Mercurial > libavcodec.hg
annotate wmadec.c @ 4096:99616a74260a libavcodec
use lambda/lmax instead of qscale/qmax as end check for vbv retry
author | michael |
---|---|
date | Sat, 28 Oct 2006 11:15:06 +0000 |
parents | c8c591fe26f8 |
children | b43bd0c56eaa |
rev | line source |
---|---|
783 | 1 /* |
2 * WMA compatible decoder | |
3 * Copyright (c) 2002 The FFmpeg Project. | |
4 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
5 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
6 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
7 * FFmpeg is free software; you can redistribute it and/or |
783 | 8 * modify it under the terms of the GNU Lesser General Public |
9 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
10 * version 2.1 of the License, or (at your option) any later version. |
783 | 11 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
12 * FFmpeg is distributed in the hope that it will be useful, |
783 | 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 * Lesser General Public License for more details. | |
16 * | |
17 * You should have received a copy of the GNU Lesser General Public | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
18 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3022
diff
changeset
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
783 | 20 */ |
1106 | 21 |
22 /** | |
23 * @file wmadec.c | |
24 * WMA compatible decoder. | |
1967
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
25 * This decoder handles Microsoft Windows Media Audio data, versions 1 & 2. |
2967 | 26 * 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
|
27 * (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
|
28 * |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
29 * 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
|
30 * bytes provided with the WMA data. These are the extra, codec-specific |
2967 | 31 * bytes at the end of a WAVEFORMATEX data structure. Transmit these bytes |
32 * 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
|
33 * should be 4 extra bytes for v1 data and 6 extra bytes for v2 data. |
1106 | 34 */ |
35 | |
783 | 36 #include "avcodec.h" |
2398
582e635cfa08
common.c -> bitstream.c (and the single non bitstream func -> utils.c)
michael
parents:
2370
diff
changeset
|
37 #include "bitstream.h" |
783 | 38 #include "dsputil.h" |
39 | |
40 /* size of blocks */ | |
41 #define BLOCK_MIN_BITS 7 | |
42 #define BLOCK_MAX_BITS 11 | |
43 #define BLOCK_MAX_SIZE (1 << BLOCK_MAX_BITS) | |
44 | |
45 #define BLOCK_NB_SIZES (BLOCK_MAX_BITS - BLOCK_MIN_BITS + 1) | |
46 | |
47 /* XXX: find exact max size */ | |
48 #define HIGH_BAND_MAX_SIZE 16 | |
49 | |
50 #define NB_LSP_COEFS 10 | |
51 | |
817 | 52 /* 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
|
53 #define MAX_CODED_SUPERFRAME_SIZE 16384 |
783 | 54 |
55 #define MAX_CHANNELS 2 | |
56 | |
57 #define NOISE_TAB_SIZE 8192 | |
58 | |
59 #define LSP_POW_BITS 7 | |
60 | |
3022 | 61 #define VLCBITS 9 |
3113 | 62 #define VLCMAX ((22+VLCBITS-1)/VLCBITS) |
63 | |
64 #define EXPVLCBITS 8 | |
65 #define EXPMAX ((19+EXPVLCBITS-1)/EXPVLCBITS) | |
66 | |
67 #define HGAINVLCBITS 9 | |
68 #define HGAINMAX ((13+HGAINVLCBITS-1)/HGAINVLCBITS) | |
3022 | 69 |
783 | 70 typedef struct WMADecodeContext { |
71 GetBitContext gb; | |
72 int sample_rate; | |
73 int nb_channels; | |
74 int bit_rate; | |
75 int version; /* 1 = 0x160 (WMAV1), 2 = 0x161 (WMAV2) */ | |
76 int block_align; | |
77 int use_bit_reservoir; | |
78 int use_variable_block_len; | |
79 int use_exp_vlc; /* exponent coding: 0 = lsp, 1 = vlc + delta */ | |
80 int use_noise_coding; /* true if perceptual noise is added */ | |
81 int byte_offset_bits; | |
82 VLC exp_vlc; | |
83 int exponent_sizes[BLOCK_NB_SIZES]; | |
84 uint16_t exponent_bands[BLOCK_NB_SIZES][25]; | |
85 int high_band_start[BLOCK_NB_SIZES]; /* index of first coef in high band */ | |
86 int coefs_start; /* first coded coef */ | |
87 int coefs_end[BLOCK_NB_SIZES]; /* max number of coded coefficients */ | |
88 int exponent_high_sizes[BLOCK_NB_SIZES]; | |
2967 | 89 int exponent_high_bands[BLOCK_NB_SIZES][HIGH_BAND_MAX_SIZE]; |
783 | 90 VLC hgain_vlc; |
2967 | 91 |
783 | 92 /* coded values in high bands */ |
93 int high_band_coded[MAX_CHANNELS][HIGH_BAND_MAX_SIZE]; | |
94 int high_band_values[MAX_CHANNELS][HIGH_BAND_MAX_SIZE]; | |
95 | |
96 /* there are two possible tables for spectral coefficients */ | |
97 VLC coef_vlc[2]; | |
98 uint16_t *run_table[2]; | |
99 uint16_t *level_table[2]; | |
100 /* frame info */ | |
101 int frame_len; /* frame length in samples */ | |
102 int frame_len_bits; /* frame_len = 1 << frame_len_bits */ | |
103 int nb_block_sizes; /* number of block sizes */ | |
104 /* block info */ | |
105 int reset_block_lengths; | |
106 int block_len_bits; /* log2 of current block length */ | |
107 int next_block_len_bits; /* log2 of next block length */ | |
108 int prev_block_len_bits; /* log2 of prev block length */ | |
109 int block_len; /* block length in samples */ | |
110 int block_num; /* block number in current frame */ | |
111 int block_pos; /* current position in frame */ | |
112 uint8_t ms_stereo; /* true if mid/side stereo mode */ | |
113 uint8_t channel_coded[MAX_CHANNELS]; /* true if channel is coded */ | |
3089 | 114 DECLARE_ALIGNED_16(float, exponents[MAX_CHANNELS][BLOCK_MAX_SIZE]); |
783 | 115 float max_exponent[MAX_CHANNELS]; |
116 int16_t coefs1[MAX_CHANNELS][BLOCK_MAX_SIZE]; | |
3089 | 117 DECLARE_ALIGNED_16(float, coefs[MAX_CHANNELS][BLOCK_MAX_SIZE]); |
783 | 118 MDCTContext mdct_ctx[BLOCK_NB_SIZES]; |
1031
19de1445beb2
use av_malloc() functions - added av_strdup and av_realloc()
bellard
parents:
1025
diff
changeset
|
119 float *windows[BLOCK_NB_SIZES]; |
3089 | 120 DECLARE_ALIGNED_16(FFTSample, mdct_tmp[BLOCK_MAX_SIZE]); /* temporary storage for imdct */ |
783 | 121 /* output buffer for one frame and the last for IMDCT windowing */ |
3089 | 122 DECLARE_ALIGNED_16(float, frame_out[MAX_CHANNELS][BLOCK_MAX_SIZE * 2]); |
783 | 123 /* last frame info */ |
124 uint8_t last_superframe[MAX_CODED_SUPERFRAME_SIZE + 4]; /* padding added */ | |
125 int last_bitoffset; | |
126 int last_superframe_len; | |
127 float noise_table[NOISE_TAB_SIZE]; | |
128 int noise_index; | |
129 float noise_mult; /* XXX: suppress that and integrate it in the noise array */ | |
130 /* lsp_to_curve tables */ | |
131 float lsp_cos_table[BLOCK_MAX_SIZE]; | |
132 float lsp_pow_e_table[256]; | |
133 float lsp_pow_m_table1[(1 << LSP_POW_BITS)]; | |
134 float lsp_pow_m_table2[(1 << LSP_POW_BITS)]; | |
3592
6a358dccf2ab
SIMD vector optimizations. 3% faster overall decoding.
