Mercurial > libavcodec.hg
annotate wmadec.c @ 3045:c5e521489fdf libavcodec
Off by one fix to prevent possible segfault. Patch by jwestfall at surrealistic dot net.
author | banan |
---|---|
date | Sat, 14 Jan 2006 17:43:22 +0000 |
parents | 0b546eab515d |
children | 072dbc669253 |
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 | |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3022
diff
changeset
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
783 | 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. |
2967 | 24 * WMA v1 is identified by audio format 0x160 in Microsoft media files |
1967
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 |
2967 | 29 * bytes at the end of a WAVEFORMATEX data structure. Transmit these bytes |
30 * to the decoder using the extradata[_size] fields in AVCodecContext. There | |
1967
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" |
2398
582e635cfa08
common.c -> bitstream.c (and the single non bitstream func -> utils.c)
michael
parents:
2370
diff
changeset
|
35 #include "bitstream.h" |
783 | 36 #include "dsputil.h" |
37 | |
38 /* size of blocks */ | |
39 #define BLOCK_MIN_BITS 7 | |
40 #define BLOCK_MAX_BITS 11 | |
41 #define BLOCK_MAX_SIZE (1 << BLOCK_MAX_BITS) | |
42 | |
43 #define BLOCK_NB_SIZES (BLOCK_MAX_BITS - BLOCK_MIN_BITS + 1) | |
44 | |
45 /* XXX: find exact max size */ | |
46 #define HIGH_BAND_MAX_SIZE 16 | |
47 | |
48 #define NB_LSP_COEFS 10 | |
49 | |
817 | 50 /* XXX: is it a suitable value ? */ |
2775
f3cdd51c9e16
WMA MAX_CODED_SUPERFRAME_SIZE too small patch by (Mark Weaver: mark-clist, npsl co uk)
michael
parents:
2398
diff
changeset
|
51 #define MAX_CODED_SUPERFRAME_SIZE 16384 |
783 | 52 |
53 #define MAX_CHANNELS 2 | |
54 | |
55 #define NOISE_TAB_SIZE 8192 | |
56 | |
57 #define LSP_POW_BITS 7 | |
58 | |
3022 | 59 #define VLCBITS 9 |
60 | |
783 | 61 typedef struct WMADecodeContext { |
62 GetBitContext gb; | |
63 int sample_rate; | |
64 int nb_channels; | |
65 int bit_rate; | |
66 int version; /* 1 = 0x160 (WMAV1), 2 = 0x161 (WMAV2) */ | |
67 int block_align; | |
68 int use_bit_reservoir; | |
69 int use_variable_block_len; | |
70 int use_exp_vlc; /* exponent coding: 0 = lsp, 1 = vlc + delta */ | |
71 int use_noise_coding; /* true if perceptual noise is added */ | |
72 int byte_offset_bits; | |
73 VLC exp_vlc; | |
74 int exponent_sizes[BLOCK_NB_SIZES]; | |
75 uint16_t exponent_bands[BLOCK_NB_SIZES][25]; | |
76 int high_band_start[BLOCK_NB_SIZES]; /* index of first coef in high band */ | |
77 int coefs_start; /* first coded coef */ | |
78 int coefs_end[BLOCK_NB_SIZES]; /* max number of coded coefficients */ | |
79 int exponent_high_sizes[BLOCK_NB_SIZES]; | |
2967 | 80 int exponent_high_bands[BLOCK_NB_SIZES][HIGH_BAND_MAX_SIZE]; |
783 | 81 VLC hgain_vlc; |
2967 | 82 |
783 | 83 /* coded values in high bands */ |
84 int high_band_coded[MAX_CHANNELS][HIGH_BAND_MAX_SIZE]; | |
85 int high_band_values[MAX_CHANNELS][HIGH_BAND_MAX_SIZE]; | |
86 | |
87 /* there are two possible tables for spectral coefficients */ | |
88 VLC coef_vlc[2]; | |
89 uint16_t *run_table[2]; | |
90 uint16_t *level_table[2]; | |
91 /* frame info */ | |
92 int frame_len; /* frame length in samples */ | |
93 int frame_len_bits; /* frame_len = 1 << frame_len_bits */ | |
94 int nb_block_sizes; /* number of block sizes */ | |
95 /* block info */ | |
96 int reset_block_lengths; | |
97 int block_len_bits; /* log2 of current block length */ | |
98 int next_block_len_bits; /* log2 of next block length */ | |
99 int prev_block_len_bits; /* log2 of prev block length */ | |
100 int block_len; /* block length in samples */ | |
101 int block_num; /* block number in current frame */ | |
102 int block_pos; /* current position in frame */ | |
103 uint8_t ms_stereo; /* true if mid/side stereo mode */ | |
104 uint8_t channel_coded[MAX_CHANNELS]; /* true if channel is coded */ | |
972 | 105 float exponents[MAX_CHANNELS][BLOCK_MAX_SIZE] __attribute__((aligned(16))); |
783 | 106 float max_exponent[MAX_CHANNELS]; |
107 int16_t coefs1[MAX_CHANNELS][BLOCK_MAX_SIZE]; | |
972 | 108 float coefs[MAX_CHANNELS][BLOCK_MAX_SIZE] __attribute__((aligned(16))); |
783 | 109 MDCTContext mdct_ctx[BLOCK_NB_SIZES]; |
1031
19de1445beb2
use av_malloc() functions - added av_strdup and av_realloc()
bellard
parents:
1025
diff
changeset
|
110 float *windows[BLOCK_NB_SIZES]; |
972 | 111 FFTSample mdct_tmp[BLOCK_MAX_SIZE] __attribute__((aligned(16))); /* temporary storage for imdct */ |
783 | 112 /* output buffer for one frame and the last for IMDCT windowing */ |
972 | 113 float frame_out[MAX_CHANNELS][BLOCK_MAX_SIZE * 2] __attribute__((aligned(16))); |
783 | 114 /* last frame info */ |
115 uint8_t last_superframe[MAX_CODED_SUPERFRAME_SIZE + 4]; /* padding added */ | |
116 int last_bitoffset; | |
117 int last_superframe_len; | |
118 float noise_table[NOISE_TAB_SIZE]; | |
119 int noise_index; | |
120 float noise_mult; /* XXX: suppress that and integrate it in the noise array */ | |
121 /* lsp_to_curve tables */ | |
122 float lsp_cos_table[BLOCK_MAX_SIZE]; | |
123 float lsp_pow_e_table[256]; | |
124 float lsp_pow_m_table1[(1 << LSP_POW_BITS)]; | |
125 float lsp_pow_m_table2[(1 << LSP_POW_BITS)]; | |
1343 | 126 |
127 #ifdef TRACE | |
128 int frame_count; | |
129 #endif | |
783 | 130 } WMADecodeContext; |
131 | |
132 typedef struct CoefVLCTable { | |
133 int n; /* total number of codes */ | |
134 const uint32_t *huffcodes; /* VLC bit values */ | |
135 const uint8_t *huffbits; /* VLC bit size */ | |
136 const uint16_t *levels; /* table to build run/level tables */ | |
137 } CoefVLCTable; | |
138 | |
139 static void wma_lsp_to_curve_init(WMADecodeContext *s, int frame_len); | |
140 | |
