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
25 #include "libavutil/attributes.h"
31 #include "wma_common.h"
33 /** current decoder limitations */
34 #define WMALL_MAX_CHANNELS 8 ///< max number of handled channels
35 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
36 #define MAX_BANDS 29 ///< max number of scale factor bands
37 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
40 #define WMALL_BLOCK_MIN_BITS 6 ///< log2 of min block size
41 #define WMALL_BLOCK_MAX_BITS 12 ///< log2 of max block size
42 #define WMALL_BLOCK_MAX_SIZE (1 << WMALL_BLOCK_MAX_BITS) ///< maximum block size
43 #define WMALL_BLOCK_SIZES (WMALL_BLOCK_MAX_BITS - WMALL_BLOCK_MIN_BITS + 1) ///< possible block sizes
47 * @brief frame-specific decoder context for a single channel
50 int16_t prev_block_len; ///< length of the previous block
51 uint8_t transmit_coefs;
52 uint8_t num_subframes;
53 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
54 uint16_t subframe_offsets[MAX_SUBFRAMES]; ///< subframe positions in the current frame
55 uint8_t cur_subframe; ///< current subframe number
56 uint16_t decoded_samples; ///< number of already processed samples
57 int quant_step; ///< quantization step for the current subframe
58 int transient_counter; ///< number of transient samples from the beginning of the transient zone
62 * @brief main decoder context
64 typedef struct WmallDecodeCtx {
65 /* generic decoder variables */
66 AVCodecContext *avctx;
68 uint8_t frame_data[MAX_FRAMESIZE + FF_INPUT_BUFFER_PADDING_SIZE]; ///< compressed frame data
69 PutBitContext pb; ///< context for filling the frame_data buffer
71 /* frame size dependent frame information (set during initialization) */
72 uint32_t decode_flags; ///< used compression features
73 int len_prefix; ///< frame is prefixed with its length
74 int dynamic_range_compression; ///< frame contains DRC data
75 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
76 uint16_t samples_per_frame; ///< number of samples to output
77 uint16_t log2_frame_size;
78 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
79 int8_t lfe_channel; ///< lfe channel index
80 uint8_t max_num_subframes;
81 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
82 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
83 uint16_t min_samples_per_subframe;
85 /* packet decode state */
86 GetBitContext pgb; ///< bitstream reader context for the packet
87 int next_packet_start; ///< start offset of the next WMA packet in the demuxer packet
88 uint8_t packet_offset; ///< offset to the frame in the packet
89 uint8_t packet_sequence_number; ///< current packet number
90 int num_saved_bits; ///< saved number of bits
91 int frame_offset; ///< frame offset in the bit reservoir
92 int subframe_offset; ///< subframe offset in the bit reservoir
93 uint8_t packet_loss; ///< set in case of bitstream error
94 uint8_t packet_done; ///< set when a packet is fully decoded
96 /* frame decode state */
97 uint32_t frame_num; ///< current frame number (not used for decoding)
98 GetBitContext gb; ///< bitstream reader context
99 int buf_bit_size; ///< buffer size in bits
100 int16_t *samples_16[WMALL_MAX_CHANNELS]; ///< current samplebuffer pointer (16-bit)
101 int32_t *samples_32[WMALL_MAX_CHANNELS]; ///< current samplebuffer pointer (24-bit)
102 uint8_t drc_gain; ///< gain for the DRC tool
103 int8_t skip_frame; ///< skip output step
104 int8_t parsed_all_subframes; ///< all subframes decoded?
106 /* subframe/block decode state */
107 int16_t subframe_len; ///< current subframe length
108 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
109 int8_t channel_indexes_for_cur_subframe[WMALL_MAX_CHANNELS];
111 WmallChannelCtx channel[WMALL_MAX_CHANNELS]; ///< per channel data
113 // WMA Lossless-specific
115 uint8_t do_arith_coding;
116 uint8_t do_ac_filter;
117 uint8_t do_inter_ch_decorr;
121 int8_t acfilter_order;
122 int8_t acfilter_scaling;
123 int64_t acfilter_coeffs[16];
124 int acfilter_prevvalues[2][16];
127 int8_t mclms_scaling;
128 int16_t mclms_coeffs[128];
129 int16_t mclms_coeffs_cur[4];
130 int16_t mclms_prevvalues[64];
131 int16_t mclms_updates[64];
142 int16_t coefs[MAX_ORDER];
143 int16_t lms_prevvalues[MAX_ORDER * 2];
144 int16_t lms_updates[MAX_ORDER * 2];
152 int is_channel_coded[2];
156 int transient_pos[2];
161 int channel_residues[2][2048];
163 int lpc_coefs[2][40];
168 int channel_coeffs[2][2048];
172 static av_cold int decode_init(AVCodecContext *avctx)
