2 * WMA compatible decoder
3 * Copyright (c) 2002 The FFmpeg Project
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * WMA compatible decoder.
25 * This decoder handles Microsoft Windows Media Audio data, versions 1 & 2.
26 * WMA v1 is identified by audio format 0x160 in Microsoft media files
27 * (ASF/AVI/WAV). WMA v2 is identified by audio format 0x161.
29 * To use this decoder, a calling application must supply the extra data
30 * bytes provided with the WMA data. These are the extra, codec-specific
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
33 * should be 4 extra bytes for v1 data and 6 extra bytes for v2 data.
36 #include "libavutil/attributes.h"
46 #define EXPMAX ((19 + EXPVLCBITS - 1) / EXPVLCBITS)
48 #define HGAINVLCBITS 9
49 #define HGAINMAX ((13 + HGAINVLCBITS - 1) / HGAINVLCBITS)
51 static void wma_lsp_to_curve_init(WMACodecContext
*s
, int frame_len
);
54 static void dump_floats(WMACodecContext
*s
, const char *name
,
55 int prec
, const float *tab
, int n
)
59 tprintf(s
->avctx
, "%s[%d]:\n", name
, n
);
60 for (i
= 0; i
< n
; i
++) {
62 tprintf(s
->avctx
, "%4d: ", i
);
63 tprintf(s
->avctx
, " %8.*f", prec
, tab
[i
]);
65 tprintf(s
->avctx
, "\n");
68 tprintf(s
->avctx
, "\n");
72 static av_cold
int wma_decode_init(AVCodecContext
*avctx
)
74 WMACodecContext
*s
= avctx
->priv_data
;
78 if (!avctx
->block_align
) {
79 av_log(avctx
, AV_LOG_ERROR
, "block_align is not set\n");
80 return AVERROR(EINVAL
);
85 /* extract flag infos */
87 extradata
= avctx
->extradata
;
88 if (avctx
->codec
->id
== AV_CODEC_ID_WMAV1
&& avctx
->extradata_size
>= 4)
89 flags2
= AV_RL16(extradata
+ 2);
90 else if (avctx
->codec
->id
== AV_CODEC_ID_WMAV2
&& avctx
->extradata_size
>= 6)
91 flags2
= AV_RL16(extradata
+ 4);
93 s
->use_exp_vlc
= flags2
& 0x0001;
94 s
->use_bit_reservoir
= flags2
& 0x0002;
95 s
->use_variable_block_len
= flags2
& 0x0004;
97 if (avctx
->codec
->id
== AV_CODEC_ID_WMAV2
&& avctx
->extradata_size
>= 8){
98 if (AV_RL16(extradata
+4)==0xd && s
->use_variable_block_len
){
99 av_log(avctx
, AV_LOG_WARNING
, "Disabling use_variable_block_len, if this fails contact the ffmpeg developers and send us the file\n");
100 s
->use_variable_block_len
= 0; // this fixes issue1503
104 for (i
=0; i
<MAX_CHANNELS
; i
++)
105 s
->max_exponent
[i
] = 1.0;
107 if (ff_wma_init(avctx
, flags2
) < 0)
111 for (i
= 0; i
< s
->nb_block_sizes
; i
++)
112 ff_mdct_init(&s
->mdct_ctx
[i
], s
->frame_len_bits
- i
+ 1, 1, 1.0 / 32768.0);
114 if (s
->use_noise_coding
) {
115 init_vlc(&s
->hgain_vlc
, HGAINVLCBITS
, sizeof(ff_wma_hgain_huffbits
),
116 ff_wma_hgain_huffbits
, 1, 1,
117 ff_wma_hgain_huffcodes
, 2, 2, 0);
121 init_vlc(&s
->exp_vlc
, EXPVLCBITS
, sizeof(ff_aac_scalefactor_bits
), // FIXME move out of context
122 ff_aac_scalefactor_bits
, 1, 1,
123 ff_aac_scalefactor_code
, 4, 4, 0);
125 wma_lsp_to_curve_init(s
, s
->frame_len
);
127 avctx
->sample_fmt
= AV_SAMPLE_FMT_FLTP
;
133 * compute x^-0.25 with an exponent and mantissa table. We use linear
134 * interpolation to reduce the mantissa table size at a small speed
135 * expense (linear interpolation approximately doubles the number of
136 * bits of precision).