141 #include "wmadata.h" | |
142 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
143 #ifdef TRACE |
783 | 144 static void dump_shorts(const char *name, const short *tab, int n) |
145 { | |
146 int i; | |
147 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
148 tprintf("%s[%d]:\n", name, n); |
783 | 149 for(i=0;i<n;i++) { |
150 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
|
151 tprintf("%4d: ", i); |
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
152 tprintf(" %5d.0", tab[i]); |
783 | 153 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
|
154 tprintf("\n"); |
783 | 155 } |
156 } | |
157 | |
158 static void dump_floats(const char *name, int prec, const float *tab, int n) | |
159 { | |
160 int i; | |
161 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
162 tprintf("%s[%d]:\n", name, n); |
783 | 163 for(i=0;i<n;i++) { |
164 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
|
165 tprintf("%4d: ", i); |
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
166 tprintf(" %8.*f", prec, tab[i]); |
783 | 167 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
|
168 tprintf("\n"); |
783 | 169 } |
170 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
|
171 tprintf("\n"); |
783 | 172 } |
173 #endif | |
174 | |
175 /* XXX: use same run/length optimization as mpeg decoders */ | |
2967 | 176 static void init_coef_vlc(VLC *vlc, |
783 | 177 uint16_t **prun_table, uint16_t **plevel_table, |
178 const CoefVLCTable *vlc_table) | |
179 { | |
180 int n = vlc_table->n; | |
181 const uint8_t *table_bits = vlc_table->huffbits; | |
182 const uint32_t *table_codes = vlc_table->huffcodes; | |
183 const uint16_t *levels_table = vlc_table->levels; | |
184 uint16_t *run_table, *level_table; | |
185 const uint16_t *p; | |
186 int i, l, j, level; | |
187 | |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2180
diff
changeset
|
188 init_vlc(vlc, 9, n, table_bits, 1, 1, table_codes, 4, 4, 0); |
783 | 189 |
1031
19de1445beb2
use av_malloc() functions - added av_strdup and av_realloc()
bellard
parents:
1025
diff
changeset
|
190 run_table = av_malloc(n * sizeof(uint16_t)); |
19de1445beb2
use av_malloc() functions - added av_strdup and av_realloc()
bellard
parents:
1025
diff
changeset
|
191 level_table = av_malloc(n * sizeof(uint16_t)); |
783 | 192 p = levels_table; |
193 i = 2; | |
194 level = 1; | |
195 while (i < n) { | |
196 l = *p++; | |
197 for(j=0;j<l;j++) { | |
198 run_table[i] = j; | |
199 level_table[i] = level; | |
200 i++; | |
201 } | |
202 level++; | |
203 } | |
204 *prun_table = run_table; | |
205 *plevel_table = level_table; | |
206 } | |
207 | |
208 static int wma_decode_init(AVCodecContext * avctx) | |
209 { | |
210 WMADecodeContext *s = avctx->priv_data; | |
211 int i, flags1, flags2; | |
212 float *window; | |
213 uint8_t *extradata; | |
2180
7524cda1604e
k6 pop fix patch by (Diego Biurrun <diego at biurrun dot de>)
michael
parents:
2101
diff
changeset
|
214 float bps1, high_freq; |
7524cda1604e
k6 pop fix patch by (Diego Biurrun <diego at biurrun dot de>)
michael
parents:
2101
diff
changeset
|
215 volatile float bps; |
783 | 216 int sample_rate1; |
217 int coef_vlc_table; | |
2967 | 218 |
783 | 219 s->sample_rate = avctx->sample_rate; |
220 s->nb_channels = avctx->channels; | |
221 s->bit_rate = avctx->bit_rate; | |
222 s->block_align = avctx->block_align; | |
223 | |
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
|
224 if (avctx->codec->id == CODEC_ID_WMAV1) { |
783 | 225 s->version = 1; |
226 } else { | |
227 s->version = 2; | |
228 } | |
2967 | 229 |
783 | 230 /* extract flag infos */ |
231 flags1 = 0; | |
232 flags2 = 0; | |
233 extradata = avctx->extradata; | |
234 if (s->version == 1 && avctx->extradata_size >= 4) { | |
235 flags1 = extradata[0] | (extradata[1] << 8); | |
236 flags2 = extradata[2] | (extradata[3] << 8); | |
237 } else if (s->version == 2 && avctx->extradata_size >= 6) { | |
2967 | 238 flags1 = extradata[0] | (extradata[1] << 8) | |
783 | 239 (extradata[2] << 16) | (extradata[3] << 24); |
240 flags2 = extradata[4] | (extradata[5] << 8); | |
241 } | |
242 s->use_exp_vlc = flags2 & 0x0001; | |
243 s->use_bit_reservoir = flags2 & 0x0002; | |
244 s->use_variable_block_len = flags2 & 0x0004; | |
245 | |
246 /* compute MDCT block size */ | |
247 if (s->sample_rate <= 16000) { | |
248 s->frame_len_bits = 9; | |
2967 | 249 } else if (s->sample_rate <= 22050 || |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
250 (s->sample_rate <= 32000 && s->version == 1)) { |
783 | 251 s->frame_len_bits = 10; |
252 } else { | |
253 s->frame_len_bits = 11; | |
254 } | |
255 s->frame_len = 1 << s->frame_len_bits; | |
256 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
|
257 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
|
258 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
|
259 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
|
260 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
|
261 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
|
262 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
|
263 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
|
264 s->nb_block_sizes = nb + 1; |
783 | 265 } else { |
266 s->nb_block_sizes = 1; | |
267 } | |
268 | |
269 /* init rate dependant parameters */ | |
270 s->use_noise_coding = 1; | |
271 high_freq = s->sample_rate * 0.5; | |
272 | |
273 /* if version 2, then the rates are normalized */ | |
274 sample_rate1 = s->sample_rate; | |
275 if (s->version == 2) { | |
2967 | 276 if (sample_rate1 >= 44100) |
783 | 277 sample_rate1 = 44100; |
2967 | 278 else if (sample_rate1 >= 22050) |
783 | 279 sample_rate1 = 22050; |
2967 | 280 else if (sample_rate1 >= 16000) |
