2 * Wmall compatible 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
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * @brief wmall decoder implementation
27 * Wmall is an MDCT based codec comparable to wma standard or AAC.
28 * The decoding therefore consists of the following steps:
29 * - bitstream decoding
30 * - reconstruction of per-channel data
31 * - rescaling and inverse quantization
33 * - windowing and overlapp-add
35 * The compressed wmall bitstream is split into individual packets.
36 * Every such packet contains one or more wma frames.
37 * The compressed frames may have a variable length and frames may
38 * cross packet boundaries.
39 * Common to all wmall frames is the number of samples that are stored in
41 * The number of samples and a few other decode flags are stored
42 * as extradata that has to be passed to the decoder.
44 * The wmall frames themselves are again split into a variable number of
45 * subframes. Every subframe contains the data for 2^N time domain samples
46 * where N varies between 7 and 12.
48 * Example wmall bitstream (in samples):
50 * || packet 0 || packet 1 || packet 2 packets
51 * ---------------------------------------------------
52 * || frame 0 || frame 1 || frame 2 || frames
53 * ---------------------------------------------------
54 * || | | || | | | || || subframes of channel 0
55 * ---------------------------------------------------
56 * || | | || | | | || || subframes of channel 1
57 * ---------------------------------------------------
59 * The frame layouts for the individual channels of a wma frame does not need
62 * However, if the offsets and lengths of several subframes of a frame are the
63 * same, the subframes of the channels can be grouped.
64 * Every group may then use special coding techniques like M/S stereo coding
65 * to improve the compression ratio. These channel transformations do not
66 * need to be applied to a whole subframe. Instead, they can also work on
67 * individual scale factor bands (see below).
68 * The coefficients that carry the audio signal in the frequency domain
69 * are transmitted as huffman-coded vectors with 4, 2 and 1 elements.
70 * In addition to that, the encoder can switch to a runlevel coding scheme
71 * by transmitting subframe_length / 128 zero coefficients.
73 * Before the audio signal can be converted to the time domain, the
74 * coefficients have to be rescaled and inverse quantized.
75 * A subframe is therefore split into several scale factor bands that get
76 * scaled individually.
77 * Scale factors are submitted for every frame but they might be shared
78 * between the subframes of a channel. Scale factors are initially DPCM-coded.
79 * Once scale factors are shared, the differences are transmitted as runlevel
81 * Every subframe length and offset combination in the frame layout shares a
82 * common quantization factor that can be adjusted for every channel by a
84 * After the inverse quantization, the coefficients get processed by an IMDCT.
85 * The resulting values are then windowed with a sine window and the first half
86 * of the values are added to the second half of the output from the previous
87 * subframe in order to reconstruct the output samples.
97 /** current decoder limitations */
98 #define WMALL_MAX_CHANNELS 8 ///< max number of handled channels
99 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
100 #define MAX_BANDS 29 ///< max number of scale factor bands
101 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
103 #define WMALL_BLOCK_MIN_BITS 6 ///< log2 of min block size
104 #define WMALL_BLOCK_MAX_BITS 12 ///< log2 of max block size
105 #define WMALL_BLOCK_MAX_SIZE (1 << WMALL_BLOCK_MAX_BITS) ///< maximum block size
106 #define WMALL_BLOCK_SIZES (WMALL_BLOCK_MAX_BITS - WMALL_BLOCK_MIN_BITS + 1) ///< possible block sizes
110 #define SCALEVLCBITS 8
111 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
112 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
113 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
114 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
115 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
117 static float sin64[33]; ///< sinus table for decorrelation
120 * @brief frame specific decoder context for a single channel
123 int16_t prev_block_len; ///< length of the previous block
124 uint8_t transmit_coefs;
125 uint8_t num_subframes;
126 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
127 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
128 uint8_t cur_subframe; ///< current subframe number
129 uint16_t decoded_samples; ///< number of already processed samples
130 uint8_t grouped; ///< channel is part of a group
131 int quant_step; ///< quantization step for the current subframe
132 int8_t reuse_sf; ///< share scale factors between subframes
133 int8_t scale_factor_step; ///< scaling step for the current subframe
134 int max_scale_factor; ///< maximum scale factor for the current subframe
135 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
136 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
137 int* scale_factors; ///< pointer to the scale factor values used for decoding
138 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
139 float* coeffs; ///< pointer to the subframe decode buffer
140 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
141 DECLARE_ALIGNED(16, float, out)[WMALL_BLOCK_MAX_SIZE + WMALL_BLOCK_MAX_SIZE / 2]; ///< output buffer
142 int transient_counter; ///< number of transient samples from the beginning of transient zone
146 * @brief channel group for channel transformations
149 uint8_t num_channels; ///< number of channels in the group
150 int8_t transform; ///< transform on / off
151 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
152 float decorrelation_matrix[WMALL_MAX_CHANNELS*WMALL_MAX_CHANNELS];
153 float* channel_data[WMALL_MAX_CHANNELS]; ///< transformation coefficients
157 * @brief main decoder context
159 typedef struct WmallDecodeCtx {
160 /* generic decoder variables */
161 AVCodecContext* avctx; ///< codec context for av_log
162 DSPContext dsp; ///< accelerated DSP functions
163 uint8_t frame_data[MAX_FRAMESIZE +
164 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
165 PutBitContext pb; ///< context for filling the frame_data buffer
166 FFTContext mdct_ctx[WMALL_BLOCK_SIZES]; ///< MDCT context per block size
167 DECLARE_ALIGNED(16, float, tmp)[WMALL_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
168 float* windows[WMALL_BLOCK_SIZES]; ///< windows for the different block sizes
170 /* frame size dependent frame information (set during initialization) */
171 uint32_t decode_flags; ///< used compression features
172 uint8_t len_prefix; ///< frame is prefixed with its length
173 uint8_t dynamic_range_compression; ///< frame contains DRC data
174 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
175 uint16_t samples_per_frame; ///< number of samples to output
176 uint16_t log2_frame_size;
177 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
178 int8_t lfe_channel; ///< lfe channel index
179 uint8_t max_num_subframes;
180 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
181 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
182 uint16_t min_samples_per_subframe;
183 int8_t num_sfb[WMALL_BLOCK_SIZES]; ///< scale factor bands per block size
184 int16_t sfb_offsets[WMALL_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
185 int8_t sf_offsets[WMALL_BLOCK_SIZES][WMALL_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
186 int16_t subwoofer_cutoffs[WMALL_BLOCK_SIZES]; ///< subwoofer cutoff values
188 /* packet decode state */
189 GetBitContext pgb; ///< bitstream reader context for the packet
190 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
191 uint8_t packet_offset; ///< frame offset in the packet
192 uint8_t packet_sequence_number; ///< current packet number
193 int num_saved_bits; ///< saved number of bits
194 int frame_offset; ///< frame offset in the bit reservoir
195 int subframe_offset; ///< subframe offset in the bit reservoir
196 uint8_t packet_loss; ///< set in case of bitstream error
197 uint8_t packet_done; ///< set when a packet is fully decoded
199 /* frame decode state */
200 uint32_t frame_num; ///< current frame number (not used for decoding)
201 GetBitContext gb; ///< bitstream reader context
202 int buf_bit_size; ///< buffer size in bits
203 float* samples; ///< current samplebuffer pointer
204 float* samples_end; ///< maximum samplebuffer pointer
205 uint8_t drc_gain; ///< gain for the DRC tool
206 int8_t skip_frame; ///< skip output step
207 int8_t parsed_all_subframes; ///< all subframes decoded?
