summaryrefslogtreecommitdiff
path: root/libavcodec/wmadec.c (plain)
blob: 78b51e5871bf8c685f9eb7524025cdc738014a4e
1/*
2 * WMA compatible decoder
3 * Copyright (c) 2002 The FFmpeg Project
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22/**
23 * @file
24 * 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.
28 *
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.
34 */
35
36#include "libavutil/attributes.h"
37#include "libavutil/ffmath.h"
38
39#include "avcodec.h"
40#include "internal.h"
41#include "wma.h"
42
43#define EXPVLCBITS 8
44#define EXPMAX ((19 + EXPVLCBITS - 1) / EXPVLCBITS)
45
46#define HGAINVLCBITS 9
47#define HGAINMAX ((13 + HGAINVLCBITS - 1) / HGAINVLCBITS)
48
49static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len);
50
51#ifdef TRACE
52static void dump_floats(WMACodecContext *s, const char *name,
53 int prec, const float *tab, int n)
54{
55 int i;
56
57 ff_tlog(s->avctx, "%s[%d]:\n", name, n);
58 for (i = 0; i < n; i++) {
59 if ((i & 7) == 0)
60 ff_tlog(s->avctx, "%4d: ", i);
61 ff_tlog(s->avctx, " %8.*f", prec, tab[i]);
62 if ((i & 7) == 7)
63 ff_tlog(s->avctx, "\n");
64 }
65 if ((i & 7) != 0)
66 ff_tlog(s->avctx, "\n");
67}
68#endif /* TRACE */
69
70static av_cold int wma_decode_init(AVCodecContext *avctx)
71{
72 WMACodecContext *s = avctx->priv_data;
73 int i, flags2;
74 uint8_t *extradata;
75
76 if (!avctx->block_align) {
77 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
78 return AVERROR(EINVAL);
79 }
80
81 s->avctx = avctx;
82
83 /* extract flag info */
84 flags2 = 0;
85 extradata = avctx->extradata;
86 if (avctx->codec->id == AV_CODEC_ID_WMAV1 && avctx->extradata_size >= 4)
87 flags2 = AV_RL16(extradata + 2);
88 else if (avctx->codec->id == AV_CODEC_ID_WMAV2 && avctx->extradata_size >= 6)
89 flags2 = AV_RL16(extradata + 4);
90
91 s->use_exp_vlc = flags2 & 0x0001;
92 s->use_bit_reservoir = flags2 & 0x0002;
93 s->use_variable_block_len = flags2 & 0x0004;
94
95 if (avctx->codec->id == AV_CODEC_ID_WMAV2 && avctx->extradata_size >= 8){
96 if (AV_RL16(extradata+4)==0xd && s->use_variable_block_len){
97 av_log(avctx, AV_LOG_WARNING, "Disabling use_variable_block_len, if this fails contact the ffmpeg developers and send us the file\n");
98 s->use_variable_block_len= 0; // this fixes issue1503
99 }
100 }
101
102 for (i=0; i<MAX_CHANNELS; i++)
103 s->max_exponent[i] = 1.0;
104
105 if (ff_wma_init(avctx, flags2) < 0)
106 return -1;
107
108 /* init MDCT */
109 for (i = 0; i < s->nb_block_sizes; i++)
110 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1, 1.0 / 32768.0);
111
112 if (s->use_noise_coding) {
113 init_vlc(&s->hgain_vlc, HGAINVLCBITS, sizeof(ff_wma_hgain_huffbits),
114 ff_wma_hgain_huffbits, 1, 1,
115 ff_wma_hgain_huffcodes, 2, 2, 0);
116 }
117
118 if (s->use_exp_vlc)
119 init_vlc(&s->exp_vlc, EXPVLCBITS, sizeof(ff_aac_scalefactor_bits), // FIXME move out of context
120 ff_aac_scalefactor_bits, 1, 1,
121 ff_aac_scalefactor_code, 4, 4, 0);
122 else
123 wma_lsp_to_curve_init(s, s->frame_len);
124
125 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
126
127 return 0;
128}
129
130/**
131 * compute x^-0.25 with an exponent and mantissa table. We use linear
132 * interpolation to reduce the mantissa table size at a small speed
133 * expense (linear interpolation approximately doubles the number of
134 * bits of precision).