138 static inline float pow_m1_4(WMACodecContext
*s
, float x
)
149 m
= (u
.v
>> (23 - LSP_POW_BITS
)) & ((1 << LSP_POW_BITS
) - 1);
150 /* build interpolation scale: 1 <= t < 2. */
151 t
.v
= ((u
.v
<< LSP_POW_BITS
) & ((1 << 23) - 1)) | (127 << 23);
152 a
= s
->lsp_pow_m_table1
[m
];
153 b
= s
->lsp_pow_m_table2
[m
];
154 return s
->lsp_pow_e_table
[e
] * (a
+ b
* t
.f
);
157 static av_cold
void wma_lsp_to_curve_init(WMACodecContext
*s
, int frame_len
)
162 wdel
= M_PI
/ frame_len
;
163 for (i
= 0; i
< frame_len
; i
++)
164 s
->lsp_cos_table
[i
] = 2.0f
* cos(wdel
* i
);
166 /* tables for x^-0.25 computation */
167 for (i
= 0; i
< 256; i
++) {
169 s
->lsp_pow_e_table
[i
] = pow(2.0, e
* -0.25);
172 /* NOTE: these two tables are needed to avoid two operations in
175 for (i
= (1 << LSP_POW_BITS
) - 1; i
>= 0; i
--) {
176 m
= (1 << LSP_POW_BITS
) + i
;
177 a
= (float) m
* (0.5 / (1 << LSP_POW_BITS
));
179 s
->lsp_pow_m_table1
[i
] = 2 * a
- b
;
180 s
->lsp_pow_m_table2
[i
] = b
- a
;
186 * NOTE: We use the same code as Vorbis here
187 * @todo optimize it further with SSE/3Dnow
189 static void wma_lsp_to_curve(WMACodecContext
*s
, float *out
, float *val_max_ptr
,
193 float p
, q
, w
, v
, val_max
;
196 for (i
= 0; i
< n
; i
++) {
199 w
= s
->lsp_cos_table
[i
];
200 for (j
= 1; j
< NB_LSP_COEFS
; j
+= 2) {
212 *val_max_ptr
= val_max
;
216 * decode exponents coded with LSP coefficients (same idea as Vorbis)
218 static void decode_exp_lsp(WMACodecContext
*s
, int ch
)
220 float lsp_coefs
[NB_LSP_COEFS
];
223 for (i
= 0; i
< NB_LSP_COEFS
; i
++) {
224 if (i
== 0 || i
>= 8)
225 val
= get_bits(&s
->gb
, 3);
227 val
= get_bits(&s
->gb
, 4);
228 lsp_coefs
[i
] = ff_wma_lsp_codebook
[i
][val
];
231 wma_lsp_to_curve(s
, s
->exponents
[ch
], &s
->max_exponent
[ch
],
232 s
->block_len
, lsp_coefs
);
235 /** pow(10, i / 16.0) for i in -60..95 */
236 static const float pow_tab
[] = {
237 1.7782794100389e-04, 2.0535250264571e-04,
238 2.3713737056617e-04, 2.7384196342644e-04,
239 3.1622776601684e-04, 3.6517412725484e-04,
240 4.2169650342858e-04, 4.8696752516586e-04,
241 5.6234132519035e-04, 6.4938163157621e-04,
242 7.4989420933246e-04, 8.6596432336006e-04,
243 1.0000000000000e-03, 1.1547819846895e-03,
244 1.3335214321633e-03, 1.5399265260595e-03,
245 1.7782794100389e-03, 2.0535250264571e-03,
246 2.3713737056617e-03, 2.7384196342644e-03,
247 3.1622776601684e-03, 3.6517412725484e-03,
248 4.2169650342858e-03, 4.8696752516586e-03,
249 5.6234132519035e-03, 6.4938163157621e-03,
250 7.4989420933246e-03, 8.6596432336006e-03,
251 1.0000000000000e-02, 1.1547819846895e-02,
252 1.3335214321633e-02, 1.5399265260595e-02,
253 1.7782794100389e-02, 2.0535250264571e-02,
254 2.3713737056617e-02, 2.7384196342644e-02,
255 3.1622776601684e-02, 3.6517412725484e-02,
256 4.2169650342858e-02, 4.8696752516586e-02,
257 5.6234132519035e-02, 6.4938163157621e-02,
258 7.4989420933246e-02, 8.6596432336007e-02,
259 1.0000000000000e-01, 1.1547819846895e-01,
260 1.3335214321633e-01, 1.5399265260595e-01,
261 1.7782794100389e-01, 2.0535250264571e-01,
262 2.3713737056617e-01, 2.7384196342644e-01,
263 3.1622776601684e-01, 3.6517412725484e-01,
264 4.2169650342858e-01, 4.8696752516586e-01,
265 5.6234132519035e-01, 6.4938163157621e-01,
