2 * Windows Media Audio Lossless decoder
3 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson
5 * Copyright (c) 2011 Andreas Ă–man
6 * Copyright (c) 2011 - 2012 Mashiat Sarker Shakkhar
8 * This file is part of Libav.
10 * Libav is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * Libav is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with Libav; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 #include "libavutil/attributes.h"
28 #include "libavutil/avassert.h"
35 #include "wma_common.h"
37 /** current decoder limitations */
38 #define WMALL_MAX_CHANNELS 8 ///< max number of handled channels
39 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
40 #define MAX_BANDS 29 ///< max number of scale factor bands
41 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
44 #define WMALL_BLOCK_MIN_BITS 6 ///< log2 of min block size
45 #define WMALL_BLOCK_MAX_BITS 14 ///< log2 of max block size
46 #define WMALL_BLOCK_MAX_SIZE (1 << WMALL_BLOCK_MAX_BITS) ///< maximum block size
47 #define WMALL_BLOCK_SIZES (WMALL_BLOCK_MAX_BITS - WMALL_BLOCK_MIN_BITS + 1) ///< possible block sizes
51 * @brief frame-specific decoder context for a single channel
54 int16_t prev_block_len; ///< length of the previous block
55 uint8_t transmit_coefs;
56 uint8_t num_subframes;
57 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
58 uint16_t subframe_offsets[MAX_SUBFRAMES]; ///< subframe positions in the current frame
59 uint8_t cur_subframe; ///< current subframe number
60 uint16_t decoded_samples; ///< number of already processed samples
61 int quant_step; ///< quantization step for the current subframe
62 int transient_counter; ///< number of transient samples from the beginning of the transient zone
66 * @brief main decoder context
68 typedef struct WmallDecodeCtx {
69 /* generic decoder variables */
70 AVCodecContext *avctx;
72 uint8_t frame_data[MAX_FRAMESIZE + FF_INPUT_BUFFER_PADDING_SIZE]; ///< compressed frame data
73 PutBitContext pb; ///< context for filling the frame_data buffer
75 /* frame size dependent frame information (set during initialization) */
76 uint32_t decode_flags; ///< used compression features
77 int len_prefix; ///< frame is prefixed with its length
78 int dynamic_range_compression; ///< frame contains DRC data
79 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
80 uint16_t samples_per_frame; ///< number of samples to output
81 uint16_t log2_frame_size;
82 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
83 int8_t lfe_channel; ///< lfe channel index
84 uint8_t max_num_subframes;
85 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
86 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
87 uint16_t min_samples_per_subframe;
89 /* packet decode state */
90 GetBitContext pgb; ///< bitstream reader context for the packet
91 int next_packet_start; ///< start offset of the next WMA packet in the demuxer packet
92 uint8_t packet_offset; ///< offset to the frame in the packet
93 uint8_t packet_sequence_number; ///< current packet number
94 int num_saved_bits; ///< saved number of bits
95 int frame_offset; ///< frame offset in the bit reservoir
96 int subframe_offset; ///< subframe offset in the bit reservoir
97 uint8_t packet_loss; ///< set in case of bitstream error
98 uint8_t packet_done; ///< set when a packet is fully decoded
100 /* frame decode state */
101 uint32_t frame_num; ///< current frame number (not used for decoding)
102 GetBitContext gb; ///< bitstream reader context
103 int buf_bit_size; ///< buffer size in bits
104 int16_t *samples_16[WMALL_MAX_CHANNELS]; ///< current samplebuffer pointer (16-bit)
105 int32_t *samples_32[WMALL_MAX_CHANNELS]; ///< current samplebuffer pointer (24-bit)
106 uint8_t drc_gain; ///< gain for the DRC tool
107 int8_t skip_frame; ///< skip output step
108 int8_t parsed_all_subframes; ///< all subframes decoded?
