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 FFmpeg.
10 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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"
34 #include "lossless_audiodsp.h"
36 #include "wma_common.h"
38 /** current decoder limitations */
39 #define WMALL_MAX_CHANNELS 8 ///< max number of handled channels
40 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
41 #define MAX_BANDS 29 ///< max number of scale factor bands
42 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
45 #define WMALL_BLOCK_MIN_BITS 6 ///< log2 of min block size
46 #define WMALL_BLOCK_MAX_BITS 14 ///< log2 of max block size
47 #define WMALL_BLOCK_MAX_SIZE (1 << WMALL_BLOCK_MAX_BITS) ///< maximum block size
48 #define WMALL_BLOCK_SIZES (WMALL_BLOCK_MAX_BITS - WMALL_BLOCK_MIN_BITS + 1) ///< possible block sizes
50 #define WMALL_COEFF_PAD_SIZE 16 ///< pad coef buffers with 0 for use with SIMD
53 * @brief frame-specific decoder context for a single channel
55 typedef struct WmallChannelCtx {
56 int16_t prev_block_len; ///< length of the previous block
57 uint8_t transmit_coefs;
58 uint8_t num_subframes;
59 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
60 uint16_t subframe_offsets[MAX_SUBFRAMES]; ///< subframe positions in the current frame
61 uint8_t cur_subframe; ///< current subframe number
62 uint16_t decoded_samples; ///< number of already processed samples
63 int quant_step; ///< quantization step for the current subframe
64 int transient_counter; ///< number of transient samples from the beginning of the transient zone
68 * @brief main decoder context
70 typedef struct WmallDecodeCtx {
71 /* generic decoder variables */
72 AVCodecContext *avctx;
74 LLAudDSPContext dsp; ///< accelerated DSP functions
75 uint8_t *frame_data; ///< compressed frame data
76 int max_frame_size; ///< max bitstream size
77 PutBitContext pb; ///< context for filling the frame_data buffer
79 /* frame size dependent frame information (set during initialization) */
80 uint32_t decode_flags; ///< used compression features
81 int len_prefix; ///< frame is prefixed with its length
82 int dynamic_range_compression; ///< frame contains DRC data
83 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
84 uint16_t samples_per_frame; ///< number of samples to output
85 uint16_t log2_frame_size;
86 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
87 int8_t lfe_channel; ///< lfe channel index
88 uint8_t max_num_subframes;
89 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
90 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
91 uint16_t min_samples_per_subframe;
93 /* packet decode state */
94 GetBitContext pgb; ///< bitstream reader context for the packet
95 int next_packet_start; ///< start offset of the next WMA packet in the demuxer packet
96 uint8_t packet_offset; ///< offset to the frame in the packet
97 uint8_t packet_sequence_number; ///< current packet number
98 int num_saved_bits; ///< saved number of bits
99 int frame_offset; ///< frame offset in the bit reservoir
100 int subframe_offset; ///< subframe offset in the bit reservoir
101 uint8_t packet_loss; ///< set in case of bitstream error
102 uint8_t packet_done; ///< set when a packet is fully decoded
104 /* frame decode state */
105 uint32_t frame_num; ///< current frame number (not used for decoding)
106 GetBitContext gb; ///< bitstream reader context
107 int buf_bit_size; ///< buffer size in bits
108 int16_t *samples_16[WMALL_MAX_CHANNELS]; ///< current sample buffer pointer (16-bit)
109 int32_t *samples_32[WMALL_MAX_CHANNELS]; ///< current sample buffer pointer (24-bit)
110 uint8_t drc_gain; ///< gain for the DRC tool
111 int8_t skip_frame; ///< skip output step
112 int8_t parsed_all_subframes; ///< all subframes decoded?
114 /* subframe/block decode state */
115 int16_t subframe_len; ///< current subframe length
116 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
117 int8_t channel_indexes_for_cur_subframe[WMALL_MAX_CHANNELS];
119 WmallChannelCtx channel[WMALL_MAX_CHANNELS]; ///< per channel data
121 // WMA Lossless-specific
123 uint8_t do_arith_coding;
124 uint8_t do_ac_filter;
125 uint8_t do_inter_ch_decorr;
129 int8_t acfilter_order;
130 int8_t acfilter_scaling;
131 int16_t acfilter_coeffs[16];
132 int acfilter_prevvalues[WMALL_MAX_CHANNELS][16];
135 int8_t mclms_scaling;
136 int16_t mclms_coeffs[WMALL_MAX_CHANNELS * WMALL_MAX_CHANNELS * 32];
137 int16_t mclms_coeffs_cur[WMALL_MAX_CHANNELS * WMALL_MAX_CHANNELS];
138 int32_t mclms_prevvalues[WMALL_MAX_CHANNELS * 2 * 32];
139 int32_t mclms_updates[WMALL_MAX_CHANNELS * 2 * 32];
150 DECLARE_ALIGNED(16, int16_t, coefs)[MAX_ORDER + WMALL_COEFF_PAD_SIZE/sizeof(int16_t)];
151 DECLARE_ALIGNED(16, int32_t, lms_prevvalues)[MAX_ORDER * 2 + WMALL_COEFF_PAD_SIZE/sizeof(int16_t)];
152 DECLARE_ALIGNED(16, int16_t, lms_updates)[MAX_ORDER * 2 + WMALL_COEFF_PAD_SIZE/sizeof(int16_t)];
154 } cdlms[WMALL_MAX_CHANNELS][9];
156 int cdlms_ttl[WMALL_MAX_CHANNELS];
160 int is_channel_coded[WMALL_MAX_CHANNELS];