135 */
136static inline float pow_m1_4(WMACodecContext *s, float x)
137{
138 union {
139 float f;
140 unsigned int v;
141 } u, t;
142 unsigned int e, m;
143 float a, b;
144
145 u.f = x;
146 e = u.v >> 23;
147 m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1);
148 /* build interpolation scale: 1 <= t < 2. */
149 t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23);
150 a = s->lsp_pow_m_table1[m];
151 b = s->lsp_pow_m_table2[m];
152 return s->lsp_pow_e_table[e] * (a + b * t.f);
153}
154
155static av_cold void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len)
156{
157 float wdel, a, b;
158 int i, e, m;
159
160 wdel = M_PI / frame_len;
161 for (i = 0; i < frame_len; i++)
162 s->lsp_cos_table[i] = 2.0f * cos(wdel * i);
163
164 /* tables for x^-0.25 computation */
165 for (i = 0; i < 256; i++) {
166 e = i - 126;
167 s->lsp_pow_e_table[i] = exp2f(e * -0.25);
168 }
169
170 /* NOTE: these two tables are needed to avoid two operations in
171 * pow_m1_4 */
172 b = 1.0;
173 for (i = (1 << LSP_POW_BITS) - 1; i >= 0; i--) {
174 m = (1 << LSP_POW_BITS) + i;
175 a = (float) m * (0.5 / (1 << LSP_POW_BITS));
176 a = 1/sqrt(sqrt(a));
177 s->lsp_pow_m_table1[i] = 2 * a - b;
178 s->lsp_pow_m_table2[i] = b - a;
179 b = a;
180 }
181}
182
183/**
184 * NOTE: We use the same code as Vorbis here
185 * @todo optimize it further with SSE/3Dnow
186 */
187static void wma_lsp_to_curve(WMACodecContext *s, float *out, float *val_max_ptr,
188 int n, float *lsp)
189{
190 int i, j;
191 float p, q, w, v, val_max;
192
193 val_max = 0;
194 for (i = 0; i < n; i++) {
195 p = 0.5f;
196 q = 0.5f;
197 w = s->lsp_cos_table[i];
198 for (j = 1; j < NB_LSP_COEFS; j += 2) {
199 q *= w - lsp[j - 1];
200 p *= w - lsp[j];
201 }
202 p *= p * (2.0f - w);
203 q *= q * (2.0f + w);
204 v = p + q;
205 v = pow_m1_4(s, v);
206 if (v > val_max)
207 val_max = v;
208 out[i] = v;
209 }
210 *val_max_ptr = val_max;
211}
212
213/**
214 * decode exponents coded with LSP coefficients (same idea as Vorbis)
215 */
216static void decode_exp_lsp(WMACodecContext *s, int ch)
217{
218 float lsp_coefs[NB_LSP_COEFS];
219 int val, i;
220
221 for (i = 0; i < NB_LSP_COEFS; i++) {
222 if (i == 0 || i >= 8)
223 val = get_bits(&s->gb, 3);
224 else
225 val = get_bits(&s->gb, 4);
226 lsp_coefs[i] = ff_wma_lsp_codebook[i][val];
227 }
228
229 wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch],
230 s->block_len, lsp_coefs);
231}
232
233/** pow(10, i / 16.0) for i in -60..95 */
234static const float pow_tab[] = {
235 1.7782794100389e-04, 2.0535250264571e-04,
236 2.3713737056617e-04, 2.7384196342644e-04,
237 3.1622776601684e-04, 3.6517412725484e-04,
238 4.2169650342858e-04, 4.8696752516586e-04,
239 5.6234132519035e-04, 6.4938163157621e-04,
240 7.4989420933246e-04, 8.6596432336006e-04,
241 1.0000000000000e-03, 1.1547819846895e-03,
242 1.3335214321633e-03, 1.5399265260595e-03,
243 1.7782794100389e-03, 2.0535250264571e-03,
244 2.3713737056617e-03, 2.7384196342644e-03,
245 3.1622776601684e-03, 3.6517412725484e-03,
246 4.2169650342858e-03, 4.8696752516586e-03,
247 5.6234132519035e-03, 6.4938163157621e-03,
248 7.4989420933246e-03, 8.6596432336006e-03,
249 1.0000000000000e-02, 1.1547819846895e-02,
250 1.3335214321633e-02, 1.5399265260595e-02,
251 1.7782794100389e-02, 2.0535250264571e-02,
252 2.3713737056617e-02, 2.7384196342644e-02,
253 3.1622776601684e-02, 3.6517412725484e-02,
254 4.2169650342858e-02, 4.8696752516586e-02,
255 5.6234132519035e-02, 6.4938163157621e-02,
256 7.4989420933246e-02, 8.6596432336007e-02,
257 1.0000000000000e-01, 1.1547819846895e-01,
258 1.3335214321633e-01, 1.5399265260595e-01,
259 1.7782794100389e-01, 2.0535250264571e-01,
260 2.3713737056617e-01, 2.7384196342644e-01,
261 3.1622776601684e-01, 3.6517412725484e-01,
262 4.2169650342858e-01, 4.8696752516586e-01,
263 5.6234132519035e-01, 6.4938163157621e-01,