110 /* subframe/block decode state */
111 int16_t subframe_len; ///< current subframe length
112 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
113 int8_t channel_indexes_for_cur_subframe[WMALL_MAX_CHANNELS];
115 WmallChannelCtx channel[WMALL_MAX_CHANNELS]; ///< per channel data
117 // WMA Lossless-specific
119 uint8_t do_arith_coding;
120 uint8_t do_ac_filter;
121 uint8_t do_inter_ch_decorr;
125 int8_t acfilter_order;
126 int8_t acfilter_scaling;
127 int64_t acfilter_coeffs[16];
128 int acfilter_prevvalues[2][16];
131 int8_t mclms_scaling;
132 int16_t mclms_coeffs[WMALL_MAX_CHANNELS * WMALL_MAX_CHANNELS * 32];
133 int16_t mclms_coeffs_cur[WMALL_MAX_CHANNELS * WMALL_MAX_CHANNELS];
134 int16_t mclms_prevvalues[WMALL_MAX_CHANNELS * 2 * 32];
135 int16_t mclms_updates[WMALL_MAX_CHANNELS * 2 * 32];
146 int16_t coefs[MAX_ORDER];
147 int16_t lms_prevvalues[MAX_ORDER * 2];
148 int16_t lms_updates[MAX_ORDER * 2];
156 int is_channel_coded[2];
160 int transient_pos[2];
165 int channel_residues[2][WMALL_BLOCK_MAX_SIZE];
167 int lpc_coefs[2][40];
172 int channel_coeffs[2][WMALL_BLOCK_MAX_SIZE];
176 static av_cold int decode_init(AVCodecContext *avctx)
178 WmallDecodeCtx *s = avctx->priv_data;
179 uint8_t *edata_ptr = avctx->extradata;
180 unsigned int channel_mask;
181 int i, log2_max_num_subframes;
184 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
186 if (avctx->extradata_size >= 18) {
187 s->decode_flags = AV_RL16(edata_ptr + 14);
188 channel_mask = AV_RL32(edata_ptr + 2);
189 s->bits_per_sample = AV_RL16(edata_ptr);
190 if (s->bits_per_sample == 16)
191 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
192 else if (s->bits_per_sample == 24) {
193 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
194 avpriv_report_missing_feature(avctx, "Bit-depth higher than 16");
195 return AVERROR_PATCHWELCOME;
197 av_log(avctx, AV_LOG_ERROR, "Unknown bit-depth: %"PRIu8"\n",
199 return AVERROR_INVALIDDATA;
201 /* dump the extradata */
202 for (i = 0; i < avctx->extradata_size; i++)
203 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
204 av_dlog(avctx, "\n");
207 avpriv_request_sample(avctx, "Unsupported extradata size");
208 return AVERROR_PATCHWELCOME;
212 s->log2_frame_size = av_log2(avctx->block_align) + 4;
215 s->skip_frame = 1; /* skip first frame */
217 s->len_prefix = s->decode_flags & 0x40;
220 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
222 av_assert0(s->samples_per_frame <= WMALL_BLOCK_MAX_SIZE);
224 /* init previous block len */
225 for (i = 0; i < avctx->channels; i++)
226 s->channel[i].prev_block_len = s->samples_per_frame;
229 log2_max_num_subframes = (s->decode_flags & 0x38) >> 3;
230 s->max_num_subframes = 1 << log2_max_num_subframes;
231 s->max_subframe_len_bit = 0;
232 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
234 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
235 s->dynamic_range_compression = s->decode_flags & 0x80;
236 s->bV3RTM = s->decode_flags & 0x100;
238 if (s->max_num_subframes > MAX_SUBFRAMES) {
239 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRIu8"\n",
240 s->max_num_subframes);
241 return AVERROR_INVALIDDATA;
244 s->num_channels = avctx->channels;
246 /* extract lfe channel position */
249 if (channel_mask & 8) {
251 for (mask = 1; mask < 16; mask <<= 1)
252 if (channel_mask & mask)
256 if (s->num_channels < 0) {
257 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %"PRId8"\n",
259 return AVERROR_INVALIDDATA;
260 } else if (s->num_channels > WMALL_MAX_CHANNELS) {
261 avpriv_request_sample(avctx,
262 "More than %d channels", WMALL_MAX_CHANNELS);
263 return AVERROR_PATCHWELCOME;
266 s->frame = av_frame_alloc();
268 return AVERROR(ENOMEM);
270 avctx->channel_layout = channel_mask;
275 * @brief Decode the subframe length.
277 * @param offset sample offset in the frame
278 * @return decoded subframe length on success, < 0 in case of an error
280 static int decode_subframe_length(WmallDecodeCtx *s, int offset)
282 int frame_len_ratio, subframe_len, len;
284 /* no need to read from the bitstream when only one length is possible */
285 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
286 return s->min_samples_per_subframe;
288 len = av_log2(s->max_num_subframes - 1) + 1;
289 frame_len_ratio = get_bits(&s->gb, len);
290 subframe_len = s->min_samples_per_subframe * (frame_len_ratio + 1);
292 /* sanity check the length */
293 if (subframe_len < s->min_samples_per_subframe ||
294 subframe_len > s->samples_per_frame) {
295 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
297 return AVERROR_INVALIDDATA;
303 * @brief Decode how the data in the frame is split into subframes.
304 * Every WMA frame contains the encoded data for a fixed number of
305 * samples per channel. The data for every channel might be split
306 * into several subframes. This function will reconstruct the list of
307 * subframes for every channel.
309 * If the subframes are not evenly split, the algorithm estimates the
310 * channels with the lowest number of total samples.
311 * Afterwards, for each of these channels a bit is read from the
312 * bitstream that indicates if the channel contains a subframe with the
313 * next subframe size that is going to be read from the bitstream or not.
314 * If a channel contains such a subframe, the subframe size gets added to
315 * the channel's subframe list.
316 * The algorithm repeats these steps until the frame is properly divided
317 * between the individual channels.