266 7.4989420933246e-01, 8.6596432336007e-01,
267 1.0000000000000e+00, 1.1547819846895e+00,
268 1.3335214321633e+00, 1.5399265260595e+00,
269 1.7782794100389e+00, 2.0535250264571e+00,
270 2.3713737056617e+00, 2.7384196342644e+00,
271 3.1622776601684e+00, 3.6517412725484e+00,
272 4.2169650342858e+00, 4.8696752516586e+00,
273 5.6234132519035e+00, 6.4938163157621e+00,
274 7.4989420933246e+00, 8.6596432336007e+00,
275 1.0000000000000e+01, 1.1547819846895e+01,
276 1.3335214321633e+01, 1.5399265260595e+01,
277 1.7782794100389e+01, 2.0535250264571e+01,
278 2.3713737056617e+01, 2.7384196342644e+01,
279 3.1622776601684e+01, 3.6517412725484e+01,
280 4.2169650342858e+01, 4.8696752516586e+01,
281 5.6234132519035e+01, 6.4938163157621e+01,
282 7.4989420933246e+01, 8.6596432336007e+01,
283 1.0000000000000e+02, 1.1547819846895e+02,
284 1.3335214321633e+02, 1.5399265260595e+02,
285 1.7782794100389e+02, 2.0535250264571e+02,
286 2.3713737056617e+02, 2.7384196342644e+02,
287 3.1622776601684e+02, 3.6517412725484e+02,
288 4.2169650342858e+02, 4.8696752516586e+02,
289 5.6234132519035e+02, 6.4938163157621e+02,
290 7.4989420933246e+02, 8.6596432336007e+02,
291 1.0000000000000e+03, 1.1547819846895e+03,
292 1.3335214321633e+03, 1.5399265260595e+03,
293 1.7782794100389e+03, 2.0535250264571e+03,
294 2.3713737056617e+03, 2.7384196342644e+03,
295 3.1622776601684e+03, 3.6517412725484e+03,
296 4.2169650342858e+03, 4.8696752516586e+03,
297 5.6234132519035e+03, 6.4938163157621e+03,
298 7.4989420933246e+03, 8.6596432336007e+03,
299 1.0000000000000e+04, 1.1547819846895e+04,
300 1.3335214321633e+04, 1.5399265260595e+04,
301 1.7782794100389e+04, 2.0535250264571e+04,
302 2.3713737056617e+04, 2.7384196342644e+04,
303 3.1622776601684e+04, 3.6517412725484e+04,
304 4.2169650342858e+04, 4.8696752516586e+04,
305 5.6234132519035e+04, 6.4938163157621e+04,
306 7.4989420933246e+04, 8.6596432336007e+04,
307 1.0000000000000e+05, 1.1547819846895e+05,
308 1.3335214321633e+05, 1.5399265260595e+05,
309 1.7782794100389e+05, 2.0535250264571e+05,
310 2.3713737056617e+05, 2.7384196342644e+05,
311 3.1622776601684e+05, 3.6517412725484e+05,
312 4.2169650342858e+05, 4.8696752516586e+05,
313 5.6234132519035e+05, 6.4938163157621e+05,
314 7.4989420933246e+05, 8.6596432336007e+05,
318 * decode exponents coded with VLC codes
320 static int decode_exp_vlc(WMACodecContext
*s
, int ch
)
322 int last_exp
, n
, code
;
325 uint32_t *q
, *q_end
, iv
;
326 const float *ptab
= pow_tab
+ 60;
327 const uint32_t *iptab
= (const uint32_t *) ptab
;
329 ptr
= s
->exponent_bands
[s
->frame_len_bits
- s
->block_len_bits
];
330 q
= (uint32_t *) s
->exponents
[ch
];
331 q_end
= q
+ s
->block_len
;
333 if (s
->version
== 1) {
334 last_exp
= get_bits(&s
->gb
, 5) + 10;
336 iv
= iptab
[last_exp
];
344 } while ((n
-= 4) > 0);
349 code
= get_vlc2(&s
->gb
, s
->exp_vlc
.table
, EXPVLCBITS
, EXPMAX
);
351 av_log(s
->avctx
, AV_LOG_ERROR
, "Exponent vlc invalid\n");
354 /* NOTE: this offset is the same as MPEG4 AAC ! */
355 last_exp
+= code
- 60;
356 if ((unsigned) last_exp
+ 60 >= FF_ARRAY_ELEMS(pow_tab
)) {
357 av_log(s
->avctx
, AV_LOG_ERROR
, "Exponent out of range: %d\n",
362 iv
= iptab
[last_exp
];
371 } while ((n
-= 4) > 0);
373 s
->max_exponent
[ch
] = max_scale
;
378 * Apply MDCT window and add into output.