264 7.4989420933246e-01, 8.6596432336007e-01,
265 1.0000000000000e+00, 1.1547819846895e+00,
266 1.3335214321633e+00, 1.5399265260595e+00,
267 1.7782794100389e+00, 2.0535250264571e+00,
268 2.3713737056617e+00, 2.7384196342644e+00,
269 3.1622776601684e+00, 3.6517412725484e+00,
270 4.2169650342858e+00, 4.8696752516586e+00,
271 5.6234132519035e+00, 6.4938163157621e+00,
272 7.4989420933246e+00, 8.6596432336007e+00,
273 1.0000000000000e+01, 1.1547819846895e+01,
274 1.3335214321633e+01, 1.5399265260595e+01,
275 1.7782794100389e+01, 2.0535250264571e+01,
276 2.3713737056617e+01, 2.7384196342644e+01,
277 3.1622776601684e+01, 3.6517412725484e+01,
278 4.2169650342858e+01, 4.8696752516586e+01,
279 5.6234132519035e+01, 6.4938163157621e+01,
280 7.4989420933246e+01, 8.6596432336007e+01,
281 1.0000000000000e+02, 1.1547819846895e+02,
282 1.3335214321633e+02, 1.5399265260595e+02,
283 1.7782794100389e+02, 2.0535250264571e+02,
284 2.3713737056617e+02, 2.7384196342644e+02,
285 3.1622776601684e+02, 3.6517412725484e+02,
286 4.2169650342858e+02, 4.8696752516586e+02,
287 5.6234132519035e+02, 6.4938163157621e+02,
288 7.4989420933246e+02, 8.6596432336007e+02,
289 1.0000000000000e+03, 1.1547819846895e+03,
290 1.3335214321633e+03, 1.5399265260595e+03,
291 1.7782794100389e+03, 2.0535250264571e+03,
292 2.3713737056617e+03, 2.7384196342644e+03,
293 3.1622776601684e+03, 3.6517412725484e+03,
294 4.2169650342858e+03, 4.8696752516586e+03,
295 5.6234132519035e+03, 6.4938163157621e+03,
296 7.4989420933246e+03, 8.6596432336007e+03,
297 1.0000000000000e+04, 1.1547819846895e+04,
298 1.3335214321633e+04, 1.5399265260595e+04,
299 1.7782794100389e+04, 2.0535250264571e+04,
300 2.3713737056617e+04, 2.7384196342644e+04,
301 3.1622776601684e+04, 3.6517412725484e+04,
302 4.2169650342858e+04, 4.8696752516586e+04,
303 5.6234132519035e+04, 6.4938163157621e+04,
304 7.4989420933246e+04, 8.6596432336007e+04,
305 1.0000000000000e+05, 1.1547819846895e+05,
306 1.3335214321633e+05, 1.5399265260595e+05,
307 1.7782794100389e+05, 2.0535250264571e+05,
308 2.3713737056617e+05, 2.7384196342644e+05,
309 3.1622776601684e+05, 3.6517412725484e+05,
310 4.2169650342858e+05, 4.8696752516586e+05,
311 5.6234132519035e+05, 6.4938163157621e+05,
312 7.4989420933246e+05, 8.6596432336007e+05,
313};
314
315/**
316 * decode exponents coded with VLC codes
317 */
318static int decode_exp_vlc(WMACodecContext *s, int ch)
319{
320 int last_exp, n, code;
321 const uint16_t *ptr;
322 float v, max_scale;
323 uint32_t *q, *q_end, iv;
324 const float *ptab = pow_tab + 60;
325 const uint32_t *iptab = (const uint32_t *) ptab;
326
327 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
328 q = (uint32_t *) s->exponents[ch];
329 q_end = q + s->block_len;
330 max_scale = 0;
331 if (s->version == 1) {
332 last_exp = get_bits(&s->gb, 5) + 10;
333 v = ptab[last_exp];
334 iv = iptab[last_exp];
335 max_scale = v;
336 n = *ptr++;
337 switch (n & 3) do {
338 case 0: *q++ = iv;
339 case 3: *q++ = iv;
340 case 2: *q++ = iv;
341 case 1: *q++ = iv;
342 } while ((n -= 4) > 0);
343 } else
344 last_exp = 36;
345
346 while (q < q_end) {
347 code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX);
348 if (code < 0) {
349 av_log(s->avctx, AV_LOG_ERROR, "Exponent vlc invalid\n");
350 return -1;
351 }
352 /* NOTE: this offset is the same as MPEG-4 AAC! */
353 last_exp += code - 60;
354 if ((unsigned) last_exp + 60 >= FF_ARRAY_ELEMS(pow_tab)) {
355 av_log(s->avctx, AV_LOG_ERROR, "Exponent out of range: %d\n",
356 last_exp);
357 return -1;
358 }
359 v = ptab[last_exp];
360 iv = iptab[last_exp];
361 if (v > max_scale)
362 max_scale = v;
363 n = *ptr++;
364 switch (n & 3) do {
365 case 0: *q++ = iv;
366 case 3: *q++ = iv;
367 case 2: *q++ = iv;
368 case 1: *q++ = iv;
369 } while ((n -= 4) > 0);
370 }
371 s->max_exponent[ch] = max_scale;
372 return 0;
373}
374
375/**
376 * Apply MDCT window and add into output.