320 * @return 0 on success, < 0 in case of an error
322 static int decode_tilehdr(WmallDecodeCtx *s)
324 uint16_t num_samples[WMALL_MAX_CHANNELS] = { 0 }; /* sum of samples for all currently known subframes of a channel */
325 uint8_t contains_subframe[WMALL_MAX_CHANNELS]; /* flag indicating if a channel contains the current subframe */
326 int channels_for_cur_subframe = s->num_channels; /* number of channels that contain the current subframe */
327 int fixed_channel_layout = 0; /* flag indicating that all channels use the same subfra2me offsets and sizes */
328 int min_channel_len = 0; /* smallest sum of samples (channels with this length will be processed first) */
331 /* reset tiling information */
332 for (c = 0; c < s->num_channels; c++)
333 s->channel[c].num_subframes = 0;
335 tile_aligned = get_bits1(&s->gb);
336 if (s->max_num_subframes == 1 || tile_aligned)
337 fixed_channel_layout = 1;
339 /* loop until the frame data is split between the subframes */
341 int subframe_len, in_use = 0;
343 /* check which channels contain the subframe */
344 for (c = 0; c < s->num_channels; c++) {
345 if (num_samples[c] == min_channel_len) {
346 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
347 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) {
348 contains_subframe[c] = in_use = 1;
350 if (get_bits1(&s->gb))
351 contains_subframe[c] = in_use = 1;
354 contains_subframe[c] = 0;
358 av_log(s->avctx, AV_LOG_ERROR,
359 "Found empty subframe\n");
360 return AVERROR_INVALIDDATA;
363 /* get subframe length, subframe_len == 0 is not allowed */
364 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
365 return AVERROR_INVALIDDATA;
366 /* add subframes to the individual channels and find new min_channel_len */
367 min_channel_len += subframe_len;
368 for (c = 0; c < s->num_channels; c++) {
369 WmallChannelCtx *chan = &s->channel[c];
371 if (contains_subframe[c]) {
372 if (chan->num_subframes >= MAX_SUBFRAMES) {
373 av_log(s->avctx, AV_LOG_ERROR,
374 "broken frame: num subframes > 31\n");
375 return AVERROR_INVALIDDATA;
377 chan->subframe_len[chan->num_subframes] = subframe_len;
378 num_samples[c] += subframe_len;
379 ++chan->num_subframes;
380 if (num_samples[c] > s->samples_per_frame) {
381 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
382 "channel len(%"PRIu16") > samples_per_frame(%"PRIu16")\n",
383 num_samples[c], s->samples_per_frame);
384 return AVERROR_INVALIDDATA;
386 } else if (num_samples[c] <= min_channel_len) {
387 if (num_samples[c] < min_channel_len) {
388 channels_for_cur_subframe = 0;
389 min_channel_len = num_samples[c];
391 ++channels_for_cur_subframe;
394 } while (min_channel_len < s->samples_per_frame);
396 for (c = 0; c < s->num_channels; c++) {
398 for (i = 0; i < s->channel[c].num_subframes; i++) {
399 s->channel[c].subframe_offsets[i] = offset;
400 offset += s->channel[c].subframe_len[i];
407 static void decode_ac_filter(WmallDecodeCtx *s)
410 s->acfilter_order = get_bits(&s->gb, 4) + 1;
411 s->acfilter_scaling = get_bits(&s->gb, 4);
413 for (i = 0; i < s->acfilter_order; i++)
414 s->acfilter_coeffs[i] = (s->acfilter_scaling ?
415 get_bits(&s->gb, s->acfilter_scaling) : 0) + 1;
418 static void decode_mclms(WmallDecodeCtx *s)
420 s->mclms_order = (get_bits(&s->gb, 4) + 1) * 2;
421 s->mclms_scaling = get_bits(&s->gb, 4);
422 if (get_bits1(&s->gb)) {
423 int i, send_coef_bits;
424 int cbits = av_log2(s->mclms_scaling + 1);
425 if (1 << cbits < s->mclms_scaling + 1)
428 send_coef_bits = (cbits ? get_bits(&s->gb, cbits) : 0) + 2;
430 for (i = 0; i < s->mclms_order * s->num_channels * s->num_channels; i++)
431 s->mclms_coeffs[i] = get_bits(&s->gb, send_coef_bits);
433 for (i = 0; i < s->num_channels; i++) {
435 for (c = 0; c < i; c++)
436 s->mclms_coeffs_cur[i * s->num_channels + c] = get_bits(&s->gb, send_coef_bits);
441 static int decode_cdlms(WmallDecodeCtx *s)
444 int cdlms_send_coef = get_bits1(&s->gb);