174 WmallDecodeCtx *s = avctx->priv_data;
175 uint8_t *edata_ptr = avctx->extradata;
176 unsigned int channel_mask;
177 int i, log2_max_num_subframes;
180 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
182 if (avctx->extradata_size >= 18) {
183 s->decode_flags = AV_RL16(edata_ptr + 14);
184 channel_mask = AV_RL32(edata_ptr + 2);
185 s->bits_per_sample = AV_RL16(edata_ptr);
186 if (s->bits_per_sample == 16)
187 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
188 else if (s->bits_per_sample == 24) {
189 avctx->sample_fmt = AV_SAMPLE_FMT_S32;
190 av_log_missing_feature(avctx, "bit-depth higher than 16", 0);
191 return AVERROR_PATCHWELCOME;
193 av_log(avctx, AV_LOG_ERROR, "Unknown bit-depth: %d\n",
195 return AVERROR_INVALIDDATA;
197 /* dump the extradata */
198 for (i = 0; i < avctx->extradata_size; i++)
199 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
200 av_dlog(avctx, "\n");
203 av_log_ask_for_sample(avctx, "Unsupported extradata size\n");
204 return AVERROR_INVALIDDATA;
208 s->log2_frame_size = av_log2(avctx->block_align) + 4;
211 s->skip_frame = 1; /* skip first frame */
213 s->len_prefix = s->decode_flags & 0x40;
216 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
219 /* init previous block len */
220 for (i = 0; i < avctx->channels; i++)
221 s->channel[i].prev_block_len = s->samples_per_frame;
224 log2_max_num_subframes = (s->decode_flags & 0x38) >> 3;
225 s->max_num_subframes = 1 << log2_max_num_subframes;
226 s->max_subframe_len_bit = 0;
227 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
229 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
230 s->dynamic_range_compression = s->decode_flags & 0x80;
231 s->bV3RTM = s->decode_flags & 0x100;
233 if (s->max_num_subframes > MAX_SUBFRAMES) {
234 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
235 s->max_num_subframes);
236 return AVERROR_INVALIDDATA;
239 s->num_channels = avctx->channels;
241 /* extract lfe channel position */
244 if (channel_mask & 8) {
246 for (mask = 1; mask < 16; mask <<= 1)
247 if (channel_mask & mask)
251 if (s->num_channels < 0) {
252 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
254 return AVERROR_INVALIDDATA;
255 } else if (s->num_channels > WMALL_MAX_CHANNELS) {
256 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
257 return AVERROR_PATCHWELCOME;
260 avcodec_get_frame_defaults(&s->frame);
261 avctx->coded_frame = &s->frame;
262 avctx->channel_layout = channel_mask;
267 * @brief Decode the subframe length.
269 * @param offset sample offset in the frame
270 * @return decoded subframe length on success, < 0 in case of an error
272 static int decode_subframe_length(WmallDecodeCtx *s, int offset)
274 int frame_len_ratio, subframe_len, len;
276 /* no need to read from the bitstream when only one length is possible */
277 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
278 return s->min_samples_per_subframe;
280 len = av_log2(s->max_num_subframes - 1) + 1;
281 frame_len_ratio = get_bits(&s->gb, len);
282 subframe_len = s->min_samples_per_subframe * (frame_len_ratio + 1);
284 /* sanity check the length */
285 if (subframe_len < s->min_samples_per_subframe ||
286 subframe_len > s->samples_per_frame) {
287 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
289 return AVERROR_INVALIDDATA;
295 * @brief Decode how the data in the frame is split into subframes.
296 * Every WMA frame contains the encoded data for a fixed number of
297 * samples per channel. The data for every channel might be split
298 * into several subframes. This function will reconstruct the list of
299 * subframes for every channel.
301 * If the subframes are not evenly split, the algorithm estimates the
302 * channels with the lowest number of total samples.
303 * Afterwards, for each of these channels a bit is read from the
304 * bitstream that indicates if the channel contains a subframe with the
305 * next subframe size that is going to be read from the bitstream or not.
306 * If a channel contains such a subframe, the subframe size gets added to
307 * the channel's subframe list.
308 * The algorithm repeats these steps until the frame is properly divided
309 * between the individual channels.
312 * @return 0 on success, < 0 in case of an error
314 static int decode_tilehdr(WmallDecodeCtx *s)
316 uint16_t num_samples[WMALL_MAX_CHANNELS] = { 0 }; /* sum of samples for all currently known subframes of a channel */
317 uint8_t contains_subframe[WMALL_MAX_CHANNELS]; /* flag indicating if a channel contains the current subframe */
