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
145 * @brief channel group for channel transformations
148 uint8_t num_channels; ///< number of channels in the group
149 int8_t transform; ///< transform on / off
150 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
151 float decorrelation_matrix[WMALL_MAX_CHANNELS*WMALL_MAX_CHANNELS];
152 float* channel_data[WMALL_MAX_CHANNELS]; ///< transformation coefficients
156 * @brief main decoder context
158 typedef struct WmallDecodeCtx {
159 /* generic decoder variables */
160 AVCodecContext* avctx; ///< codec context for av_log
161 DSPContext dsp; ///< accelerated DSP functions
162 uint8_t frame_data[MAX_FRAMESIZE +
163 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
164 PutBitContext pb; ///< context for filling the frame_data buffer
165 FFTContext mdct_ctx[WMALL_BLOCK_SIZES]; ///< MDCT context per block size
166 DECLARE_ALIGNED(16, float, tmp)[WMALL_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
167 float* windows[WMALL_BLOCK_SIZES]; ///< windows for the different block sizes
169 /* frame size dependent frame information (set during initialization) */
170 uint32_t decode_flags; ///< used compression features
171 uint8_t len_prefix; ///< frame is prefixed with its length
172 uint8_t dynamic_range_compression; ///< frame contains DRC data
173 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
174 uint16_t samples_per_frame; ///< number of samples to output
175 uint16_t log2_frame_size;
176 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
177 int8_t lfe_channel; ///< lfe channel index
178 uint8_t max_num_subframes;
179 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
180 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
181 uint16_t min_samples_per_subframe;
182 int8_t num_sfb[WMALL_BLOCK_SIZES]; ///< scale factor bands per block size
183 int16_t sfb_offsets[WMALL_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
184 int8_t sf_offsets[WMALL_BLOCK_SIZES][WMALL_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
185 int16_t subwoofer_cutoffs[WMALL_BLOCK_SIZES]; ///< subwoofer cutoff values
187 /* packet decode state */
188 GetBitContext pgb; ///< bitstream reader context for the packet
189 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
190 uint8_t packet_offset; ///< frame offset in the packet
191 uint8_t packet_sequence_number; ///< current packet number
192 int num_saved_bits; ///< saved number of bits
193 int frame_offset; ///< frame offset in the bit reservoir
194 int subframe_offset; ///< subframe offset in the bit reservoir
195 uint8_t packet_loss; ///< set in case of bitstream error
196 uint8_t packet_done; ///< set when a packet is fully decoded
198 /* frame decode state */
199 uint32_t frame_num; ///< current frame number (not used for decoding)
200 GetBitContext gb; ///< bitstream reader context
201 int buf_bit_size; ///< buffer size in bits
202 float* samples; ///< current samplebuffer pointer
203 float* samples_end; ///< maximum samplebuffer pointer
204 uint8_t drc_gain; ///< gain for the DRC tool
205 int8_t skip_frame; ///< skip output step
206 int8_t parsed_all_subframes; ///< all subframes decoded?
208 /* subframe/block decode state */
209 int16_t subframe_len; ///< current subframe length
210 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
211 int8_t channel_indexes_for_cur_subframe[WMALL_MAX_CHANNELS];
212 int8_t num_bands; ///< number of scale factor bands
213 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
214 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
215 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
216 int8_t esc_len; ///< length of escaped coefficients
218 uint8_t num_chgroups; ///< number of channel groups
219 WmallChannelGrp chgroup[WMALL_MAX_CHANNELS]; ///< channel group information
221 WmallChannelCtx channel[WMALL_MAX_CHANNELS]; ///< per channel data
225 uint8_t do_arith_coding;
226 uint8_t do_ac_filter;
227 uint8_t do_inter_ch_decorr;
231 int8_t acfilter_order;
232 int8_t acfilter_scaling;
233 int acfilter_coeffs[16];
236 int8_t mclms_scaling;
237 int16_t mclms_coeffs[128];
238 int16_t mclms_coeffs_cur[4];
256 int is_channel_coded[2];
259 int transient_pos[2];
264 int channel_residues[2][2048];
267 int lpc_coefs[2][40];
272 int channel_coeffs[2][2048];
278 #define dprintf(pctx, ...) av_log(pctx, AV_LOG_DEBUG, __VA_ARGS__)
282 *@brief helper function to print the most important members of the context
285 static void av_cold dump_context(WmallDecodeCtx *s)
287 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
288 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
290 PRINT("ed sample bit depth", s->bits_per_sample);
291 PRINT_HEX("ed decode flags", s->decode_flags);
292 PRINT("samples per frame", s->samples_per_frame);
293 PRINT("log2 frame size", s->log2_frame_size);
294 PRINT("max num subframes", s->max_num_subframes);
295 PRINT("len prefix", s->len_prefix);
296 PRINT("num channels", s->num_channels);
300 *@brief Uninitialize the decoder and free all resources.
