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 samplebuffer pointer (16-bit)
109 int32_t *samples_32[WMALL_MAX_CHANNELS]; ///< current samplebuffer 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 int 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) {
188 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
189 return AVERROR(EINVAL);
192 s->max_frame_size = MAX_FRAMESIZE * avctx->channels;
193 s->frame_data = av_mallocz(s->max_frame_size + AV_INPUT_BUFFER_PADDING_SIZE);
195 return AVERROR(ENOMEM);
198 ff_llauddsp_init(&s->dsp);
199 init_put_bits(&s->pb, s->frame_data, s->max_frame_size);
201 if (avctx->extradata_size >= 18) {
202 s->decode_flags = AV_RL16(edata_ptr + 14);
203 channel_mask = AV_RL32(edata_ptr + 2);
204 s->bits_per_sample = AV_RL16(edata_ptr);
205 if (s->bits_per_sample == 16)
206 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
207 else if (s->bits_per_sample == 24) {
208 av_log(avctx, AV_LOG_WARNING, "Decoding audio at 24 bit-depth\n");
209 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
210 avctx->bits_per_raw_sample = 24;
212 av_log(avctx, AV_LOG_ERROR, "Unknown bit-depth: %"PRIu8"\n",
214 return AVERROR_INVALIDDATA;
216 /* dump the extradata */
217 for (i = 0; i < avctx->extradata_size; i++)
218 ff_dlog(avctx, "[%x] ", avctx->extradata[i]);
219 ff_dlog(avctx, "\n");
222 avpriv_request_sample(avctx, "Unsupported extradata size");
223 return AVERROR_PATCHWELCOME;
227 s->log2_frame_size = av_log2(avctx->block_align) + 4;
230 s->skip_frame = 1; /* skip first frame */
232 s->len_prefix = s->decode_flags & 0x40;
235 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
237 av_assert0(s->samples_per_frame <= WMALL_BLOCK_MAX_SIZE);
239 /* init previous block len */
240 for (i = 0; i < avctx->channels; i++)
241 s->channel[i].prev_block_len = s->samples_per_frame;
244 log2_max_num_subframes = (s->decode_flags & 0x38) >> 3;
245 s->max_num_subframes = 1 << log2_max_num_subframes;
246 s->max_subframe_len_bit = 0;
247 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
249 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
250 s->dynamic_range_compression = s->decode_flags & 0x80;
251 s->bV3RTM = s->decode_flags & 0x100;
253 if (s->max_num_subframes > MAX_SUBFRAMES) {
254 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRIu8"\n",
255 s->max_num_subframes);
256 return AVERROR_INVALIDDATA;
259 s->num_channels = avctx->channels;
261 /* extract lfe channel position */
264 if (channel_mask & 8) {
266 for (mask = 1; mask < 16; mask <<= 1)
267 if (channel_mask & mask)
271 if (s->num_channels < 0) {
272 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %"PRId8"\n",
274 return AVERROR_INVALIDDATA;
275 } else if (s->num_channels > WMALL_MAX_CHANNELS) {
276 avpriv_request_sample(avctx,
277 "More than %d channels", WMALL_MAX_CHANNELS);
278 return AVERROR_PATCHWELCOME;
281 s->frame = av_frame_alloc();
283 return AVERROR(ENOMEM);
285 avctx->channel_layout = channel_mask;
290 * @brief Decode the subframe length.
292 * @param offset sample offset in the frame
293 * @return decoded subframe length on success, < 0 in case of an error
295 static int decode_subframe_length(WmallDecodeCtx *s, int offset)
297 int frame_len_ratio, subframe_len, len;
299 /* no need to read from the bitstream when only one length is possible */
300 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
301 return s->min_samples_per_subframe;
303 len = av_log2(s->max_num_subframes - 1) + 1;
304 frame_len_ratio = get_bits(&s->gb, len);
305 subframe_len = s->min_samples_per_subframe * (frame_len_ratio + 1);
307 /* sanity check the length */
308 if (subframe_len < s->min_samples_per_subframe ||
309 subframe_len > s->samples_per_frame) {
310 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
312 return AVERROR_INVALIDDATA;
318 * @brief Decode how the data in the frame is split into subframes.
319 * Every WMA frame contains the encoded data for a fixed number of
320 * samples per channel. The data for every channel might be split
321 * into several subframes. This function will reconstruct the list of
322 * subframes for every channel.
324 * If the subframes are not evenly split, the algorithm estimates the
325 * channels with the lowest number of total samples.
326 * Afterwards, for each of these channels a bit is read from the
327 * bitstream that indicates if the channel contains a subframe with the
328 * next subframe size that is going to be read from the bitstream or not.
329 * If a channel contains such a subframe, the subframe size gets added to
330 * the channel's subframe list.