209 /* subframe/block decode state */
210 int16_t subframe_len; ///< current subframe length
211 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
212 int8_t channel_indexes_for_cur_subframe[WMALL_MAX_CHANNELS];
213 int8_t num_bands; ///< number of scale factor bands
214 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
215 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
216 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
217 int8_t esc_len; ///< length of escaped coefficients
219 uint8_t num_chgroups; ///< number of channel groups
220 WmallChannelGrp chgroup[WMALL_MAX_CHANNELS]; ///< channel group information
222 WmallChannelCtx channel[WMALL_MAX_CHANNELS]; ///< per channel data
226 uint8_t do_arith_coding;
227 uint8_t do_ac_filter;
228 uint8_t do_inter_ch_decorr;
232 int8_t acfilter_order;
233 int8_t acfilter_scaling;
234 int acfilter_coeffs[16];
235 int acfilter_prevvalues[16];
238 int8_t mclms_scaling;
239 int16_t mclms_coeffs[128];
240 int16_t mclms_coeffs_cur[4];
241 int16_t mclms_prevvalues[64]; // FIXME: should be 32-bit / 16-bit depending on bit-depth
242 int16_t mclms_updates[64];
254 int16_t lms_prevvalues[512]; // FIXME: see above
255 int16_t lms_updates[512]; // and here too
257 } cdlms[2][9]; /* XXX: Here, 2 is the max. no. of channels allowed,
258 9 is the maximum no. of filters per channel.
259 Question is, why 2 if WMALL_MAX_CHANNELS == 8 */
266 int is_channel_coded[2]; // XXX: same question as above applies here too (and below)
270 int transient_pos[2];
275 int channel_residues[2][2048];
278 int lpc_coefs[2][40];
283 int channel_coeffs[2][2048]; // FIXME: should be 32-bit / 16-bit depending on bit-depth
289 #define dprintf(pctx, ...) av_log(pctx, AV_LOG_DEBUG, __VA_ARGS__)
292 static int num_logged_tiles = 0;
293 static int num_logged_subframes = 0;
294 static int num_lms_update_call = 0;
297 *@brief helper function to print the most important members of the context
300 static void av_cold dump_context(WmallDecodeCtx *s)
302 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
303 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
305 PRINT("ed sample bit depth", s->bits_per_sample);
306 PRINT_HEX("ed decode flags", s->decode_flags);
307 PRINT("samples per frame", s->samples_per_frame);
308 PRINT("log2 frame size", s->log2_frame_size);
309 PRINT("max num subframes", s->max_num_subframes);
310 PRINT("len prefix", s->len_prefix);
311 PRINT("num channels", s->num_channels);
314 static void dump_int_buffer(uint8_t *buffer, int size, int length, int delimiter)
318 for (i=0 ; i<length ; i++) {
320 av_log(0, 0, "\n[%d] ", i);
321 av_log(0, 0, "%d, ", *(int16_t *)(buffer + i * size));
328 *@brief Uninitialize the decoder and free all resources.
329 *@param avctx codec context
330 *@return 0 on success, < 0 otherwise
332 static av_cold int decode_end(AVCodecContext *avctx)
334 WmallDecodeCtx *s = avctx->priv_data;
337 for (i = 0; i < WMALL_BLOCK_SIZES; i++)
338 ff_mdct_end(&s->mdct_ctx[i]);
344 *@brief Initialize the decoder.
