2 * WMA compatible decoder
3 * Copyright (c) 2002 The Libav Project
5 * This file is part of Libav.
7 * Libav 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 * Libav 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 Libav; 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.
44 #define EXPMAX ((19+EXPVLCBITS-1)/EXPVLCBITS)
46 #define HGAINVLCBITS 9
47 #define HGAINMAX ((13+HGAINVLCBITS-1)/HGAINVLCBITS)
49 static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len);
52 static void dump_floats(WMACodecContext *s, const char *name, int prec, const float *tab, int n)
56 tprintf(s->avctx, "%s[%d]:\n", name, n);
59 tprintf(s->avctx, "%4d: ", i);
60 tprintf(s->avctx, " %8.*f", prec, tab[i]);
62 tprintf(s->avctx, "\n");
65 tprintf(s->avctx, "\n");
69 static int wma_decode_init(AVCodecContext * avctx)
71 WMACodecContext *s = avctx->priv_data;
75 if (!avctx->block_align) {
76 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
77 return AVERROR(EINVAL);
82 /* extract flag infos */
84 extradata = avctx->extradata;
85 if (avctx->codec->id == AV_CODEC_ID_WMAV1 && avctx->extradata_size >= 4) {
86 flags2 = AV_RL16(extradata+2);
87 } else if (avctx->codec->id == AV_CODEC_ID_WMAV2 && avctx->extradata_size >= 6) {
88 flags2 = AV_RL16(extradata+4);
91 s->use_exp_vlc = flags2 & 0x0001;
92 s->use_bit_reservoir = flags2 & 0x0002;
93 s->use_variable_block_len = flags2 & 0x0004;
95 if(ff_wma_init(avctx, flags2)<0)
99 for(i = 0; i < s->nb_block_sizes; i++)
100 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1, 1.0 / 32768.0);
102 if (s->use_noise_coding) {
103 init_vlc(&s->hgain_vlc, HGAINVLCBITS, sizeof(ff_wma_hgain_huffbits),
104 ff_wma_hgain_huffbits, 1, 1,
105 ff_wma_hgain_huffcodes, 2, 2, 0);
108 if (s->use_exp_vlc) {
109 init_vlc(&s->exp_vlc, EXPVLCBITS, sizeof(ff_aac_scalefactor_bits), //FIXME move out of context
110 ff_aac_scalefactor_bits, 1, 1,
111 ff_aac_scalefactor_code, 4, 4, 0);
113 wma_lsp_to_curve_init(s, s->frame_len);
116 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
122 * compute x^-0.25 with an exponent and mantissa table. We use linear
123 * interpolation to reduce the mantissa table size at a small speed
124 * expense (linear interpolation approximately doubles the number of
125 * bits of precision).
127 static inline float pow_m1_4(WMACodecContext *s, float x)
138 m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1);
139 /* build interpolation scale: 1 <= t < 2. */
140 t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23);
141 a = s->lsp_pow_m_table1[m];
142 b = s->lsp_pow_m_table2[m];
143 return s->lsp_pow_e_table[e] * (a + b * t.f);
146 static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len)
151 wdel = M_PI / frame_len;
152 for(i=0;i<frame_len;i++)
153 s->lsp_cos_table[i] = 2.0f * cos(wdel * i);
155 /* tables for x^-0.25 computation */
158 s->lsp_pow_e_table[i] = pow(2.0, e * -0.25);
161 /* NOTE: these two tables are needed to avoid two operations in
164 for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) {
165 m = (1 << LSP_POW_BITS) + i;
166 a = (float)m * (0.5 / (1 << LSP_POW_BITS));
168 s->lsp_pow_m_table1[i] = 2 * a - b;
169 s->lsp_pow_m_table2[i] = b - a;
175 * NOTE: We use the same code as Vorbis here
176 * @todo optimize it further with SSE/3Dnow