161 int update_speed[WMALL_MAX_CHANNELS];
163 int transient[WMALL_MAX_CHANNELS];
164 int transient_pos[WMALL_MAX_CHANNELS];
167 unsigned ave_sum[WMALL_MAX_CHANNELS];
169 int channel_residues[WMALL_MAX_CHANNELS][WMALL_BLOCK_MAX_SIZE];
171 int lpc_coefs[WMALL_MAX_CHANNELS][40];
177 /** Get sign of integer (1 for positive, -1 for negative and 0 for zero) */
178 #define WMASIGN(x) (((x) > 0) - ((x) < 0))
180 static av_cold int decode_init(AVCodecContext *avctx)
182 WmallDecodeCtx *s = avctx->priv_data;
183 uint8_t *edata_ptr = avctx->extradata;
184 unsigned int channel_mask;
185 int i, log2_max_num_subframes;
187 if (avctx->block_align <= 0 || avctx->block_align > (1<<21)) {
188 av_log(avctx, AV_LOG_ERROR, "block_align is not set or invalid\n");
189 return AVERROR(EINVAL);
192 av_assert0(avctx->channels >= 0);
193 if (avctx->channels > WMALL_MAX_CHANNELS) {
194 avpriv_request_sample(avctx,
195 "More than " AV_STRINGIFY(WMALL_MAX_CHANNELS) " channels");
196 return AVERROR_PATCHWELCOME;
199 s->max_frame_size = MAX_FRAMESIZE * avctx->channels;
200 s->frame_data = av_mallocz(s->max_frame_size + AV_INPUT_BUFFER_PADDING_SIZE);
202 return AVERROR(ENOMEM);
205 ff_llauddsp_init(&s->dsp);
206 init_put_bits(&s->pb, s->frame_data, s->max_frame_size);
208 if (avctx->extradata_size >= 18) {
209 s->decode_flags = AV_RL16(edata_ptr + 14);
210 channel_mask = AV_RL32(edata_ptr + 2);
211 s->bits_per_sample = AV_RL16(edata_ptr);
212 if (s->bits_per_sample == 16)
213 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
214 else if (s->bits_per_sample == 24) {
215 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
216 avctx->bits_per_raw_sample = 24;
218 av_log(avctx, AV_LOG_ERROR, "Unknown bit-depth: %"PRIu8"\n",
220 return AVERROR_INVALIDDATA;
222 /* dump the extradata */
223 for (i = 0; i < avctx->extradata_size; i++)
224 ff_dlog(avctx, "[%x] ", avctx->extradata[i]);
225 ff_dlog(avctx, "\n");
228 avpriv_request_sample(avctx, "Unsupported extradata size");
229 return AVERROR_PATCHWELCOME;
233 s->log2_frame_size = av_log2(avctx->block_align) + 4;
236 s->skip_frame = 1; /* skip first frame */
238 s->len_prefix = s->decode_flags & 0x40;
241 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
243 av_assert0(s->samples_per_frame <= WMALL_BLOCK_MAX_SIZE);
245 /* init previous block len */
246 for (i = 0; i < avctx->channels; i++)
247 s->channel[i].prev_block_len = s->samples_per_frame;
250 log2_max_num_subframes = (s->decode_flags & 0x38) >> 3;
251 s->max_num_subframes = 1 << log2_max_num_subframes;
252 s->max_subframe_len_bit = 0;
253 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
255 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
256 s->dynamic_range_compression = s->decode_flags & 0x80;
257 s->bV3RTM = s->decode_flags & 0x100;
259 if (s->max_num_subframes > MAX_SUBFRAMES) {
260 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRIu8"\n",
261 s->max_num_subframes);
262 return AVERROR_INVALIDDATA;
265 s->num_channels = avctx->channels;
267 /* extract lfe channel position */
270 if (channel_mask & 8) {
272 for (mask = 1; mask < 16; mask <<= 1)
273 if (channel_mask & mask)
277 s->frame = av_frame_alloc();
279 return AVERROR(ENOMEM);
281 avctx->channel_layout = channel_mask;
286 * @brief Decode the subframe length.
288 * @param offset sample offset in the frame
289 * @return decoded subframe length on success, < 0 in case of an error
291 static int decode_subframe_length(WmallDecodeCtx *s, int offset)
293 int frame_len_ratio, subframe_len, len;
295 /* no need to read from the bitstream when only one length is possible */
296 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
297 return s->min_samples_per_subframe;
299 len = av_log2(s->max_num_subframes - 1) + 1;
300 frame_len_ratio = get_bits(&s->gb, len);
301 subframe_len = s->min_samples_per_subframe * (frame_len_ratio + 1);
303 /* sanity check the length */
304 if (subframe_len < s->min_samples_per_subframe ||
305 subframe_len > s->samples_per_frame) {
306 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
308 return AVERROR_INVALIDDATA;
314 * @brief Decode how the data in the frame is split into subframes.
315 * Every WMA frame contains the encoded data for a fixed number of
316 * samples per channel. The data for every channel might be split
317 * into several subframes. This function will reconstruct the list of
318 * subframes for every channel.
320 * If the subframes are not evenly split, the algorithm estimates the
321 * channels with the lowest number of total samples.
322 * Afterwards, for each of these channels a bit is read from the
323 * bitstream that indicates if the channel contains a subframe with the
324 * next subframe size that is going to be read from the bitstream or not.
325 * If a channel contains such a subframe, the subframe size gets added to
326 * the channel's subframe list.
327 * The algorithm repeats these steps until the frame is properly divided
328 * between the individual channels.