331 * The algorithm repeats these steps until the frame is properly divided
332 * between the individual channels.
335 * @return 0 on success, < 0 in case of an error
337 static int decode_tilehdr(WmallDecodeCtx *s)
339 uint16_t num_samples[WMALL_MAX_CHANNELS] = { 0 }; /* sum of samples for all currently known subframes of a channel */
340 uint8_t contains_subframe[WMALL_MAX_CHANNELS]; /* flag indicating if a channel contains the current subframe */
341 int channels_for_cur_subframe = s->num_channels; /* number of channels that contain the current subframe */
342 int fixed_channel_layout = 0; /* flag indicating that all channels use the same subfra2me offsets and sizes */
343 int min_channel_len = 0; /* smallest sum of samples (channels with this length will be processed first) */
346 /* reset tiling information */
347 for (c = 0; c < s->num_channels; c++)
348 s->channel[c].num_subframes = 0;
350 tile_aligned = get_bits1(&s->gb);
351 if (s->max_num_subframes == 1 || tile_aligned)
352 fixed_channel_layout = 1;
354 /* loop until the frame data is split between the subframes */
356 int subframe_len, in_use = 0;
358 /* check which channels contain the subframe */
359 for (c = 0; c < s->num_channels; c++) {
360 if (num_samples[c] == min_channel_len) {
361 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
362 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) {
363 contains_subframe[c] = 1;
365 contains_subframe[c] = get_bits1(&s->gb);
367 in_use |= contains_subframe[c];
369 contains_subframe[c] = 0;
373 av_log(s->avctx, AV_LOG_ERROR,
374 "Found empty subframe\n");
375 return AVERROR_INVALIDDATA;
378 /* get subframe length, subframe_len == 0 is not allowed */
379 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
380 return AVERROR_INVALIDDATA;
381 /* add subframes to the individual channels and find new min_channel_len */
382 min_channel_len += subframe_len;
383 for (c = 0; c < s->num_channels; c++) {
384 WmallChannelCtx *chan = &s->channel[c];
386 if (contains_subframe[c]) {
387 if (chan->num_subframes >= MAX_SUBFRAMES) {
388 av_log(s->avctx, AV_LOG_ERROR,
389 "broken frame: num subframes > 31\n");
390 return AVERROR_INVALIDDATA;
392 chan->subframe_len[chan->num_subframes] = subframe_len;
393 num_samples[c] += subframe_len;
394 ++chan->num_subframes;
395 if (num_samples[c] > s->samples_per_frame) {
396 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
397 "channel len(%"PRIu16") > samples_per_frame(%"PRIu16")\n",
398 num_samples[c], s->samples_per_frame);
399 return AVERROR_INVALIDDATA;
401 } else if (num_samples[c] <= min_channel_len) {
402 if (num_samples[c] < min_channel_len) {
403 channels_for_cur_subframe = 0;
404 min_channel_len = num_samples[c];
406 ++channels_for_cur_subframe;
409 } while (min_channel_len < s->samples_per_frame);
411 for (c = 0; c < s->num_channels; c++) {
413 for (i = 0; i < s->channel[c].num_subframes; i++) {
414 s->channel[c].subframe_offsets[i] = offset;
415 offset += s->channel[c].subframe_len[i];
422 static void decode_ac_filter(WmallDecodeCtx *s)
425 s->acfilter_order = get_bits(&s->gb, 4) + 1;
426 s->acfilter_scaling = get_bits(&s->gb, 4);
428 for (i = 0; i < s->acfilter_order; i++)
429 s->acfilter_coeffs[i] = get_bitsz(&s->gb, s->acfilter_scaling) + 1;
432 static void decode_mclms(WmallDecodeCtx *s)
434 s->mclms_order = (get_bits(&s->gb, 4) + 1) * 2;
435 s->mclms_scaling = get_bits(&s->gb, 4);
436 if (get_bits1(&s->gb)) {
437 int i, send_coef_bits;
438 int cbits = av_log2(s->mclms_scaling + 1);
439 if (1 << cbits < s->mclms_scaling + 1)
442 send_coef_bits = get_bitsz(&s->gb, cbits) + 2;
444 for (i = 0; i < s->mclms_order * s->num_channels * s->num_channels; i++)
445 s->mclms_coeffs[i] = get_bits(&s->gb, send_coef_bits);
447 for (i = 0; i < s->num_channels; i++) {
449 for (c = 0; c < i; c++)
450 s->mclms_coeffs_cur[i * s->num_channels + c] = get_bits(&s->gb, send_coef_bits);
455 static int decode_cdlms(WmallDecodeCtx *s)
458 int cdlms_send_coef = get_bits1(&s->gb);
460 for (c = 0; c < s->num_channels; c++) {
461 s->cdlms_ttl[c] = get_bits(&s->gb, 3) + 1;
