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"
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[WMALL_MAX_CHANNELS][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];
150 } cdlms[WMALL_MAX_CHANNELS][9];
152 int cdlms_ttl[WMALL_MAX_CHANNELS];
156 int is_channel_coded[WMALL_MAX_CHANNELS];
157 int update_speed[WMALL_MAX_CHANNELS];
159 int transient[WMALL_MAX_CHANNELS];
160 int transient_pos[WMALL_MAX_CHANNELS];
163 int ave_sum[WMALL_MAX_CHANNELS];
165 int channel_residues[WMALL_MAX_CHANNELS][WMALL_BLOCK_MAX_SIZE];
167 int lpc_coefs[WMALL_MAX_CHANNELS][40];
172 int channel_coeffs[WMALL_MAX_CHANNELS][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;
183 if (!avctx->block_align) {
184 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
185 return AVERROR(EINVAL);
189 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
191 if (avctx->extradata_size >= 18) {
192 s->decode_flags = AV_RL16(edata_ptr + 14);
193 channel_mask = AV_RL32(edata_ptr + 2);
194 s->bits_per_sample = AV_RL16(edata_ptr);
195 if (s->bits_per_sample == 16)
196 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
197 else if (s->bits_per_sample == 24) {
198 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
199 avpriv_report_missing_feature(avctx, "Bit-depth higher than 16");
200 return AVERROR_PATCHWELCOME;
202 av_log(avctx, AV_LOG_ERROR, "Unknown bit-depth: %"PRIu8"\n",
204 return AVERROR_INVALIDDATA;
206 /* dump the extradata */
207 for (i = 0; i < avctx->extradata_size; i++)
208 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
209 av_dlog(avctx, "\n");
212 avpriv_request_sample(avctx, "Unsupported extradata size");
213 return AVERROR_PATCHWELCOME;
217 s->log2_frame_size = av_log2(avctx->block_align) + 4;
220 s->skip_frame = 1; /* skip first frame */
222 s->len_prefix = s->decode_flags & 0x40;
225 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
227 av_assert0(s->samples_per_frame <= WMALL_BLOCK_MAX_SIZE);
229 /* init previous block len */
230 for (i = 0; i < avctx->channels; i++)
231 s->channel[i].prev_block_len = s->samples_per_frame;
234 log2_max_num_subframes = (s->decode_flags & 0x38) >> 3;
235 s->max_num_subframes = 1 << log2_max_num_subframes;
236 s->max_subframe_len_bit = 0;
237 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
239 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
240 s->dynamic_range_compression = s->decode_flags & 0x80;
241 s->bV3RTM = s->decode_flags & 0x100;
243 if (s->max_num_subframes > MAX_SUBFRAMES) {
244 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRIu8"\n",
245 s->max_num_subframes);
246 return AVERROR_INVALIDDATA;
249 s->num_channels = avctx->channels;
251 /* extract lfe channel position */
254 if (channel_mask & 8) {
256 for (mask = 1; mask < 16; mask <<= 1)
257 if (channel_mask & mask)
261 if (s->num_channels < 0) {
262 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %"PRId8"\n",
264 return AVERROR_INVALIDDATA;
265 } else if (s->num_channels > WMALL_MAX_CHANNELS) {
266 avpriv_request_sample(avctx,
267 "More than %d channels", WMALL_MAX_CHANNELS);
268 return AVERROR_PATCHWELCOME;
271 s->frame = av_frame_alloc();
273 return AVERROR(ENOMEM);
275 avctx->channel_layout = channel_mask;
280 * @brief Decode the subframe length.
282 * @param offset sample offset in the frame
283 * @return decoded subframe length on success, < 0 in case of an error
285 static int decode_subframe_length(WmallDecodeCtx *s, int offset)
287 int frame_len_ratio, subframe_len, len;
289 /* no need to read from the bitstream when only one length is possible */
290 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
291 return s->min_samples_per_subframe;
293 len = av_log2(s->max_num_subframes - 1) + 1;
294 frame_len_ratio = get_bits(&s->gb, len);
295 subframe_len = s->min_samples_per_subframe * (frame_len_ratio + 1);
297 /* sanity check the length */
298 if (subframe_len < s->min_samples_per_subframe ||
299 subframe_len > s->samples_per_frame) {
300 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
302 return AVERROR_INVALIDDATA;
308 * @brief Decode how the data in the frame is split into subframes.