446 for (c = 0; c < s->num_channels; c++) {
447 s->cdlms_ttl[c] = get_bits(&s->gb, 3) + 1;
448 for (i = 0; i < s->cdlms_ttl[c]; i++) {
449 s->cdlms[c][i].order = (get_bits(&s->gb, 7) + 1) * 8;
450 if (s->cdlms[c][i].order > MAX_ORDER) {
451 av_log(s->avctx, AV_LOG_ERROR,
452 "Order[%d][%d] %d > max (%d), not supported\n",
453 c, i, s->cdlms[c][i].order, MAX_ORDER);
454 s->cdlms[0][0].order = 0;
455 return AVERROR_INVALIDDATA;
459 for (i = 0; i < s->cdlms_ttl[c]; i++)
460 s->cdlms[c][i].scaling = get_bits(&s->gb, 4);
462 if (cdlms_send_coef) {
463 for (i = 0; i < s->cdlms_ttl[c]; i++) {
464 int cbits, shift_l, shift_r, j;
465 cbits = av_log2(s->cdlms[c][i].order);
466 if ((1 << cbits) < s->cdlms[c][i].order)
468 s->cdlms[c][i].coefsend = get_bits(&s->gb, cbits) + 1;
470 cbits = av_log2(s->cdlms[c][i].scaling + 1);
471 if ((1 << cbits) < s->cdlms[c][i].scaling + 1)
474 s->cdlms[c][i].bitsend = get_bits(&s->gb, cbits) + 2;
475 shift_l = 32 - s->cdlms[c][i].bitsend;
476 shift_r = 32 - s->cdlms[c][i].scaling - 2;
477 for (j = 0; j < s->cdlms[c][i].coefsend; j++)
478 s->cdlms[c][i].coefs[j] =
479 (get_bits(&s->gb, s->cdlms[c][i].bitsend) << shift_l) >> shift_r;
487 static int decode_channel_residues(WmallDecodeCtx *s, int ch, int tile_size)
490 unsigned int ave_mean;
491 s->transient[ch] = get_bits1(&s->gb);
492 if (s->transient[ch]) {
493 s->transient_pos[ch] = get_bits(&s->gb, av_log2(tile_size));
494 if (s->transient_pos[ch])
495 s->transient[ch] = 0;
496 s->channel[ch].transient_counter =
497 FFMAX(s->channel[ch].transient_counter, s->samples_per_frame / 2);
498 } else if (s->channel[ch].transient_counter)
499 s->transient[ch] = 1;
501 if (s->seekable_tile) {
502 ave_mean = get_bits(&s->gb, s->bits_per_sample);
503 s->ave_sum[ch] = ave_mean << (s->movave_scaling + 1);
506 if (s->seekable_tile) {
507 if (s->do_inter_ch_decorr)
508 s->channel_residues[ch][0] = get_sbits(&s->gb, s->bits_per_sample + 1);
510 s->channel_residues[ch][0] = get_sbits(&s->gb, s->bits_per_sample);
513 for (; i < tile_size; i++) {
514 int quo = 0, rem, rem_bits, residue;
515 while(get_bits1(&s->gb)) {
517 if (get_bits_left(&s->gb) <= 0)
521 quo += get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1);
523 ave_mean = (s->ave_sum[ch] + (1 << s->movave_scaling)) >> (s->movave_scaling + 1);
527 rem_bits = av_ceil_log2(ave_mean);
528 rem = rem_bits ? get_bits_long(&s->gb, rem_bits) : 0;
529 residue = (quo << rem_bits) + rem;
532 s->ave_sum[ch] = residue + s->ave_sum[ch] -
533 (s->ave_sum[ch] >> s->movave_scaling);
536 residue = -(residue >> 1) - 1;
538 residue = residue >> 1;
539 s->channel_residues[ch][i] = residue;
546 static void decode_lpc(WmallDecodeCtx *s)
549 s->lpc_order = get_bits(&s->gb, 5) + 1;
550 s->lpc_scaling = get_bits(&s->gb, 4);
551 s->lpc_intbits = get_bits(&s->gb, 3) + 1;
552 cbits = s->lpc_scaling + s->lpc_intbits;
553 for (ch = 0; ch < s->num_channels; ch++)
554 for (i = 0; i < s->lpc_order; i++)
555 s->lpc_coefs[ch][i] = get_sbits(&s->gb, cbits);
558 static void clear_codec_buffers(WmallDecodeCtx *s)
562 memset(s->acfilter_coeffs, 0, sizeof(s->acfilter_coeffs));
563 memset(s->acfilter_prevvalues, 0, sizeof(s->acfilter_prevvalues));
564 memset(s->lpc_coefs, 0, sizeof(s->lpc_coefs));
566 memset(s->mclms_coeffs, 0, sizeof(s->mclms_coeffs));
567 memset(s->mclms_coeffs_cur, 0, sizeof(s->mclms_coeffs_cur));
568 memset(s->mclms_prevvalues, 0, sizeof(s->mclms_prevvalues));
569 memset(s->mclms_updates, 0, sizeof(s->mclms_updates));
571 for (ich = 0; ich < s->num_channels; ich++) {
572 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++) {
573 memset(s->cdlms[ich][ilms].coefs, 0,
574 sizeof(s->cdlms[ich][ilms].coefs));
575 memset(s->cdlms[ich][ilms].lms_prevvalues, 0,
576 sizeof(s->cdlms[ich][ilms].lms_prevvalues));
577 memset(s->cdlms[ich][ilms].lms_updates, 0,
578 sizeof(s->cdlms[ich][ilms].lms_updates));
585 * @brief Reset filter parameters and transient area at new seekable tile.