banan
parents:
3555
diff
changeset
|
135 DSPContext dsp; |
1343 | 136 |
137 #ifdef TRACE | |
138 int frame_count; | |
139 #endif | |
783 | 140 } WMADecodeContext; |
141 | |
142 typedef struct CoefVLCTable { | |
143 int n; /* total number of codes */ | |
144 const uint32_t *huffcodes; /* VLC bit values */ | |
145 const uint8_t *huffbits; /* VLC bit size */ | |
146 const uint16_t *levels; /* table to build run/level tables */ | |
147 } CoefVLCTable; | |
148 | |
149 static void wma_lsp_to_curve_init(WMADecodeContext *s, int frame_len); | |
150 | |
151 #include "wmadata.h" | |
152 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
153 #ifdef TRACE |
783 | 154 static void dump_shorts(const char *name, const short *tab, int n) |
155 { | |
156 int i; | |
157 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
158 tprintf("%s[%d]:\n", name, n); |
783 | 159 for(i=0;i<n;i++) { |
160 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
|
161 tprintf("%4d: ", i); |
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
162 tprintf(" %5d.0", tab[i]); |
783 | 163 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
|
164 tprintf("\n"); |
783 | 165 } |
166 } | |
167 | |
168 static void dump_floats(const char *name, int prec, const float *tab, int n) | |
169 { | |
170 int i; | |
171 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
172 tprintf("%s[%d]:\n", name, n); |
783 | 173 for(i=0;i<n;i++) { |
174 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
|
175 tprintf("%4d: ", i); |
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
176 tprintf(" %8.*f", prec, tab[i]); |
783 | 177 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
|
178 tprintf("\n"); |
783 | 179 } |
180 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
|
181 tprintf("\n"); |
783 | 182 } |
183 #endif | |
184 | |
185 /* XXX: use same run/length optimization as mpeg decoders */ | |
2967 | 186 static void init_coef_vlc(VLC *vlc, |
783 | 187 uint16_t **prun_table, uint16_t **plevel_table, |
188 const CoefVLCTable *vlc_table) | |
189 { | |
190 int n = vlc_table->n; | |
191 const uint8_t *table_bits = vlc_table->huffbits; | |
192 const uint32_t *table_codes = vlc_table->huffcodes; | |
193 const uint16_t *levels_table = vlc_table->levels; | |
194 uint16_t *run_table, *level_table; | |
195 const uint16_t *p; | |
196 int i, l, j, level; | |
197 | |
3113 | 198 init_vlc(vlc, VLCBITS, n, table_bits, 1, 1, table_codes, 4, 4, 0); |
783 | 199 |
1031
19de1445beb2
use av_malloc() functions - added av_strdup and av_realloc()
bellard
parents:
1025
diff
changeset
|
200 run_table = av_malloc(n * sizeof(uint16_t)); |
19de1445beb2
use av_malloc() functions - added av_strdup and av_realloc()
bellard
parents:
1025
diff
changeset
|
201 level_table = av_malloc(n * sizeof(uint16_t)); |
783 | 202 p = levels_table; |
203 i = 2; | |
204 level = 1; | |
205 while (i < n) { | |
206 l = *p++; | |
207 for(j=0;j<l;j++) { | |
208 run_table[i] = j; | |
209 level_table[i] = level; | |
210 i++; | |
211 } | |
212 level++; | |
213 } | |
214 *prun_table = run_table; | |
215 *plevel_table = level_table; | |
216 } | |
217 | |
218 static int wma_decode_init(AVCodecContext * avctx) | |
219 { | |
220 WMADecodeContext *s = avctx->priv_data; | |
221 int i, flags1, flags2; | |
222 float *window; | |
223 uint8_t *extradata; | |
3235 | 224 float bps1, high_freq; |
225 volatile float bps; | |
783 | 226 int sample_rate1; |
227 int coef_vlc_table; | |
2967 | 228 |
783 | 229 s->sample_rate = avctx->sample_rate; |
230 s->nb_channels = avctx->channels; | |
231 s->bit_rate = avctx->bit_rate; | |
232 s->block_align = avctx->block_align; | |
233 | |
3592
6a358dccf2ab
SIMD vector optimizations. 3% faster overall decoding.
banan
parents:
3555
diff
changeset
|
234 dsputil_init(&s->dsp, avctx); |
6a358dccf2ab
SIMD vector optimizations. 3% faster overall decoding.
