688 max_scale = v; | |
689 n = *ptr++; | |
690 do { | |
691 *q++ = v; | |
692 } while (--n); | |
693 } | |
694 s->max_exponent[ch] = max_scale; | |
695 return 0; | |
696 } | |
697 | |
698 /* return 0 if OK. return 1 if last block of frame. return -1 if | |
699 unrecorrable error. */ | |
700 static int wma_decode_block(WMADecodeContext *s) | |
701 { | |
702 int n, v, a, ch, code, bsize; | |
703 int coef_nb_bits, total_gain, parse_exponents; | |
704 float window[BLOCK_MAX_SIZE * 2]; | |
705 int nb_coefs[MAX_CHANNELS]; | |
706 float mdct_norm; | |
707 | |
1343 | 708 #ifdef TRACE |
709 tprintf("***decode_block: %d:%d\n", s->frame_count - 1, s->block_num); | |
710 #endif | |
783 | 711 |
712 /* compute current block length */ | |
713 if (s->use_variable_block_len) { | |
714 n = av_log2(s->nb_block_sizes - 1) + 1; | |
715 | |
716 if (s->reset_block_lengths) { | |
717 s->reset_block_lengths = 0; | |
718 v = get_bits(&s->gb, n); | |
719 if (v >= s->nb_block_sizes) | |
720 return -1; | |
721 s->prev_block_len_bits = s->frame_len_bits - v; | |
722 v = get_bits(&s->gb, n); | |
723 if (v >= s->nb_block_sizes) | |
724 return -1; | |
725 s->block_len_bits = s->frame_len_bits - v; | |
726 } else { | |
727 /* update block lengths */ | |
728 s->prev_block_len_bits = s->block_len_bits; | |
729 s->block_len_bits = s->next_block_len_bits; | |
730 } | |
731 v = get_bits(&s->gb, n); | |
732 if (v >= s->nb_block_sizes) | |
733 return -1; | |
734 s->next_block_len_bits = s->frame_len_bits - v; | |
735 } else { | |
736 /* fixed block len */ | |
737 s->next_block_len_bits = s->frame_len_bits; | |
738 s->prev_block_len_bits = s->frame_len_bits; | |
739 s->block_len_bits = s->frame_len_bits; | |
740 } | |
741 | |
742 /* now check if the block length is coherent with the frame length */ | |
743 s->block_len = 1 << s->block_len_bits; | |
744 if ((s->block_pos + s->block_len) > s->frame_len) | |
745 return -1; | |
746 | |
747 if (s->nb_channels == 2) { | |
748 s->ms_stereo = get_bits(&s->gb, 1); | |
749 } | |
750 v = 0; | |
751 for(ch = 0; ch < s->nb_channels; ch++) { | |
752 a = get_bits(&s->gb, 1); | |
753 s->channel_coded[ch] = a; | |
754 v |= a; | |
755 } | |
756 /* if no channel coded, no need to go further */ | |
757 /* XXX: fix potential framing problems */ | |
758 if (!v) | |
759 goto next; | |
760 | |
761 bsize = s->frame_len_bits - s->block_len_bits; | |
762 | |
763 /* read total gain and extract corresponding number of bits for | |
764 coef escape coding */ | |
765 total_gain = 1; | |
766 for(;;) { | |
767 a = get_bits(&s->gb, 7); | |
768 total_gain += a; | |
769 if (a != 127) | |
770 break; | |
771 } | |
772 | |
773 if (total_gain < 15) | |
774 coef_nb_bits = 13; | |
775 else if (total_gain < 32) | |
776 coef_nb_bits = 12; | |
777 else if (total_gain < 40) | |
778 coef_nb_bits = 11; | |
779 else if (total_gain < 45) | |
780 coef_nb_bits = 10; | |
781 else | |
782 coef_nb_bits = 9; | |
783 | |
784 /* compute number of coefficients */ | |
785 n = s->coefs_end[bsize] - s->coefs_start; | |
786 for(ch = 0; ch < s->nb_channels; ch++) | |
787 nb_coefs[ch] = n; | |
788 | |
789 /* complex coding */ | |
790 if (s->use_noise_coding) { | |
791 | |
792 for(ch = 0; ch < s->nb_channels; ch++) { | |
793 if (s->channel_coded[ch]) { | |
794 int i, n, a; | |
795 n = s->exponent_high_sizes[bsize]; | |
796 for(i=0;i<n;i++) { | |
797 a = get_bits(&s->gb, 1); | |
798 s->high_band_coded[ch][i] = a; | |
799 /* if noise coding, the coefficients are not transmitted */ | |