301 *@param avctx codec context
302 *@return 0 on success, < 0 otherwise
304 static av_cold int decode_end(AVCodecContext *avctx)
306 WmallDecodeCtx *s = avctx->priv_data;
309 for (i = 0; i < WMALL_BLOCK_SIZES; i++)
310 ff_mdct_end(&s->mdct_ctx[i]);
316 *@brief Initialize the decoder.
317 *@param avctx codec context
318 *@return 0 on success, -1 otherwise
320 static av_cold int decode_init(AVCodecContext *avctx)
322 WmallDecodeCtx *s = avctx->priv_data;
323 uint8_t *edata_ptr = avctx->extradata;
324 unsigned int channel_mask;
326 int log2_max_num_subframes;
327 int num_possible_block_sizes;
330 dsputil_init(&s->dsp, avctx);
331 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
333 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
335 if (avctx->extradata_size >= 18) {
336 s->decode_flags = AV_RL16(edata_ptr+14);
337 channel_mask = AV_RL32(edata_ptr+2);
338 s->bits_per_sample = AV_RL16(edata_ptr);
339 /** dump the extradata */
340 for (i = 0; i < avctx->extradata_size; i++)
341 dprintf(avctx, "[%x] ", avctx->extradata[i]);
342 dprintf(avctx, "\n");
345 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
346 return AVERROR_INVALIDDATA;
350 s->log2_frame_size = av_log2(avctx->block_align) + 4;
353 s->skip_frame = 1; /* skip first frame */
355 s->len_prefix = (s->decode_flags & 0x40);
358 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
361 /** init previous block len */
362 for (i = 0; i < avctx->channels; i++)
363 s->channel[i].prev_block_len = s->samples_per_frame;
366 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
367 s->max_num_subframes = 1 << log2_max_num_subframes;
368 s->max_subframe_len_bit = 0;
369 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
371 num_possible_block_sizes = log2_max_num_subframes + 1;
372 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
373 s->dynamic_range_compression = (s->decode_flags & 0x80);
375 s->bV3RTM = s->decode_flags & 0x100;
377 if (s->max_num_subframes > MAX_SUBFRAMES) {
378 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
379 s->max_num_subframes);
380 return AVERROR_INVALIDDATA;
383 s->num_channels = avctx->channels;
385 /** extract lfe channel position */
388 if (channel_mask & 8) {
390 for (mask = 1; mask < 16; mask <<= 1) {
391 if (channel_mask & mask)
396 if (s->num_channels < 0) {
397 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
398 return AVERROR_INVALIDDATA;
399 } else if (s->num_channels > WMALL_MAX_CHANNELS) {
400 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
401 return AVERROR_PATCHWELCOME;
404 avctx->channel_layout = channel_mask;
409 *@brief Decode the subframe length.
411 *@param offset sample offset in the frame
412 *@return decoded subframe length on success, < 0 in case of an error
414 static int decode_subframe_length(WmallDecodeCtx *s, int offset)
417 int subframe_len, len;
419 /** no need to read from the bitstream when only one length is possible */
420 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
421 return s->min_samples_per_subframe;
423 len = av_log2(s->max_num_subframes - 1) + 1;
424 frame_len_ratio = get_bits(&s->gb, len);
426 subframe_len = s->min_samples_per_subframe * (frame_len_ratio + 1);
428 /** sanity check the length */
429 if (subframe_len < s->min_samples_per_subframe ||
430 subframe_len > s->samples_per_frame) {
431 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
433 return AVERROR_INVALIDDATA;
439 *@brief Decode how the data in the frame is split into subframes.