banan
parents:
3555
diff
changeset
|
235 |
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
|
236 if (avctx->codec->id == CODEC_ID_WMAV1) { |
783 | 237 s->version = 1; |
238 } else { | |
239 s->version = 2; | |
240 } | |
2967 | 241 |
783 | 242 /* extract flag infos */ |
243 flags1 = 0; | |
244 flags2 = 0; | |
245 extradata = avctx->extradata; | |
246 if (s->version == 1 && avctx->extradata_size >= 4) { | |
247 flags1 = extradata[0] | (extradata[1] << 8); | |
248 flags2 = extradata[2] | (extradata[3] << 8); | |
249 } else if (s->version == 2 && avctx->extradata_size >= 6) { | |
2967 | 250 flags1 = extradata[0] | (extradata[1] << 8) | |
783 | 251 (extradata[2] << 16) | (extradata[3] << 24); |
252 flags2 = extradata[4] | (extradata[5] << 8); | |
253 } | |
254 s->use_exp_vlc = flags2 & 0x0001; | |
255 s->use_bit_reservoir = flags2 & 0x0002; | |
256 s->use_variable_block_len = flags2 & 0x0004; | |
257 | |
258 /* compute MDCT block size */ | |
259 if (s->sample_rate <= 16000) { | |
260 s->frame_len_bits = 9; | |
2967 | 261 } else if (s->sample_rate <= 22050 || |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
262 (s->sample_rate <= 32000 && s->version == 1)) { |
783 | 263 s->frame_len_bits = 10; |
264 } else { | |
265 s->frame_len_bits = 11; | |
266 } | |
267 s->frame_len = 1 << s->frame_len_bits; | |
268 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
|
269 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
|
270 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
|
271 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
|
272 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
|
273 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
|
274 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
|
275 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
|
276 s->nb_block_sizes = nb + 1; |
783 | 277 } else { |
278 s->nb_block_sizes = 1; | |
279 } | |
280 | |
281 /* init rate dependant parameters */ | |
282 s->use_noise_coding = 1; | |
3235 | 283 high_freq = s->sample_rate * 0.5; |
783 | 284 |
285 /* if version 2, then the rates are normalized */ | |
286 sample_rate1 = s->sample_rate; | |
287 if (s->version == 2) { | |
2967 | 288 if (sample_rate1 >= 44100) |
783 | 289 sample_rate1 = 44100; |
2967 | 290 else if (sample_rate1 >= 22050) |
783 | 291 sample_rate1 = 22050; |
2967 | 292 else if (sample_rate1 >= 16000) |
783 | 293 sample_rate1 = 16000; |
2967 | 294 else if (sample_rate1 >= 11025) |
783 | 295 sample_rate1 = 11025; |
2967 | 296 else if (sample_rate1 >= 8000) |
783 | 297 sample_rate1 = 8000; |
298 } | |
299 | |
300 bps = (float)s->bit_rate / (float)(s->nb_channels * s->sample_rate); | |
2992 | 301 s->byte_offset_bits = av_log2((int)(bps * s->frame_len / 8.0 + 0.5)) + 2; |
783 | 302 |
303 /* compute high frequency value and choose if noise coding should | |
304 be activated */ | |
305 bps1 = bps; | |
306 if (s->nb_channels == 2) | |
307 bps1 = bps * 1.6; | |
308 if (sample_rate1 == 44100) { | |
309 if (bps1 >= 0.61) | |
310 s->use_noise_coding = 0; | |
311 else | |
3235 | 312 high_freq = high_freq * 0.4; |
783 | 313 } else if (sample_rate1 == 22050) { |
314 if (bps1 >= 1.16) | |
315 s->use_noise_coding = 0; | |
2967 | 316 else if (bps1 >= 0.72) |
3235 | 317 high_freq = high_freq * 0.7; |
783 | 318 else |
3235 | 319 high_freq = high_freq * 0.6; |
783 | 320 } else if (sample_rate1 == 16000) { |
321 if (bps > 0.5) | |
3235 | 322 high_freq = high_freq * 0.5; |
783 | 323 else |
3235 | 324 high_freq = high_freq * 0.3; |
783 | 325 } else if (sample_rate1 == 11025) { |
3235 | 326 high_freq = high_freq * 0.7; |
783 | 327 } else if (sample_rate1 == 8000) { |
328 if (bps <= 0.625) { | |
3235 | 329 high_freq = high_freq * 0.5; |
783 | 330 } else if (bps > 0.75) { |
331 s->use_noise_coding = 0; | |
332 } else { | |
3235 | 333 high_freq = high_freq * 0.65; |
783 | 334 } |
335 } else { | |
336 if (bps >= 0.8) { | |
3235 | 337 high_freq = high_freq * 0.75; |
783 | 338 } else if (bps >= 0.6) { |
3235 | 339 high_freq = high_freq * 0.6; |
783 | 340 } else { |
3235 | 341 high_freq = high_freq * 0.5; |
783 | 342 } |
343 } | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
344 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
|
345 dprintf("version=%d channels=%d sample_rate=%d bitrate=%d block_align=%d\n", |
2967 | 346 s->version, s->nb_channels, s->sample_rate, s->bit_rate, |
783 | 347 s->block_align); |
3235 | 348 dprintf("bps=%f bps1=%f high_freq=%f bitoffset=%d\n", |
349 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
|
350 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
|
351 s->use_noise_coding, s->use_exp_vlc, s->nb_block_sizes); |
783 | 352 |
353 /* compute the scale factor band sizes for each MDCT block size */ | |
354 { | |
355 int a, b, pos, lpos, k, block_len, i, j, n; | |
356 const uint8_t *table; | |
2967 | 357 |
783 | 358 if (s->version == 1) { |
359 s->coefs_start = 3; | |
360 } else { | |
361 s->coefs_start = 0; | |
362 } | |
363 for(k = 0; k < s->nb_block_sizes; k++) { | |
364 block_len = s->frame_len >> k; | |
365 | |
366 if (s->version == 1) { | |
367 lpos = 0; | |
368 for(i=0;i<25;i++) { | |
369 a = wma_critical_freqs[i]; | |
370 b = s->sample_rate; | |
371 pos = ((block_len * 2 * a) + (b >> 1)) / b; | |
2967 | 372 if (pos > block_len) |
783 | 373 pos = block_len; |
374 s->exponent_bands[0][i] = pos - lpos; | |