318 int channels_for_cur_subframe = s->num_channels; /* number of channels that contain the current subframe */
319 int fixed_channel_layout = 0; /* flag indicating that all channels use the same subfra2me offsets and sizes */
320 int min_channel_len = 0; /* smallest sum of samples (channels with this length will be processed first) */
323 /* reset tiling information */
324 for (c = 0; c < s->num_channels; c++)
325 s->channel[c].num_subframes = 0;
327 tile_aligned = get_bits1(&s->gb);
328 if (s->max_num_subframes == 1 || tile_aligned)
329 fixed_channel_layout = 1;
331 /* loop until the frame data is split between the subframes */
333 int subframe_len, in_use = 0;
335 /* check which channels contain the subframe */
336 for (c = 0; c < s->num_channels; c++) {
337 if (num_samples[c] == min_channel_len) {
338 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
339 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) {
340 contains_subframe[c] = in_use = 1;
342 if (get_bits1(&s->gb))
343 contains_subframe[c] = in_use = 1;
346 contains_subframe[c] = 0;
350 av_log(s->avctx, AV_LOG_ERROR,
351 "Found empty subframe\n");
352 return AVERROR_INVALIDDATA;
355 /* get subframe length, subframe_len == 0 is not allowed */
356 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
357 return AVERROR_INVALIDDATA;
358 /* add subframes to the individual channels and find new min_channel_len */
359 min_channel_len += subframe_len;
360 for (c = 0; c < s->num_channels; c++) {
361 WmallChannelCtx *chan = &s->channel[c];
363 if (contains_subframe[c]) {
364 if (chan->num_subframes >= MAX_SUBFRAMES) {
365 av_log(s->avctx, AV_LOG_ERROR,
366 "broken frame: num subframes > 31\n");
367 return AVERROR_INVALIDDATA;
369 chan->subframe_len[chan->num_subframes] = subframe_len;
370 num_samples[c] += subframe_len;
371 ++chan->num_subframes;
372 if (num_samples[c] > s->samples_per_frame) {
373 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
374 "channel len(%d) > samples_per_frame(%d)\n",
375 num_samples[c], s->samples_per_frame);
376 return AVERROR_INVALIDDATA;
378 } else if (num_samples[c] <= min_channel_len) {
379 if (num_samples[c] < min_channel_len) {
380 channels_for_cur_subframe = 0;
381 min_channel_len = num_samples[c];
383 ++channels_for_cur_subframe;
386 } while (min_channel_len < s->samples_per_frame);
388 for (c = 0; c < s->num_channels; c++) {
390 for (i = 0; i < s->channel[c].num_subframes; i++) {
391 s->channel[c].subframe_offsets[i] = offset;
392 offset += s->channel[c].subframe_len[i];
399 static void decode_ac_filter(WmallDecodeCtx *s)
402 s->acfilter_order = get_bits(&s->gb, 4) + 1;
403 s->acfilter_scaling = get_bits(&s->gb, 4);
405 for (i = 0; i < s->acfilter_order; i++)
406 s->acfilter_coeffs[i] = get_bits(&s->gb, s->acfilter_scaling) + 1;
409 static void decode_mclms(WmallDecodeCtx *s)
411 s->mclms_order = (get_bits(&s->gb, 4) + 1) * 2;
412 s->mclms_scaling = get_bits(&s->gb, 4);
413 if (get_bits1(&s->gb)) {
414 int i, send_coef_bits;
415 int cbits = av_log2(s->mclms_scaling + 1);
416 if (1 << cbits < s->mclms_scaling + 1)
419 send_coef_bits = (cbits ? get_bits(&s->gb, cbits) : 0) + 2;
421 for (i = 0; i < s->mclms_order * s->num_channels * s->num_channels; i++)
422 s->mclms_coeffs[i] = get_bits(&s->gb, send_coef_bits);
424 for (i = 0; i < s->num_channels; i++) {
426 for (c = 0; c < i; c++)
427 s->mclms_coeffs_cur[i * s->num_channels + c] = get_bits(&s->gb, send_coef_bits);
432 static int decode_cdlms(WmallDecodeCtx *s)
435 int cdlms_send_coef = get_bits1(&s->gb);
437 for (c = 0; c < s->num_channels; c++) {
438 s->cdlms_ttl[c] = get_bits(&s->gb, 3) + 1;
439 for (i = 0; i < s->cdlms_ttl[c]; i++) {
440 s->cdlms[c][i].order = (get_bits(&s->gb, 7) + 1) * 8;
441 if (s->cdlms[c][i].order > MAX_ORDER) {
442 av_log(s->avctx, AV_LOG_ERROR,
443 "Order[%d][%d] %d > max (%d), not supported\n",
444 c, i, s->cdlms[c][i].order, MAX_ORDER);
445 s->cdlms[0][0].order = 0;
446 return AVERROR_INVALIDDATA;
450 for (i = 0; i < s->cdlms_ttl[c]; i++)
451 s->cdlms[c][i].scaling = get_bits(&s->gb, 4);
453 if (cdlms_send_coef) {
454 for (i = 0; i < s->cdlms_ttl[c]; i++) {
455 int cbits, shift_l, shift_r, j;
456 cbits = av_log2(s->cdlms[c][i].order);
457 if ((1 << cbits) < s->cdlms[c][i].order)
459 s->cdlms[c][i].coefsend = get_bits(&s->gb, cbits) + 1;
461 cbits = av_log2(s->cdlms[c][i].scaling + 1);
462 if ((1 << cbits) < s->cdlms[c][i].scaling + 1)