783 | 281 sample_rate1 = 16000; |
2967 | 282 else if (sample_rate1 >= 11025) |
783 | 283 sample_rate1 = 11025; |
2967 | 284 else if (sample_rate1 >= 8000) |
783 | 285 sample_rate1 = 8000; |
286 } | |
287 | |
288 bps = (float)s->bit_rate / (float)(s->nb_channels * s->sample_rate); | |
2992 | 289 s->byte_offset_bits = av_log2((int)(bps * s->frame_len / 8.0 + 0.5)) + 2; |
783 | 290 |
291 /* compute high frequency value and choose if noise coding should | |
292 be activated */ | |
293 bps1 = bps; | |
294 if (s->nb_channels == 2) | |
295 bps1 = bps * 1.6; | |
296 if (sample_rate1 == 44100) { | |
297 if (bps1 >= 0.61) | |
298 s->use_noise_coding = 0; | |
299 else | |
300 high_freq = high_freq * 0.4; | |
301 } else if (sample_rate1 == 22050) { | |
302 if (bps1 >= 1.16) | |
303 s->use_noise_coding = 0; | |
2967 | 304 else if (bps1 >= 0.72) |
783 | 305 high_freq = high_freq * 0.7; |
306 else | |
307 high_freq = high_freq * 0.6; | |
308 } else if (sample_rate1 == 16000) { | |
309 if (bps > 0.5) | |
310 high_freq = high_freq * 0.5; | |
311 else | |
312 high_freq = high_freq * 0.3; | |
313 } else if (sample_rate1 == 11025) { | |
314 high_freq = high_freq * 0.7; | |
315 } else if (sample_rate1 == 8000) { | |
316 if (bps <= 0.625) { | |
317 high_freq = high_freq * 0.5; | |
318 } else if (bps > 0.75) { | |
319 s->use_noise_coding = 0; | |
320 } else { | |
321 high_freq = high_freq * 0.65; | |
322 } | |
323 } else { | |
324 if (bps >= 0.8) { | |
325 high_freq = high_freq * 0.75; | |
326 } else if (bps >= 0.6) { | |
327 high_freq = high_freq * 0.6; | |
328 } else { | |
329 high_freq = high_freq * 0.5; | |
330 } | |
331 } | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
332 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
|
333 dprintf("version=%d channels=%d sample_rate=%d bitrate=%d block_align=%d\n", |
2967 | 334 s->version, s->nb_channels, s->sample_rate, s->bit_rate, |
783 | 335 s->block_align); |
2967 | 336 dprintf("bps=%f bps1=%f high_freq=%f bitoffset=%d\n", |
783 | 337 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
|
338 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
|
339 s->use_noise_coding, s->use_exp_vlc, s->nb_block_sizes); |
783 | 340 |
341 /* compute the scale factor band sizes for each MDCT block size */ | |
342 { | |
343 int a, b, pos, lpos, k, block_len, i, j, n; | |
344 const uint8_t *table; | |
2967 | 345 |
783 | 346 if (s->version == 1) { |
347 s->coefs_start = 3; | |
348 } else { | |
349 s->coefs_start = 0; | |
350 } | |
351 for(k = 0; k < s->nb_block_sizes; k++) { | |
352 block_len = s->frame_len >> k; | |
353 | |
354 if (s->version == 1) { | |
355 lpos = 0; | |
356 for(i=0;i<25;i++) { | |
357 a = wma_critical_freqs[i]; | |
358 b = s->sample_rate; | |
359 pos = ((block_len * 2 * a) + (b >> 1)) / b; | |
2967 | 360 if (pos > block_len) |
783 | 361 pos = block_len; |
362 s->exponent_bands[0][i] = pos - lpos; | |
363 if (pos >= block_len) { | |
364 i++; | |
365 break; | |
366 } | |
367 lpos = pos; | |
368 } | |
369 s->exponent_sizes[0] = i; | |
370 } else { | |
371 /* hardcoded tables */ | |
372 table = NULL; | |
373 a = s->frame_len_bits - BLOCK_MIN_BITS - k; | |
374 if (a < 3) { | |
375 if (s->sample_rate >= 44100) | |
376 table = exponent_band_44100[a]; | |
377 else if (s->sample_rate >= 32000) | |
378 table = exponent_band_32000[a]; | |
379 else if (s->sample_rate >= 22050) | |
380 table = exponent_band_22050[a]; | |
381 } | |
382 if (table) { | |
383 n = *table++; | |
384 for(i=0;i<n;i++) | |
385 s->exponent_bands[k][i] = table[i]; | |
386 s->exponent_sizes[k] = n; | |
387 } else { | |
388 j = 0; | |
389 lpos = 0; | |
390 for(i=0;i<25;i++) { | |
391 a = wma_critical_freqs[i]; | |
392 b = s->sample_rate; | |
393 pos = ((block_len * 2 * a) + (b << 1)) / (4 * b); | |
394 pos <<= 2; | |
2967 | 395 if (pos > block_len) |
783 | 396 pos = block_len; |
397 if (pos > lpos) | |
398 s->exponent_bands[k][j++] = pos - lpos; | |
399 if (pos >= block_len) | |
400 break; | |
401 lpos = pos; | |
402 } | |
403 s->exponent_sizes[k] = j; | |
404 } | |
405 } | |
406 | |
407 /* max number of coefs */ | |
408 s->coefs_end[k] = (s->frame_len - ((s->frame_len * 9) / 100)) >> k; | |
409 /* high freq computation */ | |
2967 | 410 s->high_band_start[k] = (int)((block_len * 2 * high_freq) / |
783 | 411 s->sample_rate + 0.5); |
412 n = s->exponent_sizes[k]; | |
413 j = 0; | |
414 pos = 0; | |
415 for(i=0;i<n;i++) { | |
416 int start, end; | |
417 start = pos; | |
418 pos += s->exponent_bands[k][i]; | |
419 end = pos; | |
420 if (start < s->high_band_start[k]) | |
421 start = s->high_band_start[k]; | |
422 if (end > s->coefs_end[k]) | |
423 end = s->coefs_end[k]; | |
424 if (end > start) | |
425 s->exponent_high_bands[k][j++] = end - start; | |
426 } | |
427 s->exponent_high_sizes[k] = j; | |
428 #if 0 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
429 tprintf("%5d: coefs_end=%d high_band_start=%d nb_high_bands=%d: ", |
2967 | 430 s->frame_len >> k, |
783 | 431 s->coefs_end[k], |
432 s->high_band_start[k], | |
433 s->exponent_high_sizes[k]); | |
434 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
|
435 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
|
436 tprintf("\n"); |
783 | 437 #endif |
438 } | |
439 } | |
440 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
441 #ifdef TRACE |
783 | 442 { |
443 int i, j; | |
444 for(i = 0; i < s->nb_block_sizes; i++) { | |
2967 | 445 tprintf("%5d: n=%2d:", |
446 s->frame_len >> i, | |
783 | 447 s->exponent_sizes[i]); |
448 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
|
449 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
|
450 tprintf("\n"); |
783 | 451 } |
452 } | |
453 #endif | |
454 | |