462 for (i = 0; i < s->cdlms_ttl[c]; i++) {
463 s->cdlms[c][i].order = (get_bits(&s->gb, 7) + 1) * 8;
464 if (s->cdlms[c][i].order > MAX_ORDER) {
465 av_log(s->avctx, AV_LOG_ERROR,
466 "Order[%d][%d] %d > max (%d), not supported\n",
467 c, i, s->cdlms[c][i].order, MAX_ORDER);
468 s->cdlms[0][0].order = 0;
469 return AVERROR_INVALIDDATA;
471 if(s->cdlms[c][i].order & 8 && s->bits_per_sample == 16) {
474 avpriv_request_sample(s->avctx, "CDLMS of order %d",
475 s->cdlms[c][i].order);
480 for (i = 0; i < s->cdlms_ttl[c]; i++)
481 s->cdlms[c][i].scaling = get_bits(&s->gb, 4);
483 if (cdlms_send_coef) {
484 for (i = 0; i < s->cdlms_ttl[c]; i++) {
485 int cbits, shift_l, shift_r, j;
486 cbits = av_log2(s->cdlms[c][i].order);
487 if ((1 << cbits) < s->cdlms[c][i].order)
489 s->cdlms[c][i].coefsend = get_bits(&s->gb, cbits) + 1;
491 cbits = av_log2(s->cdlms[c][i].scaling + 1);
492 if ((1 << cbits) < s->cdlms[c][i].scaling + 1)
495 s->cdlms[c][i].bitsend = get_bitsz(&s->gb, cbits) + 2;
496 shift_l = 32 - s->cdlms[c][i].bitsend;
497 shift_r = 32 - s->cdlms[c][i].scaling - 2;
498 for (j = 0; j < s->cdlms[c][i].coefsend; j++)
499 s->cdlms[c][i].coefs[j] =
500 (get_bits(&s->gb, s->cdlms[c][i].bitsend) << shift_l) >> shift_r;
504 for (i = 0; i < s->cdlms_ttl[c]; i++)
505 memset(s->cdlms[c][i].coefs + s->cdlms[c][i].order,
506 0, WMALL_COEFF_PAD_SIZE);
512 static int decode_channel_residues(WmallDecodeCtx *s, int ch, int tile_size)
515 unsigned int ave_mean;
516 s->transient[ch] = get_bits1(&s->gb);
517 if (s->transient[ch]) {
518 s->transient_pos[ch] = get_bits(&s->gb, av_log2(tile_size));
519 if (s->transient_pos[ch])
520 s->transient[ch] = 0;
521 s->channel[ch].transient_counter =
522 FFMAX(s->channel[ch].transient_counter, s->samples_per_frame / 2);
523 } else if (s->channel[ch].transient_counter)
524 s->transient[ch] = 1;
526 if (s->seekable_tile) {
527 ave_mean = get_bits(&s->gb, s->bits_per_sample);
528 s->ave_sum[ch] = ave_mean << (s->movave_scaling + 1);
531 if (s->seekable_tile) {
532 if (s->do_inter_ch_decorr)
533 s->channel_residues[ch][0] = get_sbits_long(&s->gb, s->bits_per_sample + 1);
535 s->channel_residues[ch][0] = get_sbits_long(&s->gb, s->bits_per_sample);
538 for (; i < tile_size; i++) {
539 int quo = 0, rem, rem_bits, residue;
540 while(get_bits1(&s->gb)) {
542 if (get_bits_left(&s->gb) <= 0)
546 quo += get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1);
548 ave_mean = (s->ave_sum[ch] + (1 << s->movave_scaling)) >> (s->movave_scaling + 1);
552 rem_bits = av_ceil_log2(ave_mean);
553 rem = get_bits_long(&s->gb, rem_bits);
554 residue = (quo << rem_bits) + rem;
557 s->ave_sum[ch] = residue + s->ave_sum[ch] -
558 (s->ave_sum[ch] >> s->movave_scaling);
560 residue = (residue >> 1) ^ -(residue & 1);
561 s->channel_residues[ch][i] = residue;
568 static void decode_lpc(WmallDecodeCtx *s)
571 s->lpc_order = get_bits(&s->gb, 5) + 1;
572 s->lpc_scaling = get_bits(&s->gb, 4);
573 s->lpc_intbits = get_bits(&s->gb, 3) + 1;
574 cbits = s->lpc_scaling + s->lpc_intbits;
575 for (ch = 0; ch < s->num_channels; ch++)
576 for (i = 0; i < s->lpc_order; i++)
577 s->lpc_coefs[ch][i] = get_sbits(&s->gb, cbits);
580 static void clear_codec_buffers(WmallDecodeCtx *s)
584 memset(s->acfilter_coeffs, 0, sizeof(s->acfilter_coeffs));
585 memset(s->acfilter_prevvalues, 0, sizeof(s->acfilter_prevvalues));
586 memset(s->lpc_coefs, 0, sizeof(s->lpc_coefs));
588 memset(s->mclms_coeffs, 0, sizeof(s->mclms_coeffs));
589 memset(s->mclms_coeffs_cur, 0, sizeof(s->mclms_coeffs_cur));
590 memset(s->mclms_prevvalues, 0, sizeof(s->mclms_prevvalues));
591 memset(s->mclms_updates, 0, sizeof(s->mclms_updates));
593 for (ich = 0; ich < s->num_channels; ich++) {
594 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++) {
595 memset(s->cdlms[ich][ilms].coefs, 0,
596 sizeof(s->cdlms[ich][ilms].coefs));
597 memset(s->cdlms[ich][ilms].lms_prevvalues, 0,
598 sizeof(s->cdlms[ich][ilms].lms_prevvalues));
599 memset(s->cdlms[ich][ilms].lms_updates, 0,
600 sizeof(s->cdlms[ich][ilms].lms_updates));
607 * @brief Reset filter parameters and transient area at new seekable tile.