375 if (pos >= block_len) { | |
376 i++; | |
377 break; | |
378 } | |
379 lpos = pos; | |
380 } | |
381 s->exponent_sizes[0] = i; | |
382 } else { | |
383 /* hardcoded tables */ | |
384 table = NULL; | |
385 a = s->frame_len_bits - BLOCK_MIN_BITS - k; | |
386 if (a < 3) { | |
387 if (s->sample_rate >= 44100) | |
388 table = exponent_band_44100[a]; | |
389 else if (s->sample_rate >= 32000) | |
390 table = exponent_band_32000[a]; | |
391 else if (s->sample_rate >= 22050) | |
392 table = exponent_band_22050[a]; | |
393 } | |
394 if (table) { | |
395 n = *table++; | |
396 for(i=0;i<n;i++) | |
397 s->exponent_bands[k][i] = table[i]; | |
398 s->exponent_sizes[k] = n; | |
399 } else { | |
400 j = 0; | |
401 lpos = 0; | |
402 for(i=0;i<25;i++) { | |
403 a = wma_critical_freqs[i]; | |
404 b = s->sample_rate; | |
405 pos = ((block_len * 2 * a) + (b << 1)) / (4 * b); | |
406 pos <<= 2; | |
2967 | 407 if (pos > block_len) |
783 | 408 pos = block_len; |
409 if (pos > lpos) | |
410 s->exponent_bands[k][j++] = pos - lpos; | |
411 if (pos >= block_len) | |
412 break; | |
413 lpos = pos; | |
414 } | |
415 s->exponent_sizes[k] = j; | |
416 } | |
417 } | |
418 | |
419 /* max number of coefs */ | |
420 s->coefs_end[k] = (s->frame_len - ((s->frame_len * 9) / 100)) >> k; | |
421 /* high freq computation */ | |
3235 | 422 s->high_band_start[k] = (int)((block_len * 2 * high_freq) / |
423 s->sample_rate + 0.5); | |
783 | 424 n = s->exponent_sizes[k]; |
425 j = 0; | |
426 pos = 0; | |
427 for(i=0;i<n;i++) { | |
428 int start, end; | |
429 start = pos; | |
430 pos += s->exponent_bands[k][i]; | |
431 end = pos; | |
432 if (start < s->high_band_start[k]) | |
433 start = s->high_band_start[k]; | |
434 if (end > s->coefs_end[k]) | |
435 end = s->coefs_end[k]; | |
436 if (end > start) | |
437 s->exponent_high_bands[k][j++] = end - start; | |
438 } | |
439 s->exponent_high_sizes[k] = j; | |
440 #if 0 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
441 tprintf("%5d: coefs_end=%d high_band_start=%d nb_high_bands=%d: ", |
2967 | 442 s->frame_len >> k, |
783 | 443 s->coefs_end[k], |
444 s->high_band_start[k], | |
445 s->exponent_high_sizes[k]); | |
446 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
|
447 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
|
448 tprintf("\n"); |
783 | 449 #endif |
450 } | |
451 } | |
452 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
453 #ifdef TRACE |
783 | 454 { |
455 int i, j; | |
456 for(i = 0; i < s->nb_block_sizes; i++) { | |
2967 | 457 tprintf("%5d: n=%2d:", |
458 s->frame_len >> i, | |
783 | 459 s->exponent_sizes[i]); |
460 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
|
461 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
|
462 tprintf("\n"); |
783 | 463 } |
464 } | |
465 #endif | |
466 | |
467 /* init MDCT */ | |
468 for(i = 0; i < s->nb_block_sizes; i++) | |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
469 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1); |
2967 | 470 |
783 | 471 /* init MDCT windows : simple sinus window */ |
472 for(i = 0; i < s->nb_block_sizes; i++) { | |
473 int n, j; | |
474 float alpha; | |
475 n = 1 << (s->frame_len_bits - i); | |
476 window = av_malloc(sizeof(float) * n); | |
477 alpha = M_PI / (2.0 * n); | |
478 for(j=0;j<n;j++) { | |
479 window[n - j - 1] = sin((j + 0.5) * alpha); | |
480 } | |
481 s->windows[i] = window; | |
482 } | |
483 | |
484 s->reset_block_lengths = 1; | |
2967 | 485 |
783 | 486 if (s->use_noise_coding) { |
487 | |
488 /* init the noise generator */ | |
489 if (s->use_exp_vlc) | |
490 s->noise_mult = 0.02; | |
491 else | |
492 s->noise_mult = 0.04; | |
2967 | 493 |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
494 #ifdef TRACE |
783 | 495 for(i=0;i<NOISE_TAB_SIZE;i++) |
496 s->noise_table[i] = 1.0 * s->noise_mult; | |
497 #else | |
498 { | |
499 unsigned int seed; | |
500 float norm; | |
501 seed = 1; | |
502 norm = (1.0 / (float)(1LL << 31)) * sqrt(3) * s->noise_mult; | |
503 for(i=0;i<NOISE_TAB_SIZE;i++) { | |
504 seed = seed * 314159 + 1; | |
505 s->noise_table[i] = (float)((int)seed) * norm; | |
506 } | |
507 } | |
508 #endif | |
3113 | 509 init_vlc(&s->hgain_vlc, HGAINVLCBITS, sizeof(hgain_huffbits), |
783 | 510 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
|
511 hgain_huffcodes, 2, 2, 0); |
783 | 512 } |
513 | |
514 if (s->use_exp_vlc) { | |
3113 | 515 init_vlc(&s->exp_vlc, EXPVLCBITS, sizeof(scale_huffbits), |
783 | 516 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
|
517 scale_huffcodes, 4, 4, 0); |
783 | 518 } else { |
519 wma_lsp_to_curve_init(s, s->frame_len); | |
520 } | |
521 | |
522 /* choose the VLC tables for the coefficients */ | |
523 coef_vlc_table = 2; | |
524 if (s->sample_rate >= 32000) { | |
525 if (bps1 < 0.72) | |
526 coef_vlc_table = 0; | |
527 else if (bps1 < 1.16) | |
528 coef_vlc_table = 1; | |
529 } | |
530 | |
531 init_coef_vlc(&s->coef_vlc[0], &s->run_table[0], &s->level_table[0], | |
532 &coef_vlcs[coef_vlc_table * 2]); | |
533 init_coef_vlc(&s->coef_vlc[1], &s->run_table[1], &s->level_table[1], | |
534 &coef_vlcs[coef_vlc_table * 2 + 1]); | |
535 return 0; | |
536 } | |
537 | |
538 /* interpolate values for a bigger or smaller block. The block must | |
539 have multiple sizes */ | |
540 static void interpolate_array(float *scale, int old_size, int new_size) | |
541 { | |
542 int i, j, jincr, k; | |
