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
49 static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len);
50
51 #ifdef TRACE
dump_floats(WMACodecContext * s,const char * name,int prec,const float * tab,int n)52 static 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
wma_decode_init(AVCodecContext * avctx)70 static 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 ff_init_vlc_from_lengths(&s->hgain_vlc, HGAINVLCBITS, FF_ARRAY_ELEMS(ff_wma_hgain_hufftab),
114 &ff_wma_hgain_hufftab[0][1], 2,
115 &ff_wma_hgain_hufftab[0][0], 2, 1, -18, 0, avctx);
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 */
pow_m1_4(WMACodecContext * s,float x)136 static 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
wma_lsp_to_curve_init(WMACodecContext * s,int frame_len)155 static 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 */
wma_lsp_to_curve(WMACodecContext * s,float * out,float * val_max_ptr,int n,float * lsp)187 static 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 */
decode_exp_lsp(WMACodecContext * s,int ch)216 static 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 */
234 static 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 */
decode_exp_vlc(WMACodecContext * s,int ch)318 static 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 /* NOTE: this offset is the same as MPEG-4 AAC! */
349 last_exp += code - 60;
350 if ((unsigned) last_exp + 60 >= FF_ARRAY_ELEMS(pow_tab)) {
351 av_log(s->avctx, AV_LOG_ERROR, "Exponent out of range: %d\n",
352 last_exp);
353 return -1;
354 }
355 v = ptab[last_exp];
356 iv = iptab[last_exp];
357 if (v > max_scale)
358 max_scale = v;
359 n = *ptr++;
360 switch (n & 3) do {
361 case 0: *q++ = iv;
362 case 3: *q++ = iv;
363 case 2: *q++ = iv;
364 case 1: *q++ = iv;
365 } while ((n -= 4) > 0);
366 }
367 s->max_exponent[ch] = max_scale;
368 return 0;
369 }
370
371 /**
372 * Apply MDCT window and add into output.
373 *
374 * We ensure that when the windows overlap their squared sum
375 * is always 1 (MDCT reconstruction rule).
376 */
wma_window(WMACodecContext * s,float * out)377 static void wma_window(WMACodecContext *s, float *out)
378 {
379 float *in = s->output;
380 int block_len, bsize, n;
381
382 /* left part */
383 if (s->block_len_bits <= s->prev_block_len_bits) {
384 block_len = s->block_len;
385 bsize = s->frame_len_bits - s->block_len_bits;
386
387 s->fdsp->vector_fmul_add(out, in, s->windows[bsize],
388 out, block_len);
389 } else {
390 block_len = 1 << s->prev_block_len_bits;
391 n = (s->block_len - block_len) / 2;
392 bsize = s->frame_len_bits - s->prev_block_len_bits;
393
394 s->fdsp->vector_fmul_add(out + n, in + n, s->windows[bsize],
395 out + n, block_len);
396
397 memcpy(out + n + block_len, in + n + block_len, n * sizeof(float));
398 }
399
400 out += s->block_len;
401 in += s->block_len;
402
403 /* right part */
404 if (s->block_len_bits <= s->next_block_len_bits) {
405 block_len = s->block_len;
406 bsize = s->frame_len_bits - s->block_len_bits;
407
408 s->fdsp->vector_fmul_reverse(out, in, s->windows[bsize], block_len);
409 } else {
410 block_len = 1 << s->next_block_len_bits;
411 n = (s->block_len - block_len) / 2;
412 bsize = s->frame_len_bits - s->next_block_len_bits;
413
414 memcpy(out, in, n * sizeof(float));
415
416 s->fdsp->vector_fmul_reverse(out + n, in + n, s->windows[bsize],
417 block_len);
418
419 memset(out + n + block_len, 0, n * sizeof(float));
420 }
421 }
422
423 /**
424 * @return 0 if OK. 1 if last block of frame. return -1 if
425 * unrecoverable error.