609 static void reset_codec(WmallDecodeCtx *s)
612 s->mclms_recent = s->mclms_order * s->num_channels;
613 for (ich = 0; ich < s->num_channels; ich++) {
614 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++)
615 s->cdlms[ich][ilms].recent = s->cdlms[ich][ilms].order;
616 /* first sample of a seekable subframe is considered as the starting of
617 a transient area which is samples_per_frame samples long */
618 s->channel[ich].transient_counter = s->samples_per_frame;
619 s->transient[ich] = 1;
620 s->transient_pos[ich] = 0;
624 static void mclms_update(WmallDecodeCtx *s, int icoef, int *pred)
626 int i, j, ich, pred_error;
627 int order = s->mclms_order;
628 int num_channels = s->num_channels;
629 int range = 1 << (s->bits_per_sample - 1);
631 for (ich = 0; ich < num_channels; ich++) {
632 pred_error = s->channel_residues[ich][icoef] - pred[ich];
633 if (pred_error > 0) {
634 for (i = 0; i < order * num_channels; i++)
635 s->mclms_coeffs[i + ich * order * num_channels] +=
636 s->mclms_updates[s->mclms_recent + i];
637 for (j = 0; j < ich; j++)
638 s->mclms_coeffs_cur[ich * num_channels + j] += WMASIGN(s->channel_residues[j][icoef]);
639 } else if (pred_error < 0) {
640 for (i = 0; i < order * num_channels; i++)
641 s->mclms_coeffs[i + ich * order * num_channels] -=
642 s->mclms_updates[s->mclms_recent + i];
643 for (j = 0; j < ich; j++)
644 s->mclms_coeffs_cur[ich * num_channels + j] -= WMASIGN(s->channel_residues[j][icoef]);
648 for (ich = num_channels - 1; ich >= 0; ich--) {
650 s->mclms_prevvalues[s->mclms_recent] = av_clip(s->channel_residues[ich][icoef],
652 s->mclms_updates[s->mclms_recent] = WMASIGN(s->channel_residues[ich][icoef]);
655 if (s->mclms_recent == 0) {
656 memcpy(&s->mclms_prevvalues[order * num_channels],
658 sizeof(int32_t) * order * num_channels);
659 memcpy(&s->mclms_updates[order * num_channels],
661 sizeof(int32_t) * order * num_channels);
662 s->mclms_recent = num_channels * order;
666 static void mclms_predict(WmallDecodeCtx *s, int icoef, int *pred)
669 int order = s->mclms_order;
670 int num_channels = s->num_channels;
672 for (ich = 0; ich < num_channels; ich++) {
674 if (!s->is_channel_coded[ich])
676 for (i = 0; i < order * num_channels; i++)
677 pred[ich] += s->mclms_prevvalues[i + s->mclms_recent] *
678 s->mclms_coeffs[i + order * num_channels * ich];
679 for (i = 0; i < ich; i++)
680 pred[ich] += s->channel_residues[i][icoef] *
681 s->mclms_coeffs_cur[i + num_channels * ich];
682 pred[ich] += 1 << s->mclms_scaling - 1;
683 pred[ich] >>= s->mclms_scaling;
684 s->channel_residues[ich][icoef] += pred[ich];
688 static void revert_mclms(WmallDecodeCtx *s, int tile_size)
690 int icoef, pred[WMALL_MAX_CHANNELS] = { 0 };
691 for (icoef = 0; icoef < tile_size; icoef++) {
692 mclms_predict(s, icoef, pred);
693 mclms_update(s, icoef, pred);
697 static void lms_update(WmallDecodeCtx *s, int ich, int ilms, int input)
699 int recent = s->cdlms[ich][ilms].recent;
700 int range = 1 << s->bits_per_sample - 1;
701 int order = s->cdlms[ich][ilms].order;
706 memcpy(s->cdlms[ich][ilms].lms_prevvalues + order,
707 s->cdlms[ich][ilms].lms_prevvalues, sizeof(*s->cdlms[ich][ilms].lms_prevvalues) * order);
708 memcpy(s->cdlms[ich][ilms].lms_updates + order,
709 s->cdlms[ich][ilms].lms_updates, sizeof(*s->cdlms[ich][ilms].lms_updates) * order);
713 s->cdlms[ich][ilms].lms_prevvalues[recent] = av_clip(input, -range, range - 1);
714 s->cdlms[ich][ilms].lms_updates[recent] = WMASIGN(input) * s->update_speed[ich];
716 s->cdlms[ich][ilms].lms_updates[recent + (order >> 4)] >>= 2;
717 s->cdlms[ich][ilms].lms_updates[recent + (order >> 3)] >>= 1;