377 *
378 * We ensure that when the windows overlap their squared sum
379 * is always 1 (MDCT reconstruction rule).
380 */
381static void wma_window(WMACodecContext *s, float *out)
382{
383 float *in = s->output;
384 int block_len, bsize, n;
385
386 /* left part */
387 if (s->block_len_bits <= s->prev_block_len_bits) {
388 block_len = s->block_len;
389 bsize = s->frame_len_bits - s->block_len_bits;
390
391 s->fdsp->vector_fmul_add(out, in, s->windows[bsize],
392 out, block_len);
393 } else {
394 block_len = 1 << s->prev_block_len_bits;
395 n = (s->block_len - block_len) / 2;
396 bsize = s->frame_len_bits - s->prev_block_len_bits;
397
398 s->fdsp->vector_fmul_add(out + n, in + n, s->windows[bsize],
399 out + n, block_len);
400
401 memcpy(out + n + block_len, in + n + block_len, n * sizeof(float));
402 }
403
404 out += s->block_len;
405 in += s->block_len;
406
407 /* right part */
408 if (s->block_len_bits <= s->next_block_len_bits) {
409 block_len = s->block_len;
410 bsize = s->frame_len_bits - s->block_len_bits;
411
412 s->fdsp->vector_fmul_reverse(out, in, s->windows[bsize], block_len);
413 } else {
414 block_len = 1 << s->next_block_len_bits;
415 n = (s->block_len - block_len) / 2;
416 bsize = s->frame_len_bits - s->next_block_len_bits;
417
418 memcpy(out, in, n * sizeof(float));
419
420 s->fdsp->vector_fmul_reverse(out + n, in + n, s->windows[bsize],
421 block_len);
422
423 memset(out + n + block_len, 0, n * sizeof(float));
424 }
425}
426
427/**
428 * @return 0 if OK. 1 if last block of frame. return -1 if
429 * unrecoverable error.
430 */
431static int wma_decode_block(WMACodecContext *s)
432{
433 int n, v, a, ch, bsize;
434 int coef_nb_bits, total_gain;
435 int nb_coefs[MAX_CHANNELS];
436 float mdct_norm;
437 FFTContext *mdct;
438
439#ifdef TRACE
440 ff_tlog(s->avctx, "***decode_block: %d:%d\n",
441 s->frame_count - 1, s->block_num);
442#endif /* TRACE */
443
444 /* compute current block length */
445 if (s->use_variable_block_len) {
446 n = av_log2(s->nb_block_sizes - 1) + 1;
447
448 if (s->reset_block_lengths) {
449 s->reset_block_lengths = 0;
450 v = get_bits(&s->gb, n);
451 if (v >= s->nb_block_sizes) {
452 av_log(s->avctx, AV_LOG_ERROR,
453 "prev_block_len_bits %d out of range\n",
454 s->frame_len_bits - v);
455 return -1;
456 }
457 s->prev_block_len_bits = s->frame_len_bits - v;
458 v = get_bits(&s->gb, n);
459 if (v >= s->nb_block_sizes) {
460 av_log(s->avctx, AV_LOG_ERROR,
461 "block_len_bits %d out of range\n",
462 s->frame_len_bits - v);
463 return -1;
464 }
465 s->block_len_bits = s->frame_len_bits - v;
466 } else {
467 /* update block lengths */
468 s->prev_block_len_bits = s->block_len_bits;
469 s->block_len_bits = s->next_block_len_bits;
470 }
471 v = get_bits(&s->gb, n);
472 if (v >= s->nb_block_sizes) {
473 av_log(s->avctx, AV_LOG_ERROR,
474 "next_block_len_bits %d out of range\n",
475 s->frame_len_bits - v);
476 return -1;
477 }
478 s->next_block_len_bits = s->frame_len_bits - v;
479 } else {
480 /* fixed block len */
481 s->next_block_len_bits = s->frame_len_bits;
482 s->prev_block_len_bits = s->frame_len_bits;
483 s->block_len_bits = s->frame_len_bits;
484 }
485
486 if (s->frame_len_bits - s->block_len_bits >= s->nb_block_sizes){
487 av_log(s->avctx, AV_LOG_ERROR, "block_len_bits not initialized to a valid value\n");
488 return -1;
489 }
490
491 /* now check if the block length is coherent with the frame length */
492 s->block_len = 1 << s->block_len_bits;
493 if ((s->block_pos + s->block_len) > s->frame_len) {