543 float v; | |
544 | |
545 if (new_size > old_size) { | |
546 jincr = new_size / old_size; | |
547 j = new_size; | |
548 for(i = old_size - 1; i >=0; i--) { | |
549 v = scale[i]; | |
550 k = jincr; | |
551 do { | |
552 scale[--j] = v; | |
553 } while (--k); | |
554 } | |
555 } else if (new_size < old_size) { | |
556 j = 0; | |
557 jincr = old_size / new_size; | |
558 for(i = 0; i < new_size; i++) { | |
559 scale[i] = scale[j]; | |
560 j += jincr; | |
561 } | |
562 } | |
563 } | |
564 | |
565 /* compute x^-0.25 with an exponent and mantissa table. We use linear | |
566 interpolation to reduce the mantissa table size at a small speed | |
567 expense (linear interpolation approximately doubles the number of | |
568 bits of precision). */ | |
569 static inline float pow_m1_4(WMADecodeContext *s, float x) | |
570 { | |
571 union { | |
572 float f; | |
573 unsigned int v; | |
574 } u, t; | |
575 unsigned int e, m; | |
576 float a, b; | |
577 | |
578 u.f = x; | |
579 e = u.v >> 23; | |
580 m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1); | |
581 /* build interpolation scale: 1 <= t < 2. */ | |
582 t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23); | |
583 a = s->lsp_pow_m_table1[m]; | |
584 b = s->lsp_pow_m_table2[m]; | |
585 return s->lsp_pow_e_table[e] * (a + b * t.f); | |
586 } | |
587 | |
588 static void wma_lsp_to_curve_init(WMADecodeContext *s, int frame_len) | |
2967 | 589 { |
783 | 590 float wdel, a, b; |
591 int i, e, m; | |
592 | |
593 wdel = M_PI / frame_len; | |
594 for(i=0;i<frame_len;i++) | |
595 s->lsp_cos_table[i] = 2.0f * cos(wdel * i); | |
596 | |
597 /* tables for x^-0.25 computation */ | |
598 for(i=0;i<256;i++) { | |
599 e = i - 126; | |
600 s->lsp_pow_e_table[i] = pow(2.0, e * -0.25); | |
601 } | |
602 | |
603 /* NOTE: these two tables are needed to avoid two operations in | |
604 pow_m1_4 */ | |
605 b = 1.0; | |
606 for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) { | |
607 m = (1 << LSP_POW_BITS) + i; | |
608 a = (float)m * (0.5 / (1 << LSP_POW_BITS)); | |
609 a = pow(a, -0.25); | |
610 s->lsp_pow_m_table1[i] = 2 * a - b; | |
611 s->lsp_pow_m_table2[i] = b - a; | |
612 b = a; | |
613 } | |
614 #if 0 | |
615 for(i=1;i<20;i++) { | |
616 float v, r1, r2; | |
617 v = 5.0 / i; | |
618 r1 = pow_m1_4(s, v); | |
619 r2 = pow(v,-0.25); | |
620 printf("%f^-0.25=%f e=%f\n", v, r1, r2 - r1); | |
621 } | |
622 #endif | |
623 } | |
624 | |
625 /* NOTE: We use the same code as Vorbis here */ | |
626 /* XXX: optimize it further with SSE/3Dnow */ | |
2967 | 627 static void wma_lsp_to_curve(WMADecodeContext *s, |
628 float *out, float *val_max_ptr, | |
783 | 629 int n, float *lsp) |
630 { | |
631 int i, j; | |
632 float p, q, w, v, val_max; | |
633 | |
634 val_max = 0; | |
635 for(i=0;i<n;i++) { | |
636 p = 0.5f; | |
637 q = 0.5f; | |
638 w = s->lsp_cos_table[i]; | |
639 for(j=1;j<NB_LSP_COEFS;j+=2){ | |
640 q *= w - lsp[j - 1]; | |
641 p *= w - lsp[j]; | |
642 } | |
643 p *= p * (2.0f - w); | |
644 q *= q * (2.0f + w); | |
645 v = p + q; | |
646 v = pow_m1_4(s, v); | |
647 if (v > val_max) | |
648 val_max = v; | |
649 out[i] = v; | |
650 } | |
651 *val_max_ptr = val_max; | |
652 } | |
653 | |
654 /* decode exponents coded with LSP coefficients (same idea as Vorbis) */ | |
655 static void decode_exp_lsp(WMADecodeContext *s, int ch) | |
656 { | |
657 float lsp_coefs[NB_LSP_COEFS]; | |
658 int val, i; | |
659 | |
660 for(i = 0; i < NB_LSP_COEFS; i++) { | |
661 if (i == 0 || i >= 8) | |
662 val = get_bits(&s->gb, 3); | |
663 else | |
664 val = get_bits(&s->gb, 4); | |
665 lsp_coefs[i] = lsp_codebook[i][val]; | |
666 } | |
667 | |
668 wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch], | |
669 s->block_len, lsp_coefs); | |
670 } | |
671 | |
672 /* decode exponents coded with VLC codes */ | |
673 static int decode_exp_vlc(WMADecodeContext *s, int ch) | |
674 { | |
675 int last_exp, n, code; | |
676 const uint16_t *ptr, *band_ptr; | |
677 float v, *q, max_scale, *q_end; | |
2967 | 678 |
783 | 679 band_ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits]; |
680 ptr = band_ptr; | |
681 q = s->exponents[ch]; | |
682 q_end = q + s->block_len; | |
683 max_scale = 0; | |
684 if (s->version == 1) { | |
685 last_exp = get_bits(&s->gb, 5) + 10; | |
3235 | 686 /* XXX: use a table */ |
687 v = pow(10, last_exp * (1.0 / 16.0)); | |
783 | 688 max_scale = v; |
689 n = *ptr++; | |
690 do { | |
691 *q++ = v; | |
692 } while (--n); | |
693 } | |
694 last_exp = 36; | |
695 while (q < q_end) { | |
3113 | 696 code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX); |
783 | 697 if (code < 0) |
698 return -1; | |
699 /* NOTE: this offset is the same as MPEG4 AAC ! */ | |
700 last_exp += code - 60; | |
3235 | 701 /* XXX: use a table */ |
702 v = pow(10, last_exp * (1.0 / 16.0)); | |
783 | 703 if (v > max_scale) |
704 max_scale = v; | |
705 n = *ptr++; | |
706 do { | |
707 *q++ = v; | |
708 } while (--n); | |
709 } | |
710 s->max_exponent[ch] = max_scale; | |
711 return 0; | |
712 } | |
713 | |
714 /* return 0 if OK. return 1 if last block of frame. return -1 if | |
715 unrecorrable error. */ | |
716 static int wma_decode_block(WMADecodeContext *s) | |
717 { | |
718 int n, v, a, ch, code, bsize; | |
719 int coef_nb_bits, total_gain, parse_exponents; | |
3606
e28285ddde8d
Alignment of the LT window, segfault fix patch by Baptiste Coudurier.