178 static void wma_lsp_to_curve(WMACodecContext *s,
179 float *out, float *val_max_ptr,
183 float p, q, w, v, val_max;
189 w = s->lsp_cos_table[i];
190 for(j=1;j<NB_LSP_COEFS;j+=2){
202 *val_max_ptr = val_max;
206 * decode exponents coded with LSP coefficients (same idea as Vorbis)
208 static void decode_exp_lsp(WMACodecContext *s, int ch)
210 float lsp_coefs[NB_LSP_COEFS];
213 for(i = 0; i < NB_LSP_COEFS; i++) {
214 if (i == 0 || i >= 8)
215 val = get_bits(&s->gb, 3);
217 val = get_bits(&s->gb, 4);
218 lsp_coefs[i] = ff_wma_lsp_codebook[i][val];
221 wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch],
222 s->block_len, lsp_coefs);
225 /** pow(10, i / 16.0) for i in -60..95 */
226 static const float pow_tab[] = {
227 1.7782794100389e-04, 2.0535250264571e-04,
228 2.3713737056617e-04, 2.7384196342644e-04,
229 3.1622776601684e-04, 3.6517412725484e-04,
230 4.2169650342858e-04, 4.8696752516586e-04,
231 5.6234132519035e-04, 6.4938163157621e-04,
232 7.4989420933246e-04, 8.6596432336006e-04,
233 1.0000000000000e-03, 1.1547819846895e-03,
234 1.3335214321633e-03, 1.5399265260595e-03,
235 1.7782794100389e-03, 2.0535250264571e-03,
236 2.3713737056617e-03, 2.7384196342644e-03,
237 3.1622776601684e-03, 3.6517412725484e-03,
238 4.2169650342858e-03, 4.8696752516586e-03,
239 5.6234132519035e-03, 6.4938163157621e-03,
240 7.4989420933246e-03, 8.6596432336006e-03,
241 1.0000000000000e-02, 1.1547819846895e-02,
242 1.3335214321633e-02, 1.5399265260595e-02,
243 1.7782794100389e-02, 2.0535250264571e-02,
244 2.3713737056617e-02, 2.7384196342644e-02,
245 3.1622776601684e-02, 3.6517412725484e-02,
246 4.2169650342858e-02, 4.8696752516586e-02,
247 5.6234132519035e-02, 6.4938163157621e-02,
248 7.4989420933246e-02, 8.6596432336007e-02,
249 1.0000000000000e-01, 1.1547819846895e-01,
250 1.3335214321633e-01, 1.5399265260595e-01,
251 1.7782794100389e-01, 2.0535250264571e-01,
252 2.3713737056617e-01, 2.7384196342644e-01,
253 3.1622776601684e-01, 3.6517412725484e-01,
254 4.2169650342858e-01, 4.8696752516586e-01,
255 5.6234132519035e-01, 6.4938163157621e-01,
256 7.4989420933246e-01, 8.6596432336007e-01,
257 1.0000000000000e+00, 1.1547819846895e+00,
258 1.3335214321633e+00, 1.5399265260595e+00,
259 1.7782794100389e+00, 2.0535250264571e+00,
260 2.3713737056617e+00, 2.7384196342644e+00,
261 3.1622776601684e+00, 3.6517412725484e+00,
262 4.2169650342858e+00, 4.8696752516586e+00,
263 5.6234132519035e+00, 6.4938163157621e+00,
264 7.4989420933246e+00, 8.6596432336007e+00,
265 1.0000000000000e+01, 1.1547819846895e+01,
266 1.3335214321633e+01, 1.5399265260595e+01,
267 1.7782794100389e+01, 2.0535250264571e+01,
268 2.3713737056617e+01, 2.7384196342644e+01,
269 3.1622776601684e+01, 3.6517412725484e+01,
270 4.2169650342858e+01, 4.8696752516586e+01,
271 5.6234132519035e+01, 6.4938163157621e+01,
272 7.4989420933246e+01, 8.6596432336007e+01,
273 1.0000000000000e+02, 1.1547819846895e+02,
274 1.3335214321633e+02, 1.5399265260595e+02,
275 1.7782794100389e+02, 2.0535250264571e+02,
276 2.3713737056617e+02, 2.7384196342644e+02,
277 3.1622776601684e+02, 3.6517412725484e+02,
278 4.2169650342858e+02, 4.8696752516586e+02,
279 5.6234132519035e+02, 6.4938163157621e+02,
280 7.4989420933246e+02, 8.6596432336007e+02,
281 1.0000000000000e+03, 1.1547819846895e+03,
282 1.3335214321633e+03, 1.5399265260595e+03,