440 * Every WMA frame contains the encoded data for a fixed number of
441 * samples per channel. The data for every channel might be split
442 * into several subframes. This function will reconstruct the list of
443 * subframes for every channel.
445 * If the subframes are not evenly split, the algorithm estimates the
446 * channels with the lowest number of total samples.
447 * Afterwards, for each of these channels a bit is read from the
448 * bitstream that indicates if the channel contains a subframe with the
449 * next subframe size that is going to be read from the bitstream or not.
450 * If a channel contains such a subframe, the subframe size gets added to
451 * the channel's subframe list.
452 * The algorithm repeats these steps until the frame is properly divided
453 * between the individual channels.
456 *@return 0 on success, < 0 in case of an error
458 static int decode_tilehdr(WmallDecodeCtx *s)
460 uint16_t num_samples[WMALL_MAX_CHANNELS]; /**< sum of samples for all currently known subframes of a channel */
461 uint8_t contains_subframe[WMALL_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
462 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
463 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subfra2me offsets and sizes */
464 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
467 /* Should never consume more than 3073 bits (256 iterations for the
468 * while loop when always the minimum amount of 128 samples is substracted
469 * from missing samples in the 8 channel case).
470 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
473 /** reset tiling information */
474 for (c = 0; c < s->num_channels; c++)
475 s->channel[c].num_subframes = 0;
477 memset(num_samples, 0, sizeof(num_samples));
479 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
480 fixed_channel_layout = 1;
482 /** loop until the frame data is split between the subframes */
486 /** check which channels contain the subframe */
487 for (c = 0; c < s->num_channels; c++) {
488 if (num_samples[c] == min_channel_len) {
489 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
490 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) {
491 contains_subframe[c] = 1;
494 contains_subframe[c] = get_bits1(&s->gb);
497 contains_subframe[c] = 0;
500 /** get subframe length, subframe_len == 0 is not allowed */
501 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
502 return AVERROR_INVALIDDATA;
503 /** add subframes to the individual channels and find new min_channel_len */
504 min_channel_len += subframe_len;
505 for (c = 0; c < s->num_channels; c++) {
506 WmallChannelCtx* chan = &s->channel[c];
508 if (contains_subframe[c]) {
509 if (chan->num_subframes >= MAX_SUBFRAMES) {
510 av_log(s->avctx, AV_LOG_ERROR,
511 "broken frame: num subframes > 31\n");
512 return AVERROR_INVALIDDATA;
514 chan->subframe_len[chan->num_subframes] = subframe_len;
515 num_samples[c] += subframe_len;
516 ++chan->num_subframes;
517 if (num_samples[c] > s->samples_per_frame) {
518 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
519 "channel len(%d) > samples_per_frame(%d)\n",
520 num_samples[c], s->samples_per_frame);
521 return AVERROR_INVALIDDATA;
523 } else if (num_samples[c] <= min_channel_len) {
524 if (num_samples[c] < min_channel_len) {
525 channels_for_cur_subframe = 0;
526 min_channel_len = num_samples[c];
528 ++channels_for_cur_subframe;
531 } while (min_channel_len < s->samples_per_frame);
533 for (c = 0; c < s->num_channels; c++) {
536 for (i = 0; i < s->channel[c].num_subframes; i++) {
537 s->channel[c].subframe_offset[i] = offset;
538 offset += s->channel[c].subframe_len[i];
546 static int my_log2(unsigned int i)
548 unsigned int iLog2 = 0;