494 av_log(s->avctx, AV_LOG_ERROR, "frame_len overflow\n");
495 return -1;
496 }
497
498 if (s->avctx->channels == 2)
499 s->ms_stereo = get_bits1(&s->gb);
500 v = 0;
501 for (ch = 0; ch < s->avctx->channels; ch++) {
502 a = get_bits1(&s->gb);
503 s->channel_coded[ch] = a;
504 v |= a;
505 }
506
507 bsize = s->frame_len_bits - s->block_len_bits;
508
509 /* if no channel coded, no need to go further */
510 /* XXX: fix potential framing problems */
511 if (!v)
512 goto next;
513
514 /* read total gain and extract corresponding number of bits for
515 * coef escape coding */
516 total_gain = 1;
517 for (;;) {
518 if (get_bits_left(&s->gb) < 7) {
519 av_log(s->avctx, AV_LOG_ERROR, "total_gain overread\n");
520 return AVERROR_INVALIDDATA;
521 }
522 a = get_bits(&s->gb, 7);
523 total_gain += a;
524 if (a != 127)
525 break;
526 }
527
528 coef_nb_bits = ff_wma_total_gain_to_bits(total_gain);
529
530 /* compute number of coefficients */
531 n = s->coefs_end[bsize] - s->coefs_start;
532 for (ch = 0; ch < s->avctx->channels; ch++)
533 nb_coefs[ch] = n;
534
535 /* complex coding */
536 if (s->use_noise_coding) {
537 for (ch = 0; ch < s->avctx->channels; ch++) {
538 if (s->channel_coded[ch]) {
539 int i, n, a;
540 n = s->exponent_high_sizes[bsize];
541 for (i = 0; i < n; i++) {
542 a = get_bits1(&s->gb);
543 s->high_band_coded[ch][i] = a;
544 /* if noise coding, the coefficients are not transmitted */
545 if (a)
546 nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
547 }
548 }
549 }
550 for (ch = 0; ch < s->avctx->channels; ch++) {
551 if (s->channel_coded[ch]) {
552 int i, n, val, code;
553
554 n = s->exponent_high_sizes[bsize];
555 val = (int) 0x80000000;
556 for (i = 0; i < n; i++) {
557 if (s->high_band_coded[ch][i]) {
558 if (val == (int) 0x80000000) {
559 val = get_bits(&s->gb, 7) - 19;
560 } else {
561 code = get_vlc2(&s->gb, s->hgain_vlc.table,
562 HGAINVLCBITS, HGAINMAX);
563 if (code < 0) {
564 av_log(s->avctx, AV_LOG_ERROR,
565 "hgain vlc invalid\n");
566 return -1;
567 }
568 val += code - 18;
569 }
570 s->high_band_values[ch][i] = val;
571 }
572 }
573 }
574 }
575 }
576
577 /* exponents can be reused in short blocks. */
578 if ((s->block_len_bits == s->frame_len_bits) || get_bits1(&s->gb)) {
579 for (ch = 0; ch < s->avctx->channels; ch++) {
580 if (s->channel_coded[ch]) {
581 if (s->use_exp_vlc) {
582 if (decode_exp_vlc(s, ch) < 0)
583 return -1;
584 } else {
585 decode_exp_lsp(s, ch);
586 }
587 s->exponents_bsize[ch] = bsize;
588 }
589 }
590 }
591
592 /* parse spectral coefficients : just RLE encoding */
593 for (ch = 0; ch < s->avctx->channels; ch++) {
594 if (s->channel_coded[ch]) {
595 int tindex;
596 WMACoef *ptr = &s->coefs1[ch][0];
597
598 /* special VLC tables are used for ms stereo because
599 * there is potentially less energy there */
600 tindex = (ch == 1 && s->ms_stereo);
601 memset(ptr, 0, s->block_len * sizeof(WMACoef));
602 ff_wma_run_level_decode(s->avctx, &s->gb, &s->coef_vlc[tindex],
603 s->level_table[tindex], s->run_table[tindex],
604 0, ptr, 0, nb_coefs[ch],
605 s->block_len, s->frame_len_bits, coef_nb_bits);
606 }
607 if (s->version == 1 && s->avctx->channels >= 2)
608 align_get_bits(&s->gb);
609 }
610
611 /* normalize */
612 {
613 int n4 = s->block_len / 2;
614 mdct_norm = 1.0 / (float) n4;
615 if (s->version == 1)
616 mdct_norm *= sqrt(n4);
617 }
618
619 /* finally compute the MDCT coefficients */
620 for (ch = 0; ch < s->avctx->channels; ch++) {
621 if (s->channel_coded[ch]) {
622 WMACoef *coefs1;
623 float *coefs, *exponents, mult, mult1, noise;