283 1.7782794100389e+03, 2.0535250264571e+03,
284 2.3713737056617e+03, 2.7384196342644e+03,
285 3.1622776601684e+03, 3.6517412725484e+03,
286 4.2169650342858e+03, 4.8696752516586e+03,
287 5.6234132519035e+03, 6.4938163157621e+03,
288 7.4989420933246e+03, 8.6596432336007e+03,
289 1.0000000000000e+04, 1.1547819846895e+04,
290 1.3335214321633e+04, 1.5399265260595e+04,
291 1.7782794100389e+04, 2.0535250264571e+04,
292 2.3713737056617e+04, 2.7384196342644e+04,
293 3.1622776601684e+04, 3.6517412725484e+04,
294 4.2169650342858e+04, 4.8696752516586e+04,
295 5.6234132519035e+04, 6.4938163157621e+04,
296 7.4989420933246e+04, 8.6596432336007e+04,
297 1.0000000000000e+05, 1.1547819846895e+05,
298 1.3335214321633e+05, 1.5399265260595e+05,
299 1.7782794100389e+05, 2.0535250264571e+05,
300 2.3713737056617e+05, 2.7384196342644e+05,
301 3.1622776601684e+05, 3.6517412725484e+05,
302 4.2169650342858e+05, 4.8696752516586e+05,
303 5.6234132519035e+05, 6.4938163157621e+05,
304 7.4989420933246e+05, 8.6596432336007e+05,
308 * decode exponents coded with VLC codes
310 static int decode_exp_vlc(WMACodecContext *s, int ch)
312 int last_exp, n, code;
315 uint32_t *q, *q_end, iv;
316 const float *ptab = pow_tab + 60;
317 const uint32_t *iptab = (const uint32_t*)ptab;
319 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
320 q = (uint32_t *)s->exponents[ch];
321 q_end = q + s->block_len;
323 if (s->version == 1) {
324 last_exp = get_bits(&s->gb, 5) + 10;
326 iv = iptab[last_exp];
334 } while ((n -= 4) > 0);
339 code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX);
341 av_log(s->avctx, AV_LOG_ERROR, "Exponent vlc invalid\n");
344 /* NOTE: this offset is the same as MPEG4 AAC ! */
345 last_exp += code - 60;
346 if ((unsigned)last_exp + 60 >= FF_ARRAY_ELEMS(pow_tab)) {
347 av_log(s->avctx, AV_LOG_ERROR, "Exponent out of range: %d\n",
352 iv = iptab[last_exp];
361 } while ((n -= 4) > 0);
363 s->max_exponent[ch] = max_scale;
369 * Apply MDCT window and add into output.
371 * We ensure that when the windows overlap their squared sum
372 * is always 1 (MDCT reconstruction rule).
374 static void wma_window(WMACodecContext *s, float *out)
376 float *in = s->output;
377 int block_len, bsize, n;
380 if (s->block_len_bits <= s->prev_block_len_bits) {
381 block_len = s->block_len;
382 bsize = s->frame_len_bits - s->block_len_bits;
384 s->fdsp.vector_fmul_add(out, in, s->windows[bsize],
388 block_len = 1 << s->prev_block_len_bits;
389 n = (s->block_len - block_len) / 2;
390 bsize = s->frame_len_bits - s->prev_block_len_bits;
392 s->fdsp.vector_fmul_add(out+n, in+n, s->windows[bsize],
395 memcpy(out+n+block_len, in+n+block_len, n*sizeof(float));
402 if (s->block_len_bits <= s->next_block_len_bits) {
403 block_len = s->block_len;
404 bsize = s->frame_len_bits - s->block_len_bits;
406 s->fdsp.vector_fmul_reverse(out, in, s->windows[bsize], block_len);
409 block_len = 1 << s->next_block_len_bits;
410 n = (s->block_len - block_len) / 2;
411 bsize = s->frame_len_bits - s->next_block_len_bits;
413 memcpy(out, in, n*sizeof(float));
415 s->fdsp.vector_fmul_reverse(out+n, in+n, s->windows[bsize], block_len);
417 memset(out+n+block_len, 0, n*sizeof(float));
423 * @return 0 if OK. 1 if last block of frame. return -1 if
424 * unrecorrable error.