331 * @return 0 on success, < 0 in case of an error
333 static int decode_tilehdr(WmallDecodeCtx *s)
335 uint16_t num_samples[WMALL_MAX_CHANNELS] = { 0 }; /* sum of samples for all currently known subframes of a channel */
336 uint8_t contains_subframe[WMALL_MAX_CHANNELS]; /* flag indicating if a channel contains the current subframe */
337 int channels_for_cur_subframe = s->num_channels; /* number of channels that contain the current subframe */
338 int fixed_channel_layout = 0; /* flag indicating that all channels use the same subfra2me offsets and sizes */
339 int min_channel_len = 0; /* smallest sum of samples (channels with this length will be processed first) */
342 /* reset tiling information */
343 for (c = 0; c < s->num_channels; c++)
344 s->channel[c].num_subframes = 0;
346 tile_aligned = get_bits1(&s->gb);
347 if (s->max_num_subframes == 1 || tile_aligned)
348 fixed_channel_layout = 1;
350 /* loop until the frame data is split between the subframes */
352 int subframe_len, in_use = 0;
354 /* check which channels contain the subframe */
355 for (c = 0; c < s->num_channels; c++) {
356 if (num_samples[c] == min_channel_len) {
357 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
358 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) {
359 contains_subframe[c] = 1;
361 contains_subframe[c] = get_bits1(&s->gb);
363 in_use |= contains_subframe[c];
365 contains_subframe[c] = 0;
369 av_log(s->avctx, AV_LOG_ERROR,
370 "Found empty subframe\n");
371 return AVERROR_INVALIDDATA;
374 /* get subframe length, subframe_len == 0 is not allowed */
375 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
376 return AVERROR_INVALIDDATA;
377 /* add subframes to the individual channels and find new min_channel_len */
378 min_channel_len += subframe_len;
379 for (c = 0; c < s->num_channels; c++) {
380 WmallChannelCtx *chan = &s->channel[c];
382 if (contains_subframe[c]) {
383 if (chan->num_subframes >= MAX_SUBFRAMES) {
384 av_log(s->avctx, AV_LOG_ERROR,
385 "broken frame: num subframes > 31\n");
386 return AVERROR_INVALIDDATA;
388 chan->subframe_len[chan->num_subframes] = subframe_len;
389 num_samples[c] += subframe_len;
390 ++chan->num_subframes;
391 if (num_samples[c] > s->samples_per_frame) {
392 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
393 "channel len(%"PRIu16") > samples_per_frame(%"PRIu16")\n",
394 num_samples[c], s->samples_per_frame);
395 return AVERROR_INVALIDDATA;
397 } else if (num_samples[c] <= min_channel_len) {
398 if (num_samples[c] < min_channel_len) {
399 channels_for_cur_subframe = 0;
400 min_channel_len = num_samples[c];
402 ++channels_for_cur_subframe;
405 } while (min_channel_len < s->samples_per_frame);
407 for (c = 0; c < s->num_channels; c++) {
409 for (i = 0; i < s->channel[c].num_subframes; i++) {
410 s->channel[c].subframe_offsets[i] = offset;
411 offset += s->channel[c].subframe_len[i];
418 static void decode_ac_filter(WmallDecodeCtx *s)
421 s->acfilter_order = get_bits(&s->gb, 4) + 1;
422 s->acfilter_scaling = get_bits(&s->gb, 4);
424 for (i = 0; i < s->acfilter_order; i++)
425 s->acfilter_coeffs[i] = get_bitsz(&s->gb, s->acfilter_scaling) + 1;
428 static void decode_mclms(WmallDecodeCtx *s)
430 s->mclms_order = (get_bits(&s->gb, 4) + 1) * 2;
431 s->mclms_scaling = get_bits(&s->gb, 4);
432 if (get_bits1(&s->gb)) {
433 int i, send_coef_bits;
434 int cbits = av_log2(s->mclms_scaling + 1);
435 if (1 << cbits < s->mclms_scaling + 1)
438 send_coef_bits = get_bitsz(&s->gb, cbits) + 2;
440 for (i = 0; i < s->mclms_order * s->num_channels * s->num_channels; i++)
441 s->mclms_coeffs[i] = get_bits(&s->gb, send_coef_bits);
443 for (i = 0; i < s->num_channels; i++) {
445 for (c = 0; c < i; c++)
446 s->mclms_coeffs_cur[i * s->num_channels + c] = get_bits(&s->gb, send_coef_bits);
451 static int decode_cdlms(WmallDecodeCtx *s)
454 int cdlms_send_coef = get_bits1(&s->gb);
456 for (c = 0; c < s->num_channels; c++) {
457 s->cdlms_ttl[c] = get_bits(&s->gb, 3) + 1;
458 for (i = 0; i < s->cdlms_ttl[c]; i++) {