345 *@param avctx codec context
346 *@return 0 on success, -1 otherwise
348 static av_cold int decode_init(AVCodecContext *avctx)
350 WmallDecodeCtx *s = avctx->priv_data;
351 uint8_t *edata_ptr = avctx->extradata;
352 unsigned int channel_mask;
354 int log2_max_num_subframes;
355 int num_possible_block_sizes;
358 dsputil_init(&s->dsp, avctx);
359 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
361 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
363 if (avctx->extradata_size >= 18) {
364 s->decode_flags = AV_RL16(edata_ptr+14);
365 channel_mask = AV_RL32(edata_ptr+2);
366 s->bits_per_sample = AV_RL16(edata_ptr);
367 /** dump the extradata */
368 for (i = 0; i < avctx->extradata_size; i++)
369 dprintf(avctx, "[%x] ", avctx->extradata[i]);
370 dprintf(avctx, "\n");
373 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
374 return AVERROR_INVALIDDATA;
378 s->log2_frame_size = av_log2(avctx->block_align) + 4;
381 s->skip_frame = 1; /* skip first frame */
383 s->len_prefix = (s->decode_flags & 0x40);
386 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
389 /** init previous block len */
390 for (i = 0; i < avctx->channels; i++)
391 s->channel[i].prev_block_len = s->samples_per_frame;
394 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
395 s->max_num_subframes = 1 << log2_max_num_subframes;
396 s->max_subframe_len_bit = 0;
397 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
399 num_possible_block_sizes = log2_max_num_subframes + 1;
400 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
401 s->dynamic_range_compression = (s->decode_flags & 0x80);
403 s->bV3RTM = s->decode_flags & 0x100;
405 if (s->max_num_subframes > MAX_SUBFRAMES) {
406 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
407 s->max_num_subframes);
408 return AVERROR_INVALIDDATA;
411 s->num_channels = avctx->channels;
413 /** extract lfe channel position */
416 if (channel_mask & 8) {
418 for (mask = 1; mask < 16; mask <<= 1) {
419 if (channel_mask & mask)
424 if (s->num_channels < 0) {
425 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
426 return AVERROR_INVALIDDATA;
427 } else if (s->num_channels > WMALL_MAX_CHANNELS) {
428 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
429 return AVERROR_PATCHWELCOME;
432 avctx->channel_layout = channel_mask;
437 *@brief Decode the subframe length.
439 *@param offset sample offset in the frame
440 *@return decoded subframe length on success, < 0 in case of an error
442 static int decode_subframe_length(WmallDecodeCtx *s, int offset)
445 int subframe_len, len;
447 /** no need to read from the bitstream when only one length is possible */
448 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
449 return s->min_samples_per_subframe;
451 len = av_log2(s->max_num_subframes - 1) + 1;
452 frame_len_ratio = get_bits(&s->gb, len);
454 subframe_len = s->min_samples_per_subframe * (frame_len_ratio + 1);
456 /** sanity check the length */
457 if (subframe_len < s->min_samples_per_subframe ||
458 subframe_len > s->samples_per_frame) {
459 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
461 return AVERROR_INVALIDDATA;
467 *@brief Decode how the data in the frame is split into subframes.
468 * Every WMA frame contains the encoded data for a fixed number of
469 * samples per channel. The data for every channel might be split
470 * into several subframes. This function will reconstruct the list of
471 * subframes for every channel.
473 * If the subframes are not evenly split, the algorithm estimates the
474 * channels with the lowest number of total samples.
475 * Afterwards, for each of these channels a bit is read from the
476 * bitstream that indicates if the channel contains a subframe with the
477 * next subframe size that is going to be read from the bitstream or not.
478 * If a channel contains such a subframe, the subframe size gets added to
479 * the channel's subframe list.
480 * The algorithm repeats these steps until the frame is properly divided
481 * between the individual channels.
484 *@return 0 on success, < 0 in case of an error
486 static int decode_tilehdr(WmallDecodeCtx *s)
488 uint16_t num_samples[WMALL_MAX_CHANNELS]; /**< sum of samples for all currently known subframes of a channel */
489 uint8_t contains_subframe[WMALL_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
490 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
491 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subfra2me offsets and sizes */
492 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
495 /* Should never consume more than 3073 bits (256 iterations for the
496 * while loop when always the minimum amount of 128 samples is substracted
497 * from missing samples in the 8 channel case).
498 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
501 /** reset tiling information */
502 for (c = 0; c < s->num_channels; c++)
503 s->channel[c].num_subframes = 0;
505 memset(num_samples, 0, sizeof(num_samples));
507 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
508 fixed_channel_layout = 1;
510 /** loop until the frame data is split between the subframes */
514 /** check which channels contain the subframe */
515 for (c = 0; c < s->num_channels; c++) {
516 if (num_samples[c] == min_channel_len) {
517 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
518 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) {