465 s->cdlms[c][i].bitsend = get_bits(&s->gb, cbits) + 2;
466 shift_l = 32 - s->cdlms[c][i].bitsend;
467 shift_r = 32 - s->cdlms[c][i].scaling - 2;
468 for (j = 0; j < s->cdlms[c][i].coefsend; j++)
469 s->cdlms[c][i].coefs[j] =
470 (get_bits(&s->gb, s->cdlms[c][i].bitsend) << shift_l) >> shift_r;
478 static int decode_channel_residues(WmallDecodeCtx *s, int ch, int tile_size)
481 unsigned int ave_mean;
482 s->transient[ch] = get_bits1(&s->gb);
483 if (s->transient[ch]) {
484 s->transient_pos[ch] = get_bits(&s->gb, av_log2(tile_size));
485 if (s->transient_pos[ch])
486 s->transient[ch] = 0;
487 s->channel[ch].transient_counter =
488 FFMAX(s->channel[ch].transient_counter, s->samples_per_frame / 2);
489 } else if (s->channel[ch].transient_counter)
490 s->transient[ch] = 1;
492 if (s->seekable_tile) {
493 ave_mean = get_bits(&s->gb, s->bits_per_sample);
494 s->ave_sum[ch] = ave_mean << (s->movave_scaling + 1);
497 if (s->seekable_tile) {
498 if (s->do_inter_ch_decorr)
499 s->channel_residues[ch][0] = get_sbits(&s->gb, s->bits_per_sample + 1);
501 s->channel_residues[ch][0] = get_sbits(&s->gb, s->bits_per_sample);
504 for (; i < tile_size; i++) {
505 int quo = 0, rem, rem_bits, residue;
506 while(get_bits1(&s->gb)) {
508 if (get_bits_left(&s->gb) <= 0)
512 quo += get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1);
514 ave_mean = (s->ave_sum[ch] + (1 << s->movave_scaling)) >> (s->movave_scaling + 1);
518 rem_bits = av_ceil_log2(ave_mean);
519 rem = rem_bits ? get_bits(&s->gb, rem_bits) : 0;
520 residue = (quo << rem_bits) + rem;
523 s->ave_sum[ch] = residue + s->ave_sum[ch] -
524 (s->ave_sum[ch] >> s->movave_scaling);
527 residue = -(residue >> 1) - 1;
529 residue = residue >> 1;
530 s->channel_residues[ch][i] = residue;
537 static void decode_lpc(WmallDecodeCtx *s)
540 s->lpc_order = get_bits(&s->gb, 5) + 1;
541 s->lpc_scaling = get_bits(&s->gb, 4);
542 s->lpc_intbits = get_bits(&s->gb, 3) + 1;
543 cbits = s->lpc_scaling + s->lpc_intbits;
544 for (ch = 0; ch < s->num_channels; ch++)
545 for (i = 0; i < s->lpc_order; i++)
546 s->lpc_coefs[ch][i] = get_sbits(&s->gb, cbits);
549 static void clear_codec_buffers(WmallDecodeCtx *s)
553 memset(s->acfilter_coeffs, 0, sizeof(s->acfilter_coeffs));
554 memset(s->acfilter_prevvalues, 0, sizeof(s->acfilter_prevvalues));
555 memset(s->lpc_coefs, 0, sizeof(s->lpc_coefs));
557 memset(s->mclms_coeffs, 0, sizeof(s->mclms_coeffs));
558 memset(s->mclms_coeffs_cur, 0, sizeof(s->mclms_coeffs_cur));
559 memset(s->mclms_prevvalues, 0, sizeof(s->mclms_prevvalues));
560 memset(s->mclms_updates, 0, sizeof(s->mclms_updates));
562 for (ich = 0; ich < s->num_channels; ich++) {
563 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++) {
564 memset(s->cdlms[ich][ilms].coefs, 0,
565 sizeof(s->cdlms[ich][ilms].coefs));
566 memset(s->cdlms[ich][ilms].lms_prevvalues, 0,
567 sizeof(s->cdlms[ich][ilms].lms_prevvalues));
568 memset(s->cdlms[ich][ilms].lms_updates, 0,
569 sizeof(s->cdlms[ich][ilms].lms_updates));
576 * @brief Reset filter parameters and transient area at new seekable tile.
578 static void reset_codec(WmallDecodeCtx *s)
581 s->mclms_recent = s->mclms_order * s->num_channels;
582 for (ich = 0; ich < s->num_channels; ich++) {
583 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++)
584 s->cdlms[ich][ilms].recent = s->cdlms[ich][ilms].order;
585 /* first sample of a seekable subframe is considered as the starting of
586 a transient area which is samples_per_frame samples long */
587 s->channel[ich].transient_counter = s->samples_per_frame;
588 s->transient[ich] = 1;
589 s->transient_pos[ich] = 0;
593 static void mclms_update(WmallDecodeCtx *s, int icoef, int *pred)
595 int i, j, ich, pred_error;
596 int order = s->mclms_order;
597 int num_channels = s->num_channels;
598 int range = 1 << (s->bits_per_sample - 1);
600 for (ich = 0; ich < num_channels; ich++) {
601 pred_error = s->channel_residues[ich][icoef] - pred[ich];
602 if (pred_error > 0) {
603 for (i = 0; i < order * num_channels; i++)
604 s->mclms_coeffs[i + ich * order * num_channels] +=
605 s->mclms_updates[s->mclms_recent + i];
606 for (j = 0; j < ich; j++) {
607 if (s->channel_residues[j][icoef] > 0)
608 s->mclms_coeffs_cur[ich * num_channels + j] += 1;
609 else if (s->channel_residues[j][icoef] < 0)
610 s->mclms_coeffs_cur[ich * num_channels + j] -= 1;
612 } else if (pred_error < 0) {
613 for (i = 0; i < order * num_channels; i++)
614 s->mclms_coeffs[i + ich * order * num_channels] -=