455 /* init MDCT */ | |
456 for(i = 0; i < s->nb_block_sizes; i++) | |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
457 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1); |
2967 | 458 |
783 | 459 /* init MDCT windows : simple sinus window */ |
460 for(i = 0; i < s->nb_block_sizes; i++) { | |
461 int n, j; | |
462 float alpha; | |
463 n = 1 << (s->frame_len_bits - i); | |
464 window = av_malloc(sizeof(float) * n); | |
465 alpha = M_PI / (2.0 * n); | |
466 for(j=0;j<n;j++) { | |
467 window[n - j - 1] = sin((j + 0.5) * alpha); | |
468 } | |
469 s->windows[i] = window; | |
470 } | |
471 | |
472 s->reset_block_lengths = 1; | |
2967 | 473 |
783 | 474 if (s->use_noise_coding) { |
475 | |
476 /* init the noise generator */ | |
477 if (s->use_exp_vlc) | |
478 s->noise_mult = 0.02; | |
479 else | |
480 s->noise_mult = 0.04; | |
2967 | 481 |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
482 #ifdef TRACE |
783 | 483 for(i=0;i<NOISE_TAB_SIZE;i++) |
484 s->noise_table[i] = 1.0 * s->noise_mult; | |
485 #else | |
486 { | |
487 unsigned int seed; | |
488 float norm; | |
489 seed = 1; | |
490 norm = (1.0 / (float)(1LL << 31)) * sqrt(3) * s->noise_mult; | |
491 for(i=0;i<NOISE_TAB_SIZE;i++) { | |
492 seed = seed * 314159 + 1; | |
493 s->noise_table[i] = (float)((int)seed) * norm; | |
494 } | |
495 } | |
496 #endif | |
2967 | 497 init_vlc(&s->hgain_vlc, 9, sizeof(hgain_huffbits), |
783 | 498 hgain_huffbits, 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2180
diff
changeset
|
499 hgain_huffcodes, 2, 2, 0); |
783 | 500 } |
501 | |
502 if (s->use_exp_vlc) { | |
2967 | 503 init_vlc(&s->exp_vlc, 9, sizeof(scale_huffbits), |
783 | 504 scale_huffbits, 1, 1, |
2370
26560d4fdb1f
Memory leak fix patch by (Burkhard Plaum <plaum >at< ipf.uni-stuttgart )dot( de>)
michael
parents:
2180
diff
changeset
|
505 scale_huffcodes, 4, 4, 0); |
783 | 506 } else { |
507 wma_lsp_to_curve_init(s, s->frame_len); | |
508 } | |
509 | |
510 /* choose the VLC tables for the coefficients */ | |
511 coef_vlc_table = 2; | |
512 if (s->sample_rate >= 32000) { | |
513 if (bps1 < 0.72) | |
514 coef_vlc_table = 0; | |
515 else if (bps1 < 1.16) | |
516 coef_vlc_table = 1; | |
517 } | |
518 | |
519 init_coef_vlc(&s->coef_vlc[0], &s->run_table[0], &s->level_table[0], | |
520 &coef_vlcs[coef_vlc_table * 2]); | |
521 init_coef_vlc(&s->coef_vlc[1], &s->run_table[1], &s->level_table[1], | |
522 &coef_vlcs[coef_vlc_table * 2 + 1]); | |
523 return 0; | |
524 } | |
525 | |
526 /* interpolate values for a bigger or smaller block. The block must | |
527 have multiple sizes */ | |
528 static void interpolate_array(float *scale, int old_size, int new_size) | |
529 { | |
530 int i, j, jincr, k; | |
531 float v; | |
532 | |
533 if (new_size > old_size) { | |
534 jincr = new_size / old_size; | |
535 j = new_size; | |
536 for(i = old_size - 1; i >=0; i--) { | |
537 v = scale[i]; | |
538 k = jincr; | |
539 do { | |
540 scale[--j] = v; | |
541 } while (--k); | |
542 } | |
543 } else if (new_size < old_size) { | |
544 j = 0; | |
545 jincr = old_size / new_size; | |
546 for(i = 0; i < new_size; i++) { | |
547 scale[i] = scale[j]; | |
548 j += jincr; | |
549 } | |
550 } | |
551 } | |
552 | |
553 /* compute x^-0.25 with an exponent and mantissa table. We use linear | |
554 interpolation to reduce the mantissa table size at a small speed | |
555 expense (linear interpolation approximately doubles the number of | |
556 bits of precision). */ | |
557 static inline float pow_m1_4(WMADecodeContext *s, float x) | |
558 { | |
559 union { | |
560 float f; | |
561 unsigned int v; | |
562 } u, t; | |
563 unsigned int e, m; | |
564 float a, b; | |
565 | |
566 u.f = x; | |
567 e = u.v >> 23; | |
568 m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1); | |
569 /* build interpolation scale: 1 <= t < 2. */ | |
570 t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23); | |
571 a = s->lsp_pow_m_table1[m]; | |
572 b = s->lsp_pow_m_table2[m]; | |
573 return s->lsp_pow_e_table[e] * (a + b * t.f); | |
574 } | |
575 | |
576 static void wma_lsp_to_curve_init(WMADecodeContext *s, int frame_len) | |
2967 | 577 { |
783 | 578 float wdel, a, b; |
579 int i, e, m; | |
580 | |
581 wdel = M_PI / frame_len; | |
582 for(i=0;i<frame_len;i++) | |
583 s->lsp_cos_table[i] = 2.0f * cos(wdel * i); | |
584 | |
585 /* tables for x^-0.25 computation */ | |
586 for(i=0;i<256;i++) { | |
587 e = i - 126; | |
588 s->lsp_pow_e_table[i] = pow(2.0, e * -0.25); | |
589 } | |
590 | |
591 /* NOTE: these two tables are needed to avoid two operations in | |
592 pow_m1_4 */ | |
593 b = 1.0; | |
594 for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) { | |
595 m = (1 << LSP_POW_BITS) + i; | |
596 a = (float)m * (0.5 / (1 << LSP_POW_BITS)); | |
597 a = pow(a, -0.25); | |
598 s->lsp_pow_m_table1[i] = 2 * a - b; | |
599 s->lsp_pow_m_table2[i] = b - a; | |
600 b = a; | |
601 } | |
602 #if 0 | |
603 for(i=1;i<20;i++) { | |
604 float v, r1, r2; | |
605 v = 5.0 / i; | |
606 r1 = pow_m1_4(s, v); | |
607 r2 = pow(v,-0.25); | |
608 printf("%f^-0.25=%f e=%f\n", v, r1, r2 - r1); | |
609 } | |
610 #endif | |
611 } | |
612 | |
613 /* NOTE: We use the same code as Vorbis here */ | |
614 /* XXX: optimize it further with SSE/3Dnow */ | |
2967 | 615 static void wma_lsp_to_curve(WMADecodeContext *s, |
616 float *out, float *val_max_ptr, | |
783 | 617 int n, float *lsp) |
618 { | |
619 int i, j; | |
620 float p, q, w, v, val_max; | |
621 | |
622 val_max = 0; | |
623 for(i=0;i<n;i++) { | |
624 p = 0.5f; | |
625 q = 0.5f; | |
626 w = s->lsp_cos_table[i]; | |
627 for(j=1;j<NB_LSP_COEFS;j+=2){ | |
628 q *= w - lsp[j - 1]; | |
629 p *= w - lsp[j]; | |
630 } | |
631 p *= p * (2.0f - w); | |