624 int i, j, n, n1, last_high_band, esize;
625 float exp_power[HIGH_BAND_MAX_SIZE];
626
627 coefs1 = s->coefs1[ch];
628 exponents = s->exponents[ch];
629 esize = s->exponents_bsize[ch];
630 mult = ff_exp10(total_gain * 0.05) / s->max_exponent[ch];
631 mult *= mdct_norm;
632 coefs = s->coefs[ch];
633 if (s->use_noise_coding) {
634 mult1 = mult;
635 /* very low freqs : noise */
636 for (i = 0; i < s->coefs_start; i++) {
637 *coefs++ = s->noise_table[s->noise_index] *
638 exponents[i << bsize >> esize] * mult1;
639 s->noise_index = (s->noise_index + 1) &
640 (NOISE_TAB_SIZE - 1);
641 }
642
643 n1 = s->exponent_high_sizes[bsize];
644
645 /* compute power of high bands */
646 exponents = s->exponents[ch] +
647 (s->high_band_start[bsize] << bsize >> esize);
648 last_high_band = 0; /* avoid warning */
649 for (j = 0; j < n1; j++) {
650 n = s->exponent_high_bands[s->frame_len_bits -
651 s->block_len_bits][j];
652 if (s->high_band_coded[ch][j]) {
653 float e2, v;
654 e2 = 0;
655 for (i = 0; i < n; i++) {
656 v = exponents[i << bsize >> esize];
657 e2 += v * v;
658 }
659 exp_power[j] = e2 / n;
660 last_high_band = j;
661 ff_tlog(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n);
662 }
663 exponents += n << bsize >> esize;
664 }
665
666 /* main freqs and high freqs */
667 exponents = s->exponents[ch] + (s->coefs_start << bsize >> esize);
668 for (j = -1; j < n1; j++) {
669 if (j < 0)
670 n = s->high_band_start[bsize] - s->coefs_start;
671 else
672 n = s->exponent_high_bands[s->frame_len_bits -
673 s->block_len_bits][j];
674 if (j >= 0 && s->high_band_coded[ch][j]) {
675 /* use noise with specified power */
676 mult1 = sqrt(exp_power[j] / exp_power[last_high_band]);
677 /* XXX: use a table */
678 mult1 = mult1 * ff_exp10(s->high_band_values[ch][j] * 0.05);
679 mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult);
680 mult1 *= mdct_norm;
681 for (i = 0; i < n; i++) {
682 noise = s->noise_table[s->noise_index];
683 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
684 *coefs++ = noise * exponents[i << bsize >> esize] * mult1;
685 }
686 exponents += n << bsize >> esize;
687 } else {
688 /* coded values + small noise */
689 for (i = 0; i < n; i++) {
690 noise = s->noise_table[s->noise_index];
691 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
692 *coefs++ = ((*coefs1++) + noise) *
693 exponents[i << bsize >> esize] * mult;
694 }
695 exponents += n << bsize >> esize;
696 }
697 }
698
699 /* very high freqs : noise */
700 n = s->block_len - s->coefs_end[bsize];
701 mult1 = mult * exponents[(-(1 << bsize)) >> esize];
702 for (i = 0; i < n; i++) {
703 *coefs++ = s->noise_table[s->noise_index] * mult1;
704 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
705 }
706 } else {
707 /* XXX: optimize more */
708 for (i = 0; i < s->coefs_start; i++)
709 *coefs++ = 0.0;
710 n = nb_coefs[ch];
711 for (i = 0; i < n; i++)
712 *coefs++ = coefs1[i] * exponents[i << bsize >> esize] * mult;
713 n = s->block_len - s->coefs_end[bsize];
714 for (i = 0; i < n; i++)
715 *coefs++ = 0.0;
716 }
717 }
718 }
719
720#ifdef TRACE
721 for (ch = 0; ch < s->avctx->channels; ch++) {
722 if (s->channel_coded[ch]) {
723 dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len);
724 dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len);
725 }
726 }
727#endif /* TRACE */
728
729 if (s->ms_stereo && s->channel_coded[1]) {
730 /* nominal case for ms stereo: we do it before mdct */
731 /* no need to optimize this case because it should almost
732 * never happen */
733 if (!s->channel_coded[0]) {