309 * Every WMA frame contains the encoded data for a fixed number of
310 * samples per channel. The data for every channel might be split
311 * into several subframes. This function will reconstruct the list of
312 * subframes for every channel.
314 * If the subframes are not evenly split, the algorithm estimates the
315 * channels with the lowest number of total samples.
316 * Afterwards, for each of these channels a bit is read from the
317 * bitstream that indicates if the channel contains a subframe with the
318 * next subframe size that is going to be read from the bitstream or not.
319 * If a channel contains such a subframe, the subframe size gets added to
320 * the channel's subframe list.
321 * The algorithm repeats these steps until the frame is properly divided
322 * between the individual channels.
325 * @return 0 on success, < 0 in case of an error
327 static int decode_tilehdr(WmallDecodeCtx *s)
329 uint16_t num_samples[WMALL_MAX_CHANNELS] = { 0 }; /* sum of samples for all currently known subframes of a channel */
330 uint8_t contains_subframe[WMALL_MAX_CHANNELS]; /* flag indicating if a channel contains the current subframe */
331 int channels_for_cur_subframe = s->num_channels; /* number of channels that contain the current subframe */
332 int fixed_channel_layout = 0; /* flag indicating that all channels use the same subfra2me offsets and sizes */
333 int min_channel_len = 0; /* smallest sum of samples (channels with this length will be processed first) */
336 /* reset tiling information */
337 for (c = 0; c < s->num_channels; c++)
338 s->channel[c].num_subframes = 0;
340 tile_aligned = get_bits1(&s->gb);
341 if (s->max_num_subframes == 1 || tile_aligned)
342 fixed_channel_layout = 1;
344 /* loop until the frame data is split between the subframes */
346 int subframe_len, in_use = 0;
348 /* check which channels contain the subframe */
349 for (c = 0; c < s->num_channels; c++) {
350 if (num_samples[c] == min_channel_len) {
351 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
352 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) {
353 contains_subframe[c] = 1;
355 contains_subframe[c] = get_bits1(&s->gb);
357 in_use |= contains_subframe[c];
359 contains_subframe[c] = 0;
363 av_log(s->avctx, AV_LOG_ERROR,
364 "Found empty subframe\n");
365 return AVERROR_INVALIDDATA;
368 /* get subframe length, subframe_len == 0 is not allowed */
369 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
370 return AVERROR_INVALIDDATA;
371 /* add subframes to the individual channels and find new min_channel_len */
372 min_channel_len += subframe_len;
373 for (c = 0; c < s->num_channels; c++) {
374 WmallChannelCtx *chan = &s->channel[c];
376 if (contains_subframe[c]) {
377 if (chan->num_subframes >= MAX_SUBFRAMES) {
378 av_log(s->avctx, AV_LOG_ERROR,
379 "broken frame: num subframes > 31\n");
380 return AVERROR_INVALIDDATA;
382 chan->subframe_len[chan->num_subframes] = subframe_len;
383 num_samples[c] += subframe_len;
384 ++chan->num_subframes;
385 if (num_samples[c] > s->samples_per_frame) {
386 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
387 "channel len(%"PRIu16") > samples_per_frame(%"PRIu16")\n",
388 num_samples[c], s->samples_per_frame);
389 return AVERROR_INVALIDDATA;
391 } else if (num_samples[c] <= min_channel_len) {
392 if (num_samples[c] < min_channel_len) {
393 channels_for_cur_subframe = 0;
394 min_channel_len = num_samples[c];
396 ++channels_for_cur_subframe;
399 } while (min_channel_len < s->samples_per_frame);
401 for (c = 0; c < s->num_channels; c++) {
403 for (i = 0; i < s->channel[c].num_subframes; i++) {
404 s->channel[c].subframe_offsets[i] = offset;
405 offset += s->channel[c].subframe_len[i];
412 static void decode_ac_filter(WmallDecodeCtx *s)
415 s->acfilter_order = get_bits(&s->gb, 4) + 1;
416 s->acfilter_scaling = get_bits(&s->gb, 4);
418 for (i = 0; i < s->acfilter_order; i++)
419 s->acfilter_coeffs[i] = (s->acfilter_scaling ?