718 s->cdlms[ich][ilms].recent = recent;
719 memset(s->cdlms[ich][ilms].lms_updates + recent + order, 0,
720 sizeof(s->cdlms[ich][ilms].lms_updates) - 4*(recent+order));
723 static void use_high_update_speed(WmallDecodeCtx *s, int ich)
725 int ilms, recent, icoef;
726 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
727 recent = s->cdlms[ich][ilms].recent;
728 if (s->update_speed[ich] == 16)
731 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
732 s->cdlms[ich][ilms].lms_updates[icoef + recent] *= 2;
734 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
735 s->cdlms[ich][ilms].lms_updates[icoef] *= 2;
738 s->update_speed[ich] = 16;
741 static void use_normal_update_speed(WmallDecodeCtx *s, int ich)
743 int ilms, recent, icoef;
744 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
745 recent = s->cdlms[ich][ilms].recent;
746 if (s->update_speed[ich] == 8)
749 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
750 s->cdlms[ich][ilms].lms_updates[icoef + recent] /= 2;
752 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
753 s->cdlms[ich][ilms].lms_updates[icoef] /= 2;
755 s->update_speed[ich] = 8;
758 static void revert_cdlms(WmallDecodeCtx *s, int ch,
759 int coef_begin, int coef_end)
761 int icoef, pred, ilms, num_lms, residue, input;
763 num_lms = s->cdlms_ttl[ch];
764 for (ilms = num_lms - 1; ilms >= 0; ilms--) {
765 for (icoef = coef_begin; icoef < coef_end; icoef++) {
766 pred = 1 << (s->cdlms[ch][ilms].scaling - 1);
767 residue = s->channel_residues[ch][icoef];
768 pred += s->dsp.scalarproduct_and_madd_int32(s->cdlms[ch][ilms].coefs,
769 s->cdlms[ch][ilms].lms_prevvalues
770 + s->cdlms[ch][ilms].recent,
771 s->cdlms[ch][ilms].lms_updates
772 + s->cdlms[ch][ilms].recent,
773 FFALIGN(s->cdlms[ch][ilms].order,
774 WMALL_COEFF_PAD_SIZE),
776 input = residue + (pred >> s->cdlms[ch][ilms].scaling);
777 lms_update(s, ch, ilms, input);
778 s->channel_residues[ch][icoef] = input;
784 static void revert_inter_ch_decorr(WmallDecodeCtx *s, int tile_size)
786 if (s->num_channels != 2)
788 else if (s->is_channel_coded[0] || s->is_channel_coded[1]) {
790 for (icoef = 0; icoef < tile_size; icoef++) {
791 s->channel_residues[0][icoef] -= s->channel_residues[1][icoef] >> 1;
792 s->channel_residues[1][icoef] += s->channel_residues[0][icoef];
797 static void revert_acfilter(WmallDecodeCtx *s, int tile_size)
800 int16_t *filter_coeffs = s->acfilter_coeffs;
801 int scaling = s->acfilter_scaling;
802 int order = s->acfilter_order;
804 for (ich = 0; ich < s->num_channels; ich++) {
805 int *prevvalues = s->acfilter_prevvalues[ich];
806 for (i = 0; i < order; i++) {
808 for (j = 0; j < order; j++) {
810 pred += filter_coeffs[j] * prevvalues[j - i];
812 pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
815 s->channel_residues[ich][i] += pred;
817 for (i = order; i < tile_size; i++) {
819 for (j = 0; j < order; j++)
820 pred += s->channel_residues[ich][i - j - 1] * filter_coeffs[j];
822 s->channel_residues[ich][i] += pred;
824 for (j = 0; j < order; j++)
825 prevvalues[j] = s->channel_residues[ich][tile_size - j - 1];
829 static int decode_subframe(WmallDecodeCtx *s)
831 int offset = s->samples_per_frame;
832 int subframe_len = s->samples_per_frame;
833 int total_samples = s->samples_per_frame * s->num_channels;
834 int i, j, rawpcm_tile, padding_zeroes, res;
836 s->subframe_offset = get_bits_count(&s->gb);
838 /* reset channel context and find the next block offset and size
839 == the next block of the channel with the smallest number of
841 for (i = 0; i < s->num_channels; i++) {
842 if (offset > s->channel[i].decoded_samples) {