734 ff_tlog(s->avctx, "rare ms-stereo case happened\n");
735 memset(s->coefs[0], 0, sizeof(float) * s->block_len);
736 s->channel_coded[0] = 1;
737 }
738
739 s->fdsp->butterflies_float(s->coefs[0], s->coefs[1], s->block_len);
740 }
741
742next:
743 mdct = &s->mdct_ctx[bsize];
744
745 for (ch = 0; ch < s->avctx->channels; ch++) {
746 int n4, index;
747
748 n4 = s->block_len / 2;
749 if (s->channel_coded[ch])
750 mdct->imdct_calc(mdct, s->output, s->coefs[ch]);
751 else if (!(s->ms_stereo && ch == 1))
752 memset(s->output, 0, sizeof(s->output));
753
754 /* multiply by the window and add in the frame */
755 index = (s->frame_len / 2) + s->block_pos - n4;
756 wma_window(s, &s->frame_out[ch][index]);
757 }
758
759 /* update block number */
760 s->block_num++;
761 s->block_pos += s->block_len;
762 if (s->block_pos >= s->frame_len)
763 return 1;
764 else
765 return 0;
766}
767
768/* decode a frame of frame_len samples */
769static int wma_decode_frame(WMACodecContext *s, float **samples,
770 int samples_offset)
771{
772 int ret, ch;
773
774#ifdef TRACE
775 ff_tlog(s->avctx, "***decode_frame: %d size=%d\n",
776 s->frame_count++, s->frame_len);
777#endif /* TRACE */
778
779 /* read each block */
780 s->block_num = 0;
781 s->block_pos = 0;
782 for (;;) {
783 ret = wma_decode_block(s);
784 if (ret < 0)
785 return -1;
786 if (ret)
787 break;
788 }
789
790 for (ch = 0; ch < s->avctx->channels; ch++) {
791 /* copy current block to output */
792 memcpy(samples[ch] + samples_offset, s->frame_out[ch],
793 s->frame_len * sizeof(*s->frame_out[ch]));
794 /* prepare for next block */
795 memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len],
796 s->frame_len * sizeof(*s->frame_out[ch]));
797
798#ifdef TRACE
799 dump_floats(s, "samples", 6, samples[ch] + samples_offset,
800 s->frame_len);
801#endif /* TRACE */
802 }
803
804 return 0;
805}
806
807static int wma_decode_superframe(AVCodecContext *avctx, void *data,
808 int *got_frame_ptr, AVPacket *avpkt)
809{
810 AVFrame *frame = data;
811 const uint8_t *buf = avpkt->data;
812 int buf_size = avpkt->size;
813 WMACodecContext *s = avctx->priv_data;
814 int nb_frames, bit_offset, i, pos, len, ret;
815 uint8_t *q;
816 float **samples;
817 int samples_offset;
818
819 ff_tlog(avctx, "***decode_superframe:\n");
820
821 if (buf_size == 0) {
822 s->last_superframe_len = 0;
823 return 0;
824 }
825 if (buf_size < avctx->block_align) {
826 av_log(avctx, AV_LOG_ERROR,
827 "Input packet size too small (%d < %d)\n",
828 buf_size, avctx->block_align);
829 return AVERROR_INVALIDDATA;
830 }
831 if (avctx->block_align)
832 buf_size = avctx->block_align;
833
834 init_get_bits(&s->gb, buf, buf_size * 8);
835
836 if (s->use_bit_reservoir) {
837 /* read super frame header */
838 skip_bits(&s->gb, 4); /* super frame index */
839 nb_frames = get_bits(&s->gb, 4) - (s->last_superframe_len <= 0);
840 if (nb_frames <= 0) {
841 int is_error = nb_frames < 0 || get_bits_left(&s->gb) <= 8;
842 av_log(avctx, is_error ? AV_LOG_ERROR : AV_LOG_WARNING,
843 "nb_frames is %d bits left %d\n",
844 nb_frames, get_bits_left(&s->gb));
845 if (is_error)
846 return AVERROR_INVALIDDATA;
847
848 if ((s->last_superframe_len + buf_size - 1) >
849 MAX_CODED_SUPERFRAME_SIZE)
850 goto fail;
851
852 q = s->last_superframe + s->last_superframe_len;
853 len = buf_size - 1;
854 while (len > 0) {
855 *q++ = get_bits (&s->gb, 8);
856 len --;
857 }
858 memset(q, 0, AV_INPUT_BUFFER_PADDING_SIZE);
859
860 s->last_superframe_len += 8*buf_size - 8;
861// s->reset_block_lengths = 1; //XXX is this needed ?