800 if (a) | |
801 nb_coefs[ch] -= s->exponent_high_bands[bsize][i]; | |
802 } | |
803 } | |
804 } | |
805 for(ch = 0; ch < s->nb_channels; ch++) { | |
806 if (s->channel_coded[ch]) { | |
807 int i, n, val, code; | |
808 | |
809 n = s->exponent_high_sizes[bsize]; | |
810 val = (int)0x80000000; | |
811 for(i=0;i<n;i++) { | |
812 if (s->high_band_coded[ch][i]) { | |
813 if (val == (int)0x80000000) { | |
814 val = get_bits(&s->gb, 7) - 19; | |
815 } else { | |
816 code = get_vlc(&s->gb, &s->hgain_vlc); | |
817 if (code < 0) | |
818 return -1; | |
819 val += code - 18; | |
820 } | |
821 s->high_band_values[ch][i] = val; | |
822 } | |
823 } | |
824 } | |
825 } | |
826 } | |
827 | |
828 /* exposant can be interpolated in short blocks. */ | |
829 parse_exponents = 1; | |
830 if (s->block_len_bits != s->frame_len_bits) { | |
831 parse_exponents = get_bits(&s->gb, 1); | |
832 } | |
833 | |
834 if (parse_exponents) { | |
835 for(ch = 0; ch < s->nb_channels; ch++) { | |
836 if (s->channel_coded[ch]) { | |
837 if (s->use_exp_vlc) { | |
838 if (decode_exp_vlc(s, ch) < 0) | |
839 return -1; | |
840 } else { | |
841 decode_exp_lsp(s, ch); | |
842 } | |
843 } | |
844 } | |
845 } else { | |
846 for(ch = 0; ch < s->nb_channels; ch++) { | |
847 if (s->channel_coded[ch]) { | |
848 interpolate_array(s->exponents[ch], 1 << s->prev_block_len_bits, | |
849 s->block_len); | |
850 } | |
851 } | |
852 } | |
853 | |
854 /* parse spectral coefficients : just RLE encoding */ | |
855 for(ch = 0; ch < s->nb_channels; ch++) { | |
856 if (s->channel_coded[ch]) { | |
857 VLC *coef_vlc; | |
858 int level, run, sign, tindex; | |
859 int16_t *ptr, *eptr; | |
860 const int16_t *level_table, *run_table; | |
861 | |
862 /* special VLC tables are used for ms stereo because | |
863 there is potentially less energy there */ | |
864 tindex = (ch == 1 && s->ms_stereo); | |
865 coef_vlc = &s->coef_vlc[tindex]; | |
866 run_table = s->run_table[tindex]; | |
867 level_table = s->level_table[tindex]; | |
868 /* XXX: optimize */ | |
869 ptr = &s->coefs1[ch][0]; | |
870 eptr = ptr + nb_coefs[ch]; | |
871 memset(ptr, 0, s->block_len * sizeof(int16_t)); | |
872 for(;;) { | |
873 code = get_vlc(&s->gb, coef_vlc); | |
874 if (code < 0) | |
875 return -1; | |
876 if (code == 1) { | |
877 /* EOB */ | |
878 break; | |
879 } else if (code == 0) { | |
880 /* escape */ | |
881 level = get_bits(&s->gb, coef_nb_bits); | |
882 /* NOTE: this is rather suboptimal. reading | |
883 block_len_bits would be better */ | |
884 run = get_bits(&s->gb, s->frame_len_bits); | |
885 } else { | |
886 /* normal code */ | |
887 run = run_table[code]; | |
888 level = level_table[code]; | |
889 } | |
890 sign = get_bits(&s->gb, 1); | |
891 if (!sign) | |
892 level = -level; | |
893 ptr += run; | |
894 if (ptr >= eptr) | |
895 return -1; | |
896 *ptr++ = level; | |
897 /* NOTE: EOB can be omitted */ | |
898 if (ptr >= eptr) | |
899 break; | |
900 } | |
901 } | |
902 if (s->version == 1 && s->nb_channels >= 2) { | |
903 align_get_bits(&s->gb); | |
904 } | |
905 } | |
906 | |
907 /* normalize */ | |
908 { | |
909 int n4 = s->block_len / 2; | |
910 mdct_norm = 1.0 / (float)n4; | |
911 if (s->version == 1) { | |
912 mdct_norm *= sqrt(n4); | |
913 } | |
914 } | |
915 | |
916 /* finally compute the MDCT coefficients */ | |
917 for(ch = 0; ch < s->nb_channels; ch++) { | |
918 if (s->channel_coded[ch]) { | |
919 int16_t *coefs1; | |
920 float *coefs, *exponents, mult, mult1, noise, *exp_ptr; | |
921 int i, j, n, n1, last_high_band; | |