420 get_bits(&s->gb, s->acfilter_scaling) : 0) + 1;
423 static void decode_mclms(WmallDecodeCtx *s)
425 s->mclms_order = (get_bits(&s->gb, 4) + 1) * 2;
426 s->mclms_scaling = get_bits(&s->gb, 4);
427 if (get_bits1(&s->gb)) {
428 int i, send_coef_bits;
429 int cbits = av_log2(s->mclms_scaling + 1);
430 if (1 << cbits < s->mclms_scaling + 1)
433 send_coef_bits = (cbits ? get_bits(&s->gb, cbits) : 0) + 2;
435 for (i = 0; i < s->mclms_order * s->num_channels * s->num_channels; i++)
436 s->mclms_coeffs[i] = get_bits(&s->gb, send_coef_bits);
438 for (i = 0; i < s->num_channels; i++) {
440 for (c = 0; c < i; c++)
441 s->mclms_coeffs_cur[i * s->num_channels + c] = get_bits(&s->gb, send_coef_bits);
446 static int decode_cdlms(WmallDecodeCtx *s)
449 int cdlms_send_coef = get_bits1(&s->gb);
451 for (c = 0; c < s->num_channels; c++) {
452 s->cdlms_ttl[c] = get_bits(&s->gb, 3) + 1;
453 for (i = 0; i < s->cdlms_ttl[c]; i++) {
454 s->cdlms[c][i].order = (get_bits(&s->gb, 7) + 1) * 8;
455 if (s->cdlms[c][i].order > MAX_ORDER) {
456 av_log(s->avctx, AV_LOG_ERROR,
457 "Order[%d][%d] %d > max (%d), not supported\n",
458 c, i, s->cdlms[c][i].order, MAX_ORDER);
459 s->cdlms[0][0].order = 0;
460 return AVERROR_INVALIDDATA;
464 for (i = 0; i < s->cdlms_ttl[c]; i++)
465 s->cdlms[c][i].scaling = get_bits(&s->gb, 4);
467 if (cdlms_send_coef) {
468 for (i = 0; i < s->cdlms_ttl[c]; i++) {
469 int cbits, shift_l, shift_r, j;
470 cbits = av_log2(s->cdlms[c][i].order);
471 if ((1 << cbits) < s->cdlms[c][i].order)
473 s->cdlms[c][i].coefsend = get_bits(&s->gb, cbits) + 1;
475 cbits = av_log2(s->cdlms[c][i].scaling + 1);
476 if ((1 << cbits) < s->cdlms[c][i].scaling + 1)
479 s->cdlms[c][i].bitsend = get_bits(&s->gb, cbits) + 2;
480 shift_l = 32 - s->cdlms[c][i].bitsend;
481 shift_r = 32 - s->cdlms[c][i].scaling - 2;
482 for (j = 0; j < s->cdlms[c][i].coefsend; j++)
483 s->cdlms[c][i].coefs[j] =
484 (get_bits(&s->gb, s->cdlms[c][i].bitsend) << shift_l) >> shift_r;
492 static int decode_channel_residues(WmallDecodeCtx *s, int ch, int tile_size)
495 unsigned int ave_mean;
496 s->transient[ch] = get_bits1(&s->gb);
497 if (s->transient[ch]) {
498 s->transient_pos[ch] = get_bits(&s->gb, av_log2(tile_size));
499 if (s->transient_pos[ch])
500 s->transient[ch] = 0;
501 s->channel[ch].transient_counter =
502 FFMAX(s->channel[ch].transient_counter, s->samples_per_frame / 2);
503 } else if (s->channel[ch].transient_counter)
504 s->transient[ch] = 1;
506 if (s->seekable_tile) {
507 ave_mean = get_bits(&s->gb, s->bits_per_sample);
508 s->ave_sum[ch] = ave_mean << (s->movave_scaling + 1);
511 if (s->seekable_tile) {
512 if (s->do_inter_ch_decorr)
513 s->channel_residues[ch][0] = get_sbits_long(&s->gb, s->bits_per_sample + 1);
515 s->channel_residues[ch][0] = get_sbits_long(&s->gb, s->bits_per_sample);
518 for (; i < tile_size; i++) {
519 int quo = 0, rem, rem_bits, residue;
520 while(get_bits1(&s->gb)) {
522 if (get_bits_left(&s->gb) <= 0)
526 quo += get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1);
528 ave_mean = (s->ave_sum[ch] + (1 << s->movave_scaling)) >> (s->movave_scaling + 1);
532 rem_bits = av_ceil_log2(ave_mean);
533 rem = get_bits_long(&s->gb, rem_bits);
534 residue = (quo << rem_bits) + rem;
537 s->ave_sum[ch] = residue + s->ave_sum[ch] -
538 (s->ave_sum[ch] >> s->movave_scaling);
541 residue = -(residue >> 1) - 1;
543 residue = residue >> 1;