587 static void reset_codec(WmallDecodeCtx *s)
590 s->mclms_recent = s->mclms_order * s->num_channels;
591 for (ich = 0; ich < s->num_channels; ich++) {
592 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++)
593 s->cdlms[ich][ilms].recent = s->cdlms[ich][ilms].order;
594 /* first sample of a seekable subframe is considered as the starting of
595 a transient area which is samples_per_frame samples long */
596 s->channel[ich].transient_counter = s->samples_per_frame;
597 s->transient[ich] = 1;
598 s->transient_pos[ich] = 0;
602 static void mclms_update(WmallDecodeCtx *s, int icoef, int *pred)
604 int i, j, ich, pred_error;
605 int order = s->mclms_order;
606 int num_channels = s->num_channels;
607 int range = 1 << (s->bits_per_sample - 1);
609 for (ich = 0; ich < num_channels; ich++) {
610 pred_error = s->channel_residues[ich][icoef] - pred[ich];
611 if (pred_error > 0) {
612 for (i = 0; i < order * num_channels; i++)
613 s->mclms_coeffs[i + ich * order * num_channels] +=
614 s->mclms_updates[s->mclms_recent + i];
615 for (j = 0; j < ich; j++) {
616 if (s->channel_residues[j][icoef] > 0)
617 s->mclms_coeffs_cur[ich * num_channels + j] += 1;
618 else if (s->channel_residues[j][icoef] < 0)
619 s->mclms_coeffs_cur[ich * num_channels + j] -= 1;
621 } else if (pred_error < 0) {
622 for (i = 0; i < order * num_channels; i++)
623 s->mclms_coeffs[i + ich * order * num_channels] -=
624 s->mclms_updates[s->mclms_recent + i];
625 for (j = 0; j < ich; j++) {
626 if (s->channel_residues[j][icoef] > 0)
627 s->mclms_coeffs_cur[ich * num_channels + j] -= 1;
628 else if (s->channel_residues[j][icoef] < 0)
629 s->mclms_coeffs_cur[ich * num_channels + j] += 1;
634 for (ich = num_channels - 1; ich >= 0; ich--) {
636 s->mclms_prevvalues[s->mclms_recent] = s->channel_residues[ich][icoef];
637 if (s->channel_residues[ich][icoef] > range - 1)
638 s->mclms_prevvalues[s->mclms_recent] = range - 1;
639 else if (s->channel_residues[ich][icoef] < -range)
640 s->mclms_prevvalues[s->mclms_recent] = -range;
642 s->mclms_updates[s->mclms_recent] = 0;
643 if (s->channel_residues[ich][icoef] > 0)
644 s->mclms_updates[s->mclms_recent] = 1;
645 else if (s->channel_residues[ich][icoef] < 0)
646 s->mclms_updates[s->mclms_recent] = -1;
649 if (s->mclms_recent == 0) {
650 memcpy(&s->mclms_prevvalues[order * num_channels],
652 2 * order * num_channels);
653 memcpy(&s->mclms_updates[order * num_channels],
655 2 * order * num_channels);
656 s->mclms_recent = num_channels * order;
660 static void mclms_predict(WmallDecodeCtx *s, int icoef, int *pred)
663 int order = s->mclms_order;
664 int num_channels = s->num_channels;
666 for (ich = 0; ich < num_channels; ich++) {
668 if (!s->is_channel_coded[ich])
670 for (i = 0; i < order * num_channels; i++)
671 pred[ich] += s->mclms_prevvalues[i + s->mclms_recent] *
672 s->mclms_coeffs[i + order * num_channels * ich];
673 for (i = 0; i < ich; i++)
674 pred[ich] += s->channel_residues[i][icoef] *
675 s->mclms_coeffs_cur[i + num_channels * ich];
676 pred[ich] += 1 << s->mclms_scaling - 1;
677 pred[ich] >>= s->mclms_scaling;
678 s->channel_residues[ich][icoef] += pred[ich];
682 static void revert_mclms(WmallDecodeCtx *s, int tile_size)
684 int icoef, pred[WMALL_MAX_CHANNELS] = { 0 };
685 for (icoef = 0; icoef < tile_size; icoef++) {
686 mclms_predict(s, icoef, pred);
687 mclms_update(s, icoef, pred);
691 static int lms_predict(WmallDecodeCtx *s, int ich, int ilms)
694 int recent = s->cdlms[ich][ilms].recent;
696 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
697 pred += s->cdlms[ich][ilms].coefs[icoef] *
698 s->cdlms[ich][ilms].lms_prevvalues[icoef + recent];
703 static void lms_update(WmallDecodeCtx *s, int ich, int ilms,
704 int input, int residue)
707 int recent = s->cdlms[ich][ilms].recent;
708 int range = 1 << s->bits_per_sample - 1;
711 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
712 s->cdlms[ich][ilms].coefs[icoef] -=
713 s->cdlms[ich][ilms].lms_updates[icoef + recent];
714 } else if (residue > 0) {
715 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
716 s->cdlms[ich][ilms].coefs[icoef] +=
717 s->cdlms[ich][ilms].lms_updates[icoef + recent];
723 memcpy(&s->cdlms[ich][ilms].lms_prevvalues[s->cdlms[ich][ilms].order],
724 s->cdlms[ich][ilms].lms_prevvalues,
725 2 * s->cdlms[ich][ilms].order);
726 memcpy(&s->cdlms[ich][ilms].lms_updates[s->cdlms[ich][ilms].order],
727 s->cdlms[ich][ilms].lms_updates,
728 2 * s->cdlms[ich][ilms].order);
729 recent = s->cdlms[ich][ilms].order - 1;
732 s->cdlms[ich][ilms].lms_prevvalues[recent] = av_clip(input, -range, range - 1);
734 s->cdlms[ich][ilms].lms_updates[recent] = 0;
736 s->cdlms[ich][ilms].lms_updates[recent] = -s->update_speed[ich];
738 s->cdlms[ich][ilms].lms_updates[recent] = s->update_speed[ich];
740 s->cdlms[ich][ilms].lms_updates[recent + (s->cdlms[ich][ilms].order >> 4)] >>= 2;
741 s->cdlms[ich][ilms].lms_updates[recent + (s->cdlms[ich][ilms].order >> 3)] >>= 1;