632 q *= q * (2.0f + w); | |
633 v = p + q; | |
634 v = pow_m1_4(s, v); | |
635 if (v > val_max) | |
636 val_max = v; | |
637 out[i] = v; | |
638 } | |
639 *val_max_ptr = val_max; | |
640 } | |
641 | |
642 /* decode exponents coded with LSP coefficients (same idea as Vorbis) */ | |
643 static void decode_exp_lsp(WMADecodeContext *s, int ch) | |
644 { | |
645 float lsp_coefs[NB_LSP_COEFS]; | |
646 int val, i; | |
647 | |
648 for(i = 0; i < NB_LSP_COEFS; i++) { | |
649 if (i == 0 || i >= 8) | |
650 val = get_bits(&s->gb, 3); | |
651 else | |
652 val = get_bits(&s->gb, 4); | |
653 lsp_coefs[i] = lsp_codebook[i][val]; | |
654 } | |
655 | |
656 wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch], | |
657 s->block_len, lsp_coefs); | |
658 } | |
659 | |
660 /* decode exponents coded with VLC codes */ | |
661 static int decode_exp_vlc(WMADecodeContext *s, int ch) | |
662 { | |
663 int last_exp, n, code; | |
664 const uint16_t *ptr, *band_ptr; | |
665 float v, *q, max_scale, *q_end; | |
2967 | 666 |
783 | 667 band_ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits]; |
668 ptr = band_ptr; | |
669 q = s->exponents[ch]; | |
670 q_end = q + s->block_len; | |
671 max_scale = 0; | |
672 if (s->version == 1) { | |
673 last_exp = get_bits(&s->gb, 5) + 10; | |
674 /* XXX: use a table */ | |
675 v = pow(10, last_exp * (1.0 / 16.0)); | |
676 max_scale = v; | |
677 n = *ptr++; | |
678 do { | |
679 *q++ = v; | |
680 } while (--n); | |
681 } | |
682 last_exp = 36; | |
683 while (q < q_end) { | |
3022 | 684 code = get_vlc2(&s->gb, s->exp_vlc.table, VLCBITS, 2); |
783 | 685 if (code < 0) |
686 return -1; | |
687 /* NOTE: this offset is the same as MPEG4 AAC ! */ | |
688 last_exp += code - 60; | |
689 /* XXX: use a table */ | |
690 v = pow(10, last_exp * (1.0 / 16.0)); | |
691 if (v > max_scale) | |
692 max_scale = v; | |
693 n = *ptr++; | |
694 do { | |
695 *q++ = v; | |
696 } while (--n); | |
697 } | |
698 s->max_exponent[ch] = max_scale; | |
699 return 0; | |
700 } | |
701 | |
702 /* return 0 if OK. return 1 if last block of frame. return -1 if | |
703 unrecorrable error. */ | |
704 static int wma_decode_block(WMADecodeContext *s) | |
705 { | |
706 int n, v, a, ch, code, bsize; | |
707 int coef_nb_bits, total_gain, parse_exponents; | |
708 float window[BLOCK_MAX_SIZE * 2]; | |
2101 | 709 // XXX: FIXME!! there's a bug somewhere which makes this mandatory under altivec |
710 #ifdef HAVE_ALTIVEC | |
711 volatile int nb_coefs[MAX_CHANNELS] __attribute__((aligned(16))); | |
712 #else | |
783 | 713 int nb_coefs[MAX_CHANNELS]; |
2101 | 714 #endif |
783 | 715 float mdct_norm; |
716 | |
1343 | 717 #ifdef TRACE |
718 tprintf("***decode_block: %d:%d\n", s->frame_count - 1, s->block_num); | |
719 #endif | |
783 | 720 |
721 /* compute current block length */ | |
722 if (s->use_variable_block_len) { | |
723 n = av_log2(s->nb_block_sizes - 1) + 1; | |
2967 | 724 |
783 | 725 if (s->reset_block_lengths) { |
726 s->reset_block_lengths = 0; | |
727 v = get_bits(&s->gb, n); | |
728 if (v >= s->nb_block_sizes) | |
729 return -1; | |
730 s->prev_block_len_bits = s->frame_len_bits - v; | |
731 v = get_bits(&s->gb, n); | |
732 if (v >= s->nb_block_sizes) | |
733 return -1; | |
734 s->block_len_bits = s->frame_len_bits - v; | |
735 } else { | |
736 /* update block lengths */ | |
737 s->prev_block_len_bits = s->block_len_bits; | |
738 s->block_len_bits = s->next_block_len_bits; | |
739 } | |
740 v = get_bits(&s->gb, n); | |
741 if (v >= s->nb_block_sizes) | |
742 return -1; | |
743 s->next_block_len_bits = s->frame_len_bits - v; | |
744 } else { | |
745 /* fixed block len */ | |
746 s->next_block_len_bits = s->frame_len_bits; | |
747 s->prev_block_len_bits = s->frame_len_bits; | |
748 s->block_len_bits = s->frame_len_bits; | |
749 } | |
750 | |
751 /* now check if the block length is coherent with the frame length */ | |
752 s->block_len = 1 << s->block_len_bits; | |
753 if ((s->block_pos + s->block_len) > s->frame_len) | |
754 return -1; | |
755 | |
756 if (s->nb_channels == 2) { | |
757 s->ms_stereo = get_bits(&s->gb, 1); | |
758 } | |
759 v = 0; | |
760 for(ch = 0; ch < s->nb_channels; ch++) { | |
761 a = get_bits(&s->gb, 1); | |
762 s->channel_coded[ch] = a; | |
763 v |= a; | |
764 } | |
765 /* if no channel coded, no need to go further */ | |
766 /* XXX: fix potential framing problems */ | |
767 if (!v) | |
768 goto next; | |
769 | |
770 bsize = s->frame_len_bits - s->block_len_bits; | |
771 | |
772 /* read total gain and extract corresponding number of bits for | |
773 coef escape coding */ | |
774 total_gain = 1; | |
775 for(;;) { | |
776 a = get_bits(&s->gb, 7); | |
777 total_gain += a; | |
778 if (a != 127) | |
779 break; | |
780 } | |
2967 | 781 |
783 | 782 if (total_gain < 15) |
783 coef_nb_bits = 13; | |
784 else if (total_gain < 32) | |
785 coef_nb_bits = 12; | |
786 else if (total_gain < 40) | |
787 coef_nb_bits = 11; | |
788 else if (total_gain < 45) | |
789 coef_nb_bits = 10; | |
790 else | |
791 coef_nb_bits = 9; | |
792 | |
793 /* compute number of coefficients */ | |
794 n = s->coefs_end[bsize] - s->coefs_start; | |
795 for(ch = 0; ch < s->nb_channels; ch++) | |
796 nb_coefs[ch] = n; | |
797 | |
798 /* complex coding */ | |
799 if (s->use_noise_coding) { | |
800 | |
801 for(ch = 0; ch < s->nb_channels; ch++) { | |
802 if (s->channel_coded[ch]) { | |
803 int i, n, a; | |
804 n = s->exponent_high_sizes[bsize]; | |
805 for(i=0;i<n;i++) { | |
806 a = get_bits(&s->gb, 1); | |
807 s->high_band_coded[ch][i] = a; | |
808 /* if noise coding, the coefficients are not transmitted */ | |
809 if (a) | |
810 nb_coefs[ch] -= s->exponent_high_bands[bsize][i]; | |
811 } | |
812 } | |
813 } | |