544 s->channel_residues[ch][i] = residue;
551 static void decode_lpc(WmallDecodeCtx *s)
554 s->lpc_order = get_bits(&s->gb, 5) + 1;
555 s->lpc_scaling = get_bits(&s->gb, 4);
556 s->lpc_intbits = get_bits(&s->gb, 3) + 1;
557 cbits = s->lpc_scaling + s->lpc_intbits;
558 for (ch = 0; ch < s->num_channels; ch++)
559 for (i = 0; i < s->lpc_order; i++)
560 s->lpc_coefs[ch][i] = get_sbits(&s->gb, cbits);
563 static void clear_codec_buffers(WmallDecodeCtx *s)
567 memset(s->acfilter_coeffs, 0, sizeof(s->acfilter_coeffs));
568 memset(s->acfilter_prevvalues, 0, sizeof(s->acfilter_prevvalues));
569 memset(s->lpc_coefs, 0, sizeof(s->lpc_coefs));
571 memset(s->mclms_coeffs, 0, sizeof(s->mclms_coeffs));
572 memset(s->mclms_coeffs_cur, 0, sizeof(s->mclms_coeffs_cur));
573 memset(s->mclms_prevvalues, 0, sizeof(s->mclms_prevvalues));
574 memset(s->mclms_updates, 0, sizeof(s->mclms_updates));
576 for (ich = 0; ich < s->num_channels; ich++) {
577 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++) {
578 memset(s->cdlms[ich][ilms].coefs, 0,
579 sizeof(s->cdlms[ich][ilms].coefs));
580 memset(s->cdlms[ich][ilms].lms_prevvalues, 0,
581 sizeof(s->cdlms[ich][ilms].lms_prevvalues));
582 memset(s->cdlms[ich][ilms].lms_updates, 0,
583 sizeof(s->cdlms[ich][ilms].lms_updates));
590 * @brief Reset filter parameters and transient area at new seekable tile.
592 static void reset_codec(WmallDecodeCtx *s)
595 s->mclms_recent = s->mclms_order * s->num_channels;
596 for (ich = 0; ich < s->num_channels; ich++) {
597 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++)
598 s->cdlms[ich][ilms].recent = s->cdlms[ich][ilms].order;
599 /* first sample of a seekable subframe is considered as the starting of
600 a transient area which is samples_per_frame samples long */
601 s->channel[ich].transient_counter = s->samples_per_frame;
602 s->transient[ich] = 1;
603 s->transient_pos[ich] = 0;
607 static void mclms_update(WmallDecodeCtx *s, int icoef, int *pred)
609 int i, j, ich, pred_error;
610 int order = s->mclms_order;
611 int num_channels = s->num_channels;
612 int range = 1 << (s->bits_per_sample - 1);
614 for (ich = 0; ich < num_channels; ich++) {
615 pred_error = s->channel_residues[ich][icoef] - pred[ich];
616 if (pred_error > 0) {
617 for (i = 0; i < order * num_channels; i++)
618 s->mclms_coeffs[i + ich * order * num_channels] +=
619 s->mclms_updates[s->mclms_recent + i];
620 for (j = 0; j < ich; j++) {
621 if (s->channel_residues[j][icoef] > 0)
622 s->mclms_coeffs_cur[ich * num_channels + j] += 1;
623 else if (s->channel_residues[j][icoef] < 0)
624 s->mclms_coeffs_cur[ich * num_channels + j] -= 1;
626 } else if (pred_error < 0) {
627 for (i = 0; i < order * num_channels; i++)
628 s->mclms_coeffs[i + ich * order * num_channels] -=
629 s->mclms_updates[s->mclms_recent + i];
630 for (j = 0; j < ich; j++) {
631 if (s->channel_residues[j][icoef] > 0)
632 s->mclms_coeffs_cur[ich * num_channels + j] -= 1;
633 else if (s->channel_residues[j][icoef] < 0)
634 s->mclms_coeffs_cur[ich * num_channels + j] += 1;
639 for (ich = num_channels - 1; ich >= 0; ich--) {
641 s->mclms_prevvalues[s->mclms_recent] = s->channel_residues[ich][icoef];
642 if (s->channel_residues[ich][icoef] > range - 1)
643 s->mclms_prevvalues[s->mclms_recent] = range - 1;
644 else if (s->channel_residues[ich][icoef] < -range)
645 s->mclms_prevvalues[s->mclms_recent] = -range;