459 s->cdlms[c][i].order = (get_bits(&s->gb, 7) + 1) * 8;
460 if (s->cdlms[c][i].order > MAX_ORDER) {
461 av_log(s->avctx, AV_LOG_ERROR,
462 "Order[%d][%d] %d > max (%d), not supported\n",
463 c, i, s->cdlms[c][i].order, MAX_ORDER);
464 s->cdlms[0][0].order = 0;
465 return AVERROR_INVALIDDATA;
467 if(s->cdlms[c][i].order & 8 && s->bits_per_sample == 16) {
470 avpriv_request_sample(s->avctx, "CDLMS of order %d",
471 s->cdlms[c][i].order);
476 for (i = 0; i < s->cdlms_ttl[c]; i++)
477 s->cdlms[c][i].scaling = get_bits(&s->gb, 4);
479 if (cdlms_send_coef) {
480 for (i = 0; i < s->cdlms_ttl[c]; i++) {
481 int cbits, shift_l, shift_r, j;
482 cbits = av_log2(s->cdlms[c][i].order);
483 if ((1 << cbits) < s->cdlms[c][i].order)
485 s->cdlms[c][i].coefsend = get_bits(&s->gb, cbits) + 1;
487 cbits = av_log2(s->cdlms[c][i].scaling + 1);
488 if ((1 << cbits) < s->cdlms[c][i].scaling + 1)
491 s->cdlms[c][i].bitsend = get_bitsz(&s->gb, cbits) + 2;
492 shift_l = 32 - s->cdlms[c][i].bitsend;
493 shift_r = 32 - s->cdlms[c][i].scaling - 2;
494 for (j = 0; j < s->cdlms[c][i].coefsend; j++)
495 s->cdlms[c][i].coefs[j] =
496 (get_bits(&s->gb, s->cdlms[c][i].bitsend) << shift_l) >> shift_r;
500 for (i = 0; i < s->cdlms_ttl[c]; i++)
501 memset(s->cdlms[c][i].coefs + s->cdlms[c][i].order,
502 0, WMALL_COEFF_PAD_SIZE);
508 static int decode_channel_residues(WmallDecodeCtx *s, int ch, int tile_size)
511 unsigned int ave_mean;
512 s->transient[ch] = get_bits1(&s->gb);
513 if (s->transient[ch]) {
514 s->transient_pos[ch] = get_bits(&s->gb, av_log2(tile_size));
515 if (s->transient_pos[ch])
516 s->transient[ch] = 0;
517 s->channel[ch].transient_counter =
518 FFMAX(s->channel[ch].transient_counter, s->samples_per_frame / 2);
519 } else if (s->channel[ch].transient_counter)
520 s->transient[ch] = 1;
522 if (s->seekable_tile) {
523 ave_mean = get_bits(&s->gb, s->bits_per_sample);
524 s->ave_sum[ch] = ave_mean << (s->movave_scaling + 1);
527 if (s->seekable_tile) {
528 if (s->do_inter_ch_decorr)
529 s->channel_residues[ch][0] = get_sbits_long(&s->gb, s->bits_per_sample + 1);
531 s->channel_residues[ch][0] = get_sbits_long(&s->gb, s->bits_per_sample);
534 for (; i < tile_size; i++) {
536 unsigned quo = 0, residue;
537 while(get_bits1(&s->gb)) {
539 if (get_bits_left(&s->gb) <= 0)
543 quo += get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1);
545 ave_mean = (s->ave_sum[ch] + (1 << s->movave_scaling)) >> (s->movave_scaling + 1);
549 rem_bits = av_ceil_log2(ave_mean);
550 rem = get_bits_long(&s->gb, rem_bits);
551 residue = (quo << rem_bits) + rem;
554 s->ave_sum[ch] = residue + s->ave_sum[ch] -
555 (s->ave_sum[ch] >> s->movave_scaling);
557 residue = (residue >> 1) ^ -(residue & 1);
558 s->channel_residues[ch][i] = residue;
565 static void decode_lpc(WmallDecodeCtx *s)
568 s->lpc_order = get_bits(&s->gb, 5) + 1;
569 s->lpc_scaling = get_bits(&s->gb, 4);
570 s->lpc_intbits = get_bits(&s->gb, 3) + 1;
571 cbits = s->lpc_scaling + s->lpc_intbits;
572 for (ch = 0; ch < s->num_channels; ch++)
573 for (i = 0; i < s->lpc_order; i++)
574 s->lpc_coefs[ch][i] = get_sbits(&s->gb, cbits);
577 static void clear_codec_buffers(WmallDecodeCtx *s)
581 memset(s->acfilter_coeffs, 0, sizeof(s->acfilter_coeffs));
582 memset(s->acfilter_prevvalues, 0, sizeof(s->acfilter_prevvalues));
583 memset(s->lpc_coefs, 0, sizeof(s->lpc_coefs));
585 memset(s->mclms_coeffs, 0, sizeof(s->mclms_coeffs));
586 memset(s->mclms_coeffs_cur, 0, sizeof(s->mclms_coeffs_cur));
587 memset(s->mclms_prevvalues, 0, sizeof(s->mclms_prevvalues));
588 memset(s->mclms_updates, 0, sizeof(s->mclms_updates));
590 for (ich = 0; ich < s->num_channels; ich++) {
591 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++) {
592 memset(s->cdlms[ich][ilms].coefs, 0,
593 sizeof(s->cdlms[ich][ilms].coefs));
594 memset(s->cdlms[ich][ilms].lms_prevvalues, 0,
595 sizeof(s->cdlms[ich][ilms].lms_prevvalues));
596 memset(s->cdlms[ich][ilms].lms_updates, 0,
597 sizeof(s->cdlms[ich][ilms].lms_updates));
604 * @brief Reset filter parameters and transient area at new seekable tile.