615 s->mclms_updates[s->mclms_recent + i];
616 for (j = 0; j < ich; j++) {
617 if (s->channel_residues[j][icoef] > 0)
618 s->mclms_coeffs_cur[ich * num_channels + j] -= 1;
619 else if (s->channel_residues[j][icoef] < 0)
620 s->mclms_coeffs_cur[ich * num_channels + j] += 1;
625 for (ich = num_channels - 1; ich >= 0; ich--) {
627 s->mclms_prevvalues[s->mclms_recent] = s->channel_residues[ich][icoef];
628 if (s->channel_residues[ich][icoef] > range - 1)
629 s->mclms_prevvalues[s->mclms_recent] = range - 1;
630 else if (s->channel_residues[ich][icoef] < -range)
631 s->mclms_prevvalues[s->mclms_recent] = -range;
633 s->mclms_updates[s->mclms_recent] = 0;
634 if (s->channel_residues[ich][icoef] > 0)
635 s->mclms_updates[s->mclms_recent] = 1;
636 else if (s->channel_residues[ich][icoef] < 0)
637 s->mclms_updates[s->mclms_recent] = -1;
640 if (s->mclms_recent == 0) {
641 memcpy(&s->mclms_prevvalues[order * num_channels],
643 2 * order * num_channels);
644 memcpy(&s->mclms_updates[order * num_channels],
646 2 * order * num_channels);
647 s->mclms_recent = num_channels * order;
651 static void mclms_predict(WmallDecodeCtx *s, int icoef, int *pred)
654 int order = s->mclms_order;
655 int num_channels = s->num_channels;
657 for (ich = 0; ich < num_channels; ich++) {
659 if (!s->is_channel_coded[ich])
661 for (i = 0; i < order * num_channels; i++)
662 pred[ich] += s->mclms_prevvalues[i + s->mclms_recent] *
663 s->mclms_coeffs[i + order * num_channels * ich];
664 for (i = 0; i < ich; i++)
665 pred[ich] += s->channel_residues[i][icoef] *
666 s->mclms_coeffs_cur[i + num_channels * ich];
667 pred[ich] += 1 << s->mclms_scaling - 1;
668 pred[ich] >>= s->mclms_scaling;
669 s->channel_residues[ich][icoef] += pred[ich];
673 static void revert_mclms(WmallDecodeCtx *s, int tile_size)
675 int icoef, pred[WMALL_MAX_CHANNELS] = { 0 };
676 for (icoef = 0; icoef < tile_size; icoef++) {
677 mclms_predict(s, icoef, pred);
678 mclms_update(s, icoef, pred);
682 static int lms_predict(WmallDecodeCtx *s, int ich, int ilms)
685 int recent = s->cdlms[ich][ilms].recent;
687 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
688 pred += s->cdlms[ich][ilms].coefs[icoef] *
689 s->cdlms[ich][ilms].lms_prevvalues[icoef + recent];
694 static void lms_update(WmallDecodeCtx *s, int ich, int ilms,
695 int input, int residue)
698 int recent = s->cdlms[ich][ilms].recent;
699 int range = 1 << s->bits_per_sample - 1;
702 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
703 s->cdlms[ich][ilms].coefs[icoef] -=
704 s->cdlms[ich][ilms].lms_updates[icoef + recent];
705 } else if (residue > 0) {
706 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
707 s->cdlms[ich][ilms].coefs[icoef] +=
708 s->cdlms[ich][ilms].lms_updates[icoef + recent];
714 memcpy(&s->cdlms[ich][ilms].lms_prevvalues[s->cdlms[ich][ilms].order],
715 s->cdlms[ich][ilms].lms_prevvalues,
716 2 * s->cdlms[ich][ilms].order);
717 memcpy(&s->cdlms[ich][ilms].lms_updates[s->cdlms[ich][ilms].order],
718 s->cdlms[ich][ilms].lms_updates,
719 2 * s->cdlms[ich][ilms].order);
720 recent = s->cdlms[ich][ilms].order - 1;
723 s->cdlms[ich][ilms].lms_prevvalues[recent] = av_clip(input, -range, range - 1);
725 s->cdlms[ich][ilms].lms_updates[recent] = 0;
727 s->cdlms[ich][ilms].lms_updates[recent] = -s->update_speed[ich];
729 s->cdlms[ich][ilms].lms_updates[recent] = s->update_speed[ich];
731 s->cdlms[ich][ilms].lms_updates[recent + (s->cdlms[ich][ilms].order >> 4)] >>= 2;
732 s->cdlms[ich][ilms].lms_updates[recent + (s->cdlms[ich][ilms].order >> 3)] >>= 1;
733 s->cdlms[ich][ilms].recent = recent;
736 static void use_high_update_speed(WmallDecodeCtx *s, int ich)
738 int ilms, recent, icoef;
739 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
740 recent = s->cdlms[ich][ilms].recent;
741 if (s->update_speed[ich] == 16)
744 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
745 s->cdlms[ich][ilms].lms_updates[icoef + recent] *= 2;
747 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
748 s->cdlms[ich][ilms].lms_updates[icoef] *= 2;
751 s->update_speed[ich] = 16;
754 static void use_normal_update_speed(WmallDecodeCtx *s, int ich)
756 int ilms, recent, icoef;
757 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
758 recent = s->cdlms[ich][ilms].recent;
759 if (s->update_speed[ich] == 8)
762 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
763 s->cdlms[ich][ilms].lms_updates[icoef + recent] /= 2;
765 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