742 s->cdlms[ich][ilms].recent = recent;
745 static void use_high_update_speed(WmallDecodeCtx *s, int ich)
747 int ilms, recent, icoef;
748 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
749 recent = s->cdlms[ich][ilms].recent;
750 if (s->update_speed[ich] == 16)
753 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
754 s->cdlms[ich][ilms].lms_updates[icoef + recent] *= 2;
756 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
757 s->cdlms[ich][ilms].lms_updates[icoef] *= 2;
760 s->update_speed[ich] = 16;
763 static void use_normal_update_speed(WmallDecodeCtx *s, int ich)
765 int ilms, recent, icoef;
766 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
767 recent = s->cdlms[ich][ilms].recent;
768 if (s->update_speed[ich] == 8)
771 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
772 s->cdlms[ich][ilms].lms_updates[icoef + recent] /= 2;
774 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
775 s->cdlms[ich][ilms].lms_updates[icoef] /= 2;
777 s->update_speed[ich] = 8;
780 static void revert_cdlms(WmallDecodeCtx *s, int ch,
781 int coef_begin, int coef_end)
783 int icoef, pred, ilms, num_lms, residue, input;
785 num_lms = s->cdlms_ttl[ch];
786 for (ilms = num_lms - 1; ilms >= 0; ilms--) {
787 for (icoef = coef_begin; icoef < coef_end; icoef++) {
788 pred = 1 << (s->cdlms[ch][ilms].scaling - 1);
789 residue = s->channel_residues[ch][icoef];
790 pred += lms_predict(s, ch, ilms);
791 input = residue + (pred >> s->cdlms[ch][ilms].scaling);
792 lms_update(s, ch, ilms, input, residue);
793 s->channel_residues[ch][icoef] = input;
798 static void revert_inter_ch_decorr(WmallDecodeCtx *s, int tile_size)
800 if (s->num_channels != 2)
802 else if (s->is_channel_coded[0] || s->is_channel_coded[1]) {
804 for (icoef = 0; icoef < tile_size; icoef++) {
805 s->channel_residues[0][icoef] -= s->channel_residues[1][icoef] >> 1;
806 s->channel_residues[1][icoef] += s->channel_residues[0][icoef];
811 static void revert_acfilter(WmallDecodeCtx *s, int tile_size)
814 int64_t *filter_coeffs = s->acfilter_coeffs;
815 int scaling = s->acfilter_scaling;
816 int order = s->acfilter_order;
818 for (ich = 0; ich < s->num_channels; ich++) {
819 int *prevvalues = s->acfilter_prevvalues[ich];
820 for (i = 0; i < order; i++) {
822 for (j = 0; j < order; j++) {
824 pred += filter_coeffs[j] * prevvalues[j - i];
826 pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
829 s->channel_residues[ich][i] += pred;
831 for (i = order; i < tile_size; i++) {
833 for (j = 0; j < order; j++)
834 pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
836 s->channel_residues[ich][i] += pred;
838 for (j = 0; j < order; j++)
839 prevvalues[j] = s->channel_residues[ich][tile_size - j - 1];
843 static int decode_subframe(WmallDecodeCtx *s)
845 int offset = s->samples_per_frame;
846 int subframe_len = s->samples_per_frame;
847 int total_samples = s->samples_per_frame * s->num_channels;
848 int i, j, rawpcm_tile, padding_zeroes, res;
850 s->subframe_offset = get_bits_count(&s->gb);
852 /* reset channel context and find the next block offset and size
853 == the next block of the channel with the smallest number of
855 for (i = 0; i < s->num_channels; i++) {
856 if (offset > s->channel[i].decoded_samples) {
857 offset = s->channel[i].decoded_samples;
859 s->channel[i].subframe_len[s->channel[i].cur_subframe];
863 /* get a list of all channels that contain the estimated block */
864 s->channels_for_cur_subframe = 0;
865 for (i = 0; i < s->num_channels; i++) {
866 const int cur_subframe = s->channel[i].cur_subframe;
867 /* subtract already processed samples */
868 total_samples -= s->channel[i].decoded_samples;
870 /* and count if there are multiple subframes that match our profile */
871 if (offset == s->channel[i].decoded_samples &&
872 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
873 total_samples -= s->channel[i].subframe_len[cur_subframe];
874 s->channel[i].decoded_samples +=
875 s->channel[i].subframe_len[cur_subframe];
876 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
877 ++s->channels_for_cur_subframe;
881 /* check if the frame will be complete after processing the
884 s->parsed_all_subframes = 1;
887 s->seekable_tile = get_bits1(&s->gb);
888 if (s->seekable_tile) {
889 clear_codec_buffers(s);
891 s->do_arith_coding = get_bits1(&s->gb);
892 if (s->do_arith_coding) {
893 avpriv_request_sample(s->avctx, "Arithmetic coding");
894 return AVERROR_PATCHWELCOME;
896 s->do_ac_filter = get_bits1(&s->gb);
897 s->do_inter_ch_decorr = get_bits1(&s->gb);
898 s->do_mclms = get_bits1(&s->gb);
906 if ((res = decode_cdlms(s)) < 0)
908 s->movave_scaling = get_bits(&s->gb, 3);
909 s->quant_stepsize = get_bits(&s->gb, 8) + 1;
912 } else if (!s->cdlms[0][0].order) {
913 av_log(s->avctx, AV_LOG_DEBUG,
914 "Waiting for seekable tile\n");
915 av_frame_unref(s->frame);
919 rawpcm_tile = get_bits1(&s->gb);