426 static int wma_decode_block(WMACodecContext *s)
428 int n, v, a, ch, bsize;
429 int coef_nb_bits, total_gain;
430 int nb_coefs[MAX_CHANNELS];
435 tprintf(s->avctx, "***decode_block: %d:%d\n", s->frame_count - 1, s->block_num);
438 /* compute current block length */
439 if (s->use_variable_block_len) {
440 n = av_log2(s->nb_block_sizes - 1) + 1;
442 if (s->reset_block_lengths) {
443 s->reset_block_lengths = 0;
444 v = get_bits(&s->gb, n);
445 if (v >= s->nb_block_sizes){
446 av_log(s->avctx, AV_LOG_ERROR, "prev_block_len_bits %d out of range\n", s->frame_len_bits - v);
449 s->prev_block_len_bits = s->frame_len_bits - v;
450 v = get_bits(&s->gb, n);
451 if (v >= s->nb_block_sizes){
452 av_log(s->avctx, AV_LOG_ERROR, "block_len_bits %d out of range\n", s->frame_len_bits - v);
455 s->block_len_bits = s->frame_len_bits - v;
457 /* update block lengths */
458 s->prev_block_len_bits = s->block_len_bits;
459 s->block_len_bits = s->next_block_len_bits;
461 v = get_bits(&s->gb, n);
462 if (v >= s->nb_block_sizes){
463 av_log(s->avctx, AV_LOG_ERROR, "next_block_len_bits %d out of range\n", s->frame_len_bits - v);
466 s->next_block_len_bits = s->frame_len_bits - v;
468 /* fixed block len */
469 s->next_block_len_bits = s->frame_len_bits;
470 s->prev_block_len_bits = s->frame_len_bits;
471 s->block_len_bits = s->frame_len_bits;
474 /* now check if the block length is coherent with the frame length */
475 s->block_len = 1 << s->block_len_bits;
476 if ((s->block_pos + s->block_len) > s->frame_len){
477 av_log(s->avctx, AV_LOG_ERROR, "frame_len overflow\n");
481 if (s->avctx->channels == 2) {
482 s->ms_stereo = get_bits1(&s->gb);
485 for(ch = 0; ch < s->avctx->channels; ch++) {
486 a = get_bits1(&s->gb);
487 s->channel_coded[ch] = a;
491 bsize = s->frame_len_bits - s->block_len_bits;
493 /* if no channel coded, no need to go further */
494 /* XXX: fix potential framing problems */
498 /* read total gain and extract corresponding number of bits for
499 coef escape coding */
502 a = get_bits(&s->gb, 7);
508 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
510 /* compute number of coefficients */
511 n = s->coefs_end[bsize] - s->coefs_start;
512 for(ch = 0; ch < s->avctx->channels; ch++)
516 if (s->use_noise_coding) {
518 for(ch = 0; ch < s->avctx->channels; ch++) {
519 if (s->channel_coded[ch]) {
521 n = s->exponent_high_sizes[bsize];
523 a = get_bits1(&s->gb);
524 s->high_band_coded[ch][i] = a;
525 /* if noise coding, the coefficients are not transmitted */
527 nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
531 for(ch = 0; ch < s->avctx->channels; ch++) {
532 if (s->channel_coded[ch]) {
535 n = s->exponent_high_sizes[bsize];
536 val = (int)0x80000000;
538 if (s->high_band_coded[ch][i]) {
539 if (val == (int)0x80000000) {
540 val = get_bits(&s->gb, 7) - 19;
542 code = get_vlc2(&s->gb, s->hgain_vlc.table, HGAINVLCBITS, HGAINMAX);
544 av_log(s->avctx, AV_LOG_ERROR, "hgain vlc invalid\n");
549 s->high_band_values[ch][i] = val;
556 /* exponents can be reused in short blocks. */
557 if ((s->block_len_bits == s->frame_len_bits) ||
559 for(ch = 0; ch < s->avctx->channels; ch++) {
560 if (s->channel_coded[ch]) {
561 if (s->use_exp_vlc) {
562 if (decode_exp_vlc(s, ch) < 0)