766 s->cdlms[ich][ilms].lms_updates[icoef] /= 2;
768 s->update_speed[ich] = 8;
771 static void revert_cdlms(WmallDecodeCtx *s, int ch,
772 int coef_begin, int coef_end)
774 int icoef, pred, ilms, num_lms, residue, input;
776 num_lms = s->cdlms_ttl[ch];
777 for (ilms = num_lms - 1; ilms >= 0; ilms--) {
778 for (icoef = coef_begin; icoef < coef_end; icoef++) {
779 pred = 1 << (s->cdlms[ch][ilms].scaling - 1);
780 residue = s->channel_residues[ch][icoef];
781 pred += lms_predict(s, ch, ilms);
782 input = residue + (pred >> s->cdlms[ch][ilms].scaling);
783 lms_update(s, ch, ilms, input, residue);
784 s->channel_residues[ch][icoef] = input;
789 static void revert_inter_ch_decorr(WmallDecodeCtx *s, int tile_size)
791 if (s->num_channels != 2)
793 else if (s->is_channel_coded[0] || s->is_channel_coded[1]) {
795 for (icoef = 0; icoef < tile_size; icoef++) {
796 s->channel_residues[0][icoef] -= s->channel_residues[1][icoef] >> 1;
797 s->channel_residues[1][icoef] += s->channel_residues[0][icoef];
802 static void revert_acfilter(WmallDecodeCtx *s, int tile_size)
805 int64_t *filter_coeffs = s->acfilter_coeffs;
806 int scaling = s->acfilter_scaling;
807 int order = s->acfilter_order;
809 for (ich = 0; ich < s->num_channels; ich++) {
810 int *prevvalues = s->acfilter_prevvalues[ich];
811 for (i = 0; i < order; i++) {
813 for (j = 0; j < order; j++) {
815 pred += filter_coeffs[j] * prevvalues[j - i];
817 pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
820 s->channel_residues[ich][i] += pred;
822 for (i = order; i < tile_size; i++) {
824 for (j = 0; j < order; j++)
825 pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
827 s->channel_residues[ich][i] += pred;
829 for (j = 0; j < order; j++)
830 prevvalues[j] = s->channel_residues[ich][tile_size - j - 1];
834 static int decode_subframe(WmallDecodeCtx *s)
836 int offset = s->samples_per_frame;
837 int subframe_len = s->samples_per_frame;
838 int total_samples = s->samples_per_frame * s->num_channels;
839 int i, j, rawpcm_tile, padding_zeroes, res;
841 s->subframe_offset = get_bits_count(&s->gb);
843 /* reset channel context and find the next block offset and size
844 == the next block of the channel with the smallest number of
846 for (i = 0; i < s->num_channels; i++) {
847 if (offset > s->channel[i].decoded_samples) {
848 offset = s->channel[i].decoded_samples;
850 s->channel[i].subframe_len[s->channel[i].cur_subframe];
854 /* get a list of all channels that contain the estimated block */
855 s->channels_for_cur_subframe = 0;
856 for (i = 0; i < s->num_channels; i++) {
857 const int cur_subframe = s->channel[i].cur_subframe;
858 /* subtract already processed samples */
859 total_samples -= s->channel[i].decoded_samples;
861 /* and count if there are multiple subframes that match our profile */
862 if (offset == s->channel[i].decoded_samples &&
863 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
864 total_samples -= s->channel[i].subframe_len[cur_subframe];
865 s->channel[i].decoded_samples +=
866 s->channel[i].subframe_len[cur_subframe];
867 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
868 ++s->channels_for_cur_subframe;
872 /* check if the frame will be complete after processing the
875 s->parsed_all_subframes = 1;
878 s->seekable_tile = get_bits1(&s->gb);
879 if (s->seekable_tile) {
880 clear_codec_buffers(s);
882 s->do_arith_coding = get_bits1(&s->gb);
883 if (s->do_arith_coding) {
884 av_log_missing_feature(s->avctx, "arithmetic coding", 1);
885 return AVERROR_PATCHWELCOME;
887 s->do_ac_filter = get_bits1(&s->gb);
888 s->do_inter_ch_decorr = get_bits1(&s->gb);
889 s->do_mclms = get_bits1(&s->gb);
897 if ((res = decode_cdlms(s)) < 0)
899 s->movave_scaling = get_bits(&s->gb, 3);
900 s->quant_stepsize = get_bits(&s->gb, 8) + 1;
903 } else if (!s->cdlms[0][0].order) {
904 av_log(s->avctx, AV_LOG_DEBUG,
905 "Waiting for seekable tile\n");
906 s->frame.nb_samples = 0;
910 rawpcm_tile = get_bits1(&s->gb);
912 for (i = 0; i < s->num_channels; i++)
913 s->is_channel_coded[i] = 1;
916 for (i = 0; i < s->num_channels; i++)
917 s->is_channel_coded[i] = get_bits1(&s->gb);
921 s->do_lpc = get_bits1(&s->gb);
924 av_log_ask_for_sample(s->avctx, "Inverse LPC filter not "
925 "implemented. Expect wrong output.\n");
932 if (get_bits1(&s->gb))
933 padding_zeroes = get_bits(&s->gb, 5);
938 int bits = s->bits_per_sample - padding_zeroes;
940 av_log(s->avctx, AV_LOG_ERROR,
941 "Invalid number of padding bits in raw PCM tile\n");