426 */
wma_decode_block(WMACodecContext * s)427 static int wma_decode_block(WMACodecContext *s)
428 {
429 int n, v, a, ch, bsize;
430 int coef_nb_bits, total_gain;
431 int nb_coefs[MAX_CHANNELS];
432 float mdct_norm;
433 FFTContext *mdct;
434
435 #ifdef TRACE
436 ff_tlog(s->avctx, "***decode_block: %d:%d\n",
437 s->frame_count - 1, s->block_num);
438 #endif /* TRACE */
439
440 /* compute current block length */
441 if (s->use_variable_block_len) {
442 n = av_log2(s->nb_block_sizes - 1) + 1;
443
444 if (s->reset_block_lengths) {
445 s->reset_block_lengths = 0;
446 v = get_bits(&s->gb, n);
447 if (v >= s->nb_block_sizes) {
448 av_log(s->avctx, AV_LOG_ERROR,
449 "prev_block_len_bits %d out of range\n",
450 s->frame_len_bits - v);
451 return -1;
452 }
453 s->prev_block_len_bits = s->frame_len_bits - v;
454 v = get_bits(&s->gb, n);
455 if (v >= s->nb_block_sizes) {
456 av_log(s->avctx, AV_LOG_ERROR,
457 "block_len_bits %d out of range\n",
458 s->frame_len_bits - v);
459 return -1;
460 }
461 s->block_len_bits = s->frame_len_bits - v;
462 } else {
463 /* update block lengths */
464 s->prev_block_len_bits = s->block_len_bits;
465 s->block_len_bits = s->next_block_len_bits;
466 }
467 v = get_bits(&s->gb, n);
468 if (v >= s->nb_block_sizes) {
469 av_log(s->avctx, AV_LOG_ERROR,
470 "next_block_len_bits %d out of range\n",
471 s->frame_len_bits - v);
472 return -1;
473 }
474 s->next_block_len_bits = s->frame_len_bits - v;
475 } else {
476 /* fixed block len */
477 s->next_block_len_bits = s->frame_len_bits;
478 s->prev_block_len_bits = s->frame_len_bits;
479 s->block_len_bits = s->frame_len_bits;
480 }
481
482 if (s->frame_len_bits - s->block_len_bits >= s->nb_block_sizes){
483 av_log(s->avctx, AV_LOG_ERROR, "block_len_bits not initialized to a valid value\n");
484 return -1;
485 }
486
487 /* now check if the block length is coherent with the frame length */
488 s->block_len = 1 << s->block_len_bits;
489 if ((s->block_pos + s->block_len) > s->frame_len) {
490 av_log(s->avctx, AV_LOG_ERROR, "frame_len overflow\n");
491 return -1;
492 }
493
494 if (s->avctx->channels == 2)
495 s->ms_stereo = get_bits1(&s->gb);
496 v = 0;
497 for (ch = 0; ch < s->avctx->channels; ch++) {
498 a = get_bits1(&s->gb);
499 s->channel_coded[ch] = a;
500 v |= a;
501 }
502
503 bsize = s->frame_len_bits - s->block_len_bits;
504
505 /* if no channel coded, no need to go further */
506 /* XXX: fix potential framing problems */
507 if (!v)
508 goto next;
509
510 /* read total gain and extract corresponding number of bits for
511 * coef escape coding */
512 total_gain = 1;
513 for (;;) {
514 if (get_bits_left(&s->gb) < 7) {
515 av_log(s->avctx, AV_LOG_ERROR, "total_gain overread\n");
516 return AVERROR_INVALIDDATA;
517 }
518 a = get_bits(&s->gb, 7);
519 total_gain += a;
520 if (a != 127)
521 break;
522 }
523
524 coef_nb_bits = ff_wma_total_gain_to_bits(total_gain);
525
526 /* compute number of coefficients */
527 n = s->coefs_end[bsize] - s->coefs_start;
528 for (ch = 0; ch < s->avctx->channels; ch++)
529 nb_coefs[ch] = n;
530
531 /* complex coding */
532 if (s->use_noise_coding) {
533 for (ch = 0; ch < s->avctx->channels; ch++) {
534 if (s->channel_coded[ch]) {
535 int i, n, a;
536 n = s->exponent_high_sizes[bsize];