519 contains_subframe[c] = 1;
522 contains_subframe[c] = get_bits1(&s->gb);
525 contains_subframe[c] = 0;
528 /** get subframe length, subframe_len == 0 is not allowed */
529 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
530 return AVERROR_INVALIDDATA;
531 /** add subframes to the individual channels and find new min_channel_len */
532 min_channel_len += subframe_len;
533 for (c = 0; c < s->num_channels; c++) {
534 WmallChannelCtx* chan = &s->channel[c];
536 if (contains_subframe[c]) {
537 if (chan->num_subframes >= MAX_SUBFRAMES) {
538 av_log(s->avctx, AV_LOG_ERROR,
539 "broken frame: num subframes > 31\n");
540 return AVERROR_INVALIDDATA;
542 chan->subframe_len[chan->num_subframes] = subframe_len;
543 num_samples[c] += subframe_len;
544 ++chan->num_subframes;
545 if (num_samples[c] > s->samples_per_frame) {
546 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
547 "channel len(%d) > samples_per_frame(%d)\n",
548 num_samples[c], s->samples_per_frame);
549 return AVERROR_INVALIDDATA;
551 } else if (num_samples[c] <= min_channel_len) {
552 if (num_samples[c] < min_channel_len) {
553 channels_for_cur_subframe = 0;
554 min_channel_len = num_samples[c];
556 ++channels_for_cur_subframe;
559 } while (min_channel_len < s->samples_per_frame);
561 for (c = 0; c < s->num_channels; c++) {
564 for (i = 0; i < s->channel[c].num_subframes; i++) {
565 s->channel[c].subframe_offset[i] = offset;
566 offset += s->channel[c].subframe_len[i];
574 static int my_log2(unsigned int i)
576 unsigned int iLog2 = 0;
577 while ((i >> iLog2) > 1)
586 static void decode_ac_filter(WmallDecodeCtx *s)
589 s->acfilter_order = get_bits(&s->gb, 4) + 1;
590 s->acfilter_scaling = get_bits(&s->gb, 4);
592 for(i = 0; i < s->acfilter_order; i++) {
593 s->acfilter_coeffs[i] = get_bits(&s->gb, s->acfilter_scaling) + 1;
601 static void decode_mclms(WmallDecodeCtx *s)
603 s->mclms_order = (get_bits(&s->gb, 4) + 1) * 2;
604 s->mclms_scaling = get_bits(&s->gb, 4);
605 if(get_bits1(&s->gb)) {
609 int cbits = av_log2(s->mclms_scaling + 1);
610 assert(cbits == my_log2(s->mclms_scaling + 1));
611 if(1 << cbits < s->mclms_scaling + 1)
614 send_coef_bits = (cbits ? get_bits(&s->gb, cbits) : 0) + 2;
616 for(i = 0; i < s->mclms_order * s->num_channels * s->num_channels; i++) {
617 s->mclms_coeffs[i] = get_bits(&s->gb, send_coef_bits);
620 for(i = 0; i < s->num_channels; i++) {
622 for(c = 0; c < i; c++) {
623 s->mclms_coeffs_cur[i * s->num_channels + c] = get_bits(&s->gb, send_coef_bits);
633 static void decode_cdlms(WmallDecodeCtx *s)
636 int cdlms_send_coef = get_bits1(&s->gb);
638 for(c = 0; c < s->num_channels; c++) {
639 s->cdlms_ttl[c] = get_bits(&s->gb, 3) + 1;
640 for(i = 0; i < s->cdlms_ttl[c]; i++) {
641 s->cdlms[c][i].order = (get_bits(&s->gb, 7) + 1) * 8;
644 for(i = 0; i < s->cdlms_ttl[c]; i++) {
645 s->cdlms[c][i].scaling = get_bits(&s->gb, 4);
648 if(cdlms_send_coef) {
649 for(i = 0; i < s->cdlms_ttl[c]; i++) {
650 int cbits, shift_l, shift_r, j;
651 cbits = av_log2(s->cdlms[c][i].order);
652 if(1 << cbits < s->cdlms[c][i].order)
654 s->cdlms[c][i].coefsend = get_bits(&s->gb, cbits) + 1;
656 cbits = av_log2(s->cdlms[c][i].scaling + 1);
657 if(1 << cbits < s->cdlms[c][i].scaling + 1)
660 s->cdlms[c][i].bitsend = get_bits(&s->gb, cbits) + 2;
661 shift_l = 32 - s->cdlms[c][i].bitsend;
662 shift_r = 32 - 2 - s->cdlms[c][i].scaling;
663 for(j = 0; j < s->cdlms[c][i].coefsend; j++) {
664 s->cdlms[c][i].coefs[j] =
665 (get_bits(&s->gb, s->cdlms[c][i].bitsend) << shift_l) >> shift_r;
675 static int decode_channel_residues(WmallDecodeCtx *s, int ch, int tile_size)
678 unsigned int ave_mean;
679 s->transient[ch] = get_bits1(&s->gb);
680 if(s->transient[ch]) {
681 s->transient_pos[ch] = get_bits(&s->gb, av_log2(tile_size));
682 if (s->transient_pos[ch])
683 s->transient[ch] = 0;
684 s->channel[ch].transient_counter =
685 FFMAX(s->channel[ch].transient_counter, s->samples_per_frame / 2);
686 } else if (s->channel[ch].transient_counter)
687 s->transient[ch] = 1;
689 if(s->seekable_tile) {
690 ave_mean = get_bits(&s->gb, s->bits_per_sample);
691 s->ave_sum[ch] = ave_mean << (s->movave_scaling + 1);
692 // s->ave_sum[ch] *= 2;
695 if(s->seekable_tile) {
696 if(s->do_inter_ch_decorr)
697 s->channel_residues[ch][0] = get_sbits(&s->gb, s->bits_per_sample + 1);
699 s->channel_residues[ch][0] = get_sbits(&s->gb, s->bits_per_sample);
702 //av_log(0, 0, "%8d: ", num_logged_tiles++);
703 for(; i < tile_size; i++) {
704 int quo = 0, rem, rem_bits, residue;
705 while(get_bits1(&s->gb))
708 quo += get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1);
710 ave_mean = (s->ave_sum[ch] + (1 << s->movave_scaling)) >> (s->movave_scaling + 1);
711 rem_bits = av_ceil_log2(ave_mean);
712 rem = rem_bits ? get_bits(&s->gb, rem_bits) : 0;
713 residue = (quo << rem_bits) + rem;
715 s->ave_sum[ch] = residue + s->ave_sum[ch] - (s->ave_sum[ch] >> s->movave_scaling);
718 residue = -(residue >> 1) - 1;
720 residue = residue >> 1;
721 s->channel_residues[ch][i] = residue;
723 dump_int_buffer(s->channel_residues[ch], 4, tile_size, 16);
734 decode_lpc(WmallDecodeCtx *s)
737 s->lpc_order = get_bits(&s->gb, 5) + 1;
738 s->lpc_scaling = get_bits(&s->gb, 4);
739 s->lpc_intbits = get_bits(&s->gb, 3) + 1;
740 cbits = s->lpc_scaling + s->lpc_intbits;
741 for(ch = 0; ch < s->num_channels; ch++) {
742 for(i = 0; i < s->lpc_order; i++) {