380 * We ensure that when the windows overlap their squared sum
381 * is always 1 (MDCT reconstruction rule).
383 static void wma_window(WMACodecContext
*s
, float *out
)
385 float *in
= s
->output
;
386 int block_len
, bsize
, n
;
389 if (s
->block_len_bits
<= s
->prev_block_len_bits
) {
390 block_len
= s
->block_len
;
391 bsize
= s
->frame_len_bits
- s
->block_len_bits
;
393 s
->fdsp
->vector_fmul_add(out
, in
, s
->windows
[bsize
],
396 block_len
= 1 << s
->prev_block_len_bits
;
397 n
= (s
->block_len
- block_len
) / 2;
398 bsize
= s
->frame_len_bits
- s
->prev_block_len_bits
;
400 s
->fdsp
->vector_fmul_add(out
+ n
, in
+ n
, s
->windows
[bsize
],
403 memcpy(out
+ n
+ block_len
, in
+ n
+ block_len
, n
* sizeof(float));
410 if (s
->block_len_bits
<= s
->next_block_len_bits
) {
411 block_len
= s
->block_len
;
412 bsize
= s
->frame_len_bits
- s
->block_len_bits
;
414 s
->fdsp
->vector_fmul_reverse(out
, in
, s
->windows
[bsize
], block_len
);
416 block_len
= 1 << s
->next_block_len_bits
;
417 n
= (s
->block_len
- block_len
) / 2;
418 bsize
= s
->frame_len_bits
- s
->next_block_len_bits
;
420 memcpy(out
, in
, n
* sizeof(float));
422 s
->fdsp
->vector_fmul_reverse(out
+ n
, in
+ n
, s
->windows
[bsize
],
425 memset(out
+ n
+ block_len
, 0, n
* sizeof(float));
430 * @return 0 if OK. 1 if last block of frame. return -1 if
431 * unrecorrable error.
433 static int wma_decode_block(WMACodecContext
*s
)
435 int n
, v
, a
, ch
, bsize
;
436 int coef_nb_bits
, total_gain
;
437 int nb_coefs
[MAX_CHANNELS
];
442 tprintf(s
->avctx
, "***decode_block: %d:%d\n",
443 s
->frame_count
- 1, s
->block_num
);
446 /* compute current block length */
447 if (s
->use_variable_block_len
) {
448 n
= av_log2(s
->nb_block_sizes
- 1) + 1;
450 if (s
->reset_block_lengths
) {
451 s
->reset_block_lengths
= 0;
452 v
= get_bits(&s
->gb
, n
);
453 if (v
>= s
->nb_block_sizes
) {
454 av_log(s
->avctx
, AV_LOG_ERROR
,
455 "prev_block_len_bits %d out of range\n",
456 s
->frame_len_bits
- v
);
459 s
->prev_block_len_bits
= s
->frame_len_bits
- v
;
460 v
= get_bits(&s
->gb
, n
);
461 if (v
>= s
->nb_block_sizes
) {
462 av_log(s
->avctx
, AV_LOG_ERROR
,
463 "block_len_bits %d out of range\n",
464 s
->frame_len_bits
- v
);
467 s
->block_len_bits
= s
->frame_len_bits
- v
;
469 /* update block lengths */
470 s
->prev_block_len_bits
= s
->block_len_bits
;
471 s
->block_len_bits
= s
->next_block_len_bits
;
473 v
= get_bits(&s
->gb
, n
);
474 if (v
>= s
->nb_block_sizes
) {
475 av_log(s
->avctx
, AV_LOG_ERROR
,
476 "next_block_len_bits %d out of range\n",
477 s
->frame_len_bits
- v
);
480 s
->next_block_len_bits
= s
->frame_len_bits
- v
;
482 /* fixed block len */
483 s
->next_block_len_bits
= s
->frame_len_bits
;
484 s
->prev_block_len_bits
= s
->frame_len_bits
;
485 s
->block_len_bits
= s
->frame_len_bits
;
488 if (s
->frame_len_bits
- s
->block_len_bits
>= s
->nb_block_sizes
){
489 av_log(s
->avctx
, AV_LOG_ERROR
, "block_len_bits not initialized to a valid value\n");
493 /* now check if the block length is coherent with the frame length */
494 s
->block_len
= 1 << s
->block_len_bits
;
495 if ((s
->block_pos
+ s
->block_len
) > s
->frame_len