922 float exp_power[HIGH_BAND_MAX_SIZE]; | |
923 | |
924 coefs1 = s->coefs1[ch]; | |
925 exponents = s->exponents[ch]; | |
926 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch]; | |
927 mult *= mdct_norm; | |
928 coefs = s->coefs[ch]; | |
929 if (s->use_noise_coding) { | |
930 mult1 = mult; | |
931 /* very low freqs : noise */ | |
932 for(i = 0;i < s->coefs_start; i++) { | |
933 *coefs++ = s->noise_table[s->noise_index] * (*exponents++) * mult1; | |
934 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
935 } | |
936 | |
937 n1 = s->exponent_high_sizes[bsize]; | |
938 | |
939 /* compute power of high bands */ | |
940 exp_ptr = exponents + | |
941 s->high_band_start[bsize] - | |
942 s->coefs_start; | |
943 last_high_band = 0; /* avoid warning */ | |
944 for(j=0;j<n1;j++) { | |
945 n = s->exponent_high_bands[s->frame_len_bits - | |
946 s->block_len_bits][j]; | |
947 if (s->high_band_coded[ch][j]) { | |
948 float e2, v; | |
949 e2 = 0; | |
950 for(i = 0;i < n; i++) { | |
951 v = exp_ptr[i]; | |
952 e2 += v * v; | |
953 } | |
954 exp_power[j] = e2 / n; | |
955 last_high_band = j; | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
956 tprintf("%d: power=%f (%d)\n", j, exp_power[j], n); |
783 | 957 } |
958 exp_ptr += n; | |
959 } | |
960 | |
961 /* main freqs and high freqs */ | |
962 for(j=-1;j<n1;j++) { | |
963 if (j < 0) { | |
964 n = s->high_band_start[bsize] - | |
965 s->coefs_start; | |
966 } else { | |
967 n = s->exponent_high_bands[s->frame_len_bits - | |
968 s->block_len_bits][j]; | |
969 } | |
970 if (j >= 0 && s->high_band_coded[ch][j]) { | |
971 /* use noise with specified power */ | |
972 mult1 = sqrt(exp_power[j] / exp_power[last_high_band]); | |
973 /* XXX: use a table */ | |
974 mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05); | |
975 mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult); | |
976 mult1 *= mdct_norm; | |
977 for(i = 0;i < n; i++) { | |
978 noise = s->noise_table[s->noise_index]; | |
979 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
980 *coefs++ = (*exponents++) * noise * mult1; | |
981 } | |
982 } else { | |
983 /* coded values + small noise */ | |
984 for(i = 0;i < n; i++) { | |
985 noise = s->noise_table[s->noise_index]; | |
986 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
987 *coefs++ = ((*coefs1++) + noise) * (*exponents++) * mult; | |
988 } | |
989 } | |
990 } | |
991 | |
992 /* very high freqs : noise */ | |
993 n = s->block_len - s->coefs_end[bsize]; | |
994 mult1 = mult * exponents[-1]; | |
995 for(i = 0; i < n; i++) { | |
996 *coefs++ = s->noise_table[s->noise_index] * mult1; | |
997 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
998 } | |
999 } else { | |
1000 /* XXX: optimize more */ | |
1001 for(i = 0;i < s->coefs_start; i++) | |
1002 *coefs++ = 0.0; | |
1003 n = nb_coefs[ch]; | |
1004 for(i = 0;i < n; i++) { | |
1005 *coefs++ = coefs1[i] * exponents[i] * mult; | |
1006 } | |
1007 n = s->block_len - s->coefs_end[bsize]; | |
1008 for(i = 0;i < n; i++) | |
1009 *coefs++ = 0.0; | |
1010 } | |
1011 } | |
1012 } | |
1013 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
1014 #ifdef TRACE |
783 | 1015 for(ch = 0; ch < s->nb_channels; ch++) { |
1016 if (s->channel_coded[ch]) { | |
1017 dump_floats("exponents", 3, s->exponents[ch], s->block_len); | |
1018 dump_floats("coefs", 1, s->coefs[ch], s->block_len); | |
1019 } | |
1020 } | |
1021 #endif | |
1022 | |
1023 if (s->ms_stereo && s->channel_coded[1]) { | |
1024 float a, b; | |
1025 int i; | |