549 while ((i >> iLog2) > 1)
558 static void decode_ac_filter(WmallDecodeCtx *s)
561 s->acfilter_order = get_bits(&s->gb, 4) + 1;
562 s->acfilter_scaling = get_bits(&s->gb, 4);
564 for(i = 0; i < s->acfilter_order; i++) {
565 s->acfilter_coeffs[i] = get_bits(&s->gb, s->acfilter_scaling) + 1;
573 static void decode_mclms(WmallDecodeCtx *s)
575 s->mclms_order = (get_bits(&s->gb, 4) + 1) * 2;
576 s->mclms_scaling = get_bits(&s->gb, 4);
577 if(get_bits1(&s->gb)) {
581 int cbits = av_log2(s->mclms_scaling + 1);
582 assert(cbits == my_log2(s->mclms_scaling + 1));
583 if(1 << cbits < s->mclms_scaling + 1)
586 send_coef_bits = (cbits ? get_bits(&s->gb, cbits) : 0) + 2;
588 for(i = 0; i < s->mclms_order * s->num_channels * s->num_channels; i++) {
589 s->mclms_coeffs[i] = get_bits(&s->gb, send_coef_bits);
592 for(i = 0; i < s->num_channels; i++) {
594 for(c = 0; c < i; c++) {
595 s->mclms_coeffs_cur[i * s->num_channels + c] = get_bits(&s->gb, send_coef_bits);
605 static void decode_cdlms(WmallDecodeCtx *s)
608 int cdlms_send_coef = get_bits1(&s->gb);
610 for(c = 0; c < s->num_channels; c++) {
611 s->cdlms_ttl[c] = get_bits(&s->gb, 3) + 1;
612 for(i = 0; i < s->cdlms_ttl[c]; i++) {
613 s->cdlms[c][i].order = (get_bits(&s->gb, 7) + 1) * 8;
616 for(i = 0; i < s->cdlms_ttl[c]; i++) {
617 s->cdlms[c][i].scaling = get_bits(&s->gb, 4);
620 if(cdlms_send_coef) {
621 for(i = 0; i < s->cdlms_ttl[c]; i++) {
622 int cbits, shift_l, shift_r, j;
623 cbits = av_log2(s->cdlms[c][i].order);
624 if(1 << cbits < s->cdlms[c][i].order)
626 s->cdlms[c][i].coefsend = get_bits(&s->gb, cbits) + 1;
628 cbits = av_log2(s->cdlms[c][i].scaling + 1);
629 if(1 << cbits < s->cdlms[c][i].scaling + 1)
632 s->cdlms[c][i].bitsend = get_bits(&s->gb, cbits) + 2;
633 shift_l = 32 - s->cdlms[c][i].bitsend;
634 shift_r = 32 - 2 - s->cdlms[c][i].scaling;
635 for(j = 0; j < s->cdlms[c][i].coefsend; j++) {
636 s->cdlms[c][i].coefs[j] =
637 (get_bits(&s->gb, s->cdlms[c][i].bitsend) << shift_l) >> shift_r;
647 static int decode_channel_residues(WmallDecodeCtx *s, int ch, int tile_size)
650 unsigned int ave_mean;
651 s->transient[ch] = get_bits1(&s->gb);
653 s->transient_pos[ch] = get_bits(&s->gb, av_log2(tile_size));
655 if(s->seekable_tile) {
656 ave_mean = get_bits(&s->gb, s->bits_per_sample);
657 s->ave_sum[ch] = ave_mean << (s->movave_scaling + 1);
658 // s->ave_sum[ch] *= 2;
661 if(s->seekable_tile) {
662 if(s->do_inter_ch_decorr)
663 s->channel_residues[ch][0] = get_sbits(&s->gb, s->bits_per_sample + 1);
665 s->channel_residues[ch][0] = get_sbits(&s->gb, s->bits_per_sample);
668 for(; i < tile_size; i++) {
669 int quo = 0, rem, rem_bits, residue;
670 while(get_bits1(&s->gb))
673 quo += get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1);
675 ave_mean = (s->ave_sum[ch] + (1 << s->movave_scaling)) >> (s->movave_scaling + 1);
676 rem_bits = av_ceil_log2(ave_mean);
677 rem = rem_bits ? get_bits(&s->gb, rem_bits) : 0;
678 residue = (quo << rem_bits) + rem;
680 s->ave_sum[ch] = residue + s->ave_sum[ch] - (s->ave_sum[ch] >> s->movave_scaling);
683 residue = -(residue >> 1) - 1;
685 residue = residue >> 1;
686 s->channel_residues[ch][i] = residue;
688 // dprintf(s->avctx, "%5d: %5d %10d %12d %12d %5d %-16d %04x\n",i, quo, ave_mean, s->ave_sum[ch], rem, rem_bits, s->channel_residues[ch][i], show_bits(&s->gb, 16));
700 decode_lpc(WmallDecodeCtx *s)
703 s->lpc_order = get_bits(&s->gb, 5) + 1;
704 s->lpc_scaling = get_bits(&s->gb, 4);
705 s->lpc_intbits = get_bits(&s->gb, 3) + 1;
706 cbits = s->lpc_scaling + s->lpc_intbits;
707 for(ch = 0; ch < s->num_channels; ch++) {
708 for(i = 0; i < s->lpc_order; i++) {
709 s->lpc_coefs[ch][i] = get_sbits(&s->gb, cbits);
717 *@brief Decode a single subframe (block).