743 s->lpc_coefs[ch][i] = get_sbits(&s->gb, cbits);
749 static void clear_codec_buffers(WmallDecodeCtx *s)
753 memset(s->acfilter_coeffs, 0, 16 * sizeof(int));
754 memset(s->lpc_coefs , 0, 40 * 2 * sizeof(int));
756 memset(s->mclms_coeffs , 0, 128 * sizeof(int16_t));
757 memset(s->mclms_coeffs_cur, 0, 4 * sizeof(int16_t));
758 memset(s->mclms_prevvalues, 0, 64 * sizeof(int));
759 memset(s->mclms_updates , 0, 64 * sizeof(int16_t));
761 for (ich = 0; ich < s->num_channels; ich++) {
762 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++) {
763 memset(s->cdlms[ich][ilms].coefs , 0, 256 * sizeof(int16_t));
764 memset(s->cdlms[ich][ilms].lms_prevvalues, 0, 512 * sizeof(int16_t));
765 memset(s->cdlms[ich][ilms].lms_updates , 0, 512 * sizeof(int16_t));
772 *@brief Resets filter parameters and transient area at new seekable tile
774 static void reset_codec(WmallDecodeCtx *s)
777 s->mclms_recent = s->mclms_order * s->num_channels;
778 for (ich = 0; ich < s->num_channels; ich++) {
779 for (ilms = 0; ilms < s->cdlms_ttl[ich]; ilms++)
780 s->cdlms[ich][ilms].recent = s->cdlms[ich][ilms].order;
781 /* first sample of a seekable subframe is considered as the starting of
782 a transient area which is samples_per_frame samples long */
783 s->channel[ich].transient_counter = s->samples_per_frame;
784 s->transient[ich] = 1;
785 s->transient_pos[ich] = 0;
791 static void mclms_update(WmallDecodeCtx *s, int icoef, int *pred)
795 int order = s->mclms_order;
796 int num_channels = s->num_channels;
797 int range = 1 << (s->bits_per_sample - 1);
798 int bps = s->bits_per_sample > 16 ? 4 : 2; // bytes per sample
800 for (ich = 0; ich < num_channels; ich++) {
801 pred_error = s->channel_residues[ich][icoef] - pred[ich];
802 if (pred_error > 0) {
803 for (i = 0; i < order * num_channels; i++)
804 s->mclms_coeffs[i + ich * order * num_channels] +=
805 s->mclms_updates[s->mclms_recent + i];
806 for (j = 0; j < ich; j++) {
807 if (s->channel_residues[j][icoef] > 0)
808 s->mclms_coeffs_cur[ich * num_channels + j] += 1;
809 else if (s->channel_residues[j][icoef] < 0)
810 s->mclms_coeffs_cur[ich * num_channels + j] -= 1;
812 } else if (pred_error < 0) {
813 for (i = 0; i < order * num_channels; i++)
814 s->mclms_coeffs[i + ich * order * num_channels] -=
815 s->mclms_updates[s->mclms_recent + i];
816 for (j = 0; j < ich; j++) {
817 if (s->channel_residues[j][icoef] > 0)
818 s->mclms_coeffs_cur[ich * num_channels + j] -= 1;
819 else if (s->channel_residues[j][icoef] < 0)
820 s->mclms_coeffs_cur[ich * num_channels + j] += 1;
825 for (ich = num_channels - 1; ich >= 0; ich--) {
827 s->mclms_prevvalues[s->mclms_recent] = s->channel_residues[ich][icoef];
828 if (s->channel_residues[ich][icoef] > range - 1)
829 s->mclms_prevvalues[s->mclms_recent] = range - 1;
830 else if (s->channel_residues[ich][icoef] < -range)
831 s->mclms_prevvalues[s->mclms_recent] = -range;
833 s->mclms_updates[s->mclms_recent] = 0;
834 if (s->channel_residues[ich][icoef] > 0)
835 s->mclms_updates[s->mclms_recent] = 1;
836 else if (s->channel_residues[ich][icoef] < 0)
837 s->mclms_updates[s->mclms_recent] = -1;
840 if (s->mclms_recent == 0) {
841 memcpy(&s->mclms_prevvalues[order * num_channels],
843 bps * order * num_channels);
844 memcpy(&s->mclms_updates[order * num_channels],
846 bps * order * num_channels);
847 s->mclms_recent = num_channels * order;
851 static void mclms_predict(WmallDecodeCtx *s, int icoef, int *pred)
854 int order = s->mclms_order;
855 int num_channels = s->num_channels;
857 for (ich = 0; ich < num_channels; ich++) {
858 if (!s->is_channel_coded[ich])
861 for (i = 0; i < order * num_channels; i++)
862 pred[ich] += s->mclms_prevvalues[i + s->mclms_recent] *
863 s->mclms_coeffs[i + order * num_channels * ich];
864 for (i = 0; i < ich; i++)
865 pred[ich] += s->channel_residues[i][icoef] *
866 s->mclms_coeffs_cur[i + num_channels * ich];
867 pred[ich] += 1 << s->mclms_scaling - 1;
868 pred[ich] >>= s->mclms_scaling;
869 s->channel_residues[ich][icoef] += pred[ich];
873 static void revert_mclms(WmallDecodeCtx *s, int tile_size)
875 int icoef, pred[s->num_channels];
876 for (icoef = 0; icoef < tile_size; icoef++) {
877 mclms_predict(s, icoef, pred);
878 mclms_update(s, icoef, pred);
882 static int lms_predict(WmallDecodeCtx *s, int ich, int ilms)
886 int recent = s->cdlms[ich][ilms].recent;
888 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
889 pred += s->cdlms[ich][ilms].coefs[icoef] *
890 s->cdlms[ich][ilms].lms_prevvalues[icoef + recent];
892 //pred += (1 << (s->cdlms[ich][ilms].scaling - 1));
893 /* XXX: Table 29 has:
894 iPred >= cdlms[iCh][ilms].scaling;
895 seems to me like a missing > */
896 //pred >>= s->cdlms[ich][ilms].scaling;
900 static void lms_update(WmallDecodeCtx *s, int ich, int ilms, int input, int residue)
903 int recent = s->cdlms[ich][ilms].recent;
904 int range = 1 << s->bits_per_sample - 1;
905 int bps = s->bits_per_sample > 16 ? 4 : 2; // bytes per sample
908 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
909 s->cdlms[ich][ilms].coefs[icoef] -=
910 s->cdlms[ich][ilms].lms_updates[icoef + recent];
911 } else if (residue > 0) {
912 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
913 s->cdlms[ich][ilms].coefs[icoef] +=
914 s->cdlms[ich][ilms].lms_updates[icoef + recent]; /* spec mistakenly
915 dropped the recent */
921 /* XXX: This memcpy()s will probably fail if a fixed 32-bit buffer is used.