1026 | |
1027 /* nominal case for ms stereo: we do it before mdct */ | |
1028 /* no need to optimize this case because it should almost | |
1029 never happen */ | |
1030 if (!s->channel_coded[0]) { | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
1031 tprintf("rare ms-stereo case happened\n"); |
783 | 1032 memset(s->coefs[0], 0, sizeof(float) * s->block_len); |
1033 s->channel_coded[0] = 1; | |
1034 } | |
1035 | |
1036 for(i = 0; i < s->block_len; i++) { | |
1037 a = s->coefs[0][i]; | |
1038 b = s->coefs[1][i]; | |
1039 s->coefs[0][i] = a + b; | |
1040 s->coefs[1][i] = a - b; | |
1041 } | |
1042 } | |
1043 | |
1044 /* build the window : we ensure that when the windows overlap | |
1045 their squared sum is always 1 (MDCT reconstruction rule) */ | |
1046 /* XXX: merge with output */ | |
1047 { | |
1048 int i, next_block_len, block_len, prev_block_len, n; | |
1049 float *wptr; | |
1050 | |
1051 block_len = s->block_len; | |
1052 prev_block_len = 1 << s->prev_block_len_bits; | |
1053 next_block_len = 1 << s->next_block_len_bits; | |
1054 | |
1055 /* right part */ | |
1056 wptr = window + block_len; | |
1057 if (block_len <= next_block_len) { | |
1058 for(i=0;i<block_len;i++) | |
1059 *wptr++ = s->windows[bsize][i]; | |
1060 } else { | |
1061 /* overlap */ | |
1062 n = (block_len / 2) - (next_block_len / 2); | |
1063 for(i=0;i<n;i++) | |
1064 *wptr++ = 1.0; | |
1065 for(i=0;i<next_block_len;i++) | |
1066 *wptr++ = s->windows[s->frame_len_bits - s->next_block_len_bits][i]; | |
1067 for(i=0;i<n;i++) | |
1068 *wptr++ = 0.0; | |
1069 } | |
1070 | |
1071 /* left part */ | |
1072 wptr = window + block_len; | |
1073 if (block_len <= prev_block_len) { | |
1074 for(i=0;i<block_len;i++) | |
1075 *--wptr = s->windows[bsize][i]; | |
1076 } else { | |
1077 /* overlap */ | |
1078 n = (block_len / 2) - (prev_block_len / 2); | |
1079 for(i=0;i<n;i++) | |
1080 *--wptr = 1.0; | |
1081 for(i=0;i<prev_block_len;i++) | |
1082 *--wptr = s->windows[s->frame_len_bits - s->prev_block_len_bits][i]; | |
1083 for(i=0;i<n;i++) | |
1084 *--wptr = 0.0; | |
1085 } | |
1086 } | |
1087 | |
1088 | |
1089 for(ch = 0; ch < s->nb_channels; ch++) { | |
1090 if (s->channel_coded[ch]) { | |
972 | 1091 FFTSample output[BLOCK_MAX_SIZE * 2] __attribute__((aligned(16))); |
783 | 1092 float *ptr; |
1093 int i, n4, index, n; | |
1094 | |
1095 n = s->block_len; | |
1096 n4 = s->block_len / 2; | |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
1097 ff_imdct_calc(&s->mdct_ctx[bsize], |
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
1098 output, s->coefs[ch], s->mdct_tmp); |
783 | 1099 |
1100 /* XXX: optimize all that by build the window and | |
1101 multipying/adding at the same time */ | |
1102 /* multiply by the window */ | |
1103 for(i=0;i<n * 2;i++) { | |
1104 output[i] *= window[i]; | |
1105 } | |
1106 | |
1107 /* add in the frame */ | |
1108 index = (s->frame_len / 2) + s->block_pos - n4; | |
1109 ptr = &s->frame_out[ch][index]; | |
1110 for(i=0;i<n * 2;i++) { | |
1111 *ptr += output[i]; | |
1112 ptr++; | |
1113 } | |
1114 | |
1115 /* specific fast case for ms-stereo : add to second | |
1116 channel if it is not coded */ | |
1117 if (s->ms_stereo && !s->channel_coded[1]) { | |
1118 ptr = &s->frame_out[1][index]; | |
1119 for(i=0;i<n * 2;i++) { | |
1120 *ptr += output[i]; | |
1121 ptr++; | |
1122 } | |
1123 } | |
1124 } | |
1125 } | |
1126 next: | |
1127 /* update block number */ | |
1128 s->block_num++; | |
1129 s->block_pos += s->block_len; | |