606 static void reset_codec(WmallDecodeCtx *s)
609 s->mclms_recent = s->mclms_order * s->num_channels;
610 for (ich = 0; ich < s->num_channels; ich++) {
611 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++)
612 s->cdlms[ich][ilms].recent = s->cdlms[ich][ilms].order;
613 /* first sample of a seekable subframe is considered as the starting of
614 a transient area which is samples_per_frame samples long */
615 s->channel[ich].transient_counter = s->samples_per_frame;
616 s->transient[ich] = 1;
617 s->transient_pos[ich] = 0;
621 static void mclms_update(WmallDecodeCtx *s, int icoef, int *pred)
623 int i, j, ich, pred_error;
624 int order = s->mclms_order;
625 int num_channels = s->num_channels;
626 int range = 1 << (s->bits_per_sample - 1);
628 for (ich = 0; ich < num_channels; ich++) {
629 pred_error = s->channel_residues[ich][icoef] - (unsigned)pred[ich];
630 if (pred_error > 0) {
631 for (i = 0; i < order * num_channels; i++)
632 s->mclms_coeffs[i + ich * order * num_channels] +=
633 s->mclms_updates[s->mclms_recent + i];
634 for (j = 0; j < ich; j++)
635 s->mclms_coeffs_cur[ich * num_channels + j] += WMASIGN(s->channel_residues[j][icoef]);
636 } else if (pred_error < 0) {
637 for (i = 0; i < order * num_channels; i++)
638 s->mclms_coeffs[i + ich * order * num_channels] -=
639 s->mclms_updates[s->mclms_recent + i];
640 for (j = 0; j < ich; j++)
641 s->mclms_coeffs_cur[ich * num_channels + j] -= WMASIGN(s->channel_residues[j][icoef]);
645 for (ich = num_channels - 1; ich >= 0; ich--) {
647 s->mclms_prevvalues[s->mclms_recent] = av_clip(s->channel_residues[ich][icoef],
649 s->mclms_updates[s->mclms_recent] = WMASIGN(s->channel_residues[ich][icoef]);
652 if (s->mclms_recent == 0) {
653 memcpy(&s->mclms_prevvalues[order * num_channels],
655 sizeof(int32_t) * order * num_channels);
656 memcpy(&s->mclms_updates[order * num_channels],
658 sizeof(int32_t) * order * num_channels);
659 s->mclms_recent = num_channels * order;
663 static void mclms_predict(WmallDecodeCtx *s, int icoef, int *pred)
666 int order = s->mclms_order;
667 int num_channels = s->num_channels;
669 for (ich = 0; ich < num_channels; ich++) {
671 if (!s->is_channel_coded[ich])
673 for (i = 0; i < order * num_channels; i++)
674 pred[ich] += (uint32_t)s->mclms_prevvalues[i + s->mclms_recent] *
675 s->mclms_coeffs[i + order * num_channels * ich];
676 for (i = 0; i < ich; i++)
677 pred[ich] += (uint32_t)s->channel_residues[i][icoef] *
678 s->mclms_coeffs_cur[i + num_channels * ich];
679 pred[ich] += (1U << s->mclms_scaling) >> 1;
680 pred[ich] >>= s->mclms_scaling;
681 s->channel_residues[ich][icoef] += (unsigned)pred[ich];
685 static void revert_mclms(WmallDecodeCtx *s, int tile_size)
687 int icoef, pred[WMALL_MAX_CHANNELS] = { 0 };
688 for (icoef = 0; icoef < tile_size; icoef++) {
689 mclms_predict(s, icoef, pred);
690 mclms_update(s, icoef, pred);
694 static void use_high_update_speed(WmallDecodeCtx *s, int ich)
696 int ilms, recent, icoef;
697 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
698 recent = s->cdlms[ich][ilms].recent;
699 if (s->update_speed[ich] == 16)
702 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
703 s->cdlms[ich][ilms].lms_updates[icoef + recent] *= 2;
705 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
706 s->cdlms[ich][ilms].lms_updates[icoef] *= 2;
709 s->update_speed[ich] = 16;
712 static void use_normal_update_speed(WmallDecodeCtx *s, int ich)
714 int ilms, recent, icoef;
715 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
716 recent = s->cdlms[ich][ilms].recent;
717 if (s->update_speed[ich] == 8)
720 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
721 s->cdlms[ich][ilms].lms_updates[icoef + recent] /= 2;
723 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
724 s->cdlms[ich][ilms].lms_updates[icoef] /= 2;
726 s->update_speed[ich] = 8;
729 #define CD_LMS(bits, ROUND) \
730 static void lms_update ## bits (WmallDecodeCtx *s, int ich, int ilms, int input) \
732 int recent = s->cdlms[ich][ilms].recent; \
733 int range = 1 << s->bits_per_sample - 1; \
734 int order = s->cdlms[ich][ilms].order; \
735 int ##bits##_t *prev = (int##bits##_t *)s->cdlms[ich][ilms].lms_prevvalues; \
740 memcpy(prev + order, prev, (bits/8) * order); \
741 memcpy(s->cdlms[ich][ilms].lms_updates + order, \
742 s->cdlms[ich][ilms].lms_updates, \
743 sizeof(*s->cdlms[ich][ilms].lms_updates) * order); \
744 recent = order - 1; \
747 prev[recent] = av_clip(input, -range, range - 1); \
748 s->cdlms[ich][ilms].lms_updates[recent] = WMASIGN(input) * s->update_speed[ich]; \
750 s->cdlms[ich][ilms].lms_updates[recent + (order >> 4)] >>= 2; \
751 s->cdlms[ich][ilms].lms_updates[recent + (order >> 3)] >>= 1; \
752 s->cdlms[ich][ilms].recent = recent; \
753 memset(s->cdlms[ich][ilms].lms_updates + recent + order, 0, \
754 sizeof(s->cdlms[ich][ilms].lms_updates) - \
755 sizeof(*s->cdlms[ich][ilms].lms_updates)*(recent+order)); \