843 offset = s->channel[i].decoded_samples;
845 s->channel[i].subframe_len[s->channel[i].cur_subframe];
849 /* get a list of all channels that contain the estimated block */
850 s->channels_for_cur_subframe = 0;
851 for (i = 0; i < s->num_channels; i++) {
852 const int cur_subframe = s->channel[i].cur_subframe;
853 /* subtract already processed samples */
854 total_samples -= s->channel[i].decoded_samples;
856 /* and count if there are multiple subframes that match our profile */
857 if (offset == s->channel[i].decoded_samples &&
858 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
859 total_samples -= s->channel[i].subframe_len[cur_subframe];
860 s->channel[i].decoded_samples +=
861 s->channel[i].subframe_len[cur_subframe];
862 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
863 ++s->channels_for_cur_subframe;
867 /* check if the frame will be complete after processing the
870 s->parsed_all_subframes = 1;
873 s->seekable_tile = get_bits1(&s->gb);
874 if (s->seekable_tile) {
875 clear_codec_buffers(s);
877 s->do_arith_coding = get_bits1(&s->gb);
878 if (s->do_arith_coding) {
879 avpriv_request_sample(s->avctx, "Arithmetic coding");
880 return AVERROR_PATCHWELCOME;
882 s->do_ac_filter = get_bits1(&s->gb);
883 s->do_inter_ch_decorr = get_bits1(&s->gb);
884 s->do_mclms = get_bits1(&s->gb);
892 if ((res = decode_cdlms(s)) < 0)
894 s->movave_scaling = get_bits(&s->gb, 3);
895 s->quant_stepsize = get_bits(&s->gb, 8) + 1;
900 rawpcm_tile = get_bits1(&s->gb);
902 if (!rawpcm_tile && !s->cdlms[0][0].order) {
903 av_log(s->avctx, AV_LOG_DEBUG,
904 "Waiting for seekable tile\n");
905 av_frame_unref(s->frame);
910 for (i = 0; i < s->num_channels; i++)
911 s->is_channel_coded[i] = 1;
914 for (i = 0; i < s->num_channels; i++)
915 s->is_channel_coded[i] = get_bits1(&s->gb);
919 s->do_lpc = get_bits1(&s->gb);
922 avpriv_request_sample(s->avctx, "Expect wrong output since "
923 "inverse LPC filter");
930 if (get_bits1(&s->gb))
931 padding_zeroes = get_bits(&s->gb, 5);
936 int bits = s->bits_per_sample - padding_zeroes;
938 av_log(s->avctx, AV_LOG_ERROR,
939 "Invalid number of padding bits in raw PCM tile\n");
940 return AVERROR_INVALIDDATA;
942 ff_dlog(s->avctx, "RAWPCM %d bits per sample. "
943 "total %d bits, remain=%d\n", bits,
944 bits * s->num_channels * subframe_len, get_bits_count(&s->gb));
945 for (i = 0; i < s->num_channels; i++)
946 for (j = 0; j < subframe_len; j++)
947 s->channel_residues[i][j] = get_sbits_long(&s->gb, bits);
949 for (i = 0; i < s->num_channels; i++) {
950 if (s->is_channel_coded[i]) {
951 decode_channel_residues(s, i, subframe_len);
952 if (s->seekable_tile)
953 use_high_update_speed(s, i);
955 use_normal_update_speed(s, i);
956 revert_cdlms(s, i, 0, subframe_len);
958 memset(s->channel_residues[i], 0, sizeof(**s->channel_residues) * subframe_len);
963 revert_mclms(s, subframe_len);
964 if (s->do_inter_ch_decorr)
965 revert_inter_ch_decorr(s, subframe_len);
967 revert_acfilter(s, subframe_len);
970 if (s->quant_stepsize != 1)
971 for (i = 0; i < s->num_channels; i++)
972 for (j = 0; j < subframe_len; j++)
973 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;
985 *s->samples_32[c]++ = s->channel_residues[c][j] << (padding_zeroes + 8);
990 /* handled one subframe */
991 for (i = 0; i < s->channels_for_cur_subframe; i++) {
992 int c = s->channel_indexes_for_cur_subframe[i];
993 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
994 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
995 return AVERROR_INVALIDDATA;
997 ++s->channel[c].cur_subframe;
1003 * @brief Decode one WMA frame.