942 return AVERROR_INVALIDDATA;
944 av_dlog(s->avctx, "RAWPCM %d bits per sample. "
945 "total %d bits, remain=%d\n", bits,
946 bits * s->num_channels * subframe_len, get_bits_count(&s->gb));
947 for (i = 0; i < s->num_channels; i++)
948 for (j = 0; j < subframe_len; j++)
949 s->channel_coeffs[i][j] = get_sbits(&s->gb, bits);
951 for (i = 0; i < s->num_channels; i++)
952 if (s->is_channel_coded[i]) {
953 decode_channel_residues(s, i, subframe_len);
954 if (s->seekable_tile)
955 use_high_update_speed(s, i);
957 use_normal_update_speed(s, i);
958 revert_cdlms(s, i, 0, subframe_len);
960 memset(s->channel_residues[i], 0, sizeof(**s->channel_residues) * subframe_len);
964 revert_mclms(s, subframe_len);
965 if (s->do_inter_ch_decorr)
966 revert_inter_ch_decorr(s, subframe_len);
968 revert_acfilter(s, subframe_len);
971 if (s->quant_stepsize != 1)
972 for (i = 0; i < s->num_channels; i++)
973 for (j = 0; j < subframe_len; j++)
974 s->channel_residues[i][j] *= s->quant_stepsize;
976 /* Write to proper output buffer depending on bit-depth */
977 for (i = 0; i < s->channels_for_cur_subframe; i++) {
978 int c = s->channel_indexes_for_cur_subframe[i];
979 int subframe_len = s->channel[c].subframe_len[s->channel[c].cur_subframe];
981 for (j = 0; j < subframe_len; j++) {
982 if (s->bits_per_sample == 16) {
983 *s->samples_16[c] = (int16_t) s->channel_residues[c][j] << padding_zeroes;
984 s->samples_16[c] += s->num_channels;
986 *s->samples_32[c] = s->channel_residues[c][j] << padding_zeroes;
987 s->samples_32[c] += s->num_channels;
992 /* handled one subframe */
993 for (i = 0; i < s->channels_for_cur_subframe; i++) {
994 int c = s->channel_indexes_for_cur_subframe[i];
995 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
996 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
997 return AVERROR_INVALIDDATA;
999 ++s->channel[c].cur_subframe;
1005 * @brief Decode one WMA frame.
1006 * @param s codec context
1007 * @return 0 if the trailer bit indicates that this is the last frame,
1008 * 1 if there are additional frames
1010 static int decode_frame(WmallDecodeCtx *s)
1012 GetBitContext* gb = &s->gb;
1013 int more_frames = 0, len = 0, i, ret;
1015 s->frame.nb_samples = s->samples_per_frame;
1016 if ((ret = s->avctx->get_buffer(s->avctx, &s->frame)) < 0) {
1017 /* return an error if no frame could be decoded at all */
1018 av_log(s->avctx, AV_LOG_ERROR,
1019 "not enough space for the output samples\n");
1023 for (i = 0; i < s->num_channels; i++) {
1024 s->samples_16[i] = (int16_t *)s->frame.data[0] + i;
1025 s->samples_32[i] = (int32_t *)s->frame.data[0] + i;
1028 /* get frame length */
1030 len = get_bits(gb, s->log2_frame_size);
1032 /* decode tile information */
1033 if (decode_tilehdr(s)) {
1039 if (s->dynamic_range_compression)
1040 s->drc_gain = get_bits(gb, 8);
1042 /* no idea what these are for, might be the number of samples
1043 that need to be skipped at the beginning or end of a stream */
1044 if (get_bits1(gb)) {
1047 /* usually true for the first frame */
1048 if (get_bits1(gb)) {
1049 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1050 av_dlog(s->avctx, "start skip: %i\n", skip);
1053 /* sometimes true for the last frame */
1054 if (get_bits1(gb)) {
1055 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1056 av_dlog(s->avctx, "end skip: %i\n", skip);
1061 /* reset subframe states */
1062 s->parsed_all_subframes = 0;
1063 for (i = 0; i < s->num_channels; i++) {
1064 s->channel[i].decoded_samples = 0;
1065 s->channel[i].cur_subframe = 0;
1068 /* decode all subframes */
1069 while (!s->parsed_all_subframes) {
1070 if (decode_subframe(s) < 0) {
1076 av_dlog(s->avctx, "Frame done\n");
1081 if (s->len_prefix) {
1082 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1083 /* FIXME: not sure if this is always an error */
1084 av_log(s->avctx, AV_LOG_ERROR,
1085 "frame[%i] would have to skip %i bits\n", s->frame_num,
1086 len - (get_bits_count(gb) - s->frame_offset) - 1);
1091 /* skip the rest of the frame data */
1092 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1095 /* decode trailer bit */
1096 more_frames = get_bits1(gb);
1102 * @brief Calculate remaining input buffer length.
1103 * @param s codec context
1104 * @param gb bitstream reader context
1105 * @return remaining size in bits
1107 static int remaining_bits(WmallDecodeCtx *s, GetBitContext *gb)
1109 return s->buf_bit_size - get_bits_count(gb);
1113 * @brief Fill the bit reservoir with a (partial) frame.