537 for (i = 0; i < n; i++) {
538 a = get_bits1(&s->gb);
539 s->high_band_coded[ch][i] = a;
540 /* if noise coding, the coefficients are not transmitted */
541 if (a)
542 nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
543 }
544 }
545 }
546 for (ch = 0; ch < s->avctx->channels; ch++) {
547 if (s->channel_coded[ch]) {
548 int i, n, val;
549
550 n = s->exponent_high_sizes[bsize];
551 val = (int) 0x80000000;
552 for (i = 0; i < n; i++) {
553 if (s->high_band_coded[ch][i]) {
554 if (val == (int) 0x80000000) {
555 val = get_bits(&s->gb, 7) - 19;
556 } else {
557 val += get_vlc2(&s->gb, s->hgain_vlc.table,
558 HGAINVLCBITS, HGAINMAX);
559 }
560 s->high_band_values[ch][i] = val;
561 }
562 }
563 }
564 }
565 }
566
567 /* exponents can be reused in short blocks. */
568 if ((s->block_len_bits == s->frame_len_bits) || get_bits1(&s->gb)) {
569 for (ch = 0; ch < s->avctx->channels; ch++) {
570 if (s->channel_coded[ch]) {
571 if (s->use_exp_vlc) {
572 if (decode_exp_vlc(s, ch) < 0)
573 return -1;
574 } else {
575 decode_exp_lsp(s, ch);
576 }
577 s->exponents_bsize[ch] = bsize;
578 s->exponents_initialized[ch] = 1;
579 }
580 }
581 }
582
583 for (ch = 0; ch < s->avctx->channels; ch++) {
584 if (s->channel_coded[ch] && !s->exponents_initialized[ch])
585 return AVERROR_INVALIDDATA;
586 }
587
588 /* parse spectral coefficients : just RLE encoding */
589 for (ch = 0; ch < s->avctx->channels; ch++) {
590 if (s->channel_coded[ch]) {
591 int tindex;
592 WMACoef *ptr = &s->coefs1[ch][0];
593 int ret;
594
595 /* special VLC tables are used for ms stereo because
596 * there is potentially less energy there */
597 tindex = (ch == 1 && s->ms_stereo);
598 memset(ptr, 0, s->block_len * sizeof(WMACoef));
599 ret = ff_wma_run_level_decode(s->avctx, &s->gb, &s->coef_vlc[tindex],
600 s->level_table[tindex], s->run_table[tindex],
601 0, ptr, 0, nb_coefs[ch],
602 s->block_len, s->frame_len_bits, coef_nb_bits);
603 if (ret < 0)
604 return ret;
605 }
606 if (s->version == 1 && s->avctx->channels >= 2)
607 align_get_bits(&s->gb);
608 }
609
610 /* normalize */
611 {
612 int n4 = s->block_len / 2;
613 mdct_norm = 1.0 / (float) n4;
614 if (s->version == 1)
615 mdct_norm *= sqrt(n4);
616 }
617
618 /* finally compute the MDCT coefficients */
619 for (ch = 0; ch < s->avctx->channels; ch++) {
620 if (s->channel_coded[ch]) {
621 WMACoef *coefs1;
622 float *coefs, *exponents, mult, mult1, noise;
623 int i, j, n, n1, last_high_band, esize;
624 float exp_power[HIGH_BAND_MAX_SIZE];
625
626 coefs1 = s->coefs1[ch];
627 exponents = s->exponents[ch];
628 esize = s->exponents_bsize[ch];
629 mult = ff_exp10(total_gain * 0.05) / s->max_exponent[ch];
630 mult *= mdct_norm;
631 coefs = s->coefs[ch];
632 if (s->use_noise_coding) {
633 mult1 = mult;
634 /* very low freqs : noise */
635 for (i = 0; i < s->coefs_start; i++) {
636 *coefs++ = s->noise_table[s->noise_index] *
637 exponents[i << bsize >> esize] * mult1;
638 s->noise_index = (s->noise_index + 1) &
639 (NOISE_TAB_SIZE - 1);
640 }
641
642 n1 = s->exponent_high_sizes[bsize];
643
644 /* compute power of high bands */
645 exponents = s->exponents[ch] +
646 (s->high_band_start[bsize] << bsize >> esize);
647 last_high_band = 0; /* avoid warning */