814 for(ch = 0; ch < s->nb_channels; ch++) { | |
815 if (s->channel_coded[ch]) { | |
816 int i, n, val, code; | |
817 | |
818 n = s->exponent_high_sizes[bsize]; | |
819 val = (int)0x80000000; | |
820 for(i=0;i<n;i++) { | |
821 if (s->high_band_coded[ch][i]) { | |
822 if (val == (int)0x80000000) { | |
823 val = get_bits(&s->gb, 7) - 19; | |
824 } else { | |
3022 | 825 code = get_vlc2(&s->gb, s->hgain_vlc.table, VLCBITS, 2); |
783 | 826 if (code < 0) |
827 return -1; | |
828 val += code - 18; | |
829 } | |
830 s->high_band_values[ch][i] = val; | |
831 } | |
832 } | |
833 } | |
834 } | |
835 } | |
2967 | 836 |
783 | 837 /* exposant can be interpolated in short blocks. */ |
838 parse_exponents = 1; | |
839 if (s->block_len_bits != s->frame_len_bits) { | |
840 parse_exponents = get_bits(&s->gb, 1); | |
841 } | |
2967 | 842 |
783 | 843 if (parse_exponents) { |
844 for(ch = 0; ch < s->nb_channels; ch++) { | |
845 if (s->channel_coded[ch]) { | |
846 if (s->use_exp_vlc) { | |
847 if (decode_exp_vlc(s, ch) < 0) | |
848 return -1; | |
849 } else { | |
850 decode_exp_lsp(s, ch); | |
851 } | |
852 } | |
853 } | |
854 } else { | |
855 for(ch = 0; ch < s->nb_channels; ch++) { | |
856 if (s->channel_coded[ch]) { | |
2967 | 857 interpolate_array(s->exponents[ch], 1 << s->prev_block_len_bits, |
783 | 858 s->block_len); |
859 } | |
860 } | |
861 } | |
862 | |
863 /* parse spectral coefficients : just RLE encoding */ | |
864 for(ch = 0; ch < s->nb_channels; ch++) { | |
865 if (s->channel_coded[ch]) { | |
866 VLC *coef_vlc; | |
867 int level, run, sign, tindex; | |
868 int16_t *ptr, *eptr; | |
869 const int16_t *level_table, *run_table; | |
870 | |
871 /* special VLC tables are used for ms stereo because | |
872 there is potentially less energy there */ | |
873 tindex = (ch == 1 && s->ms_stereo); | |
874 coef_vlc = &s->coef_vlc[tindex]; | |
875 run_table = s->run_table[tindex]; | |
876 level_table = s->level_table[tindex]; | |
877 /* XXX: optimize */ | |
878 ptr = &s->coefs1[ch][0]; | |
879 eptr = ptr + nb_coefs[ch]; | |
880 memset(ptr, 0, s->block_len * sizeof(int16_t)); | |
881 for(;;) { | |
3022 | 882 code = get_vlc2(&s->gb, coef_vlc->table, VLCBITS, 3); |
783 | 883 if (code < 0) |
884 return -1; | |
885 if (code == 1) { | |
886 /* EOB */ | |
887 break; | |
888 } else if (code == 0) { | |
889 /* escape */ | |
890 level = get_bits(&s->gb, coef_nb_bits); | |
891 /* NOTE: this is rather suboptimal. reading | |
892 block_len_bits would be better */ | |
893 run = get_bits(&s->gb, s->frame_len_bits); | |
894 } else { | |
895 /* normal code */ | |
896 run = run_table[code]; | |
897 level = level_table[code]; | |
898 } | |
899 sign = get_bits(&s->gb, 1); | |
900 if (!sign) | |
901 level = -level; | |
902 ptr += run; | |
903 if (ptr >= eptr) | |
904 return -1; | |
905 *ptr++ = level; | |
906 /* NOTE: EOB can be omitted */ | |
907 if (ptr >= eptr) | |
908 break; | |
909 } | |
910 } | |
911 if (s->version == 1 && s->nb_channels >= 2) { | |
912 align_get_bits(&s->gb); | |
913 } | |
914 } | |
2967 | 915 |
783 | 916 /* normalize */ |
917 { | |
918 int n4 = s->block_len / 2; | |
919 mdct_norm = 1.0 / (float)n4; | |
920 if (s->version == 1) { | |
921 mdct_norm *= sqrt(n4); | |
922 } | |
923 } | |
924 | |
925 /* finally compute the MDCT coefficients */ | |
926 for(ch = 0; ch < s->nb_channels; ch++) { | |
927 if (s->channel_coded[ch]) { | |
928 int16_t *coefs1; | |
929 float *coefs, *exponents, mult, mult1, noise, *exp_ptr; | |
930 int i, j, n, n1, last_high_band; | |
931 float exp_power[HIGH_BAND_MAX_SIZE]; | |
932 | |
933 coefs1 = s->coefs1[ch]; | |
934 exponents = s->exponents[ch]; | |
935 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch]; | |
936 mult *= mdct_norm; | |
937 coefs = s->coefs[ch]; | |
938 if (s->use_noise_coding) { | |
939 mult1 = mult; | |
940 /* very low freqs : noise */ | |
941 for(i = 0;i < s->coefs_start; i++) { | |
942 *coefs++ = s->noise_table[s->noise_index] * (*exponents++) * mult1; | |
943 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
944 } | |
2967 | 945 |
783 | 946 n1 = s->exponent_high_sizes[bsize]; |
947 | |
948 /* compute power of high bands */ | |
2967 | 949 exp_ptr = exponents + |
950 s->high_band_start[bsize] - | |
783 | 951 s->coefs_start; |
952 last_high_band = 0; /* avoid warning */ | |
953 for(j=0;j<n1;j++) { | |
2967 | 954 n = s->exponent_high_bands[s->frame_len_bits - |
783 | 955 s->block_len_bits][j]; |
956 if (s->high_band_coded[ch][j]) { | |
957 float e2, v; | |
958 e2 = 0; | |
959 for(i = 0;i < n; i++) { | |
960 v = exp_ptr[i]; | |
961 e2 += v * v; | |
962 } | |
963 exp_power[j] = e2 / n; | |
964 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
|
965 tprintf("%d: power=%f (%d)\n", j, exp_power[j], n); |
783 | 966 } |
967 exp_ptr += n; | |
968 } | |
969 | |
970 /* main freqs and high freqs */ | |
971 for(j=-1;j<n1;j++) { | |
972 if (j < 0) { | |
2967 | 973 n = s->high_band_start[bsize] - |
783 | 974 s->coefs_start; |
975 } else { | |
2967 | 976 n = s->exponent_high_bands[s->frame_len_bits - |
783 | 977 s->block_len_bits][j]; |
978 } | |
979 if (j >= 0 && s->high_band_coded[ch][j]) { | |
980 /* use noise with specified power */ | |
981 mult1 = sqrt(exp_power[j] / exp_power[last_high_band]); | |
982 /* XXX: use a table */ | |
983 mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05); | |
984 mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult); | |
985 mult1 *= mdct_norm; | |
986 for(i = 0;i < n; i++) { | |
987 noise = s->noise_table[s->noise_index]; | |
988 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
989 *coefs++ = (*exponents++) * noise * mult1; | |
990 } | |
991 } else { | |
992 /* coded values + small noise */ | |
993 for(i = 0;i < n; i++) { | |