1114 * @param s codec context
1115 * @param gb bitstream reader context
1116 * @param len length of the partial frame
1117 * @param append decides whether to reset the buffer or not
1119 static void save_bits(WmallDecodeCtx *s, GetBitContext* gb, int len,
1125 /* when the frame data does not need to be concatenated, the input buffer
1126 is reset and additional bits from the previous frame are copied
1127 and skipped later so that a fast byte copy is possible */
1130 s->frame_offset = get_bits_count(gb) & 7;
1131 s->num_saved_bits = s->frame_offset;
1132 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1135 buflen = (s->num_saved_bits + len + 8) >> 3;
1137 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1138 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1143 s->num_saved_bits += len;
1145 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1148 int align = 8 - (get_bits_count(gb) & 7);
1149 align = FFMIN(align, len);
1150 put_bits(&s->pb, align, get_bits(gb, align));
1152 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1154 skip_bits_long(gb, len);
1157 flush_put_bits(&tmp);
1159 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1160 skip_bits(&s->gb, s->frame_offset);
1163 static int decode_packet(AVCodecContext *avctx, void *data, int *got_frame_ptr,
1166 WmallDecodeCtx *s = avctx->priv_data;
1167 GetBitContext* gb = &s->pgb;
1168 const uint8_t* buf = avpkt->data;
1169 int buf_size = avpkt->size;
1170 int num_bits_prev_frame, packet_sequence_number, spliced_packet;
1172 s->frame.nb_samples = 0;
1174 if (s->packet_done || s->packet_loss) {
1177 /* sanity check for the buffer length */
1178 if (buf_size < avctx->block_align)
1181 s->next_packet_start = buf_size - avctx->block_align;
1182 buf_size = avctx->block_align;
1183 s->buf_bit_size = buf_size << 3;
1185 /* parse packet header */
1186 init_get_bits(gb, buf, s->buf_bit_size);
1187 packet_sequence_number = get_bits(gb, 4);
1188 skip_bits(gb, 1); // Skip seekable_frame_in_packet, currently ununused
1189 spliced_packet = get_bits1(gb);
1191 av_log_missing_feature(avctx, "Bitstream splicing", 1);
1193 /* get number of bits that need to be added to the previous frame */
1194 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1196 /* check for packet loss */
1197 if (!s->packet_loss &&
1198 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1200 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1201 s->packet_sequence_number, packet_sequence_number);
1203 s->packet_sequence_number = packet_sequence_number;
1205 if (num_bits_prev_frame > 0) {
1206 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1207 if (num_bits_prev_frame >= remaining_packet_bits) {
1208 num_bits_prev_frame = remaining_packet_bits;
1212 /* Append the previous frame data to the remaining data from the
1213 * previous packet to create a full frame. */
1214 save_bits(s, gb, num_bits_prev_frame, 1);
1216 /* decode the cross packet frame if it is valid */
1217 if (num_bits_prev_frame < remaining_packet_bits && !s->packet_loss)
1219 } else if (s->num_saved_bits - s->frame_offset) {
1220 av_dlog(avctx, "ignoring %x previously saved bits\n",
1221 s->num_saved_bits - s->frame_offset);
1224 if (s->packet_loss) {
1225 /* Reset number of saved bits so that the decoder does not start
1226 * to decode incomplete frames in the s->len_prefix == 0 case. */
1227 s->num_saved_bits = 0;
1234 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1235 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1236 skip_bits(gb, s->packet_offset);
1238 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1239 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1240 frame_size <= remaining_bits(s, gb)) {
1241 save_bits(s, gb, frame_size, 0);
1242 s->packet_done = !decode_frame(s);
1243 } else if (!s->len_prefix
1244 && s->num_saved_bits > get_bits_count(&s->gb)) {
1245 /* when the frames do not have a length prefix, we don't know the
1246 * compressed length of the individual frames however, we know what
1247 * part of a new packet belongs to the previous frame therefore we
1248 * save the incoming packet first, then we append the "previous
1249 * frame" data from the next packet so that we get a buffer that
1250 * only contains full frames */
1251 s->packet_done = !decode_frame(s);
1257 if (s->packet_done && !s->packet_loss &&
1258 remaining_bits(s, gb) > 0) {
1259 /* save the rest of the data so that it can be decoded
1260 * with the next packet */
1261 save_bits(s, gb, remaining_bits(s, gb), 0);
1264 *(AVFrame *)data = s->frame;
1265 *got_frame_ptr = s->frame.nb_samples > 0;
1266 s->packet_offset = get_bits_count(gb) & 7;
1268 return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3;
1271 static void flush(AVCodecContext *avctx)
1273 WmallDecodeCtx *s = avctx->priv_data;
1276 s->num_saved_bits = 0;
1277 s->frame_offset = 0;
1278 s->next_packet_start = 0;
1279 s->cdlms[0][0].order = 0;
1280 s->frame.nb_samples = 0;
1283 AVCodec ff_wmalossless_decoder = {
1284 .name = "wmalossless",
1285 .type = AVMEDIA_TYPE_AUDIO,
1286 .id = CODEC_ID_WMALOSSLESS,
1287 .priv_data_size = sizeof(WmallDecodeCtx),
1288 .init = decode_init,
1289 .decode = decode_packet,
1291 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1292 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio Lossless"),