647 s->mclms_updates[s->mclms_recent] = 0;
648 if (s->channel_residues[ich][icoef] > 0)
649 s->mclms_updates[s->mclms_recent] = 1;
650 else if (s->channel_residues[ich][icoef] < 0)
651 s->mclms_updates[s->mclms_recent] = -1;
654 if (s->mclms_recent == 0) {
655 memcpy(&s->mclms_prevvalues[order * num_channels],
657 sizeof(int16_t) * order * num_channels);
658 memcpy(&s->mclms_updates[order * num_channels],
660 sizeof(int16_t) * order * num_channels);
661 s->mclms_recent = num_channels * order;
665 static void mclms_predict(WmallDecodeCtx *s, int icoef, int *pred)
668 int order = s->mclms_order;
669 int num_channels = s->num_channels;
671 for (ich = 0; ich < num_channels; ich++) {
673 if (!s->is_channel_coded[ich])
675 for (i = 0; i < order * num_channels; i++)
676 pred[ich] += s->mclms_prevvalues[i + s->mclms_recent] *
677 s->mclms_coeffs[i + order * num_channels * ich];
678 for (i = 0; i < ich; i++)
679 pred[ich] += s->channel_residues[i][icoef] *
680 s->mclms_coeffs_cur[i + num_channels * ich];
681 pred[ich] += 1 << s->mclms_scaling - 1;
682 pred[ich] >>= s->mclms_scaling;
683 s->channel_residues[ich][icoef] += pred[ich];
687 static void revert_mclms(WmallDecodeCtx *s, int tile_size)
689 int icoef, pred[WMALL_MAX_CHANNELS] = { 0 };
690 for (icoef = 0; icoef < tile_size; icoef++) {
691 mclms_predict(s, icoef, pred);
692 mclms_update(s, icoef, pred);
696 static int lms_predict(WmallDecodeCtx *s, int ich, int ilms)
699 int recent = s->cdlms[ich][ilms].recent;
701 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
702 pred += s->cdlms[ich][ilms].coefs[icoef] *
703 s->cdlms[ich][ilms].lms_prevvalues[icoef + recent];
708 static void lms_update(WmallDecodeCtx *s, int ich, int ilms,
709 int input, int residue)
712 int recent = s->cdlms[ich][ilms].recent;
713 int range = 1 << s->bits_per_sample - 1;
716 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
717 s->cdlms[ich][ilms].coefs[icoef] -=
718 s->cdlms[ich][ilms].lms_updates[icoef + recent];
719 } else if (residue > 0) {
720 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
721 s->cdlms[ich][ilms].coefs[icoef] +=
722 s->cdlms[ich][ilms].lms_updates[icoef + recent];
728 memcpy(&s->cdlms[ich][ilms].lms_prevvalues[s->cdlms[ich][ilms].order],
729 s->cdlms[ich][ilms].lms_prevvalues,
730 2 * s->cdlms[ich][ilms].order);
731 memcpy(&s->cdlms[ich][ilms].lms_updates[s->cdlms[ich][ilms].order],
732 s->cdlms[ich][ilms].lms_updates,
733 2 * s->cdlms[ich][ilms].order);
734 recent = s->cdlms[ich][ilms].order - 1;
737 s->cdlms[ich][ilms].lms_prevvalues[recent] = av_clip(input, -range, range - 1);
739 s->cdlms[ich][ilms].lms_updates[recent] = 0;
741 s->cdlms[ich][ilms].lms_updates[recent] = -s->update_speed[ich];
743 s->cdlms[ich][ilms].lms_updates[recent] = s->update_speed[ich];
745 s->cdlms[ich][ilms].lms_updates[recent + (s->cdlms[ich][ilms].order >> 4)] >>= 2;
746 s->cdlms[ich][ilms].lms_updates[recent + (s->cdlms[ich][ilms].order >> 3)] >>= 1;
747 s->cdlms[ich][ilms].recent = recent;
750 static void use_high_update_speed(WmallDecodeCtx *s, int ich)
752 int ilms, recent, icoef;
753 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
754 recent = s->cdlms[ich][ilms].recent;
755 if (s->update_speed[ich] == 16)
758 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
759 s->cdlms[ich][ilms].lms_updates[icoef + recent] *= 2;
761 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