) {
496 av_log(s
->avctx
, AV_LOG_ERROR
, "frame_len overflow\n");
500 if (s
->avctx
->channels
== 2)
501 s
->ms_stereo
= get_bits1(&s
->gb
);
503 for (ch
= 0; ch
< s
->avctx
->channels
; ch
++) {
504 a
= get_bits1(&s
->gb
);
505 s
->channel_coded
[ch
] = a
;
509 bsize
= s
->frame_len_bits
- s
->block_len_bits
;
511 /* if no channel coded, no need to go further */
512 /* XXX: fix potential framing problems */
516 /* read total gain and extract corresponding number of bits for
517 * coef escape coding */
520 if (get_bits_left(&s
->gb
) < 7) {
521 av_log(s
->avctx
, AV_LOG_ERROR
, "total_gain overread\n");
522 return AVERROR_INVALIDDATA
;
524 a
= get_bits(&s
->gb
, 7);
530 coef_nb_bits
= ff_wma_total_gain_to_bits(total_gain
);
532 /* compute number of coefficients */
533 n
= s
->coefs_end
[bsize
] - s
->coefs_start
;
534 for (ch
= 0; ch
< s
->avctx
->channels
; ch
++)
538 if (s
->use_noise_coding
) {
539 for (ch
= 0; ch
< s
->avctx
->channels
; ch
++) {
540 if (s
->channel_coded
[ch
]) {
542 n
= s
->exponent_high_sizes
[bsize
];
543 for (i
= 0; i
< n
; i
++) {
544 a
= get_bits1(&s
->gb
);
545 s
->high_band_coded
[ch
][i
] = a
;
546 /* if noise coding, the coefficients are not transmitted */
548 nb_coefs
[ch
] -= s
->exponent_high_bands
[bsize
][i
];
552 for (ch
= 0; ch
< s
->avctx
->channels
; ch
++) {
553 if (s
->channel_coded
[ch
]) {
556 n
= s
->exponent_high_sizes
[bsize
];
557 val
= (int) 0x80000000;
558 for (i
= 0; i
< n
; i
++) {
559 if (s
->high_band_coded
[ch
][i
]) {
560 if (val
== (int) 0x80000000) {
561 val
= get_bits(&s
->gb
, 7) - 19;
563 code
= get_vlc2(&s
->gb
, s
->hgain_vlc
.table
,
564 HGAINVLCBITS
, HGAINMAX
);
566 av_log(s
->avctx
, AV_LOG_ERROR
,
567 "hgain vlc invalid\n");
572 s
->high_band_values
[ch
][i
] = val
;
579 /* exponents can be reused in short blocks. */
580 if ((s
->block_len_bits
== s
->frame_len_bits
) || get_bits1(&s
->gb
)) {
581 for (ch
= 0; ch
< s
->avctx
->channels
; ch
++) {
582 if (s
->channel_coded
[ch
]) {
583 if (s
->use_exp_vlc
) {
584 if (decode_exp_vlc(s
, ch
) < 0)
587 decode_exp_lsp(s
, ch
);
589 s
->exponents_bsize
[ch
] = bsize
;
594 /* parse spectral coefficients : just RLE encoding */
595 for (ch
= 0; ch
< s
->avctx
->channels
; ch
++) {
596 if (s
->channel_coded
[ch
]) {
598 WMACoef
*ptr
= &s
->coefs1
[ch
][0];
600 /* special VLC tables are used for ms stereo because
601 * there is potentially less energy there */
602 tindex
= (ch
== 1 && s
->ms_stereo
);
603 memset(ptr
, 0, s
->block_len
* sizeof(WMACoef
));
604 ff_wma_run_level_decode(s
->avctx
, &s
->gb
, &s
->coef_vlc
[tindex
],
605 s
->level_table
[tindex
], s
->run_table
[tindex
],
606 0, ptr
, 0, nb_coefs
[ch
],
607 s
->block_len
, s
->frame_len_bits
, coef_nb_bits
);
609 if (s
->version
== 1 && s
->avctx
->channels
>= 2)
610 align_get_bits(&s
->gb
);
615 int n4
= s
->block_len
/ 2;
616 mdct_norm
= 1.0 / (float) n4
;
618 mdct_norm
*= sqrt(n4
);
621 /* finally compute the MDCT coefficients */
622 for (ch
= 0; ch
< s
->avctx
->channels
; ch
++) {
623 if (s
->channel_coded
[ch
]) {
625 float *coefs
, *exponents
, mult
, mult1
, noise
;
626 int i
, j
, n
, n1
, last_high_band
, esize
;