565 decode_exp_lsp(s, ch);
567 s->exponents_bsize[ch] = bsize;
572 /* parse spectral coefficients : just RLE encoding */
573 for (ch = 0; ch < s->avctx->channels; ch++) {
574 if (s->channel_coded[ch]) {
576 WMACoef* ptr = &s->coefs1[ch][0];
578 /* special VLC tables are used for ms stereo because
579 there is potentially less energy there */
580 tindex = (ch == 1 && s->ms_stereo);
581 memset(ptr, 0, s->block_len * sizeof(WMACoef));
582 ff_wma_run_level_decode(s->avctx, &s->gb, &s->coef_vlc[tindex],
583 s->level_table[tindex], s->run_table[tindex],
584 0, ptr, 0, nb_coefs[ch],
585 s->block_len, s->frame_len_bits, coef_nb_bits);
587 if (s->version == 1 && s->avctx->channels >= 2) {
588 align_get_bits(&s->gb);
594 int n4 = s->block_len / 2;
595 mdct_norm = 1.0 / (float)n4;
596 if (s->version == 1) {
597 mdct_norm *= sqrt(n4);
601 /* finally compute the MDCT coefficients */
602 for (ch = 0; ch < s->avctx->channels; ch++) {
603 if (s->channel_coded[ch]) {
605 float *coefs, *exponents, mult, mult1, noise;
606 int i, j, n, n1, last_high_band, esize;
607 float exp_power[HIGH_BAND_MAX_SIZE];
609 coefs1 = s->coefs1[ch];
610 exponents = s->exponents[ch];
611 esize = s->exponents_bsize[ch];
612 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
614 coefs = s->coefs[ch];
615 if (s->use_noise_coding) {
617 /* very low freqs : noise */
618 for(i = 0;i < s->coefs_start; i++) {
619 *coefs++ = s->noise_table[s->noise_index] *
620 exponents[i<<bsize>>esize] * mult1;
621 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
624 n1 = s->exponent_high_sizes[bsize];
626 /* compute power of high bands */
627 exponents = s->exponents[ch] +
628 (s->high_band_start[bsize]<<bsize>>esize);
629 last_high_band = 0; /* avoid warning */
631 n = s->exponent_high_bands[s->frame_len_bits -
632 s->block_len_bits][j];
633 if (s->high_band_coded[ch][j]) {
636 for(i = 0;i < n; i++) {
637 v = exponents[i<<bsize>>esize];
640 exp_power[j] = e2 / n;
642 tprintf(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n);
644 exponents += n<<bsize>>esize;
647 /* main freqs and high freqs */
648 exponents = s->exponents[ch] + (s->coefs_start<<bsize>>esize);
651 n = s->high_band_start[bsize] -
654 n = s->exponent_high_bands[s->frame_len_bits -
655 s->block_len_bits][j];
657 if (j >= 0 && s->high_band_coded[ch][j]) {
658 /* use noise with specified power */
659 mult1 = sqrt(exp_power[j] / exp_power[last_high_band]);
660 /* XXX: use a table */
661 mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05);
662 mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult);
664 for(i = 0;i < n; i++) {
665 noise = s->noise_table[s->noise_index];
666 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
668 exponents[i<<bsize>>esize] * mult1;
670 exponents += n<<bsize>>esize;
672 /* coded values + small noise */
673 for(i = 0;i < n; i++) {
674 noise = s->noise_table[s->noise_index];
675 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
676 *coefs++ = ((*coefs1++) + noise) *
677 exponents[i<<bsize>>esize] * mult;
679 exponents += n<<bsize>>esize;
683 /* very high freqs : noise */
684 n = s->block_len - s->coefs_end[bsize];
685 mult1 = mult * exponents[((-1<<bsize))>>esize];
686 for(i = 0; i < n; i++) {