762 s->cdlms[ich][ilms].lms_updates[icoef] *= 2;
765 s->update_speed[ich] = 16;
768 static void use_normal_update_speed(WmallDecodeCtx *s, int ich)
770 int ilms, recent, icoef;
771 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
772 recent = s->cdlms[ich][ilms].recent;
773 if (s->update_speed[ich] == 8)
776 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
777 s->cdlms[ich][ilms].lms_updates[icoef + recent] /= 2;
779 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
780 s->cdlms[ich][ilms].lms_updates[icoef] /= 2;
782 s->update_speed[ich] = 8;
785 static void revert_cdlms(WmallDecodeCtx *s, int ch,
786 int coef_begin, int coef_end)
788 int icoef, pred, ilms, num_lms, residue, input;
790 num_lms = s->cdlms_ttl[ch];
791 for (ilms = num_lms - 1; ilms >= 0; ilms--) {
792 for (icoef = coef_begin; icoef < coef_end; icoef++) {
793 pred = 1 << (s->cdlms[ch][ilms].scaling - 1);
794 residue = s->channel_residues[ch][icoef];
795 pred += lms_predict(s, ch, ilms);
796 input = residue + (pred >> s->cdlms[ch][ilms].scaling);
797 lms_update(s, ch, ilms, input, residue);
798 s->channel_residues[ch][icoef] = input;
803 static void revert_inter_ch_decorr(WmallDecodeCtx *s, int tile_size)
805 if (s->num_channels != 2)
807 else if (s->is_channel_coded[0] || s->is_channel_coded[1]) {
809 for (icoef = 0; icoef < tile_size; icoef++) {
810 s->channel_residues[0][icoef] -= s->channel_residues[1][icoef] >> 1;
811 s->channel_residues[1][icoef] += s->channel_residues[0][icoef];
816 static void revert_acfilter(WmallDecodeCtx *s, int tile_size)
819 int64_t *filter_coeffs = s->acfilter_coeffs;
820 int scaling = s->acfilter_scaling;
821 int order = s->acfilter_order;
823 for (ich = 0; ich < s->num_channels; ich++) {
824 int *prevvalues = s->acfilter_prevvalues[ich];
825 for (i = 0; i < order; i++) {
827 for (j = 0; j < order; j++) {
829 pred += filter_coeffs[j] * prevvalues[j - i];
831 pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
834 s->channel_residues[ich][i] += pred;
836 for (i = order; i < tile_size; i++) {
838 for (j = 0; j < order; j++)
839 pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
841 s->channel_residues[ich][i] += pred;
843 for (j = 0; j < order; j++)
844 prevvalues[j] = s->channel_residues[ich][tile_size - j - 1];
848 static int decode_subframe(WmallDecodeCtx *s)
850 int offset = s->samples_per_frame;
851 int subframe_len = s->samples_per_frame;
852 int total_samples = s->samples_per_frame * s->num_channels;
853 int i, j, rawpcm_tile, padding_zeroes, res;
855 s->subframe_offset = get_bits_count(&s->gb);
857 /* reset channel context and find the next block offset and size
858 == the next block of the channel with the smallest number of
860 for (i = 0; i < s->num_channels; i++) {
861 if (offset > s->channel[i].decoded_samples) {
862 offset = s->channel[i].decoded_samples;
864 s->channel[i].subframe_len[s->channel[i].cur_subframe];
868 /* get a list of all channels that contain the estimated block */
869 s->channels_for_cur_subframe = 0;
870 for (i = 0; i < s->num_channels; i++) {
871 const int cur_subframe = s->channel[i].cur_subframe;
872 /* subtract already processed samples */
873 total_samples -= s->channel[i].decoded_samples;
875 /* and count if there are multiple subframes that match our profile */
876 if (offset == s->channel[i].decoded_samples &&
877 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