627 float exp_power
[HIGH_BAND_MAX_SIZE
];
629 coefs1
= s
->coefs1
[ch
];
630 exponents
= s
->exponents
[ch
];
631 esize
= s
->exponents_bsize
[ch
];
632 mult
= pow(10, total_gain
* 0.05) / s
->max_exponent
[ch
];
634 coefs
= s
->coefs
[ch
];
635 if (s
->use_noise_coding
) {
637 /* very low freqs : noise */
638 for (i
= 0; i
< s
->coefs_start
; i
++) {
639 *coefs
++ = s
->noise_table
[s
->noise_index
] *
640 exponents
[i
<< bsize
>> esize
] * mult1
;
641 s
->noise_index
= (s
->noise_index
+ 1) &
642 (NOISE_TAB_SIZE
- 1);
645 n1
= s
->exponent_high_sizes
[bsize
];
647 /* compute power of high bands */
648 exponents
= s
->exponents
[ch
] +
649 (s
->high_band_start
[bsize
] << bsize
>> esize
);
650 last_high_band
= 0; /* avoid warning */
651 for (j
= 0; j
< n1
; j
++) {
652 n
= s
->exponent_high_bands
[s
->frame_len_bits
-
653 s
->block_len_bits
][j
];
654 if (s
->high_band_coded
[ch
][j
]) {
657 for (i
= 0; i
< n
; i
++) {
658 v
= exponents
[i
<< bsize
>> esize
];
661 exp_power
[j
] = e2
/ n
;
663 tprintf(s
->avctx
, "%d: power=%f (%d)\n", j
, exp_power
[j
], n
);
665 exponents
+= n
<< bsize
>> esize
;
668 /* main freqs and high freqs */
669 exponents
= s
->exponents
[ch
] + (s
->coefs_start
<< bsize
>> esize
);
670 for (j
= -1; j
< n1
; j
++) {
672 n
= s
->high_band_start
[bsize
] - s
->coefs_start
;
674 n
= s
->exponent_high_bands
[s
->frame_len_bits
-
675 s
->block_len_bits
][j
];
676 if (j
>= 0 && s
->high_band_coded
[ch
][j
]) {
677 /* use noise with specified power */
678 mult1
= sqrt(exp_power
[j
] / exp_power
[last_high_band
]);
679 /* XXX: use a table */
680 mult1
= mult1
* pow(10, s
->high_band_values
[ch
][j
] * 0.05);
681 mult1
= mult1
/ (s
->max_exponent
[ch
] * s
->noise_mult
);
683 for (i
= 0; i
< n
; i
++) {
684 noise
= s
->noise_table
[s
->noise_index
];
685 s
->noise_index
= (s
->noise_index
+ 1) & (NOISE_TAB_SIZE
- 1);
686 *coefs
++ = noise
* exponents
[i
<< bsize
>> esize
] * mult1
;
688 exponents
+= n
<< bsize
>> esize
;
690 /* coded values + small noise */
691 for (i
= 0; i
< n
; i
++) {
692 noise
= s
->noise_table
[s
->noise_index
];
693 s
->noise_index
= (s
->noise_index
+ 1) & (NOISE_TAB_SIZE
- 1);
694 *coefs
++ = ((*coefs1
++) + noise
) *
695 exponents
[i
<< bsize
>> esize
] * mult
;
697 exponents
+= n
<< bsize
>> esize
;
701 /* very high freqs : noise */
702 n
= s
->block_len
- s
->coefs_end
[bsize
];
703 mult1
= mult
* exponents
[((-1 << bsize
)) >> esize
];
704 for (i
= 0; i
< n
; i
++) {
705 *coefs
++ = s
->noise_table
[s
->noise_index
] * mult1
;
706 s
->noise_index
= (s
->noise_index
+ 1) & (NOISE_TAB_SIZE
- 1);
709 /* XXX: optimize more */
710 for (i
= 0; i
< s
->coefs_start
; i
++)
713 for (i
= 0; i
< n
; i
++)
714 *coefs
++ = coefs1
[i
] * exponents
[i
<< bsize
>> esize
] * mult
;
715 n
= s
->block_len
- s
->coefs_end
[bsize
];
716 for (i
= 0; i
< n
; i
++)
723 for (ch
= 0; ch
< s
->avctx
->channels
; ch
++) {
724 if (s
->channel_coded
[ch
]) {
725 dump_floats(s
, "exponents", 3, s
->exponents
[ch
], s
->block_len
);
726 dump_floats(s
, "coefs", 1, s
->coefs
[ch
], s
->block_len
);
731 if (s
->ms_stereo
&& s
->channel_coded
[1]) {
732 /* nominal case for ms stereo: we do it before mdct */