1130 if (s->block_pos >= s->frame_len) | |
1131 return 1; | |
1132 else | |
1133 return 0; | |
1134 } | |
1135 | |
1136 /* decode a frame of frame_len samples */ | |
1137 static int wma_decode_frame(WMADecodeContext *s, int16_t *samples) | |
1138 { | |
1139 int ret, i, n, a, ch, incr; | |
1140 int16_t *ptr; | |
1141 float *iptr; | |
1142 | |
1343 | 1143 #ifdef TRACE |
1144 tprintf("***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len); | |
1145 #endif | |
783 | 1146 |
1147 /* read each block */ | |
1148 s->block_num = 0; | |
1149 s->block_pos = 0; | |
1150 for(;;) { | |
1151 ret = wma_decode_block(s); | |
1152 if (ret < 0) | |
1153 return -1; | |
1154 if (ret) | |
1155 break; | |
1156 } | |
1157 | |
1158 /* convert frame to integer */ | |
1159 n = s->frame_len; | |
1160 incr = s->nb_channels; | |
1161 for(ch = 0; ch < s->nb_channels; ch++) { | |
1162 ptr = samples + ch; | |
1163 iptr = s->frame_out[ch]; | |
1164 | |
1165 for(i=0;i<n;i++) { | |
797 | 1166 a = lrintf(*iptr++); |
783 | 1167 if (a > 32767) |
1168 a = 32767; | |
1169 else if (a < -32768) | |
1170 a = -32768; | |
1171 *ptr = a; | |
1172 ptr += incr; | |
1173 } | |
1174 /* prepare for next block */ | |
1175 memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len], | |
1176 s->frame_len * sizeof(float)); | |
1177 /* XXX: suppress this */ | |
1178 memset(&s->frame_out[ch][s->frame_len], 0, | |
1179 s->frame_len * sizeof(float)); | |
1180 } | |
1181 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
1182 #ifdef TRACE |
783 | 1183 dump_shorts("samples", samples, n * s->nb_channels); |
1184 #endif | |
1185 return 0; | |
1186 } | |
1187 | |
1188 static int wma_decode_superframe(AVCodecContext *avctx, | |
1189 void *data, int *data_size, | |
1064 | 1190 uint8_t *buf, int buf_size) |
783 | 1191 { |
1192 WMADecodeContext *s = avctx->priv_data; | |
1193 int nb_frames, bit_offset, i, pos, len; | |
1194 uint8_t *q; | |
1195 int16_t *samples; | |
1196 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
1197 tprintf("***decode_superframe:\n"); |
783 | 1198 |
1750 | 1199 if(buf_size==0){ |
1200 s->last_superframe_len = 0; | |
1201 return 0; | |
1202 } | |
1203 | |
783 | 1204 samples = data; |
1205 | |
1025
1f9afd8b9131
GetBitContext.size is allways multiplied by 8 -> use size_in_bits to avoid useless *8 in a few inner loops
michaelni
parents:
972
diff
changeset
|
1206 init_get_bits(&s->gb, buf, buf_size*8); |
783 | 1207 |
1208 if (s->use_bit_reservoir) { | |
1209 /* read super frame header */ | |
1210 get_bits(&s->gb, 4); /* super frame index */ | |
1211 nb_frames = get_bits(&s->gb, 4) - 1; | |
1212 | |
1213 bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3); | |
1214 | |
1215 if (s->last_superframe_len > 0) { | |
1216 // printf("skip=%d\n", s->last_bitoffset); | |
1217 /* add bit_offset bits to last frame */ | |
1218 if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) > | |
1219 MAX_CODED_SUPERFRAME_SIZE) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1220 goto fail; |
783 | 1221 q = s->last_superframe + s->last_superframe_len; |
1222 len = bit_offset; | |
1223 while (len > 0) { | |
1224 *q++ = (get_bits)(&s->gb, 8); | |
1225 len -= 8; | |
1226 } | |
1227 if (len > 0) { | |
1228 *q++ = (get_bits)(&s->gb, len) << (8 - len); | |
1229 } | |
1230 | |
1231 /* XXX: bit_offset bits into last frame */ | |
1025
1f9afd8b9131
GetBitContext.size is allways multiplied by 8 -> use size_in_bits to avoid useless *8 in a few inner loops
michaelni
parents:
972
diff
changeset
|