648 for (j = 0; j < n1; j++) {
649 n = s->exponent_high_bands[s->frame_len_bits -
650 s->block_len_bits][j];
651 if (s->high_band_coded[ch][j]) {
652 float e2, v;
653 e2 = 0;
654 for (i = 0; i < n; i++) {
655 v = exponents[i << bsize >> esize];
656 e2 += v * v;
657 }
658 exp_power[j] = e2 / n;
659 last_high_band = j;
660 ff_tlog(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n);
661 }
662 exponents += n << bsize >> esize;
663 }
664
665 /* main freqs and high freqs */
666 exponents = s->exponents[ch] + (s->coefs_start << bsize >> esize);
667 for (j = -1; j < n1; j++) {
668 if (j < 0)
669 n = s->high_band_start[bsize] - s->coefs_start;
670 else
671 n = s->exponent_high_bands[s->frame_len_bits -
672 s->block_len_bits][j];
673 if (j >= 0 && s->high_band_coded[ch][j]) {
674 /* use noise with specified power */
675 mult1 = sqrt(exp_power[j] / exp_power[last_high_band]);
676 /* XXX: use a table */
677 mult1 = mult1 * ff_exp10(s->high_band_values[ch][j] * 0.05);
678 mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult);
679 mult1 *= mdct_norm;
680 for (i = 0; i < n; i++) {
681 noise = s->noise_table[s->noise_index];
682 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
683 *coefs++ = noise * exponents[i << bsize >> esize] * mult1;
684 }
685 exponents += n << bsize >> esize;
686 } else {
687 /* coded values + small noise */
688 for (i = 0; i < n; i++) {
689 noise = s->noise_table[s->noise_index];
690 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
691 *coefs++ = ((*coefs1++) + noise) *
692 exponents[i << bsize >> esize] * mult;
693 }
694 exponents += n << bsize >> esize;
695 }
696 }
697
698 /* very high freqs : noise */
699 n = s->block_len - s->coefs_end[bsize];
700 mult1 = mult * exponents[(-(1 << bsize)) >> esize];
701 for (i = 0; i < n; i++) {
702 *coefs++ = s->noise_table[s->noise_index] * mult1;
703 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
704 }
705 } else {
706 /* XXX: optimize more */
707 for (i = 0; i < s->coefs_start; i++)
708 *coefs++ = 0.0;
709 n = nb_coefs[ch];
710 for (i = 0; i < n; i++)
711 *coefs++ = coefs1[i] * exponents[i << bsize >> esize] * mult;
712 n = s->block_len - s->coefs_end[bsize];
713 for (i = 0; i < n; i++)
714 *coefs++ = 0.0;
715 }
716 }
717 }
718
719 #ifdef TRACE
720 for (ch = 0; ch < s->avctx->channels; ch++) {
721 if (s->channel_coded[ch]) {
722 dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len);
723 dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len);
724 }
725 }
726 #endif /* TRACE */
727
728 if (s->ms_stereo && s->channel_coded[1]) {
729 /* nominal case for ms stereo: we do it before mdct */
730 /* no need to optimize this case because it should almost
731 * never happen */
732 if (!s->channel_coded[0]) {
733 ff_tlog(s->avctx, "rare ms-stereo case happened\n");
734 memset(s->coefs[0], 0, sizeof(float) * s->block_len);
735 s->channel_coded[0] = 1;
736 }
737
738 s->fdsp->butterflies_float(s->coefs[0], s->coefs[1], s->block_len);
739 }
740
741 next:
742 mdct = &s->mdct_ctx[bsize];
743
744 for (ch = 0; ch < s->avctx->channels; ch++) {
745 int n4, index;
746
747 n4 = s->block_len / 2;
748 if (s->channel_coded[ch])
749 mdct->imdct_calc(mdct, s->output, s->coefs[ch]);
750 else if (!(s->ms_stereo && ch == 1))