733 /* no need to optimize this case because it should almost
735 if (!s
->channel_coded
[0]) {
736 tprintf(s
->avctx
, "rare ms-stereo case happened\n");
737 memset(s
->coefs
[0], 0, sizeof(float) * s
->block_len
);
738 s
->channel_coded
[0] = 1;
741 s
->fdsp
->butterflies_float(s
->coefs
[0], s
->coefs
[1], s
->block_len
);
745 mdct
= &s
->mdct_ctx
[bsize
];
747 for (ch
= 0; ch
< s
->avctx
->channels
; ch
++) {
750 n4
= s
->block_len
/ 2;
751 if (s
->channel_coded
[ch
])
752 mdct
->imdct_calc(mdct
, s
->output
, s
->coefs
[ch
]);
753 else if (!(s
->ms_stereo
&& ch
== 1))
754 memset(s
->output
, 0, sizeof(s
->output
));
756 /* multiply by the window and add in the frame */
757 index
= (s
->frame_len
/ 2) + s
->block_pos
- n4
;
758 wma_window(s
, &s
->frame_out
[ch
][index
]);
761 /* update block number */
763 s
->block_pos
+= s
->block_len
;
764 if (s
->block_pos
>= s
->frame_len
)
770 /* decode a frame of frame_len samples */
771 static int wma_decode_frame(WMACodecContext
*s
, float **samples
,
777 tprintf(s
->avctx
, "***decode_frame: %d size=%d\n",
778 s
->frame_count
++, s
->frame_len
);
781 /* read each block */
785 ret
= wma_decode_block(s
);
792 for (ch
= 0; ch
< s
->avctx
->channels
; ch
++) {
793 /* copy current block to output */
794 memcpy(samples
[ch
] + samples_offset
, s
->frame_out
[ch
],
795 s
->frame_len
* sizeof(*s
->frame_out
[ch
]));
796 /* prepare for next block */
797 memmove(&s
->frame_out
[ch
][0], &s
->frame_out
[ch
][s
->frame_len
],
798 s
->frame_len
* sizeof(*s
->frame_out
[ch
]));
801 dump_floats(s
, "samples", 6, samples
[ch
] + samples_offset
,
809 static int wma_decode_superframe(AVCodecContext
*avctx
, void *data
,
810 int *got_frame_ptr
, AVPacket
*avpkt
)
812 AVFrame
*frame
= data
;
813 const uint8_t *buf
= avpkt
->data
;
814 int buf_size
= avpkt
->size
;
815 WMACodecContext
*s
= avctx
->priv_data
;
816 int nb_frames
, bit_offset
, i
, pos
, len
, ret
;
821 tprintf(avctx
, "***decode_superframe:\n");
824 s
->last_superframe_len
= 0;
827 if (buf_size
< avctx
->block_align
) {
828 av_log(avctx
, AV_LOG_ERROR
,
829 "Input packet size too small (%d < %d)\n",
830 buf_size
, avctx
->block_align
);
831 return AVERROR_INVALIDDATA
;
833 if (avctx
->block_align
)
834 buf_size
= avctx
->block_align
;
836 init_get_bits(&s
->gb
, buf
, buf_size
* 8);
838 if (s
->use_bit_reservoir
) {
839 /* read super frame header */
840 skip_bits(&s
->gb
, 4); /* super frame index */
841 nb_frames
= get_bits(&s
->gb
, 4) - (s
->last_superframe_len
<= 0);
842 if (nb_frames
<= 0) {
843 av_log(avctx
, AV_LOG_ERROR
, "nb_frames is %d\n", nb_frames
);
844 return AVERROR_INVALIDDATA
;
849 /* get output buffer */
850 frame
->nb_samples
= nb_frames
* s
->frame_len
;
851 if ((ret
= ff_get_buffer(avctx
, frame
, 0)) < 0)
853 samples
= (float **) frame
->extended_data
;
856 if (s
->use_bit_reservoir
) {
857 bit_offset
= get_bits(&s
->gb
, s
->byte_offset_bits
+ 3);
858 if (bit_offset
> get_bits_left(&s
->gb
)) {
859 av_log(avctx
, AV_LOG_ERROR
,
860 "Invalid last frame bit offset %d > buf size %d (%d)\n",
861 bit_offset
, get_bits_left(&s
->gb
), buf_size
);
865 if (s