1004 * @param s codec context
1005 * @return 0 if the trailer bit indicates that this is the last frame,
1006 * 1 if there are additional frames
1008 static int decode_frame(WmallDecodeCtx *s)
1010 GetBitContext* gb = &s->gb;
1011 int more_frames = 0, len = 0, i, ret;
1013 s->frame->nb_samples = s->samples_per_frame;
1014 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) {
1015 /* return an error if no frame could be decoded at all */
1017 s->frame->nb_samples = 0;
1020 for (i = 0; i < s->num_channels; i++) {
1021 s->samples_16[i] = (int16_t *)s->frame->extended_data[i];
1022 s->samples_32[i] = (int32_t *)s->frame->extended_data[i];
1025 /* get frame length */
1027 len = get_bits(gb, s->log2_frame_size);
1029 /* decode tile information */
1030 if ((ret = decode_tilehdr(s))) {
1032 av_frame_unref(s->frame);
1037 if (s->dynamic_range_compression)
1038 s->drc_gain = get_bits(gb, 8);
1040 /* no idea what these are for, might be the number of samples
1041 that need to be skipped at the beginning or end of a stream */
1042 if (get_bits1(gb)) {
1045 /* usually true for the first frame */
1046 if (get_bits1(gb)) {
1047 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1048 ff_dlog(s->avctx, "start skip: %i\n", skip);
1051 /* sometimes true for the last frame */
1052 if (get_bits1(gb)) {
1053 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1054 ff_dlog(s->avctx, "end skip: %i\n", skip);
1055 s->frame->nb_samples -= skip;
1056 if (s->frame->nb_samples <= 0)
1057 return AVERROR_INVALIDDATA;
1062 /* reset subframe states */
1063 s->parsed_all_subframes = 0;
1064 for (i = 0; i < s->num_channels; i++) {
1065 s->channel[i].decoded_samples = 0;
1066 s->channel[i].cur_subframe = 0;
1069 /* decode all subframes */
1070 while (!s->parsed_all_subframes) {
1071 int decoded_samples = s->channel[0].decoded_samples;
1072 if (decode_subframe(s) < 0) {
1074 if (s->frame->nb_samples)
1075 s->frame->nb_samples = decoded_samples;
1080 ff_dlog(s->avctx, "Frame done\n");
1084 if (s->len_prefix) {
1085 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1086 /* FIXME: not sure if this is always an error */
1087 av_log(s->avctx, AV_LOG_ERROR,
1088 "frame[%"PRIu32"] would have to skip %i bits\n",
1090 len - (get_bits_count(gb) - s->frame_offset) - 1);
1095 /* skip the rest of the frame data */
1096 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1099 /* decode trailer bit */
1100 more_frames = get_bits1(gb);
1106 * @brief Calculate remaining input buffer length.
1107 * @param s codec context
1108 * @param gb bitstream reader context
1109 * @return remaining size in bits
1111 static int remaining_bits(WmallDecodeCtx *s, GetBitContext *gb)
1113 return s->buf_bit_size - get_bits_count(gb);
1117 * @brief Fill the bit reservoir with a (partial) frame.
1118 * @param s codec context
1119 * @param gb bitstream reader context
1120 * @param len length of the partial frame
1121 * @param append decides whether to reset the buffer or not
1123 static void save_bits(WmallDecodeCtx *s, GetBitContext* gb, int len,
1129 /* when the frame data does not need to be concatenated, the input buffer
1130 is reset and additional bits from the previous frame are copied
1131 and skipped later so that a fast byte copy is possible */
1134 s->frame_offset = get_bits_count(gb) & 7;
1135 s->num_saved_bits = s->frame_offset;
1136 init_put_bits(&s->pb, s->frame_data, s->max_frame_size);
1139 buflen = (s->num_saved_bits + len + 8) >> 3;
1141 if (len <= 0 || buflen > s->max_frame_size) {
1142 avpriv_request_sample(s->avctx, "Too small input buffer");
1147 s->num_saved_bits += len;
1149 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1152 int align = 8 - (get_bits_count(gb) & 7);
1153 align = FFMIN(align, len);
1154 put_bits(&s->pb, align, get_bits(gb, align));
1156 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1158 skip_bits_long(gb, len);
1161 flush_put_bits(&tmp);
1163 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1164 skip_bits(&s->gb, s->frame_offset);
1167 static int decode_packet(AVCodecContext *avctx, void *data, int *got_frame_ptr,
1170 WmallDecodeCtx *s = avctx->priv_data;
1171 GetBitContext* gb = &s->pgb;
1172 const uint8_t* buf = avpkt->data;
1173 int buf_size = avpkt->size;
1174 int num_bits_prev_frame, packet_sequence_number, spliced_packet;