878 total_samples -= s->channel[i].subframe_len[cur_subframe];
879 s->channel[i].decoded_samples +=
880 s->channel[i].subframe_len[cur_subframe];
881 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
882 ++s->channels_for_cur_subframe;
886 /* check if the frame will be complete after processing the
889 s->parsed_all_subframes = 1;
892 s->seekable_tile = get_bits1(&s->gb);
893 if (s->seekable_tile) {
894 clear_codec_buffers(s);
896 s->do_arith_coding = get_bits1(&s->gb);
897 if (s->do_arith_coding) {
898 avpriv_request_sample(s->avctx, "Arithmetic coding");
899 return AVERROR_PATCHWELCOME;
901 s->do_ac_filter = get_bits1(&s->gb);
902 s->do_inter_ch_decorr = get_bits1(&s->gb);
903 s->do_mclms = get_bits1(&s->gb);
911 if ((res = decode_cdlms(s)) < 0)
913 s->movave_scaling = get_bits(&s->gb, 3);
914 s->quant_stepsize = get_bits(&s->gb, 8) + 1;
917 } else if (!s->cdlms[0][0].order) {
918 av_log(s->avctx, AV_LOG_DEBUG,
919 "Waiting for seekable tile\n");
920 av_frame_unref(s->frame);
924 rawpcm_tile = get_bits1(&s->gb);
926 for (i = 0; i < s->num_channels; i++)
927 s->is_channel_coded[i] = 1;
930 for (i = 0; i < s->num_channels; i++)
931 s->is_channel_coded[i] = get_bits1(&s->gb);
935 s->do_lpc = get_bits1(&s->gb);
938 avpriv_request_sample(s->avctx, "Expect wrong output since "
939 "inverse LPC filter");
946 if (get_bits1(&s->gb))
947 padding_zeroes = get_bits(&s->gb, 5);
952 int bits = s->bits_per_sample - padding_zeroes;
954 av_log(s->avctx, AV_LOG_ERROR,
955 "Invalid number of padding bits in raw PCM tile\n");
956 return AVERROR_INVALIDDATA;
958 av_dlog(s->avctx, "RAWPCM %d bits per sample. "
959 "total %d bits, remain=%d\n", bits,
960 bits * s->num_channels * subframe_len, get_bits_count(&s->gb));
961 for (i = 0; i < s->num_channels; i++)
962 for (j = 0; j < subframe_len; j++)
963 s->channel_coeffs[i][j] = get_sbits_long(&s->gb, bits);
965 for (i = 0; i < s->num_channels; i++)
966 if (s->is_channel_coded[i]) {
967 decode_channel_residues(s, i, subframe_len);
968 if (s->seekable_tile)
969 use_high_update_speed(s, i);
971 use_normal_update_speed(s, i);
972 revert_cdlms(s, i, 0, subframe_len);
974 memset(s->channel_residues[i], 0, sizeof(**s->channel_residues) * subframe_len);
978 revert_mclms(s, subframe_len);
979 if (s->do_inter_ch_decorr)
980 revert_inter_ch_decorr(s, subframe_len);
982 revert_acfilter(s, subframe_len);
985 if (s->quant_stepsize != 1)
986 for (i = 0; i < s->num_channels; i++)
987 for (j = 0; j < subframe_len; j++)
988 s->channel_residues[i][j] *= s->quant_stepsize;
990 /* Write to proper output buffer depending on bit-depth */
991 for (i = 0; i < s->channels_for_cur_subframe; i++) {
992 int c = s->channel_indexes_for_cur_subframe[i];
993 int subframe_len = s->channel[c].subframe_len[s->channel[c].cur_subframe];
995 for (j = 0; j < subframe_len; j++) {
996 if (s->bits_per_sample == 16) {
997 *s->samples_16[c]++ = (int16_t) s->channel_residues[c][j] << padding_zeroes;
999 *s->samples_32[c]++ = s->channel_residues[c][j] << padding_zeroes;
1004 /* handled one subframe */
1005 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1006 int c = s->channel_indexes_for_cur_subframe[i];
1007 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1008 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1009 return AVERROR_INVALIDDATA;
1011 ++s->channel[c].cur_subframe;
1017 * @brief Decode one WMA frame.