751 memset(s->output, 0, sizeof(s->output));
752
753 /* multiply by the window and add in the frame */
754 index = (s->frame_len / 2) + s->block_pos - n4;
755 wma_window(s, &s->frame_out[ch][index]);
756 }
757
758 /* update block number */
759 s->block_num++;
760 s->block_pos += s->block_len;
761 if (s->block_pos >= s->frame_len)
762 return 1;
763 else
764 return 0;
765 }
766
767 /* decode a frame of frame_len samples */
wma_decode_frame(WMACodecContext * s,float ** samples,int samples_offset)768 static int wma_decode_frame(WMACodecContext *s, float **samples,
769 int samples_offset)
770 {
771 int ret, ch;
772
773 #ifdef TRACE
774 ff_tlog(s->avctx, "***decode_frame: %d size=%d\n",
775 s->frame_count++, s->frame_len);
776 #endif /* TRACE */
777
778 /* read each block */
779 s->block_num = 0;
780 s->block_pos = 0;
781 for (;;) {
782 ret = wma_decode_block(s);
783 if (ret < 0)
784 return -1;
785 if (ret)
786 break;
787 }
788
789 for (ch = 0; ch < s->avctx->channels; ch++) {
790 /* copy current block to output */
791 memcpy(samples[ch] + samples_offset, s->frame_out[ch],
792 s->frame_len * sizeof(*s->frame_out[ch]));
793 /* prepare for next block */
794 memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len],
795 s->frame_len * sizeof(*s->frame_out[ch]));
796
797 #ifdef TRACE
798 dump_floats(s, "samples", 6, samples[ch] + samples_offset,
799 s->frame_len);
800 #endif /* TRACE */
801 }
802
803 return 0;
804 }
805
wma_decode_superframe(AVCodecContext * avctx,void * data,int * got_frame_ptr,AVPacket * avpkt)806 static int wma_decode_superframe(AVCodecContext *avctx, void *data,
807 int *got_frame_ptr, AVPacket *avpkt)
808 {
809 AVFrame *frame = data;
810 const uint8_t *buf = avpkt->data;
811 int buf_size = avpkt->size;
812 WMACodecContext *s = avctx->priv_data;
813 int nb_frames, bit_offset, i, pos, len, ret;
814 uint8_t *q;
815 float **samples;
816 int samples_offset;
817
818 ff_tlog(avctx, "***decode_superframe:\n");
819
820 if (buf_size == 0) {
821 s->last_superframe_len = 0;
822 return 0;
823 }
824 if (buf_size < avctx->block_align) {
825 av_log(avctx, AV_LOG_ERROR,
826 "Input packet size too small (%d < %d)\n",
827 buf_size, avctx->block_align);
828 return AVERROR_INVALIDDATA;
829 }
830 if (avctx->block_align)
831 buf_size = avctx->block_align;
832
833 init_get_bits(&s->gb, buf, buf_size * 8);
834
835 if (s->use_bit_reservoir) {
836 /* read super frame header */
837 skip_bits(&s->gb, 4); /* super frame index */
838 nb_frames = get_bits(&s->gb, 4) - (s->last_superframe_len <= 0);
839 if (nb_frames <= 0) {
840 int is_error = nb_frames < 0 || get_bits_left(&s->gb) <= 8;
841 av_log(avctx, is_error ? AV_LOG_ERROR : AV_LOG_WARNING,
842 "nb_frames is %d bits left %d\n",
843 nb_frames, get_bits_left(&s->gb));
844 if (is_error)
845 return AVERROR_INVALIDDATA;
846
847 if ((s->last_superframe_len + buf_size - 1) >
848 MAX_CODED_SUPERFRAME_SIZE)
849 goto fail;
850
851 q = s->last_superframe + s->last_superframe_len;
852 len = buf_size - 1;
853 while (len > 0) {
854 *q++ = get_bits (&s->gb, 8);
855 len --;
856 }
857 memset(q, 0, AV_INPUT_BUFFER_PADDING_SIZE);
858
859 s->last_superframe_len += 8*buf_size - 8;
860 // s->reset_block_lengths = 1; //XXX is this needed ?