banan
parents:
3592
diff
changeset
|
720 DECLARE_ALIGNED_16(float, window[BLOCK_MAX_SIZE * 2]); |
783 | 721 int nb_coefs[MAX_CHANNELS]; |
722 float mdct_norm; | |
723 | |
1343 | 724 #ifdef TRACE |
725 tprintf("***decode_block: %d:%d\n", s->frame_count - 1, s->block_num); | |
726 #endif | |
783 | 727 |
728 /* compute current block length */ | |
729 if (s->use_variable_block_len) { | |
730 n = av_log2(s->nb_block_sizes - 1) + 1; | |
2967 | 731 |
783 | 732 if (s->reset_block_lengths) { |
733 s->reset_block_lengths = 0; | |
734 v = get_bits(&s->gb, n); | |
735 if (v >= s->nb_block_sizes) | |
736 return -1; | |
737 s->prev_block_len_bits = s->frame_len_bits - v; | |
738 v = get_bits(&s->gb, n); | |
739 if (v >= s->nb_block_sizes) | |
740 return -1; | |
741 s->block_len_bits = s->frame_len_bits - v; | |
742 } else { | |
743 /* update block lengths */ | |
744 s->prev_block_len_bits = s->block_len_bits; | |
745 s->block_len_bits = s->next_block_len_bits; | |
746 } | |
747 v = get_bits(&s->gb, n); | |
748 if (v >= s->nb_block_sizes) | |
749 return -1; | |
750 s->next_block_len_bits = s->frame_len_bits - v; | |
751 } else { | |
752 /* fixed block len */ | |
753 s->next_block_len_bits = s->frame_len_bits; | |
754 s->prev_block_len_bits = s->frame_len_bits; | |
755 s->block_len_bits = s->frame_len_bits; | |
756 } | |
757 | |
758 /* now check if the block length is coherent with the frame length */ | |
759 s->block_len = 1 << s->block_len_bits; | |
760 if ((s->block_pos + s->block_len) > s->frame_len) | |
761 return -1; | |
762 | |
763 if (s->nb_channels == 2) { | |
764 s->ms_stereo = get_bits(&s->gb, 1); | |
765 } | |
766 v = 0; | |
767 for(ch = 0; ch < s->nb_channels; ch++) { | |
768 a = get_bits(&s->gb, 1); | |
769 s->channel_coded[ch] = a; | |
770 v |= a; | |
771 } | |
772 /* if no channel coded, no need to go further */ | |
773 /* XXX: fix potential framing problems */ | |
774 if (!v) | |
775 goto next; | |
776 | |
777 bsize = s->frame_len_bits - s->block_len_bits; | |
778 | |
779 /* read total gain and extract corresponding number of bits for | |
780 coef escape coding */ | |
781 total_gain = 1; | |
782 for(;;) { | |
783 a = get_bits(&s->gb, 7); | |
784 total_gain += a; | |
785 if (a != 127) | |
786 break; | |
787 } | |
2967 | 788 |
783 | 789 if (total_gain < 15) |
790 coef_nb_bits = 13; | |
791 else if (total_gain < 32) | |
792 coef_nb_bits = 12; | |
793 else if (total_gain < 40) | |
794 coef_nb_bits = 11; | |
795 else if (total_gain < 45) | |
796 coef_nb_bits = 10; | |
797 else | |
798 coef_nb_bits = 9; | |
799 | |
800 /* compute number of coefficients */ | |
801 n = s->coefs_end[bsize] - s->coefs_start; | |
802 for(ch = 0; ch < s->nb_channels; ch++) | |
803 nb_coefs[ch] = n; | |
804 | |
805 /* complex coding */ | |
806 if (s->use_noise_coding) { | |
807 | |
808 for(ch = 0; ch < s->nb_channels; ch++) { | |
809 if (s->channel_coded[ch]) { | |
810 int i, n, a; | |
811 n = s->exponent_high_sizes[bsize]; | |
812 for(i=0;i<n;i++) { | |
813 a = get_bits(&s->gb, 1); | |
814 s->high_band_coded[ch][i] = a; | |
815 /* if noise coding, the coefficients are not transmitted */ | |
816 if (a) | |
817 nb_coefs[ch] -= s->exponent_high_bands[bsize][i]; | |
818 } | |
819 } | |
820 } | |
821 for(ch = 0; ch < s->nb_channels; ch++) { | |
822 if (s->channel_coded[ch]) { | |
823 int i, n, val, code; | |
824 | |
825 n = s->exponent_high_sizes[bsize]; | |
826 val = (int)0x80000000; | |
827 for(i=0;i<n;i++) { | |
828 if (s->high_band_coded[ch][i]) { | |
829 if (val == (int)0x80000000) { | |
830 val = get_bits(&s->gb, 7) - 19; | |
831 } else { | |
3113 | 832 code = get_vlc2(&s->gb, s->hgain_vlc.table, HGAINVLCBITS, HGAINMAX); |
783 | 833 if (code < 0) |
834 return -1; | |
835 val += code - 18; | |
836 } | |
837 s->high_band_values[ch][i] = val; | |
838 } | |
839 } | |
840 } | |
841 } | |
842 } | |
2967 | 843 |
783 | 844 /* exposant can be interpolated in short blocks. */ |
845 parse_exponents = 1; | |
846 if (s->block_len_bits != s->frame_len_bits) { | |
847 parse_exponents = get_bits(&s->gb, 1); | |
848 } | |
2967 | 849 |
783 | 850 if (parse_exponents) { |
851 for(ch = 0; ch < s->nb_channels; ch++) { | |
852 if (s->channel_coded[ch]) { | |
853 if (s->use_exp_vlc) { | |
854 if (decode_exp_vlc(s, ch) < 0) | |
855 return -1; | |
856 } else { | |
857 decode_exp_lsp(s, ch); | |
858 } | |
859 } | |
860 } | |
861 } else { | |
862 for(ch = 0; ch < s->nb_channels; ch++) { | |
863 if (s->channel_coded[ch]) { | |
2967 | 864 interpolate_array(s->exponents[ch], 1 << s->prev_block_len_bits, |
783 | 865 s->block_len); |
866 } | |
867 } | |
868 } | |
869 | |
870 /* parse spectral coefficients : just RLE encoding */ | |
871 for(ch = 0; ch < s->nb_channels; ch++) { | |
872 if (s->channel_coded[ch]) { | |
873 VLC *coef_vlc; | |