->last_superframe_len
> 0) {
866 /* add bit_offset bits to last frame */
867 if ((s
->last_superframe_len
+ ((bit_offset
+ 7) >> 3)) >
868 MAX_CODED_SUPERFRAME_SIZE
)
870 q
= s
->last_superframe
+ s
->last_superframe_len
;
873 *q
++ = (get_bits
) (&s
->gb
, 8);
877 *q
++ = (get_bits
) (&s
->gb
, len
) << (8 - len
);
878 memset(q
, 0, FF_INPUT_BUFFER_PADDING_SIZE
);
880 /* XXX: bit_offset bits into last frame */
881 init_get_bits(&s
->gb
, s
->last_superframe
,
882 s
->last_superframe_len
* 8 + bit_offset
);
883 /* skip unused bits */
884 if (s
->last_bitoffset
> 0)
885 skip_bits(&s
->gb
, s
->last_bitoffset
);
886 /* this frame is stored in the last superframe and in the
888 if (wma_decode_frame(s
, samples
, samples_offset
) < 0)
890 samples_offset
+= s
->frame_len
;
894 /* read each frame starting from bit_offset */
895 pos
= bit_offset
+ 4 + 4 + s
->byte_offset_bits
+ 3;
896 if (pos
>= MAX_CODED_SUPERFRAME_SIZE
* 8 || pos
> buf_size
* 8)
897 return AVERROR_INVALIDDATA
;
898 init_get_bits(&s
->gb
, buf
+ (pos
>> 3), (buf_size
- (pos
>> 3)) * 8);
901 skip_bits(&s
->gb
, len
);
903 s
->reset_block_lengths
= 1;
904 for (i
= 0; i
< nb_frames
; i
++) {
905 if (wma_decode_frame(s
, samples
, samples_offset
) < 0)
907 samples_offset
+= s
->frame_len
;
910 /* we copy the end of the frame in the last frame buffer */
911 pos
= get_bits_count(&s
->gb
) +
912 ((bit_offset
+ 4 + 4 + s
->byte_offset_bits
+ 3) & ~7);
913 s
->last_bitoffset
= pos
& 7;
915 len
= buf_size
- pos
;
916 if (len
> MAX_CODED_SUPERFRAME_SIZE
|| len
< 0) {
917 av_log(s
->avctx
, AV_LOG_ERROR
, "len %d invalid\n", len
);
920 s
->last_superframe_len
= len
;
921 memcpy(s
->last_superframe
, buf
+ pos
, len
);
923 /* single frame decode */
924 if (wma_decode_frame(s
, samples
, samples_offset
) < 0)
926 samples_offset
+= s
->frame_len
;
929 av_dlog(s
->avctx
, "%d %d %d %d outbytes:%"PTRDIFF_SPECIFIER
" eaten:%d\n",
930 s
->frame_len_bits
, s
->block_len_bits
, s
->frame_len
, s
->block_len
,
931 (int8_t *) samples
- (int8_t *) data
, avctx
->block_align
);
938 /* when error, we reset the bit reservoir */
939 s
->last_superframe_len
= 0;
943 static av_cold
void flush(AVCodecContext
*avctx
)
945 WMACodecContext
*s
= avctx
->priv_data
;
948 s
->last_superframe_len
= 0;
951 #if CONFIG_WMAV1_DECODER
952 AVCodec ff_wmav1_decoder
= {
954 .long_name
= NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
955 .type
= AVMEDIA_TYPE_AUDIO
,
956 .id
= AV_CODEC_ID_WMAV1
,
957 .priv_data_size
= sizeof(WMACodecContext
),
958 .init
= wma_decode_init
,
960 .decode
= wma_decode_superframe
,
962 .capabilities
= CODEC_CAP_DR1
,
963 .sample_fmts
= (const enum AVSampleFormat
[]) { AV_SAMPLE_FMT_FLTP
,
964 AV_SAMPLE_FMT_NONE
},
967 #if CONFIG_WMAV2_DECODER
968 AVCodec ff_wmav2_decoder
= {
970 .long_name
= NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),
971 .type
= AVMEDIA_TYPE_AUDIO
,
972 .id
= AV_CODEC_ID_WMAV2
,
973 .priv_data_size
= sizeof(WMACodecContext
),
974 .init
= wma_decode_init
,
976 .decode
= wma_decode_superframe
,
978 .capabilities
= CODEC_CAP_DR1
,
979 .sample_fmts
= (const enum AVSampleFormat
[]) { AV_SAMPLE_FMT_FLTP
,
980 AV_SAMPLE_FMT_NONE
},