718 *@param s codec context
719 *@return 0 on success, < 0 when decoding failed
721 static int decode_subframe(WmallDecodeCtx *s)
723 int offset = s->samples_per_frame;
724 int subframe_len = s->samples_per_frame;
726 int total_samples = s->samples_per_frame * s->num_channels;
730 s->subframe_offset = get_bits_count(&s->gb);
732 /** reset channel context and find the next block offset and size
733 == the next block of the channel with the smallest number of
736 for (i = 0; i < s->num_channels; i++) {
737 s->channel[i].grouped = 0;
738 if (offset > s->channel[i].decoded_samples) {
739 offset = s->channel[i].decoded_samples;
741 s->channel[i].subframe_len[s->channel[i].cur_subframe];
745 /** get a list of all channels that contain the estimated block */
746 s->channels_for_cur_subframe = 0;
747 for (i = 0; i < s->num_channels; i++) {
748 const int cur_subframe = s->channel[i].cur_subframe;
749 /** substract already processed samples */
750 total_samples -= s->channel[i].decoded_samples;
752 /** and count if there are multiple subframes that match our profile */
753 if (offset == s->channel[i].decoded_samples &&
754 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
755 total_samples -= s->channel[i].subframe_len[cur_subframe];
756 s->channel[i].decoded_samples +=
757 s->channel[i].subframe_len[cur_subframe];
758 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
759 ++s->channels_for_cur_subframe;
763 /** check if the frame will be complete after processing the
766 s->parsed_all_subframes = 1;
769 s->seekable_tile = get_bits1(&s->gb);
770 if(s->seekable_tile) {
771 s->do_arith_coding = get_bits1(&s->gb);
772 if(s->do_arith_coding) {
773 dprintf(s->avctx, "do_arith_coding == 1");
776 s->do_ac_filter = get_bits1(&s->gb);
777 s->do_inter_ch_decorr = get_bits1(&s->gb);
778 s->do_mclms = get_bits1(&s->gb);
787 s->movave_scaling = get_bits(&s->gb, 3);
788 s->quant_stepsize = get_bits(&s->gb, 8) + 1;
791 rawpcm_tile = get_bits1(&s->gb);
793 for(i = 0; i < s->num_channels; i++) {
794 s->is_channel_coded[i] = 1;
799 for(i = 0; i < s->num_channels; i++) {
800 s->is_channel_coded[i] = get_bits1(&s->gb);
805 s->do_lpc = get_bits1(&s->gb);
815 if(get_bits1(&s->gb)) {
816 padding_zeroes = get_bits(&s->gb, 5);
823 int bits = s->bits_per_sample - padding_zeroes;
825 dprintf(s->avctx, "RAWPCM %d bits per sample. total %d bits, remain=%d\n", bits,
826 bits * s->num_channels * subframe_len, get_bits_count(&s->gb));
827 for(i = 0; i < s->num_channels; i++) {
828 for(j = 0; j < subframe_len; j++) {
829 s->channel_coeffs[i][j] = get_sbits(&s->gb, bits);
830 // dprintf(s->avctx, "PCM[%d][%d] = 0x%04x\n", i, j, s->channel_coeffs[i][j]);
834 for(i = 0; i < s->num_channels; i++)
835 if(s->is_channel_coded[i])
836 decode_channel_residues(s, i, subframe_len);
839 /** handled one subframe */
841 for (i = 0; i < s->channels_for_cur_subframe; i++) {
842 int c = s->channel_indexes_for_cur_subframe[i];
843 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
844 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
845 return AVERROR_INVALIDDATA;
847 ++s->channel[c].cur_subframe;
853 *@brief Decode one WMA frame.