687 *coefs++ = s->noise_table[s->noise_index] * mult1;
688 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1);
691 /* XXX: optimize more */
692 for(i = 0;i < s->coefs_start; i++)
695 for(i = 0;i < n; i++) {
696 *coefs++ = coefs1[i] * exponents[i<<bsize>>esize] * mult;
698 n = s->block_len - s->coefs_end[bsize];
699 for(i = 0;i < n; i++)
706 for (ch = 0; ch < s->avctx->channels; ch++) {
707 if (s->channel_coded[ch]) {
708 dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len);
709 dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len);
714 if (s->ms_stereo && s->channel_coded[1]) {
715 /* nominal case for ms stereo: we do it before mdct */
716 /* no need to optimize this case because it should almost
718 if (!s->channel_coded[0]) {
719 tprintf(s->avctx, "rare ms-stereo case happened\n");
720 memset(s->coefs[0], 0, sizeof(float) * s->block_len);
721 s->channel_coded[0] = 1;
724 s->fdsp.butterflies_float(s->coefs[0], s->coefs[1], s->block_len);
728 mdct = &s->mdct_ctx[bsize];
730 for (ch = 0; ch < s->avctx->channels; ch++) {
733 n4 = s->block_len / 2;
734 if(s->channel_coded[ch]){
735 mdct->imdct_calc(mdct, s->output, s->coefs[ch]);
736 }else if(!(s->ms_stereo && ch==1))
737 memset(s->output, 0, sizeof(s->output));
739 /* multiply by the window and add in the frame */
740 index = (s->frame_len / 2) + s->block_pos - n4;
741 wma_window(s, &s->frame_out[ch][index]);
744 /* update block number */
746 s->block_pos += s->block_len;
747 if (s->block_pos >= s->frame_len)
753 /* decode a frame of frame_len samples */
754 static int wma_decode_frame(WMACodecContext *s, float **samples,
760 tprintf(s->avctx, "***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len);
763 /* read each block */
767 ret = wma_decode_block(s);
774 for (ch = 0; ch < s->avctx->channels; ch++) {
775 /* copy current block to output */
776 memcpy(samples[ch] + samples_offset, s->frame_out[ch],
777 s->frame_len * sizeof(*s->frame_out[ch]));
778 /* prepare for next block */
779 memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len],
780 s->frame_len * sizeof(*s->frame_out[ch]));
783 dump_floats(s, "samples", 6, samples[ch] + samples_offset, s->frame_len);
790 static int wma_decode_superframe(AVCodecContext *avctx, void *data,
791 int *got_frame_ptr, AVPacket *avpkt)
793 AVFrame *frame = data;
794 const uint8_t *buf = avpkt->data;
795 int buf_size = avpkt->size;
796 WMACodecContext *s = avctx->priv_data;
797 int nb_frames, bit_offset, i, pos, len, ret;
802 tprintf(avctx, "***decode_superframe:\n");
805 s->last_superframe_len = 0;
808 if (buf_size < avctx->block_align) {
809 av_log(avctx, AV_LOG_ERROR,
810 "Input packet size too small (%d < %d)\n",
811 buf_size, avctx->block_align);
812 return AVERROR_INVALIDDATA;
814 buf_size = avctx->block_align;
816 init_get_bits(&s->gb, buf, buf_size*8);
818 if (s->use_bit_reservoir) {
819 /* read super frame header */
820 skip_bits(&s->gb, 4); /* super frame index */
821 nb_frames = get_bits(&s->gb, 4) - (s->last_superframe_len <= 0);
826 /* get output buffer */
827 frame->nb_samples = nb_frames * s->frame_len;
828 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
829 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
832 samples = (float **)frame->extended_data;