861 *got_frame_ptr = 0;
862 return buf_size;
863 }
864 } else
865 nb_frames = 1;
866
867 /* get output buffer */
868 frame->nb_samples = nb_frames * s->frame_len;
869 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
870 return ret;
871 samples = (float **) frame->extended_data;
872 samples_offset = 0;
873
874 if (s->use_bit_reservoir) {
875 bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3);
876 if (bit_offset > get_bits_left(&s->gb)) {
877 av_log(avctx, AV_LOG_ERROR,
878 "Invalid last frame bit offset %d > buf size %d (%d)\n",
879 bit_offset, get_bits_left(&s->gb), buf_size);
880 goto fail;
881 }
882
883 if (s->last_superframe_len > 0) {
884 /* add bit_offset bits to last frame */
885 if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) >
886 MAX_CODED_SUPERFRAME_SIZE)
887 goto fail;
888 q = s->last_superframe + s->last_superframe_len;
889 len = bit_offset;
890 while (len > 7) {
891 *q++ = get_bits(&s->gb, 8);
892 len -= 8;
893 }
894 if (len > 0)
895 *q++ = get_bits(&s->gb, len) << (8 - len);
896 memset(q, 0, AV_INPUT_BUFFER_PADDING_SIZE);
897
898 /* XXX: bit_offset bits into last frame */
899 init_get_bits(&s->gb, s->last_superframe,
900 s->last_superframe_len * 8 + bit_offset);
901 /* skip unused bits */
902 if (s->last_bitoffset > 0)
903 skip_bits(&s->gb, s->last_bitoffset);
904 /* this frame is stored in the last superframe and in the
905 * current one */
906 if (wma_decode_frame(s, samples, samples_offset) < 0)
907 goto fail;
908 samples_offset += s->frame_len;
909 nb_frames--;
910 }
911
912 /* read each frame starting from bit_offset */
913 pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3;
914 if (pos >= MAX_CODED_SUPERFRAME_SIZE * 8 || pos > buf_size * 8)
915 return AVERROR_INVALIDDATA;
916 init_get_bits(&s->gb, buf + (pos >> 3), (buf_size - (pos >> 3)) * 8);
917 len = pos & 7;
918 if (len > 0)
919 skip_bits(&s->gb, len);
920
921 s->reset_block_lengths = 1;
922 for (i = 0; i < nb_frames; i++) {
923 if (wma_decode_frame(s, samples, samples_offset) < 0)
924 goto fail;
925 samples_offset += s->frame_len;
926 }
927
928 /* we copy the end of the frame in the last frame buffer */
929 pos = get_bits_count(&s->gb) +
930 ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7);
931 s->last_bitoffset = pos & 7;
932 pos >>= 3;
933 len = buf_size - pos;
934 if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) {
935 av_log(s->avctx, AV_LOG_ERROR, "len %d invalid\n", len);
936 goto fail;
937 }
938 s->last_superframe_len = len;
939 memcpy(s->last_superframe, buf + pos, len);
940 } else {
941 /* single frame decode */
942 if (wma_decode_frame(s, samples, samples_offset) < 0)
943 goto fail;
944 samples_offset += s->frame_len;
945 }
946
947 ff_dlog(s->avctx, "%d %d %d %d outbytes:%"PTRDIFF_SPECIFIER" eaten:%d\n",
948 s->frame_len_bits, s->block_len_bits, s->frame_len, s->block_len,
949 (int8_t *) samples - (int8_t *) data, avctx->block_align);
950
951 *got_frame_ptr = 1;
952
953 return buf_size;
954
955 fail:
956 /* when error, we reset the bit reservoir */
957 s->last_superframe_len = 0;
958 return -1;
959 }
960
flush(AVCodecContext * avctx)961 static av_cold void flush(AVCodecContext *avctx)
962 {
963 WMACodecContext *s = avctx->priv_data;
964
965 s->last_bitoffset =
966 s->last_superframe_len = 0;
967 }
968
969 #if CONFIG_WMAV1_DECODER
970 AVCodec ff_wmav1_decoder = {
971 .name = "wmav1",
972 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
973 .type = AVMEDIA_TYPE_AUDIO,
974 .id = AV_CODEC_ID_WMAV1,
975 .priv_data_size = sizeof(WMACodecContext),
976 .init = wma_decode_init,
977 .close = ff_wma_end,
978 .decode = wma_decode_superframe,
979 .flush = flush,
980 .capabilities = AV_CODEC_CAP_DR1,
981 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
982 AV_SAMPLE_FMT_NONE },
983 };
984 #endif
985 #if CONFIG_WMAV2_DECODER
986 AVCodec ff_wmav2_decoder = {
987 .name = "wmav2",
988 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),
989 .type = AVMEDIA_TYPE_AUDIO,
990 .id = AV_CODEC_ID_WMAV2,
991 .priv_data_size = sizeof(WMACodecContext),
992 .init = wma_decode_init,
993 .close = ff_wma_end,
994 .decode = wma_decode_superframe,
995 .flush = flush,
996 .capabilities = AV_CODEC_CAP_DR1,
997 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
998 AV_SAMPLE_FMT_NONE },
999 };
1000 #endif
1001