922 follow kshishkov's suggestion of using a union. */
923 memcpy(&s->cdlms[ich][ilms].lms_prevvalues[s->cdlms[ich][ilms].order],
924 s->cdlms[ich][ilms].lms_prevvalues,
925 bps * s->cdlms[ich][ilms].order);
926 memcpy(&s->cdlms[ich][ilms].lms_updates[s->cdlms[ich][ilms].order],
927 s->cdlms[ich][ilms].lms_updates,
928 bps * s->cdlms[ich][ilms].order);
929 recent = s->cdlms[ich][ilms].order - 1;
932 s->cdlms[ich][ilms].lms_prevvalues[recent] = av_clip(input, -range, range - 1);
934 s->cdlms[ich][ilms].lms_updates[recent] = 0;
936 s->cdlms[ich][ilms].lms_updates[recent] = -s->update_speed[ich];
938 s->cdlms[ich][ilms].lms_updates[recent] = s->update_speed[ich];
941 cdlms[iCh][ilms].updates[iRecent + cdlms[iCh][ilms].order >> 4] >>= 2;
942 lms_updates[iCh][ilms][iRecent + cdlms[iCh][ilms].order >> 3] >>= 1;
944 Questions is - are cdlms[iCh][ilms].updates[] and lms_updates[][][] two
945 seperate buffers? Here I've assumed that the two are same which makes
948 s->cdlms[ich][ilms].lms_updates[recent + (s->cdlms[ich][ilms].order >> 4)] >>= 2;
949 s->cdlms[ich][ilms].lms_updates[recent + (s->cdlms[ich][ilms].order >> 3)] >>= 1;
950 s->cdlms[ich][ilms].recent = recent;
953 static void use_high_update_speed(WmallDecodeCtx *s, int ich)
955 int ilms, recent, icoef;
956 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
957 recent = s->cdlms[ich][ilms].recent;
958 if (s->update_speed[ich] == 16)
961 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
962 s->cdlms[ich][ilms].lms_updates[icoef + recent] *= 2;
964 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
965 s->cdlms[ich][ilms].lms_updates[icoef] *= 2;
968 s->update_speed[ich] = 16;
971 static void use_normal_update_speed(WmallDecodeCtx *s, int ich)
973 int ilms, recent, icoef;
974 for (ilms = s->cdlms_ttl[ich] - 1; ilms >= 0; ilms--) {
975 recent = s->cdlms[ich][ilms].recent;
976 if (s->update_speed[ich] == 8)
979 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
980 s->cdlms[ich][ilms].lms_updates[icoef + recent] /= 2;
982 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
983 s->cdlms[ich][ilms].lms_updates[icoef] /= 2;
986 s->update_speed[ich] = 8;
989 static void revert_cdlms(WmallDecodeCtx *s, int ch, int coef_begin, int coef_end)
996 num_lms = s->cdlms_ttl[ch];
997 for (ilms = num_lms - 1; ilms >= 0; ilms--) {
998 //s->cdlms[ch][ilms].recent = s->cdlms[ch][ilms].order;
999 for (icoef = coef_begin; icoef < coef_end; icoef++) {
1000 pred = 1 << (s->cdlms[ch][ilms].scaling - 1);
1001 residue = s->channel_residues[ch][icoef];
1002 pred += lms_predict(s, ch, ilms);
1003 input = residue + (pred >> s->cdlms[ch][ilms].scaling);
1004 lms_update(s, ch, ilms, input, residue);
1005 s->channel_residues[ch][icoef] = input;
1010 static void revert_inter_ch_decorr(WmallDecodeCtx *s, int tile_size)
1013 if (s->num_channels != 2)
1016 for (icoef = 0; icoef < tile_size; icoef++) {
1017 s->channel_residues[0][icoef] -= s->channel_residues[1][icoef] >> 1;
1018 s->channel_residues[1][icoef] += s->channel_residues[0][icoef];
1023 static void revert_acfilter(WmallDecodeCtx *s, int tile_size)
1027 int **ch_coeffs = s->channel_residues;
1028 int *filter_coeffs = s->acfilter_coeffs;
1029 int *prevvalues = s->acfilter_prevvalues;
1030 int scaling = s->acfilter_scaling;
1031 int order = s->acfilter_order;
1033 for (ich = 0; ich < s->num_channels; ich++) {
1034 for (icoef = 0; icoef < tile_size; icoef++) {
1035 for (itap = 0; itap < order; itap++)
1036 pred += filter_coeffs[itap] * prevvalues[itap];
1038 ch_coeffs[ich][icoef] += pred;
1039 for (itap = 1; itap < order; itap++)
1040 prevvalues[itap] = prevvalues[itap - 1];
1041 prevvalues[0] = ch_coeffs[ich][icoef];
1047 *@brief Decode a single subframe (block).