835 if (s->use_bit_reservoir) {
836 bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3);
837 if (bit_offset > get_bits_left(&s->gb)) {
838 av_log(avctx, AV_LOG_ERROR,
839 "Invalid last frame bit offset %d > buf size %d (%d)\n",
840 bit_offset, get_bits_left(&s->gb), buf_size);
844 if (s->last_superframe_len > 0) {
845 /* add bit_offset bits to last frame */
846 if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) >
847 MAX_CODED_SUPERFRAME_SIZE)
849 q = s->last_superframe + s->last_superframe_len;
852 *q++ = (get_bits)(&s->gb, 8);
856 *q++ = (get_bits)(&s->gb, len) << (8 - len);
858 memset(q, 0, FF_INPUT_BUFFER_PADDING_SIZE);
860 /* XXX: bit_offset bits into last frame */
861 init_get_bits(&s->gb, s->last_superframe, s->last_superframe_len * 8 + bit_offset);
862 /* skip unused bits */
863 if (s->last_bitoffset > 0)
864 skip_bits(&s->gb, s->last_bitoffset);
865 /* this frame is stored in the last superframe and in the
867 if (wma_decode_frame(s, samples, samples_offset) < 0)
869 samples_offset += s->frame_len;
873 /* read each frame starting from bit_offset */
874 pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3;
875 if (pos >= MAX_CODED_SUPERFRAME_SIZE * 8 || pos > buf_size * 8)
876 return AVERROR_INVALIDDATA;
877 init_get_bits(&s->gb, buf + (pos >> 3), (buf_size - (pos >> 3))*8);
880 skip_bits(&s->gb, len);
882 s->reset_block_lengths = 1;
883 for(i=0;i<nb_frames;i++) {
884 if (wma_decode_frame(s, samples, samples_offset) < 0)
886 samples_offset += s->frame_len;
889 /* we copy the end of the frame in the last frame buffer */
890 pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7);
891 s->last_bitoffset = pos & 7;
893 len = buf_size - pos;
894 if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) {
895 av_log(s->avctx, AV_LOG_ERROR, "len %d invalid\n", len);
898 s->last_superframe_len = len;
899 memcpy(s->last_superframe, buf + pos, len);
901 /* single frame decode */
902 if (wma_decode_frame(s, samples, samples_offset) < 0)
904 samples_offset += s->frame_len;
907 av_dlog(s->avctx, "%d %d %d %d outbytes:%td eaten:%d\n",
908 s->frame_len_bits, s->block_len_bits, s->frame_len, s->block_len,
909 (int8_t *)samples - (int8_t *)data, avctx->block_align);
913 return avctx->block_align;
915 /* when error, we reset the bit reservoir */
916 s->last_superframe_len = 0;
920 static av_cold void flush(AVCodecContext *avctx)
922 WMACodecContext *s = avctx->priv_data;
925 s->last_superframe_len= 0;
928 AVCodec ff_wmav1_decoder = {
930 .type = AVMEDIA_TYPE_AUDIO,
931 .id = AV_CODEC_ID_WMAV1,
932 .priv_data_size = sizeof(WMACodecContext),
933 .init = wma_decode_init,
935 .decode = wma_decode_superframe,
937 .capabilities = CODEC_CAP_DR1,
938 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
939 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
940 AV_SAMPLE_FMT_NONE },
943 AVCodec ff_wmav2_decoder = {
945 .type = AVMEDIA_TYPE_AUDIO,
946 .id = AV_CODEC_ID_WMAV2,
947 .priv_data_size = sizeof(WMACodecContext),
948 .init = wma_decode_init,
950 .decode = wma_decode_superframe,
952 .capabilities = CODEC_CAP_DR1,
953 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),
954 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
955 AV_SAMPLE_FMT_NONE },