862 *got_frame_ptr = 0;
863 return buf_size;
864 }
865 } else
866 nb_frames = 1;
867
868 /* get output buffer */
869 frame->nb_samples = nb_frames * s->frame_len;
870 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
871 return ret;
872 samples = (float **) frame->extended_data;
873 samples_offset = 0;
874
875 if (s->use_bit_reservoir) {
876 bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3);
877 if (bit_offset > get_bits_left(&s->gb)) {
878 av_log(avctx, AV_LOG_ERROR,
879 "Invalid last frame bit offset %d > buf size %d (%d)\n",
880 bit_offset, get_bits_left(&s->gb), buf_size);
881 goto fail;
882 }
883
884 if (s->last_superframe_len > 0) {
885 /* add bit_offset bits to last frame */
886 if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) >
887 MAX_CODED_SUPERFRAME_SIZE)
888 goto fail;
889 q = s->last_superframe + s->last_superframe_len;
890 len = bit_offset;
891 while (len > 7) {
892 *q++ = (get_bits) (&s->gb, 8);
893 len -= 8;
894 }
895 if (len > 0)
896 *q++ = (get_bits) (&s->gb, len) << (8 - len);
897 memset(q, 0, AV_INPUT_BUFFER_PADDING_SIZE);
898
899 /* XXX: bit_offset bits into last frame */
900 init_get_bits(&s->gb, s->last_superframe,
901 s->last_superframe_len * 8 + bit_offset);
902 /* skip unused bits */
903 if (s->last_bitoffset > 0)
904 skip_bits(&s->gb, s->last_bitoffset);
905 /* this frame is stored in the last superframe and in the
906 * current one */
907 if (wma_decode_frame(s, samples, samples_offset) < 0)
908 goto fail;
909 samples_offset += s->frame_len;
910 nb_frames--;
911 }
912
913 /* read each frame starting from bit_offset */
914 pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3;
915 if (pos >= MAX_CODED_SUPERFRAME_SIZE * 8 || pos > buf_size * 8)
916 return AVERROR_INVALIDDATA;
917 init_get_bits(&s->gb, buf + (pos >> 3), (buf_size - (pos >> 3)) * 8);
918 len = pos & 7;
919 if (len > 0)
920 skip_bits(&s->gb, len);
921
922 s->reset_block_lengths = 1;
923 for (i = 0; i < nb_frames; i++) {
924 if (wma_decode_frame(s, samples, samples_offset) < 0)
925 goto fail;
926 samples_offset += s->frame_len;
927 }
928
929 /* we copy the end of the frame in the last frame buffer */
930 pos = get_bits_count(&s->gb) +
931 ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7);
932 s->last_bitoffset = pos & 7;
933 pos >>= 3;
934 len = buf_size - pos;
935 if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) {
936 av_log(s->avctx, AV_LOG_ERROR, "len %d invalid\n", len);
937 goto fail;
938 }
939 s->last_superframe_len = len;
940 memcpy(s->last_superframe, buf + pos, len);
941 } else {
942 /* single frame decode */
943 if (wma_decode_frame(s, samples, samples_offset) < 0)
944 goto fail;
945 samples_offset += s->frame_len;
946 }
947
948 ff_dlog(s->avctx, "%d %d %d %d outbytes:%"PTRDIFF_SPECIFIER" eaten:%d\n",
949 s->frame_len_bits, s->block_len_bits, s->frame_len, s->block_len,
950 (int8_t *) samples - (int8_t *) data, avctx->block_align);
951
952 *got_frame_ptr = 1;
953
954 return buf_size;
955
956fail:
957 /* when error, we reset the bit reservoir */
958 s->last_superframe_len = 0;
959 return -1;
960}
961
962static av_cold void flush(AVCodecContext *avctx)
963{
964 WMACodecContext *s = avctx->priv_data;
965
966 s->last_bitoffset =
967 s->last_superframe_len = 0;
968}
969
970#if CONFIG_WMAV1_DECODER
971AVCodec ff_wmav1_decoder = {
972 .name = "wmav1",
973 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
974 .type = AVMEDIA_TYPE_AUDIO,
975 .id = AV_CODEC_ID_WMAV1,
976 .priv_data_size = sizeof(WMACodecContext),
977 .init = wma_decode_init,
978 .close = ff_wma_end,
979 .decode = wma_decode_superframe,
980 .flush = flush,
981 .capabilities = AV_CODEC_CAP_DR1,
982 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
983 AV_SAMPLE_FMT_NONE },
984};
985#endif
986#if CONFIG_WMAV2_DECODER
987AVCodec ff_wmav2_decoder = {
988 .name = "wmav2",
989 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),
990 .type = AVMEDIA_TYPE_AUDIO,
991 .id = AV_CODEC_ID_WMAV2,
992 .priv_data_size = sizeof(WMACodecContext),
993 .init = wma_decode_init,
994 .close = ff_wma_end,
995 .decode = wma_decode_superframe,
996 .flush = flush,
997 .capabilities = AV_CODEC_CAP_DR1,
998 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
999 AV_SAMPLE_FMT_NONE },
1000};
1001#endif
1002
generated by cgit at 2024-05-17 03:21:01 (GMT)