1018 * @param s codec context
1019 * @return 0 if the trailer bit indicates that this is the last frame,
1020 * 1 if there are additional frames
1022 static int decode_frame(WmallDecodeCtx *s)
1024 GetBitContext* gb = &s->gb;
1025 int more_frames = 0, len = 0, i, ret;
1027 s->frame->nb_samples = s->samples_per_frame;
1028 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) {
1029 /* return an error if no frame could be decoded at all */
1033 for (i = 0; i < s->num_channels; i++) {
1034 s->samples_16[i] = (int16_t *)s->frame->extended_data[i];
1035 s->samples_32[i] = (int32_t *)s->frame->extended_data[i];
1038 /* get frame length */
1040 len = get_bits(gb, s->log2_frame_size);
1042 /* decode tile information */
1043 if ((ret = decode_tilehdr(s))) {
1045 av_frame_unref(s->frame);
1050 if (s->dynamic_range_compression)
1051 s->drc_gain = get_bits(gb, 8);
1053 /* no idea what these are for, might be the number of samples
1054 that need to be skipped at the beginning or end of a stream */
1055 if (get_bits1(gb)) {
1058 /* usually true for the first frame */
1059 if (get_bits1(gb)) {
1060 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1061 av_dlog(s->avctx, "start skip: %i\n", skip);
1064 /* sometimes true for the last frame */
1065 if (get_bits1(gb)) {
1066 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1067 av_dlog(s->avctx, "end skip: %i\n", skip);
1072 /* reset subframe states */
1073 s->parsed_all_subframes = 0;
1074 for (i = 0; i < s->num_channels; i++) {
1075 s->channel[i].decoded_samples = 0;
1076 s->channel[i].cur_subframe = 0;
1079 /* decode all subframes */
1080 while (!s->parsed_all_subframes) {
1081 int decoded_samples = s->channel[0].decoded_samples;
1082 if (decode_subframe(s) < 0) {
1084 if (s->frame->nb_samples)
1085 s->frame->nb_samples = decoded_samples;
1090 av_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, MAX_FRAMESIZE);
1150 buflen = (s->num_saved_bits + len + 8) >> 3;
1152 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1153 avpriv_request_sample(s->avctx, "Too small input buffer");
1158 s->num_saved_bits += len;
1160 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1163 int align = 8 - (get_bits_count(gb) & 7);
1164 align = FFMIN(align, len);
1165 put_bits(&s->pb, align, get_bits(gb, align));
1167 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1169 skip_bits_long(gb, len);
1172 flush_put_bits(&tmp);
1174 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1175 skip_bits(&s->gb, s->frame_offset);
1178 static int decode_packet(AVCodecContext *avctx, void *data, int *got_frame_ptr,
1181 WmallDecodeCtx *s = avctx->priv_data;
1182 GetBitContext* gb = &s->pgb;
1183 const uint8_t* buf = avpkt->data;
1184 int buf_size = avpkt->size;
1185 int num_bits_prev_frame, packet_sequence_number, spliced_packet;
1187 s->frame->nb_samples = 0;
1189 if (s->packet_done || s->packet_loss) {
1194 /* sanity check for the buffer length */
1195 if (buf_size < avctx->block_align) {
1196 av_log(avctx, AV_LOG_ERROR, "buf size %d invalid\n", buf_size);
1197 return AVERROR_INVALIDDATA;
1200 s->next_packet_start = buf_size - avctx->block_align;
1201 buf_size = avctx->block_align;
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 ununused
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 av_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, MAX_FRAMESIZE);
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);
1263 s->packet_done = !decode_frame(s);
1264 } else if (!s->len_prefix
1265 && s->num_saved_bits > get_bits_count(&s->gb)) {
1266 /* when the frames do not have a length prefix, we don't know the
1267 * compressed length of the individual frames however, we know what
1268 * part of a new packet belongs to the previous frame therefore we
1269 * save the incoming packet first, then we append the "previous
1270 * frame" data from the next packet so that we get a buffer that
1271 * only contains full frames */
1272 s->packet_done = !decode_frame(s);
1278 if (s->packet_done && !s->packet_loss &&
1279 remaining_bits(s, gb) > 0) {
1280 /* save the rest of the data so that it can be decoded
1281 * with the next packet */
1282 save_bits(s, gb, remaining_bits(s, gb), 0);
1285 *got_frame_ptr = s->frame->nb_samples > 0;
1286 av_frame_move_ref(data, s->frame);
1288 s->packet_offset = get_bits_count(gb) & 7;
1290 return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3;
1293 static void flush(AVCodecContext *avctx)
1295 WmallDecodeCtx *s = avctx->priv_data;
1298 s->num_saved_bits = 0;
1299 s->frame_offset = 0;
1300 s->next_packet_start = 0;
1301 s->cdlms[0][0].order = 0;
1302 s->frame->nb_samples = 0;
1303 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1306 static av_cold int decode_close(AVCodecContext *avctx)
1308 WmallDecodeCtx *s = avctx->priv_data;
1310 av_frame_free(&s->frame);
1315 AVCodec ff_wmalossless_decoder = {
1316 .name = "wmalossless",
1317 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio Lossless"),
1318 .type = AVMEDIA_TYPE_AUDIO,
1319 .id = AV_CODEC_ID_WMALOSSLESS,
1320 .priv_data_size = sizeof(WmallDecodeCtx),
1321 .init = decode_init,
1322 .close = decode_close,
1323 .decode = decode_packet,
1325 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1 | CODEC_CAP_DELAY,
1326 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
1328 AV_SAMPLE_FMT_NONE },