854 *@param s codec context
855 *@return 0 if the trailer bit indicates that this is the last frame,
856 * 1 if there are additional frames
858 static int decode_frame(WmallDecodeCtx *s)
860 GetBitContext* gb = &s->gb;
865 /** check for potential output buffer overflow */
866 if (s->num_channels * s->samples_per_frame > s->samples_end - s->samples) {
867 /** return an error if no frame could be decoded at all */
868 av_log(s->avctx, AV_LOG_ERROR,
869 "not enough space for the output samples\n");
874 /** get frame length */
876 len = get_bits(gb, s->log2_frame_size);
878 /** decode tile information */
879 if (decode_tilehdr(s)) {
885 if (s->dynamic_range_compression) {
886 s->drc_gain = get_bits(gb, 8);
889 /** no idea what these are for, might be the number of samples
890 that need to be skipped at the beginning or end of a stream */
894 /** usually true for the first frame */
896 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
897 dprintf(s->avctx, "start skip: %i\n", skip);
900 /** sometimes true for the last frame */
902 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
903 dprintf(s->avctx, "end skip: %i\n", skip);
908 /** reset subframe states */
909 s->parsed_all_subframes = 0;
910 for (i = 0; i < s->num_channels; i++) {
911 s->channel[i].decoded_samples = 0;
912 s->channel[i].cur_subframe = 0;
913 s->channel[i].reuse_sf = 0;
916 /** decode all subframes */
917 while (!s->parsed_all_subframes) {
918 if (decode_subframe(s) < 0) {
924 dprintf(s->avctx, "Frame done\n");
929 s->samples += s->num_channels * s->samples_per_frame;
932 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
933 /** FIXME: not sure if this is always an error */
934 av_log(s->avctx, AV_LOG_ERROR,
935 "frame[%i] would have to skip %i bits\n", s->frame_num,
936 len - (get_bits_count(gb) - s->frame_offset) - 1);
941 /** skip the rest of the frame data */
942 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
945 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
946 dprintf(s->avctx, "skip1\n");
951 /** decode trailer bit */
952 more_frames = get_bits1(gb);
958 *@brief Calculate remaining input buffer length.
959 *@param s codec context
960 *@param gb bitstream reader context
961 *@return remaining size in bits
963 static int remaining_bits(WmallDecodeCtx *s, GetBitContext *gb)
965 return s->buf_bit_size - get_bits_count(gb);
969 *@brief Fill the bit reservoir with a (partial) frame.
970 *@param s codec context
971 *@param gb bitstream reader context
972 *@param len length of the partial frame
973 *@param append decides wether to reset the buffer or not
975 static void save_bits(WmallDecodeCtx *s, GetBitContext* gb, int len,
980 /** when the frame data does not need to be concatenated, the input buffer
981 is resetted and additional bits from the previous frame are copyed
982 and skipped later so that a fast byte copy is possible */
985 s->frame_offset = get_bits_count(gb) & 7;
986 s->num_saved_bits = s->frame_offset;
987 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
990 buflen = (s->num_saved_bits + len + 8) >> 3;
992 if (len <= 0 || buflen > MAX_FRAMESIZE) {
993 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
998 s->num_saved_bits += len;
1000 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1003 int align = 8 - (get_bits_count(gb) & 7);
1004 align = FFMIN(align, len);
1005 put_bits(&s->pb, align, get_bits(gb, align));
1007 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1009 skip_bits_long(gb, len);
1012 PutBitContext tmp = s->pb;
1013 flush_put_bits(&tmp);
1016 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1017 skip_bits(&s->gb, s->frame_offset);
1021 *@brief Decode a single WMA packet.