1176 s->frame->nb_samples = 0;
1178 if (!buf_size && s->num_saved_bits > get_bits_count(&s->gb)) {
1180 if (!decode_frame(s))
1181 s->num_saved_bits = 0;
1182 } else if (s->packet_done || s->packet_loss) {
1188 s->next_packet_start = buf_size - FFMIN(avctx->block_align, buf_size);
1189 buf_size = FFMIN(avctx->block_align, buf_size);
1190 s->buf_bit_size = buf_size << 3;
1192 /* parse packet header */
1193 init_get_bits(gb, buf, s->buf_bit_size);
1194 packet_sequence_number = get_bits(gb, 4);
1195 skip_bits(gb, 1); // Skip seekable_frame_in_packet, currently ununused
1196 spliced_packet = get_bits1(gb);
1198 avpriv_request_sample(avctx, "Bitstream splicing");
1200 /* get number of bits that need to be added to the previous frame */
1201 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1203 /* check for packet loss */
1204 if (!s->packet_loss &&
1205 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1207 av_log(avctx, AV_LOG_ERROR,
1208 "Packet loss detected! seq %"PRIx8" vs %x\n",
1209 s->packet_sequence_number, packet_sequence_number);
1211 s->packet_sequence_number = packet_sequence_number;
1213 if (num_bits_prev_frame > 0) {
1214 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1215 if (num_bits_prev_frame >= remaining_packet_bits) {
1216 num_bits_prev_frame = remaining_packet_bits;
1220 /* Append the previous frame data to the remaining data from the
1221 * previous packet to create a full frame. */
1222 save_bits(s, gb, num_bits_prev_frame, 1);
1224 /* decode the cross packet frame if it is valid */
1225 if (num_bits_prev_frame < remaining_packet_bits && !s->packet_loss)
1227 } else if (s->num_saved_bits - s->frame_offset) {
1228 ff_dlog(avctx, "ignoring %x previously saved bits\n",
1229 s->num_saved_bits - s->frame_offset);
1232 if (s->packet_loss) {
1233 /* Reset number of saved bits so that the decoder does not start
1234 * to decode incomplete frames in the s->len_prefix == 0 case. */
1235 s->num_saved_bits = 0;
1237 init_put_bits(&s->pb, s->frame_data, s->max_frame_size);
1243 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1244 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1245 skip_bits(gb, s->packet_offset);
1247 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1248 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1249 frame_size <= remaining_bits(s, gb)) {
1250 save_bits(s, gb, frame_size, 0);
1251 s->packet_done = !decode_frame(s);
1252 } else if (!s->len_prefix
1253 && s->num_saved_bits > get_bits_count(&s->gb)) {
1254 /* when the frames do not have a length prefix, we don't know the
1255 * compressed length of the individual frames however, we know what
1256 * part of a new packet belongs to the previous frame therefore we
1257 * save the incoming packet first, then we append the "previous
1258 * frame" data from the next packet so that we get a buffer that
1259 * only contains full frames */
1260 s->packet_done = !decode_frame(s);
1266 if (s->packet_done && !s->packet_loss &&
1267 remaining_bits(s, gb) > 0) {
1268 /* save the rest of the data so that it can be decoded
1269 * with the next packet */
1270 save_bits(s, gb, remaining_bits(s, gb), 0);
1273 *got_frame_ptr = s->frame->nb_samples > 0;
1274 av_frame_move_ref(data, s->frame);
1276 s->packet_offset = get_bits_count(gb) & 7;
1278 return (s->packet_loss) ? AVERROR_INVALIDDATA : buf_size ? get_bits_count(gb) >> 3 : 0;
1281 static void flush(AVCodecContext *avctx)
1283 WmallDecodeCtx *s = avctx->priv_data;
1286 s->num_saved_bits = 0;
1287 s->frame_offset = 0;
1288 s->next_packet_start = 0;
1289 s->cdlms[0][0].order = 0;
1290 s->frame->nb_samples = 0;
1291 init_put_bits(&s->pb, s->frame_data, s->max_frame_size);
1294 static av_cold int decode_close(AVCodecContext *avctx)
1296 WmallDecodeCtx *s = avctx->priv_data;
1298 av_frame_free(&s->frame);
1299 av_freep(&s->frame_data);
1304 AVCodec ff_wmalossless_decoder = {
1305 .name = "wmalossless",
1306 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio Lossless"),
1307 .type = AVMEDIA_TYPE_AUDIO,
1308 .id = AV_CODEC_ID_WMALOSSLESS,
1309 .priv_data_size = sizeof(WmallDecodeCtx),
1310 .init = decode_init,
1311 .close = decode_close,
1312 .decode = decode_packet,
1314 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
1315 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
1317 AV_SAMPLE_FMT_NONE },