1048 *@param s codec context
1049 *@return 0 on success, < 0 when decoding failed
1051 static int decode_subframe(WmallDecodeCtx *s)
1053 int offset = s->samples_per_frame;
1054 int subframe_len = s->samples_per_frame;
1056 int total_samples = s->samples_per_frame * s->num_channels;
1059 int quant_stepsize = s->quant_stepsize;
1061 s->subframe_offset = get_bits_count(&s->gb);
1063 /** reset channel context and find the next block offset and size
1064 == the next block of the channel with the smallest number of
1067 for (i = 0; i < s->num_channels; i++) {
1068 s->channel[i].grouped = 0;
1069 if (offset > s->channel[i].decoded_samples) {
1070 offset = s->channel[i].decoded_samples;
1072 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1076 /** get a list of all channels that contain the estimated block */
1077 s->channels_for_cur_subframe = 0;
1078 for (i = 0; i < s->num_channels; i++) {
1079 const int cur_subframe = s->channel[i].cur_subframe;
1080 /** substract already processed samples */
1081 total_samples -= s->channel[i].decoded_samples;
1083 /** and count if there are multiple subframes that match our profile */
1084 if (offset == s->channel[i].decoded_samples &&
1085 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1086 total_samples -= s->channel[i].subframe_len[cur_subframe];
1087 s->channel[i].decoded_samples +=
1088 s->channel[i].subframe_len[cur_subframe];
1089 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1090 ++s->channels_for_cur_subframe;
1094 /** check if the frame will be complete after processing the
1097 s->parsed_all_subframes = 1;
1100 s->seekable_tile = get_bits1(&s->gb);
1101 if(s->seekable_tile) {
1102 clear_codec_buffers(s);
1104 s->do_arith_coding = get_bits1(&s->gb);
1105 if(s->do_arith_coding) {
1106 dprintf(s->avctx, "do_arith_coding == 1");
1109 s->do_ac_filter = get_bits1(&s->gb);
1110 s->do_inter_ch_decorr = get_bits1(&s->gb);
1111 s->do_mclms = get_bits1(&s->gb);
1114 decode_ac_filter(s);
1120 s->movave_scaling = get_bits(&s->gb, 3);
1121 s->quant_stepsize = get_bits(&s->gb, 8) + 1;
1126 rawpcm_tile = get_bits1(&s->gb);
1128 for(i = 0; i < s->num_channels; i++) {
1129 s->is_channel_coded[i] = 1;
1134 for(i = 0; i < s->num_channels; i++) {
1135 s->is_channel_coded[i] = get_bits1(&s->gb);
1140 s->do_lpc = get_bits1(&s->gb);
1150 if(get_bits1(&s->gb)) {
1151 padding_zeroes = get_bits(&s->gb, 5);
1158 int bits = s->bits_per_sample - padding_zeroes;
1159 dprintf(s->avctx, "RAWPCM %d bits per sample. total %d bits, remain=%d\n", bits,
1160 bits * s->num_channels * subframe_len, get_bits_count(&s->gb));
1161 for(i = 0; i < s->num_channels; i++) {
1162 for(j = 0; j < subframe_len; j++) {
1163 s->channel_coeffs[i][j] = get_sbits(&s->gb, bits);
1164 // dprintf(s->avctx, "PCM[%d][%d] = 0x%04x\n", i, j, s->channel_coeffs[i][j]);
1168 for(i = 0; i < s->num_channels; i++)
1169 if(s->is_channel_coded[i]) {
1170 decode_channel_residues(s, i, subframe_len);
1171 if (s->seekable_tile)
1172 use_high_update_speed(s, i);
1174 use_normal_update_speed(s, i);
1175 revert_cdlms(s, i, 0, subframe_len);
1179 revert_mclms(s, subframe_len);
1180 if (s->do_inter_ch_decorr)
1181 revert_inter_ch_decorr(s, subframe_len);
1183 revert_acfilter(s, subframe_len);
1186 for (i = 0; i < s->num_channels; i++)
1187 for (j = 0; j < subframe_len; j++)
1188 s->channel_residues[i][j] *= quant_stepsize;
1190 /** handled one subframe */
1192 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1193 int c = s->channel_indexes_for_cur_subframe[i];
1194 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1195 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1196 return AVERROR_INVALIDDATA;
1198 ++s->channel[c].cur_subframe;
1200 num_logged_subframes++;
1205 *@brief Decode one WMA frame.
1206 *@param s codec context
1207 *@return 0 if the trailer bit indicates that this is the last frame,
1208 * 1 if there are additional frames
1210 static int decode_frame(WmallDecodeCtx *s)
1212 GetBitContext* gb = &s->gb;
1213 int more_frames = 0;
1217 /** check for potential output buffer overflow */
1218 if (s->num_channels * s->samples_per_frame > s->samples_end - s->samples) {
1219 /** return an error if no frame could be decoded at all */
1220 av_log(s->avctx, AV_LOG_ERROR,
1221 "not enough space for the output samples\n");
1226 /** get frame length */
1228 len = get_bits(gb, s->log2_frame_size);
1230 /** decode tile information */
1231 if (decode_tilehdr(s)) {
1236 /** read drc info */
1237 if (s->dynamic_range_compression) {
1238 s->drc_gain = get_bits(gb, 8);
1241 /** no idea what these are for, might be the number of samples
1242 that need to be skipped at the beginning or end of a stream */
1243 if (get_bits1(gb)) {
1246 /** usually true for the first frame */
1247 if (get_bits1(gb)) {
1248 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1249 dprintf(s->avctx, "start skip: %i\n", skip);
1252 /** sometimes true for the last frame */
1253 if (get_bits1(gb)) {
1254 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1255 dprintf(s->avctx, "end skip: %i\n", skip);
1260 /** reset subframe states */
1261 s->parsed_all_subframes = 0;
1262 for (i = 0; i < s->num_channels; i++) {
1263 s->channel[i].decoded_samples = 0;
1264 s->channel[i].cur_subframe = 0;
1265 s->channel[i].reuse_sf = 0;
1268 /** decode all subframes */
1269 while (!s->parsed_all_subframes) {
1270 if (decode_subframe(s) < 0) {
1276 dprintf(s->avctx, "Frame done\n");
1278 if (s->skip_frame) {
1281 s->samples += s->num_channels * s->samples_per_frame;
1283 if (s->len_prefix) {
1284 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1285 /** FIXME: not sure if this is always an error */
1286 av_log(s->avctx, AV_LOG_ERROR,
1287 "frame[%i] would have to skip %i bits\n", s->frame_num,
1288 len - (get_bits_count(gb) - s->frame_offset) - 1);
1293 /** skip the rest of the frame data */
1294 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1297 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1298 dprintf(s->avctx, "skip1\n");
1303 /** decode trailer bit */
1304 more_frames = get_bits1(gb);
1310 *@brief Calculate remaining input buffer length.