921 for (i = 0; i < s->num_channels; i++)
922 s->is_channel_coded[i] = 1;
925 for (i = 0; i < s->num_channels; i++)
926 s->is_channel_coded[i] = get_bits1(&s->gb);
930 s->do_lpc = get_bits1(&s->gb);
933 avpriv_request_sample(s->avctx, "Expect wrong output since "
934 "inverse LPC filter");
941 if (get_bits1(&s->gb))
942 padding_zeroes = get_bits(&s->gb, 5);
947 int bits = s->bits_per_sample - padding_zeroes;
949 av_log(s->avctx, AV_LOG_ERROR,
950 "Invalid number of padding bits in raw PCM tile\n");
951 return AVERROR_INVALIDDATA;
953 av_dlog(s->avctx, "RAWPCM %d bits per sample. "
954 "total %d bits, remain=%d\n", bits,
955 bits * s->num_channels * subframe_len, get_bits_count(&s->gb));
956 for (i = 0; i < s->num_channels; i++)
957 for (j = 0; j < subframe_len; j++)
958 s->channel_coeffs[i][j] = get_sbits(&s->gb, bits);
960 for (i = 0; i < s->num_channels; i++)
961 if (s->is_channel_coded[i]) {
962 decode_channel_residues(s, i, subframe_len);
963 if (s->seekable_tile)
964 use_high_update_speed(s, i);
966 use_normal_update_speed(s, i);
967 revert_cdlms(s, i, 0, subframe_len);
969 memset(s->channel_residues[i], 0, sizeof(**s->channel_residues) * subframe_len);
973 revert_mclms(s, subframe_len);
974 if (s->do_inter_ch_decorr)
975 revert_inter_ch_decorr(s, subframe_len);
977 revert_acfilter(s, subframe_len);
980 if (s->quant_stepsize != 1)
981 for (i = 0; i < s->num_channels; i++)
982 for (j = 0; j < subframe_len; j++)
983 s->channel_residues[i][j] *= s->quant_stepsize;
985 /* Write to proper output buffer depending on bit-depth */
986 for (i = 0; i < s->channels_for_cur_subframe; i++) {
987 int c = s->channel_indexes_for_cur_subframe[i];
988 int subframe_len = s->channel[c].subframe_len[s->channel[c].cur_subframe];
990 for (j = 0; j < subframe_len; j++) {
991 if (s->bits_per_sample == 16) {
992 *s->samples_16[c]++ = (int16_t) s->channel_residues[c][j] << padding_zeroes;
994 *s->samples_32[c]++ = s->channel_residues[c][j] << padding_zeroes;
999 /* handled one subframe */
1000 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1001 int c = s->channel_indexes_for_cur_subframe[i];
1002 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1003 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1004 return AVERROR_INVALIDDATA;
1006 ++s->channel[c].cur_subframe;
1012 * @brief Decode one WMA frame.
1013 * @param s codec context
1014 * @return 0 if the trailer bit indicates that this is the last frame,
1015 * 1 if there are additional frames
1017 static int decode_frame(WmallDecodeCtx *s)
1019 GetBitContext* gb = &s->gb;
1020 int more_frames = 0, len = 0, i, ret;
1022 s->frame->nb_samples = s->samples_per_frame;
1023 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) {
1024 /* return an error if no frame could be decoded at all */
1025 av_log(s->avctx, AV_LOG_ERROR,
1026 "not enough space for the output samples\n");
1030 for (i = 0; i < s->num_channels; i++) {
1031 s->samples_16[i] = (int16_t *)s->frame->extended_data[i];
1032 s->samples_32[i] = (int32_t *)s->frame->extended_data[i];
1035 /* get frame length */
1037 len = get_bits(gb, s->log2_frame_size);
1039 /* decode tile information */
1040 if (decode_tilehdr(s)) {
1046 if (s->dynamic_range_compression)
1047 s->drc_gain = get_bits(gb, 8);
1049 /* no idea what these are for, might be the number of samples
1050 that need to be skipped at the beginning or end of a stream */
1051 if (get_bits1(gb)) {
1054 /* usually true for the first frame */
1055 if (get_bits1(gb)) {
1056 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1057 av_dlog(s->avctx, "start skip: %i\n", skip);
1060 /* sometimes true for the last frame */
1061 if (get_bits1(gb)) {
1062 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1063 av_dlog(s->avctx, "end skip: %i\n", skip);
1068 /* reset subframe states */
1069 s->parsed_all_subframes = 0;
1070 for (i = 0; i < s->num_channels; i++) {
1071 s->channel[i].decoded_samples = 0;
1072 s->channel[i].cur_subframe = 0;
1075 /* decode all subframes */
1076 while (!s->parsed_all_subframes) {
1077 if (decode_subframe(s) < 0) {
1083 av_dlog(s->avctx, "Frame done\n");
1088 if (s->len_prefix) {
1089 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1090 /* FIXME: not sure if this is always an error */
1091 av_log(s->avctx, AV_LOG_ERROR,
1092 "frame[%"PRIu32"] would have to skip %i bits\n",
1094 len - (get_bits_count(gb) - s->frame_offset) - 1);
1099 /* skip the rest of the frame data */
1100 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1103 /* decode trailer bit */
1104 more_frames = get_bits1(gb);
1110 * @brief Calculate remaining input buffer length.
1111 * @param s codec context
1112 * @param gb bitstream reader context
1113 * @return remaining size in bits
1115 static int remaining_bits(WmallDecodeCtx *s, GetBitContext *gb)
1117 return s->buf_bit_size - get_bits_count(gb);
1121 * @brief Fill the bit reservoir with a (partial) frame.