758 static void revert_cdlms ## bits (WmallDecodeCtx *s, int ch, \
759 int coef_begin, int coef_end) \
761 int icoef, ilms, num_lms, residue, input; \
764 num_lms = s->cdlms_ttl[ch]; \
765 for (ilms = num_lms - 1; ilms >= 0; ilms--) { \
766 for (icoef = coef_begin; icoef < coef_end; icoef++) { \
767 int##bits##_t *prevvalues = (int##bits##_t *)s->cdlms[ch][ilms].lms_prevvalues; \
768 pred = (1 << s->cdlms[ch][ilms].scaling) >> 1; \
769 residue = s->channel_residues[ch][icoef]; \
770 pred += s->dsp.scalarproduct_and_madd_int## bits (s->cdlms[ch][ilms].coefs, \
771 prevvalues + s->cdlms[ch][ilms].recent, \
772 s->cdlms[ch][ilms].lms_updates + \
773 s->cdlms[ch][ilms].recent, \
774 FFALIGN(s->cdlms[ch][ilms].order, ROUND), \
776 input = residue + (unsigned)((int)pred >> s->cdlms[ch][ilms].scaling); \
777 lms_update ## bits(s, ch, ilms, input); \
778 s->channel_residues[ch][icoef] = input; \
781 if (bits <= 16) emms_c(); \
784 CD_LMS(16, WMALL_COEFF_PAD_SIZE)
787 static void revert_inter_ch_decorr(WmallDecodeCtx *s, int tile_size)
789 if (s->num_channels != 2)
791 else if (s->is_channel_coded[0] || s->is_channel_coded[1]) {
793 for (icoef = 0; icoef < tile_size; icoef++) {
794 s->channel_residues[0][icoef] -= (unsigned)(s->channel_residues[1][icoef] >> 1);
795 s->channel_residues[1][icoef] += (unsigned) s->channel_residues[0][icoef];
800 static void revert_acfilter(WmallDecodeCtx *s, int tile_size)
803 int16_t *filter_coeffs = s->acfilter_coeffs;
804 int scaling = s->acfilter_scaling;
805 int order = s->acfilter_order;
807 for (ich = 0; ich < s->num_channels; ich++) {
808 int *prevvalues = s->acfilter_prevvalues[ich];
809 for (i = 0; i < order; i++) {
811 for (j = 0; j < order; j++) {
813 pred += (uint32_t)filter_coeffs[j] * prevvalues[j - i];
815 pred += (uint32_t)s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
818 s->channel_residues[ich][i] += (unsigned)pred;
820 for (i = order; i < tile_size; i++) {
822 for (j = 0; j < order; j++)
823 pred += (uint32_t)s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
825 s->channel_residues[ich][i] += (unsigned)pred;
827 for (j = order - 1; j >= 0; j--)
828 if (tile_size <= j) {
829 prevvalues[j] = prevvalues[j - tile_size];
831 prevvalues[j] = s->channel_residues[ich][tile_size - j - 1];
835 static int decode_subframe(WmallDecodeCtx *s)
837 int offset = s->samples_per_frame;
838 int subframe_len = s->samples_per_frame;
839 int total_samples = s->samples_per_frame * s->num_channels;
840 int i, j, rawpcm_tile, padding_zeroes, res;
842 s->subframe_offset = get_bits_count(&s->gb);
844 /* reset channel context and find the next block offset and size
845 == the next block of the channel with the smallest number of
847 for (i = 0; i < s->num_channels; i++) {
848 if (offset > s->channel[i].decoded_samples) {
849 offset = s->channel[i].decoded_samples;
851 s->channel[i].subframe_len[s->channel[i].cur_subframe];
855 /* get a list of all channels that contain the estimated block */
856 s->channels_for_cur_subframe = 0;
857 for (i = 0; i < s->num_channels; i++) {
858 const int cur_subframe = s->channel[i].cur_subframe;
859 /* subtract already processed samples */
860 total_samples -= s->channel[i].decoded_samples;
862 /* and count if there are multiple subframes that match our profile */
863 if (offset == s->channel[i].decoded_samples &&
864 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
865 total_samples -= s->channel[i].subframe_len[cur_subframe];
866 s->channel[i].decoded_samples +=
867 s->channel[i].subframe_len[cur_subframe];
868 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
869 ++s->channels_for_cur_subframe;
873 /* check if the frame will be complete after processing the
876 s->parsed_all_subframes = 1;
879 s->seekable_tile = get_bits1(&s->gb);
880 if (s->seekable_tile) {
881 clear_codec_buffers(s);
883 s->do_arith_coding = get_bits1(&s->gb);
884 if (s->do_arith_coding) {
885 avpriv_request_sample(s->avctx, "Arithmetic coding");
886 return AVERROR_PATCHWELCOME;
888 s->do_ac_filter = get_bits1(&s->gb);
889 s->do_inter_ch_decorr = get_bits1(&s->gb);
890 s->do_mclms = get_bits1(&s->gb);
898 if ((res = decode_cdlms(s)) < 0)
900 s->movave_scaling = get_bits(&s->gb, 3);
901 s->quant_stepsize = get_bits(&s->gb, 8) + 1;
906 rawpcm_tile = get_bits1(&s->gb);
908 if (!rawpcm_tile && !s->cdlms[0][0].order) {
909 av_log(s->avctx, AV_LOG_DEBUG,
910 "Waiting for seekable tile\n");
911 av_frame_unref(s->frame);
916 for (i = 0; i < s->num_channels; i++)
917 s->is_channel_coded[i] = 1;
920 for (i = 0; i < s->num_channels; i++)
921 s->is_channel_coded[i] = get_bits1(&s->gb);
925 s->do_lpc = get_bits1(&s->gb);
928 avpriv_request_sample(s->avctx, "Expect wrong output since "
929 "inverse LPC filter");
936 if (get_bits1(&s->gb))
937 padding_zeroes = get_bits(&s->gb, 5);
942 int bits = s->bits_per_sample - padding_zeroes;