1311 *@param s codec context
1312 *@param gb bitstream reader context
1313 *@return remaining size in bits
1315 static int remaining_bits(WmallDecodeCtx *s, GetBitContext *gb)
1317 return s->buf_bit_size - get_bits_count(gb);
1321 *@brief Fill the bit reservoir with a (partial) frame.
1322 *@param s codec context
1323 *@param gb bitstream reader context
1324 *@param len length of the partial frame
1325 *@param append decides wether to reset the buffer or not
1327 static void save_bits(WmallDecodeCtx *s, GetBitContext* gb, int len,
1332 /** when the frame data does not need to be concatenated, the input buffer
1333 is resetted and additional bits from the previous frame are copyed
1334 and skipped later so that a fast byte copy is possible */
1337 s->frame_offset = get_bits_count(gb) & 7;
1338 s->num_saved_bits = s->frame_offset;
1339 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1342 buflen = (s->num_saved_bits + len + 8) >> 3;
1344 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1345 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1350 s->num_saved_bits += len;
1352 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1355 int align = 8 - (get_bits_count(gb) & 7);
1356 align = FFMIN(align, len);
1357 put_bits(&s->pb, align, get_bits(gb, align));
1359 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1361 skip_bits_long(gb, len);
1364 PutBitContext tmp = s->pb;
1365 flush_put_bits(&tmp);
1368 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1369 skip_bits(&s->gb, s->frame_offset);
1373 *@brief Decode a single WMA packet.
1374 *@param avctx codec context
1375 *@param data the output buffer
1376 *@param data_size number of bytes that were written to the output buffer
1377 *@param avpkt input packet
1378 *@return number of bytes that were read from the input buffer
1380 static int decode_packet(AVCodecContext *avctx,
1381 void *data, int *data_size, AVPacket* avpkt)
1383 WmallDecodeCtx *s = avctx->priv_data;
1384 GetBitContext* gb = &s->pgb;
1385 const uint8_t* buf = avpkt->data;
1386 int buf_size = avpkt->size;
1387 int num_bits_prev_frame;
1388 int packet_sequence_number;
1391 s->samples_end = (float*)((int8_t*)data + *data_size);
1394 if (s->packet_done || s->packet_loss) {
1395 int seekable_frame_in_packet, spliced_packet;
1398 /** sanity check for the buffer length */
1399 if (buf_size < avctx->block_align)
1402 s->next_packet_start = buf_size - avctx->block_align;
1403 buf_size = avctx->block_align;
1404 s->buf_bit_size = buf_size << 3;
1406 /** parse packet header */
1407 init_get_bits(gb, buf, s->buf_bit_size);
1408 packet_sequence_number = get_bits(gb, 4);
1409 seekable_frame_in_packet = get_bits1(gb);
1410 spliced_packet = get_bits1(gb);
1412 /** get number of bits that need to be added to the previous frame */
1413 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1415 /** check for packet loss */
1416 if (!s->packet_loss &&
1417 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1419 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1420 s->packet_sequence_number, packet_sequence_number);
1422 s->packet_sequence_number = packet_sequence_number;
1424 if (num_bits_prev_frame > 0) {
1425 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1426 if (num_bits_prev_frame >= remaining_packet_bits) {
1427 num_bits_prev_frame = remaining_packet_bits;
1431 /** append the previous frame data to the remaining data from the
1432 previous packet to create a full frame */
1433 save_bits(s, gb, num_bits_prev_frame, 1);
1435 /** decode the cross packet frame if it is valid */
1436 if (!s->packet_loss)
1438 } else if (s->num_saved_bits - s->frame_offset) {
1439 dprintf(avctx, "ignoring %x previously saved bits\n",
1440 s->num_saved_bits - s->frame_offset);
1443 if (s->packet_loss) {
1444 /** reset number of saved bits so that the decoder
1445 does not start to decode incomplete frames in the
1446 s->len_prefix == 0 case */
1447 s->num_saved_bits = 0;
1454 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1455 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1456 skip_bits(gb, s->packet_offset);
1458 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1459 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1460 frame_size <= remaining_bits(s, gb)) {
1461 save_bits(s, gb, frame_size, 0);
1462 s->packet_done = !decode_frame(s);
1463 } else if (!s->len_prefix
1464 && s->num_saved_bits > get_bits_count(&s->gb)) {
1465 /** when the frames do not have a length prefix, we don't know
1466 the compressed length of the individual frames
1467 however, we know what part of a new packet belongs to the
1469 therefore we save the incoming packet first, then we append
1470 the "previous frame" data from the next packet so that
1471 we get a buffer that only contains full frames */
1472 s->packet_done = !decode_frame(s);
1478 if (s->packet_done && !s->packet_loss &&
1479 remaining_bits(s, gb) > 0) {
1480 /** save the rest of the data so that it can be decoded
1481 with the next packet */
1482 save_bits(s, gb, remaining_bits(s, gb), 0);
1485 *data_size = 0; // (int8_t *)s->samples - (int8_t *)data;
1486 s->packet_offset = get_bits_count(gb) & 7;
1488 return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3;
1492 *@brief Clear decoder buffers (for seeking).
1493 *@param avctx codec context
1495 static void flush(AVCodecContext *avctx)
1497 WmallDecodeCtx *s = avctx->priv_data;
1499 /** reset output buffer as a part of it is used during the windowing of a
1501 for (i = 0; i < s->num_channels; i++)
1502 memset(s->channel[i].out, 0, s->samples_per_frame *
1503 sizeof(*s->channel[i].out));
1509 *@brief wmall decoder
1511 AVCodec ff_wmalossless_decoder = {
1514 CODEC_ID_WMALOSSLESS,
1515 sizeof(WmallDecodeCtx),
1520 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_EXPERIMENTAL,
1522 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Lossless"),