874 int level, run, sign, tindex; | |
875 int16_t *ptr, *eptr; | |
3776 | 876 const uint16_t *level_table, *run_table; |
783 | 877 |
878 /* special VLC tables are used for ms stereo because | |
879 there is potentially less energy there */ | |
880 tindex = (ch == 1 && s->ms_stereo); | |
881 coef_vlc = &s->coef_vlc[tindex]; | |
882 run_table = s->run_table[tindex]; | |
883 level_table = s->level_table[tindex]; | |
884 /* XXX: optimize */ | |
885 ptr = &s->coefs1[ch][0]; | |
886 eptr = ptr + nb_coefs[ch]; | |
887 memset(ptr, 0, s->block_len * sizeof(int16_t)); | |
888 for(;;) { | |
3113 | 889 code = get_vlc2(&s->gb, coef_vlc->table, VLCBITS, VLCMAX); |
783 | 890 if (code < 0) |
891 return -1; | |
892 if (code == 1) { | |
893 /* EOB */ | |
894 break; | |
895 } else if (code == 0) { | |
896 /* escape */ | |
897 level = get_bits(&s->gb, coef_nb_bits); | |
898 /* NOTE: this is rather suboptimal. reading | |
899 block_len_bits would be better */ | |
900 run = get_bits(&s->gb, s->frame_len_bits); | |
901 } else { | |
902 /* normal code */ | |
903 run = run_table[code]; | |
904 level = level_table[code]; | |
905 } | |
906 sign = get_bits(&s->gb, 1); | |
907 if (!sign) | |
908 level = -level; | |
909 ptr += run; | |
910 if (ptr >= eptr) | |
3361 | 911 { |
912 av_log(NULL, AV_LOG_ERROR, "overflow in spectral RLE, ignoring\n"); | |
913 break; | |
914 } | |
783 | 915 *ptr++ = level; |
916 /* NOTE: EOB can be omitted */ | |
917 if (ptr >= eptr) | |
918 break; | |
919 } | |
920 } | |
921 if (s->version == 1 && s->nb_channels >= 2) { | |
922 align_get_bits(&s->gb); | |
923 } | |
924 } | |
2967 | 925 |
783 | 926 /* normalize */ |
927 { | |
928 int n4 = s->block_len / 2; | |
929 mdct_norm = 1.0 / (float)n4; | |
930 if (s->version == 1) { | |
931 mdct_norm *= sqrt(n4); | |
932 } | |
933 } | |
934 | |
935 /* finally compute the MDCT coefficients */ | |
936 for(ch = 0; ch < s->nb_channels; ch++) { | |
937 if (s->channel_coded[ch]) { | |
938 int16_t *coefs1; | |
939 float *coefs, *exponents, mult, mult1, noise, *exp_ptr; | |
940 int i, j, n, n1, last_high_band; | |
941 float exp_power[HIGH_BAND_MAX_SIZE]; | |
942 | |
943 coefs1 = s->coefs1[ch]; | |
944 exponents = s->exponents[ch]; | |
3235 | 945 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch]; |
783 | 946 mult *= mdct_norm; |
947 coefs = s->coefs[ch]; | |
948 if (s->use_noise_coding) { | |
949 mult1 = mult; | |
950 /* very low freqs : noise */ | |
951 for(i = 0;i < s->coefs_start; i++) { | |
952 *coefs++ = s->noise_table[s->noise_index] * (*exponents++) * mult1; | |
953 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
954 } | |
2967 | 955 |
783 | 956 n1 = s->exponent_high_sizes[bsize]; |
957 | |
958 /* compute power of high bands */ | |
2967 | 959 exp_ptr = exponents + |
960 s->high_band_start[bsize] - | |
783 | 961 s->coefs_start; |
962 last_high_band = 0; /* avoid warning */ | |
963 for(j=0;j<n1;j++) { | |
2967 | 964 n = s->exponent_high_bands[s->frame_len_bits - |
783 | 965 s->block_len_bits][j]; |
966 if (s->high_band_coded[ch][j]) { | |
967 float e2, v; | |
968 e2 = 0; | |
969 for(i = 0;i < n; i++) { | |
970 v = exp_ptr[i]; | |
971 e2 += v * v; | |
972 } | |
973 exp_power[j] = e2 / n; | |
974 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
|
975 tprintf("%d: power=%f (%d)\n", j, exp_power[j], n); |
783 | 976 } |
977 exp_ptr += n; | |
978 } | |
979 | |
980 /* main freqs and high freqs */ | |
981 for(j=-1;j<n1;j++) { | |
982 if (j < 0) { | |
2967 | 983 n = s->high_band_start[bsize] - |
783 | 984 s->coefs_start; |
985 } else { | |
2967 | 986 n = s->exponent_high_bands[s->frame_len_bits - |
783 | 987 s->block_len_bits][j]; |
988 } | |
989 if (j >= 0 && s->high_band_coded[ch][j]) { | |
990 /* use noise with specified power */ | |
991 mult1 = sqrt(exp_power[j] / exp_power[last_high_band]); | |
3235 | 992 /* XXX: use a table */ |
993 mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05); | |
783 | 994 mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult); |
995 mult1 *= mdct_norm; | |
996 for(i = 0;i < n; i++) { | |
997 noise = s->noise_table[s->noise_index]; | |
998 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
999 *coefs++ = (*exponents++) * noise * mult1; | |
1000 } | |
1001 } else { | |
1002 /* coded values + small noise */ | |
1003 for(i = 0;i < n; i++) { | |
1004 noise = s->noise_table[s->noise_index]; | |
1005 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
1006 *coefs++ = ((*coefs1++) + noise) * (*exponents++) * mult; | |
1007 } | |
1008 } | |
1009 } | |
1010 | |
1011 /* very high freqs : noise */ | |
1012 n = s->block_len - s->coefs_end[bsize]; | |
1013 mult1 = mult * exponents[-1]; | |
1014 for(i = 0; i < n; i++) { | |
1015 *coefs++ = s->noise_table[s->noise_index] * mult1; | |
1016 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