944 av_log(s->avctx, AV_LOG_ERROR,
945 "Invalid number of padding bits in raw PCM tile\n");
946 return AVERROR_INVALIDDATA;
948 ff_dlog(s->avctx, "RAWPCM %d bits per sample. "
949 "total %d bits, remain=%d\n", bits,
950 bits * s->num_channels * subframe_len, get_bits_count(&s->gb));
951 for (i = 0; i < s->num_channels; i++)
952 for (j = 0; j < subframe_len; j++)
953 s->channel_residues[i][j] = get_sbits_long(&s->gb, bits);
955 if (s->bits_per_sample < padding_zeroes)
956 return AVERROR_INVALIDDATA;
957 for (i = 0; i < s->num_channels; i++) {
958 if (s->is_channel_coded[i]) {
959 decode_channel_residues(s, i, subframe_len);
960 if (s->seekable_tile)
961 use_high_update_speed(s, i);
963 use_normal_update_speed(s, i);
964 if (s->bits_per_sample > 16)
965 revert_cdlms32(s, i, 0, subframe_len);
967 revert_cdlms16(s, i, 0, subframe_len);
969 memset(s->channel_residues[i], 0, sizeof(**s->channel_residues) * subframe_len);
974 revert_mclms(s, subframe_len);
975 if (s->do_inter_ch_decorr)
976 revert_inter_ch_decorr(s, subframe_len);
978 revert_acfilter(s, subframe_len);
981 if (s->quant_stepsize != 1)
982 for (i = 0; i < s->num_channels; i++)
983 for (j = 0; j < subframe_len; j++)
984 s->channel_residues[i][j] *= (unsigned)s->quant_stepsize;
987 /* Write to proper output buffer depending on bit-depth */
988 for (i = 0; i < s->channels_for_cur_subframe; i++) {
989 int c = s->channel_indexes_for_cur_subframe[i];
990 int subframe_len = s->channel[c].subframe_len[s->channel[c].cur_subframe];
992 for (j = 0; j < subframe_len; j++) {
993 if (s->bits_per_sample == 16) {
994 *s->samples_16[c]++ = (int16_t) s->channel_residues[c][j] * (1 << padding_zeroes);
996 *s->samples_32[c]++ = s->channel_residues[c][j] * (256U << padding_zeroes);
1001 /* handled one subframe */
1002 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1003 int c = s->channel_indexes_for_cur_subframe[i];
1004 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1005 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1006 return AVERROR_INVALIDDATA;
1008 ++s->channel[c].cur_subframe;
1014 * @brief Decode one WMA frame.
1015 * @param s codec context
1016 * @return 0 if the trailer bit indicates that this is the last frame,
1017 * 1 if there are additional frames
1019 static int decode_frame(WmallDecodeCtx *s)
1021 GetBitContext* gb = &s->gb;
1022 int more_frames = 0, len = 0, i, ret;
1024 s->frame->nb_samples = s->samples_per_frame;
1025 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) {
1026 /* return an error if no frame could be decoded at all */
1028 s->frame->nb_samples = 0;
1031 for (i = 0; i < s->num_channels; i++) {
1032 s->samples_16[i] = (int16_t *)s->frame->extended_data[i];
1033 s->samples_32[i] = (int32_t *)s->frame->extended_data[i];
1036 /* get frame length */
1038 len = get_bits(gb, s->log2_frame_size);
1040 /* decode tile information */
1041 if ((ret = decode_tilehdr(s))) {
1043 av_frame_unref(s->frame);
1048 if (s->dynamic_range_compression)
1049 s->drc_gain = get_bits(gb, 8);
1051 /* no idea what these are for, might be the number of samples
1052 that need to be skipped at the beginning or end of a stream */
1053 if (get_bits1(gb)) {
1056 /* usually true for the first frame */
1057 if (get_bits1(gb)) {
1058 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1059 ff_dlog(s->avctx, "start skip: %i\n", skip);
1062 /* sometimes true for the last frame */
1063 if (get_bits1(gb)) {
1064 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1065 ff_dlog(s->avctx, "end skip: %i\n", skip);
1066 s->frame->nb_samples -= skip;
1067 if (s->frame->nb_samples <= 0)
1068 return AVERROR_INVALIDDATA;
1073 /* reset subframe states */
1074 s->parsed_all_subframes = 0;
1075 for (i = 0; i < s->num_channels; i++) {
1076 s->channel[i].decoded_samples = 0;
1077 s->channel[i].cur_subframe = 0;
1080 /* decode all subframes */
1081 while (!s->parsed_all_subframes) {
1082 int decoded_samples = s->channel[0].decoded_samples;
1083 if (decode_subframe(s) < 0) {
1085 if (s->frame->nb_samples)
1086 s->frame->nb_samples = decoded_samples;
1091 ff_dlog(s->avctx, "Frame done\n");
1095 if (s->len_prefix) {
1096 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1097 /* FIXME: not sure if this is always an error */
1098 av_log(s->avctx, AV_LOG_ERROR,
1099 "frame[%"PRIu32"] would have to skip %i bits\n",
1101 len - (get_bits_count(gb) - s->frame_offset) - 1);
1106 /* skip the rest of the frame data */
1107 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1110 /* decode trailer bit */
1111 more_frames = get_bits1(gb);
1117 * @brief Calculate remaining input buffer length.
1118 * @param s codec context
1119 * @param gb bitstream reader context
1120 * @return remaining size in bits
1122 static int remaining_bits(WmallDecodeCtx *s, GetBitContext *gb)
1124 return s->buf_bit_size - get_bits_count(gb);
1128 * @brief Fill the bit reservoir with a (partial) frame.