1022 *@param avctx codec context
1023 *@param data the output buffer
1024 *@param data_size number of bytes that were written to the output buffer
1025 *@param avpkt input packet
1026 *@return number of bytes that were read from the input buffer
1028 static int decode_packet(AVCodecContext *avctx,
1029 void *data, int *data_size, AVPacket* avpkt)
1031 WmallDecodeCtx *s = avctx->priv_data;
1032 GetBitContext* gb = &s->pgb;
1033 const uint8_t* buf = avpkt->data;
1034 int buf_size = avpkt->size;
1035 int num_bits_prev_frame;
1036 int packet_sequence_number;
1039 s->samples_end = (float*)((int8_t*)data + *data_size);
1042 if (s->packet_done || s->packet_loss) {
1045 /** sanity check for the buffer length */
1046 if (buf_size < avctx->block_align)
1049 s->next_packet_start = buf_size - avctx->block_align;
1050 buf_size = avctx->block_align;
1051 s->buf_bit_size = buf_size << 3;
1053 /** parse packet header */
1054 init_get_bits(gb, buf, s->buf_bit_size);
1055 packet_sequence_number = get_bits(gb, 4);
1056 int seekable_frame_in_packet = get_bits1(gb);
1057 int spliced_packet = get_bits1(gb);
1059 /** get number of bits that need to be added to the previous frame */
1060 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1062 /** check for packet loss */
1063 if (!s->packet_loss &&
1064 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1066 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1067 s->packet_sequence_number, packet_sequence_number);
1069 s->packet_sequence_number = packet_sequence_number;
1071 if (num_bits_prev_frame > 0) {
1072 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1073 if (num_bits_prev_frame >= remaining_packet_bits) {
1074 num_bits_prev_frame = remaining_packet_bits;
1078 /** append the previous frame data to the remaining data from the
1079 previous packet to create a full frame */
1080 save_bits(s, gb, num_bits_prev_frame, 1);
1082 /** decode the cross packet frame if it is valid */
1083 if (!s->packet_loss)
1085 } else if (s->num_saved_bits - s->frame_offset) {
1086 dprintf(avctx, "ignoring %x previously saved bits\n",
1087 s->num_saved_bits - s->frame_offset);
1090 if (s->packet_loss) {
1091 /** reset number of saved bits so that the decoder
1092 does not start to decode incomplete frames in the
1093 s->len_prefix == 0 case */
1094 s->num_saved_bits = 0;
1101 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1102 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1103 skip_bits(gb, s->packet_offset);
1105 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1106 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1107 frame_size <= remaining_bits(s, gb)) {
1108 save_bits(s, gb, frame_size, 0);
1109 s->packet_done = !decode_frame(s);
1110 } else if (!s->len_prefix
1111 && s->num_saved_bits > get_bits_count(&s->gb)) {
1112 /** when the frames do not have a length prefix, we don't know
1113 the compressed length of the individual frames
1114 however, we know what part of a new packet belongs to the
1116 therefore we save the incoming packet first, then we append
1117 the "previous frame" data from the next packet so that
1118 we get a buffer that only contains full frames */
1119 s->packet_done = !decode_frame(s);
1125 if (s->packet_done && !s->packet_loss &&
1126 remaining_bits(s, gb) > 0) {
1127 /** save the rest of the data so that it can be decoded
1128 with the next packet */
1129 save_bits(s, gb, remaining_bits(s, gb), 0);
1132 *data_size = 0; // (int8_t *)s->samples - (int8_t *)data;
1133 s->packet_offset = get_bits_count(gb) & 7;
1135 return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3;
1139 *@brief Clear decoder buffers (for seeking).
1140 *@param avctx codec context
1142 static void flush(AVCodecContext *avctx)
1144 WmallDecodeCtx *s = avctx->priv_data;
1146 /** reset output buffer as a part of it is used during the windowing of a
1148 for (i = 0; i < s->num_channels; i++)
1149 memset(s->channel[i].out, 0, s->samples_per_frame *
1150 sizeof(*s->channel[i].out));
1156 *@brief wmall decoder
1158 AVCodec ff_wmalossless_decoder = {
1161 CODEC_ID_WMALOSSLESS,
1162 sizeof(WmallDecodeCtx),
1167 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_EXPERIMENTAL,
1169 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Lossless"),