1232 init_get_bits(&s->gb, s->last_superframe, MAX_CODED_SUPERFRAME_SIZE*8); |
783 | 1233 /* skip unused bits */ |
1234 if (s->last_bitoffset > 0) | |
1235 skip_bits(&s->gb, s->last_bitoffset); | |
1236 /* this frame is stored in the last superframe and in the | |
1237 current one */ | |
1238 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1239 goto fail; |
783 | 1240 samples += s->nb_channels * s->frame_len; |
1241 } | |
1242 | |
1243 /* read each frame starting from bit_offset */ | |
1244 pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3; | |
1025
1f9afd8b9131
GetBitContext.size is allways multiplied by 8 -> use size_in_bits to avoid useless *8 in a few inner loops
michaelni
parents:
972
diff
changeset
|
1245 init_get_bits(&s->gb, buf + (pos >> 3), (MAX_CODED_SUPERFRAME_SIZE - (pos >> 3))*8); |
783 | 1246 len = pos & 7; |
1247 if (len > 0) | |
1248 skip_bits(&s->gb, len); | |
1249 | |
1250 s->reset_block_lengths = 1; | |
1251 for(i=0;i<nb_frames;i++) { | |
1252 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1253 goto fail; |
783 | 1254 samples += s->nb_channels * s->frame_len; |
1255 } | |
1256 | |
1257 /* we copy the end of the frame in the last frame buffer */ | |
1258 pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7); | |
1259 s->last_bitoffset = pos & 7; | |
1260 pos >>= 3; | |
1261 len = buf_size - pos; | |
819 | 1262 if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) { |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1263 goto fail; |
783 | 1264 } |
1265 s->last_superframe_len = len; | |
1266 memcpy(s->last_superframe, buf + pos, len); | |
1267 } else { | |
1268 /* single frame decode */ | |
1269 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1270 goto fail; |
783 | 1271 samples += s->nb_channels * s->frame_len; |
1272 } | |
1273 *data_size = (int8_t *)samples - (int8_t *)data; | |
1274 return s->block_align; | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1275 fail: |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1276 /* when error, we reset the bit reservoir */ |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1277 s->last_superframe_len = 0; |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1278 return -1; |
783 | 1279 } |
1280 | |
1281 static int wma_decode_end(AVCodecContext *avctx) | |
1282 { | |
1283 WMADecodeContext *s = avctx->priv_data; | |
1284 int i; | |
1285 | |
1286 for(i = 0; i < s->nb_block_sizes; i++) | |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
1287 ff_mdct_end(&s->mdct_ctx[i]); |
783 | 1288 for(i = 0; i < s->nb_block_sizes; i++) |
1289 av_free(s->windows[i]); | |
1290 | |
1291 if (s->use_exp_vlc) { | |
1292 free_vlc(&s->exp_vlc); | |
1293 } | |
1294 if (s->use_noise_coding) { | |
1295 free_vlc(&s->hgain_vlc); | |
1296 } | |
1297 for(i = 0;i < 2; i++) { | |
1298 free_vlc(&s->coef_vlc[i]); | |
1299 av_free(s->run_table[i]); | |
1300 av_free(s->level_table[i]); | |
1301 } | |
1302 | |
1303 return 0; | |
1304 } | |
1305 | |
1306 AVCodec wmav1_decoder = | |
1307 { | |
1308 "wmav1", | |
1309 CODEC_TYPE_AUDIO, | |
1310 CODEC_ID_WMAV1, | |
1311 sizeof(WMADecodeContext), | |
1312 wma_decode_init, | |
1313 NULL, | |
1314 wma_decode_end, | |
1315 wma_decode_superframe, | |
1316 }; | |
1317 | |
1318 AVCodec wmav2_decoder = | |
1319 { | |
1320 "wmav2", | |
1321 CODEC_TYPE_AUDIO, | |
1322 CODEC_ID_WMAV2, | |
1323 sizeof(WMADecodeContext), | |
1324 wma_decode_init, | |
1325 NULL, | |
1326 wma_decode_end, | |
1327 wma_decode_superframe, | |
1328 }; |