994 noise = s->noise_table[s->noise_index]; | |
995 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
996 *coefs++ = ((*coefs1++) + noise) * (*exponents++) * mult; | |
997 } | |
998 } | |
999 } | |
1000 | |
1001 /* very high freqs : noise */ | |
1002 n = s->block_len - s->coefs_end[bsize]; | |
1003 mult1 = mult * exponents[-1]; | |
1004 for(i = 0; i < n; i++) { | |
1005 *coefs++ = s->noise_table[s->noise_index] * mult1; | |
1006 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
1007 } | |
1008 } else { | |
1009 /* XXX: optimize more */ | |
1010 for(i = 0;i < s->coefs_start; i++) | |
1011 *coefs++ = 0.0; | |
1012 n = nb_coefs[ch]; | |
1013 for(i = 0;i < n; i++) { | |
1014 *coefs++ = coefs1[i] * exponents[i] * mult; | |
1015 } | |
1016 n = s->block_len - s->coefs_end[bsize]; | |
1017 for(i = 0;i < n; i++) | |
1018 *coefs++ = 0.0; | |
1019 } | |
1020 } | |
1021 } | |
1022 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
1023 #ifdef TRACE |
783 | 1024 for(ch = 0; ch < s->nb_channels; ch++) { |
1025 if (s->channel_coded[ch]) { | |
1026 dump_floats("exponents", 3, s->exponents[ch], s->block_len); | |
1027 dump_floats("coefs", 1, s->coefs[ch], s->block_len); | |
1028 } | |
1029 } | |
1030 #endif | |
2967 | 1031 |
783 | 1032 if (s->ms_stereo && s->channel_coded[1]) { |
1033 float a, b; | |
1034 int i; | |
1035 | |
1036 /* nominal case for ms stereo: we do it before mdct */ | |
1037 /* no need to optimize this case because it should almost | |
1038 never happen */ | |
1039 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
|
1040 tprintf("rare ms-stereo case happened\n"); |
783 | 1041 memset(s->coefs[0], 0, sizeof(float) * s->block_len); |
1042 s->channel_coded[0] = 1; | |
1043 } | |
2967 | 1044 |
783 | 1045 for(i = 0; i < s->block_len; i++) { |
1046 a = s->coefs[0][i]; | |
1047 b = s->coefs[1][i]; | |
1048 s->coefs[0][i] = a + b; | |
1049 s->coefs[1][i] = a - b; | |
1050 } | |
1051 } | |
1052 | |
1053 /* build the window : we ensure that when the windows overlap | |
1054 their squared sum is always 1 (MDCT reconstruction rule) */ | |
1055 /* XXX: merge with output */ | |
1056 { | |
1057 int i, next_block_len, block_len, prev_block_len, n; | |
1058 float *wptr; | |
1059 | |
1060 block_len = s->block_len; | |
1061 prev_block_len = 1 << s->prev_block_len_bits; | |
1062 next_block_len = 1 << s->next_block_len_bits; | |
1063 | |
1064 /* right part */ | |
1065 wptr = window + block_len; | |
1066 if (block_len <= next_block_len) { | |
1067 for(i=0;i<block_len;i++) | |
1068 *wptr++ = s->windows[bsize][i]; | |
1069 } else { | |
1070 /* overlap */ | |
1071 n = (block_len / 2) - (next_block_len / 2); | |
1072 for(i=0;i<n;i++) | |
1073 *wptr++ = 1.0; | |
1074 for(i=0;i<next_block_len;i++) | |
1075 *wptr++ = s->windows[s->frame_len_bits - s->next_block_len_bits][i]; | |
1076 for(i=0;i<n;i++) | |
1077 *wptr++ = 0.0; | |
1078 } | |
1079 | |
1080 /* left part */ | |
1081 wptr = window + block_len; | |
1082 if (block_len <= prev_block_len) { | |
1083 for(i=0;i<block_len;i++) | |
1084 *--wptr = s->windows[bsize][i]; | |
1085 } else { | |
1086 /* overlap */ | |
1087 n = (block_len / 2) - (prev_block_len / 2); | |
1088 for(i=0;i<n;i++) | |
1089 *--wptr = 1.0; | |
1090 for(i=0;i<prev_block_len;i++) | |
1091 *--wptr = s->windows[s->frame_len_bits - s->prev_block_len_bits][i]; | |
1092 for(i=0;i<n;i++) | |
1093 *--wptr = 0.0; | |
1094 } | |
1095 } | |
1096 | |
2967 | 1097 |
783 | 1098 for(ch = 0; ch < s->nb_channels; ch++) { |
1099 if (s->channel_coded[ch]) { | |
972 | 1100 FFTSample output[BLOCK_MAX_SIZE * 2] __attribute__((aligned(16))); |
783 | 1101 float *ptr; |
1102 int i, n4, index, n; | |
1103 | |
1104 n = s->block_len; | |
1105 n4 = s->block_len / 2; | |
2967 | 1106 ff_imdct_calc(&s->mdct_ctx[bsize], |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
1107 output, s->coefs[ch], s->mdct_tmp); |
783 | 1108 |
1109 /* XXX: optimize all that by build the window and | |
1110 multipying/adding at the same time */ | |
1111 /* multiply by the window */ | |
1112 for(i=0;i<n * 2;i++) { | |
1113 output[i] *= window[i]; | |
1114 } | |
1115 | |
1116 /* add in the frame */ | |
1117 index = (s->frame_len / 2) + s->block_pos - n4; | |
1118 ptr = &s->frame_out[ch][index]; | |
1119 for(i=0;i<n * 2;i++) { | |
1120 *ptr += output[i]; | |
1121 ptr++; | |
1122 } | |
1123 | |
1124 /* specific fast case for ms-stereo : add to second | |
1125 channel if it is not coded */ | |
1126 if (s->ms_stereo && !s->channel_coded[1]) { | |
1127 ptr = &s->frame_out[1][index]; | |
1128 for(i=0;i<n * 2;i++) { | |
1129 *ptr += output[i]; | |
1130 ptr++; | |
1131 } | |
1132 } | |
1133 } | |
1134 } | |
1135 next: | |
1136 /* update block number */ | |
1137 s->block_num++; | |
1138 s->block_pos += s->block_len; | |
1139 if (s->block_pos >= s->frame_len) | |
1140 return 1; | |
1141 else | |
1142 return 0; | |
1143 } | |
1144 | |
1145 /* decode a frame of frame_len samples */ | |
1146 static int wma_decode_frame(WMADecodeContext *s, int16_t *samples) | |
1147 { | |
1148 int ret, i, n, a, ch, incr; | |
1149 int16_t *ptr; | |
1150 float *iptr; | |
1151 | |
1343 | 1152 #ifdef TRACE |
1153 tprintf("***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len); | |
1154 #endif | |
783 | 1155 |
1156 /* read each block */ | |
1157 s->block_num = 0; | |
1158 s->block_pos = 0; | |
1159 for(;;) { | |
1160 ret = wma_decode_block(s); | |
2967 | 1161 if (ret < 0) |
783 | 1162 return -1; |
1163 if (ret) | |
1164 break; | |
1165 } | |
1166 | |
1167 /* convert frame to integer */ | |
1168 n = s->frame_len; | |
1169 incr = s->nb_channels; | |
1170 for(ch = 0; ch < s->nb_channels; ch++) { | |
1171 ptr = samples + ch; | |