1122 * @param s codec context
1123 * @param gb bitstream reader context
1124 * @param len length of the partial frame
1125 * @param append decides whether to reset the buffer or not
1127 static void save_bits(WmallDecodeCtx *s, GetBitContext* gb, int len,
1133 /* when the frame data does not need to be concatenated, the input buffer
1134 is reset and additional bits from the previous frame are copied
1135 and skipped later so that a fast byte copy is possible */
1138 s->frame_offset = get_bits_count(gb) & 7;
1139 s->num_saved_bits = s->frame_offset;
1140 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1143 buflen = (s->num_saved_bits + len + 8) >> 3;
1145 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1146 avpriv_request_sample(s->avctx, "Too small input buffer");
1151 s->num_saved_bits += len;
1153 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1156 int align = 8 - (get_bits_count(gb) & 7);
1157 align = FFMIN(align, len);
1158 put_bits(&s->pb, align, get_bits(gb, align));
1160 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1162 skip_bits_long(gb, len);
1165 flush_put_bits(&tmp);
1167 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1168 skip_bits(&s->gb, s->frame_offset);
1171 static int decode_packet(AVCodecContext *avctx, void *data, int *got_frame_ptr,
1174 WmallDecodeCtx *s = avctx->priv_data;
1175 GetBitContext* gb = &s->pgb;
1176 const uint8_t* buf = avpkt->data;
1177 int buf_size = avpkt->size;
1178 int num_bits_prev_frame, packet_sequence_number, spliced_packet;
1180 s->frame->nb_samples = 0;
1182 if (s->packet_done || s->packet_loss) {
1185 /* sanity check for the buffer length */
1186 if (buf_size < avctx->block_align)
1189 s->next_packet_start = buf_size - avctx->block_align;
1190 buf_size = avctx->block_align;
1191 s->buf_bit_size = buf_size << 3;
1193 /* parse packet header */
1194 init_get_bits(gb, buf, s->buf_bit_size);
1195 packet_sequence_number = get_bits(gb, 4);
1196 skip_bits(gb, 1); // Skip seekable_frame_in_packet, currently ununused
1197 spliced_packet = get_bits1(gb);
1199 avpriv_request_sample(avctx, "Bitstream splicing");
1201 /* get number of bits that need to be added to the previous frame */
1202 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1204 /* check for packet loss */
1205 if (!s->packet_loss &&
1206 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1208 av_log(avctx, AV_LOG_ERROR,
1209 "Packet loss detected! seq %"PRIx8" vs %x\n",
1210 s->packet_sequence_number, packet_sequence_number);
1212 s->packet_sequence_number = packet_sequence_number;
1214 if (num_bits_prev_frame > 0) {
1215 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1216 if (num_bits_prev_frame >= remaining_packet_bits) {
1217 num_bits_prev_frame = remaining_packet_bits;
1221 /* Append the previous frame data to the remaining data from the
1222 * previous packet to create a full frame. */
1223 save_bits(s, gb, num_bits_prev_frame, 1);
1225 /* decode the cross packet frame if it is valid */
1226 if (num_bits_prev_frame < remaining_packet_bits && !s->packet_loss)
1228 } else if (s->num_saved_bits - s->frame_offset) {
1229 av_dlog(avctx, "ignoring %x previously saved bits\n",
1230 s->num_saved_bits - s->frame_offset);
1233 if (s->packet_loss) {
1234 /* Reset number of saved bits so that the decoder does not start
1235 * to decode incomplete frames in the s->len_prefix == 0 case. */
1236 s->num_saved_bits = 0;
1238 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1244 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1245 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1246 skip_bits(gb, s->packet_offset);
1248 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1249 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1250 frame_size <= remaining_bits(s, gb)) {
1251 save_bits(s, gb, frame_size, 0);
1252 s->packet_done = !decode_frame(s);
1253 } else if (!s->len_prefix
1254 && s->num_saved_bits > get_bits_count(&s->gb)) {
1255 /* when the frames do not have a length prefix, we don't know the
1256 * compressed length of the individual frames however, we know what
1257 * part of a new packet belongs to the previous frame therefore we
1258 * save the incoming packet first, then we append the "previous
1259 * frame" data from the next packet so that we get a buffer that
1260 * only contains full frames */
1261 s->packet_done = !decode_frame(s);
1267 if (s->packet_done && !s->packet_loss &&
1268 remaining_bits(s, gb) > 0) {
1269 /* save the rest of the data so that it can be decoded
1270 * with the next packet */
1271 save_bits(s, gb, remaining_bits(s, gb), 0);
1274 *got_frame_ptr = s->frame->nb_samples > 0;
1275 av_frame_move_ref(data, s->frame);
1277 s->packet_offset = get_bits_count(gb) & 7;
1279 return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3;
1282 static void flush(AVCodecContext *avctx)
1284 WmallDecodeCtx *s = avctx->priv_data;
1287 s->num_saved_bits = 0;
1288 s->frame_offset = 0;
1289 s->next_packet_start = 0;
1290 s->cdlms[0][0].order = 0;
1291 s->frame->nb_samples = 0;
1292 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1295 static av_cold int decode_close(AVCodecContext *avctx)
1297 WmallDecodeCtx *s = avctx->priv_data;
1299 av_frame_free(&s->frame);
1304 AVCodec ff_wmalossless_decoder = {
1305 .name = "wmalossless",
1306 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio Lossless"),
1307 .type = AVMEDIA_TYPE_AUDIO,
1308 .id = AV_CODEC_ID_WMALOSSLESS,
1309 .priv_data_size = sizeof(WmallDecodeCtx),
1310 .init = decode_init,
1311 .close = decode_close,
1312 .decode = decode_packet,
1314 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1315 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
1317 AV_SAMPLE_FMT_NONE },