1129 * @param s codec context
1130 * @param gb bitstream reader context
1131 * @param len length of the partial frame
1132 * @param append decides whether to reset the buffer or not
1134 static void save_bits(WmallDecodeCtx *s, GetBitContext* gb, int len,
1140 /* when the frame data does not need to be concatenated, the input buffer
1141 is reset and additional bits from the previous frame are copied
1142 and skipped later so that a fast byte copy is possible */
1145 s->frame_offset = get_bits_count(gb) & 7;
1146 s->num_saved_bits = s->frame_offset;
1147 init_put_bits(&s->pb, s->frame_data, s->max_frame_size);
1150 buflen = (s->num_saved_bits + len + 8) >> 3;
1152 if (len <= 0 || buflen > s->max_frame_size) {
1153 avpriv_request_sample(s->avctx, "Too small input buffer");
1155 s->num_saved_bits = 0;
1159 s->num_saved_bits += len;
1161 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1164 int align = 8 - (get_bits_count(gb) & 7);
1165 align = FFMIN(align, len);
1166 put_bits(&s->pb, align, get_bits(gb, align));
1168 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1170 skip_bits_long(gb, len);
1173 flush_put_bits(&tmp);
1175 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1176 skip_bits(&s->gb, s->frame_offset);
1179 static int decode_packet(AVCodecContext *avctx, void *data, int *got_frame_ptr,
1182 WmallDecodeCtx *s = avctx->priv_data;
1183 GetBitContext* gb = &s->pgb;
1184 const uint8_t* buf = avpkt->data;
1185 int buf_size = avpkt->size;
1186 int num_bits_prev_frame, packet_sequence_number, spliced_packet;
1188 s->frame->nb_samples = 0;
1190 if (!buf_size && s->num_saved_bits > get_bits_count(&s->gb)) {
1192 if (!decode_frame(s))
1193 s->num_saved_bits = 0;
1194 } else if (s->packet_done || s->packet_loss) {
1200 s->next_packet_start = buf_size - FFMIN(avctx->block_align, buf_size);
1201 buf_size = FFMIN(avctx->block_align, buf_size);
1202 s->buf_bit_size = buf_size << 3;
1204 /* parse packet header */
1205 init_get_bits(gb, buf, s->buf_bit_size);
1206 packet_sequence_number = get_bits(gb, 4);
1207 skip_bits(gb, 1); // Skip seekable_frame_in_packet, currently unused
1208 spliced_packet = get_bits1(gb);
1210 avpriv_request_sample(avctx, "Bitstream splicing");
1212 /* get number of bits that need to be added to the previous frame */
1213 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1215 /* check for packet loss */
1216 if (!s->packet_loss &&
1217 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1219 av_log(avctx, AV_LOG_ERROR,
1220 "Packet loss detected! seq %"PRIx8" vs %x\n",
1221 s->packet_sequence_number, packet_sequence_number);
1223 s->packet_sequence_number = packet_sequence_number;
1225 if (num_bits_prev_frame > 0) {
1226 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1227 if (num_bits_prev_frame >= remaining_packet_bits) {
1228 num_bits_prev_frame = remaining_packet_bits;
1232 /* Append the previous frame data to the remaining data from the
1233 * previous packet to create a full frame. */
1234 save_bits(s, gb, num_bits_prev_frame, 1);
1236 /* decode the cross packet frame if it is valid */
1237 if (num_bits_prev_frame < remaining_packet_bits && !s->packet_loss)
1239 } else if (s->num_saved_bits - s->frame_offset) {
1240 ff_dlog(avctx, "ignoring %x previously saved bits\n",
1241 s->num_saved_bits - s->frame_offset);
1244 if (s->packet_loss) {
1245 /* Reset number of saved bits so that the decoder does not start
1246 * to decode incomplete frames in the s->len_prefix == 0 case. */
1247 s->num_saved_bits = 0;
1249 init_put_bits(&s->pb, s->frame_data, s->max_frame_size);
1255 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1256 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1257 skip_bits(gb, s->packet_offset);
1259 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1260 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1261 frame_size <= remaining_bits(s, gb)) {
1262 save_bits(s, gb, frame_size, 0);
1264 if (!s->packet_loss)
1265 s->packet_done = !decode_frame(s);
1266 } else if (!s->len_prefix
1267 && s->num_saved_bits > get_bits_count(&s->gb)) {
1268 /* when the frames do not have a length prefix, we don't know the
1269 * compressed length of the individual frames however, we know what
1270 * part of a new packet belongs to the previous frame therefore we
1271 * save the incoming packet first, then we append the "previous
1272 * frame" data from the next packet so that we get a buffer that
1273 * only contains full frames */
1274 s->packet_done = !decode_frame(s);
1280 if (remaining_bits(s, gb) < 0) {
1281 av_log(avctx, AV_LOG_ERROR, "Overread %d\n", -remaining_bits(s, gb));
1285 if (s->packet_done && !s->packet_loss &&
1286 remaining_bits(s, gb) > 0) {
1287 /* save the rest of the data so that it can be decoded
1288 * with the next packet */
1289 save_bits(s, gb, remaining_bits(s, gb), 0);
1292 *got_frame_ptr = s->frame->nb_samples > 0;
1293 av_frame_move_ref(data, s->frame);
1295 s->packet_offset = get_bits_count(gb) & 7;
1297 return (s->packet_loss) ? AVERROR_INVALIDDATA : buf_size ? get_bits_count(gb) >> 3 : 0;
1300 static void flush(AVCodecContext *avctx)
1302 WmallDecodeCtx *s = avctx->priv_data;
1305 s->num_saved_bits = 0;
1306 s->frame_offset = 0;
1307 s->next_packet_start = 0;
1308 s->cdlms[0][0].order = 0;
1309 s->frame->nb_samples = 0;
1310 init_put_bits(&s->pb, s->frame_data, s->max_frame_size);
1313 static av_cold int decode_close(AVCodecContext *avctx)
1315 WmallDecodeCtx *s = avctx->priv_data;
1317 av_frame_free(&s->frame);
1318 av_freep(&s->frame_data);
1323 AVCodec ff_wmalossless_decoder = {
1324 .name = "wmalossless",
1325 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio Lossless"),
1326 .type = AVMEDIA_TYPE_AUDIO,
1327 .id = AV_CODEC_ID_WMALOSSLESS,
1328 .priv_data_size = sizeof(WmallDecodeCtx),
1329 .init = decode_init,
1330 .close = decode_close,
1331 .decode = decode_packet,
1333 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
1334 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
1335 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
1337 AV_SAMPLE_FMT_NONE },