1017 } | |
1018 } else { | |
1019 /* XXX: optimize more */ | |
1020 for(i = 0;i < s->coefs_start; i++) | |
1021 *coefs++ = 0.0; | |
1022 n = nb_coefs[ch]; | |
1023 for(i = 0;i < n; i++) { | |
1024 *coefs++ = coefs1[i] * exponents[i] * mult; | |
1025 } | |
1026 n = s->block_len - s->coefs_end[bsize]; | |
1027 for(i = 0;i < n; i++) | |
1028 *coefs++ = 0.0; | |
1029 } | |
1030 } | |
1031 } | |
1032 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
1033 #ifdef TRACE |
783 | 1034 for(ch = 0; ch < s->nb_channels; ch++) { |
1035 if (s->channel_coded[ch]) { | |
1036 dump_floats("exponents", 3, s->exponents[ch], s->block_len); | |
1037 dump_floats("coefs", 1, s->coefs[ch], s->block_len); | |
1038 } | |
1039 } | |
1040 #endif | |
2967 | 1041 |
783 | 1042 if (s->ms_stereo && s->channel_coded[1]) { |
1043 float a, b; | |
1044 int i; | |
1045 | |
1046 /* nominal case for ms stereo: we do it before mdct */ | |
1047 /* no need to optimize this case because it should almost | |
1048 never happen */ | |
1049 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
|
1050 tprintf("rare ms-stereo case happened\n"); |
783 | 1051 memset(s->coefs[0], 0, sizeof(float) * s->block_len); |
1052 s->channel_coded[0] = 1; | |
1053 } | |
2967 | 1054 |
783 | 1055 for(i = 0; i < s->block_len; i++) { |
1056 a = s->coefs[0][i]; | |
1057 b = s->coefs[1][i]; | |
1058 s->coefs[0][i] = a + b; | |
1059 s->coefs[1][i] = a - b; | |
1060 } | |
1061 } | |
1062 | |
1063 /* build the window : we ensure that when the windows overlap | |
1064 their squared sum is always 1 (MDCT reconstruction rule) */ | |
1065 /* XXX: merge with output */ | |
1066 { | |
1067 int i, next_block_len, block_len, prev_block_len, n; | |
1068 float *wptr; | |
1069 | |
1070 block_len = s->block_len; | |
1071 prev_block_len = 1 << s->prev_block_len_bits; | |
1072 next_block_len = 1 << s->next_block_len_bits; | |
1073 | |
1074 /* right part */ | |
1075 wptr = window + block_len; | |
1076 if (block_len <= next_block_len) { | |
1077 for(i=0;i<block_len;i++) | |
1078 *wptr++ = s->windows[bsize][i]; | |
1079 } else { | |
1080 /* overlap */ | |
1081 n = (block_len / 2) - (next_block_len / 2); | |
1082 for(i=0;i<n;i++) | |
1083 *wptr++ = 1.0; | |
1084 for(i=0;i<next_block_len;i++) | |
1085 *wptr++ = s->windows[s->frame_len_bits - s->next_block_len_bits][i]; | |
1086 for(i=0;i<n;i++) | |
1087 *wptr++ = 0.0; | |
1088 } | |
1089 | |
1090 /* left part */ | |
1091 wptr = window + block_len; | |
1092 if (block_len <= prev_block_len) { | |
1093 for(i=0;i<block_len;i++) | |
1094 *--wptr = s->windows[bsize][i]; | |
1095 } else { | |
1096 /* overlap */ | |
1097 n = (block_len / 2) - (prev_block_len / 2); | |
1098 for(i=0;i<n;i++) | |
1099 *--wptr = 1.0; | |
1100 for(i=0;i<prev_block_len;i++) | |
1101 *--wptr = s->windows[s->frame_len_bits - s->prev_block_len_bits][i]; | |
1102 for(i=0;i<n;i++) | |
1103 *--wptr = 0.0; | |
1104 } | |
1105 } | |
1106 | |
2967 | 1107 |
783 | 1108 for(ch = 0; ch < s->nb_channels; ch++) { |
1109 if (s->channel_coded[ch]) { | |
3089 | 1110 DECLARE_ALIGNED_16(FFTSample, output[BLOCK_MAX_SIZE * 2]); |
783 | 1111 float *ptr; |
3592
6a358dccf2ab
SIMD vector optimizations. 3% faster overall decoding.
banan
parents:
3555
diff
changeset
|
1112 int n4, index, n; |
783 | 1113 |
1114 n = s->block_len; | |
1115 n4 = s->block_len / 2; | |
3555 | 1116 s->mdct_ctx[bsize].fft.imdct_calc(&s->mdct_ctx[bsize], |
795
55add0e7eafb
avoid name clash - fixed again block size selection
bellard
parents:
785
diff
changeset
|
1117 output, s->coefs[ch], s->mdct_tmp); |
783 | 1118 |
1119 /* XXX: optimize all that by build the window and | |
1120 multipying/adding at the same time */ | |
1121 | |
3592
6a358dccf2ab
SIMD vector optimizations. 3% faster overall decoding.
banan
parents:
3555
diff
changeset
|
1122 /* multiply by the window and add in the frame */ |
783 | 1123 index = (s->frame_len / 2) + s->block_pos - n4; |
1124 ptr = &s->frame_out[ch][index]; | |
3592
6a358dccf2ab
SIMD vector optimizations. 3% faster overall decoding.
banan
parents:
3555
diff
changeset
|
1125 s->dsp.vector_fmul_add_add(ptr,window,output,ptr,0,2*n,1); |
783 | 1126 |
1127 /* specific fast case for ms-stereo : add to second | |
1128 channel if it is not coded */ | |
1129 if (s->ms_stereo && !s->channel_coded[1]) { | |
1130 ptr = &s->frame_out[1][index]; | |
3592
6a358dccf2ab
SIMD vector optimizations. 3% faster overall decoding.
banan
parents:
3555
diff
changeset
|
1131 s->dsp.vector_fmul_add_add(ptr,window,output,ptr,0,2*n,1); |
783 | 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; | |
3362
c43fcf831f7c
Do not read full byte when less than 8 bits are still to be read.
reimar
parents:
3361
diff
changeset
|
1232 while (len > 7) { |
783 | 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 }; |