1172 iptr = s->frame_out[ch]; | |
1173 | |
1174 for(i=0;i<n;i++) { | |
797 | 1175 a = lrintf(*iptr++); |
783 | 1176 if (a > 32767) |
1177 a = 32767; | |
1178 else if (a < -32768) | |
1179 a = -32768; | |
1180 *ptr = a; | |
1181 ptr += incr; | |
1182 } | |
1183 /* prepare for next block */ | |
1184 memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len], | |
1185 s->frame_len * sizeof(float)); | |
1186 /* XXX: suppress this */ | |
2967 | 1187 memset(&s->frame_out[ch][s->frame_len], 0, |
783 | 1188 s->frame_len * sizeof(float)); |
1189 } | |
1190 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
1191 #ifdef TRACE |
783 | 1192 dump_shorts("samples", samples, n * s->nb_channels); |
1193 #endif | |
1194 return 0; | |
1195 } | |
1196 | |
2967 | 1197 static int wma_decode_superframe(AVCodecContext *avctx, |
783 | 1198 void *data, int *data_size, |
1064 | 1199 uint8_t *buf, int buf_size) |
783 | 1200 { |
1201 WMADecodeContext *s = avctx->priv_data; | |
1202 int nb_frames, bit_offset, i, pos, len; | |
1203 uint8_t *q; | |
1204 int16_t *samples; | |
2967 | 1205 |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
1206 tprintf("***decode_superframe:\n"); |
783 | 1207 |
1750 | 1208 if(buf_size==0){ |
1209 s->last_superframe_len = 0; | |
1210 return 0; | |
1211 } | |
2967 | 1212 |
783 | 1213 samples = data; |
1214 | |
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
|
1215 init_get_bits(&s->gb, buf, buf_size*8); |
2967 | 1216 |
783 | 1217 if (s->use_bit_reservoir) { |
1218 /* read super frame header */ | |
1219 get_bits(&s->gb, 4); /* super frame index */ | |
1220 nb_frames = get_bits(&s->gb, 4) - 1; | |
1221 | |
1222 bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3); | |
1223 | |
1224 if (s->last_superframe_len > 0) { | |
1225 // printf("skip=%d\n", s->last_bitoffset); | |
1226 /* add bit_offset bits to last frame */ | |
2967 | 1227 if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) > |
783 | 1228 MAX_CODED_SUPERFRAME_SIZE) |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1229 goto fail; |
783 | 1230 q = s->last_superframe + s->last_superframe_len; |
1231 len = bit_offset; | |
1232 while (len > 0) { | |
1233 *q++ = (get_bits)(&s->gb, 8); | |
1234 len -= 8; | |
1235 } | |
1236 if (len > 0) { | |
1237 *q++ = (get_bits)(&s->gb, len) << (8 - len); | |
1238 } | |
2967 | 1239 |
783 | 1240 /* 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
|
1241 init_get_bits(&s->gb, s->last_superframe, MAX_CODED_SUPERFRAME_SIZE*8); |
783 | 1242 /* skip unused bits */ |
1243 if (s->last_bitoffset > 0) | |
1244 skip_bits(&s->gb, s->last_bitoffset); | |
1245 /* this frame is stored in the last superframe and in the | |
1246 current one */ | |
1247 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1248 goto fail; |
783 | 1249 samples += s->nb_channels * s->frame_len; |
1250 } | |
1251 | |
1252 /* read each frame starting from bit_offset */ | |
1253 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
|
1254 init_get_bits(&s->gb, buf + (pos >> 3), (MAX_CODED_SUPERFRAME_SIZE - (pos >> 3))*8); |
783 | 1255 len = pos & 7; |
1256 if (len > 0) | |
1257 skip_bits(&s->gb, len); | |
2967 | 1258 |
783 | 1259 s->reset_block_lengths = 1; |
1260 for(i=0;i<nb_frames;i++) { | |
1261 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1262 goto fail; |
783 | 1263 samples += s->nb_channels * s->frame_len; |
1264 } | |
1265 | |
1266 /* we copy the end of the frame in the last frame buffer */ | |
1267 pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7); | |
1268 s->last_bitoffset = pos & 7; | |
1269 pos >>= 3; | |
1270 len = buf_size - pos; | |
819 | 1271 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
|
1272 goto fail; |
783 | 1273 } |
1274 s->last_superframe_len = len; | |
1275 memcpy(s->last_superframe, buf + pos, len); | |
1276 } else { | |
1277 /* single frame decode */ | |
1278 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1279 goto fail; |
783 | 1280 samples += s->nb_channels * s->frame_len; |
1281 } | |
1282 *data_size = (int8_t *)samples - (int8_t *)data; | |
1283 return s->block_align; | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1284 fail: |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1285 /* when error, we reset the bit reservoir */ |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1286 s->last_superframe_len = 0; |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
1287 return -1; |
783 | 1288 } |
1289 | |
1290 static int wma_decode_end(AVCodecContext *avctx) | |
1291 { | |
1292 WMADecodeContext *s = avctx->priv_data; | |
1293 int i; | |
1294 | |
1295 for(i = 0; i < s->nb_block_sizes; i++) | |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
1296 ff_mdct_end(&s->mdct_ctx[i]); |
783 | 1297 for(i = 0; i < s->nb_block_sizes; i++) |
1298 av_free(s->windows[i]); | |
1299 | |
1300 if (s->use_exp_vlc) { | |
1301 free_vlc(&s->exp_vlc); | |
1302 } | |
1303 if (s->use_noise_coding) { | |
1304 free_vlc(&s->hgain_vlc); | |
1305 } | |
1306 for(i = 0;i < 2; i++) { | |
1307 free_vlc(&s->coef_vlc[i]); | |
1308 av_free(s->run_table[i]); | |
1309 av_free(s->level_table[i]); | |
1310 } | |
2967 | 1311 |
783 | 1312 return 0; |
1313 } | |
1314 | |
1315 AVCodec wmav1_decoder = | |
1316 { | |
1317 "wmav1", | |
1318 CODEC_TYPE_AUDIO, | |
1319 CODEC_ID_WMAV1, | |
1320 sizeof(WMADecodeContext), | |
1321 wma_decode_init, | |
1322 NULL, | |
1323 wma_decode_end, | |
1324 wma_decode_superframe, | |
1325 }; | |
1326 | |
1327 AVCodec wmav2_decoder = | |
1328 { | |
1329 "wmav2", | |
1330 CODEC_TYPE_AUDIO, | |
1331 CODEC_ID_WMAV2, | |
1332 sizeof(WMADecodeContext), | |
1333 wma_decode_init, | |
1334 NULL, | |
1335 wma_decode_end, | |
1336 wma_decode_superframe, | |
1337 }; |