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];
237 int8_t mclms_scaling;
238 int16_t mclms_coeffs[128];
239 int16_t mclms_coeffs_cur[4];
240 int mclms_prevvalues[64]; // FIXME: should be 32-bit / 16-bit depending on bit-depth
241 int16_t mclms_updates[64];
253 int lms_prevvalues[512]; // FIXME: see above
254 int16_t lms_updates[512]; // and here too
256 } cdlms[2][9]; /* XXX: Here, 2 is the max. no. of channels allowed,
257 9 is the maximum no. of filters per channel.
258 Question is, why 2 if WMALL_MAX_CHANNELS == 8 */
265 int is_channel_coded[2]; // XXX: same question as above applies here too (and below)
269 int transient_pos[2];
274 int channel_residues[2][2048];
277 int lpc_coefs[2][40];
282 int channel_coeffs[2][2048];
288 #define dprintf(pctx, ...) av_log(pctx, AV_LOG_DEBUG, __VA_ARGS__)
291 static int num_logged_tiles = 0;
294 *@brief helper function to print the most important members of the context
297 static void av_cold dump_context(WmallDecodeCtx *s)
299 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
300 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
302 PRINT("ed sample bit depth", s->bits_per_sample);
303 PRINT_HEX("ed decode flags", s->decode_flags);
304 PRINT("samples per frame", s->samples_per_frame);
305 PRINT("log2 frame size", s->log2_frame_size);
306 PRINT("max num subframes", s->max_num_subframes);
307 PRINT("len prefix", s->len_prefix);
308 PRINT("num channels", s->num_channels);
311 static int dump_int_buffer(int *buffer, int length, int delimiter)
315 for (i=0 ; i<length ; i++) {
317 av_log(0, 0, "\n[%d] ", i);
318 av_log(0, 0, "%d, ", buffer[i]);
325 *@brief Uninitialize the decoder and free all resources.
326 *@param avctx codec context
327 *@return 0 on success, < 0 otherwise
329 static av_cold int decode_end(AVCodecContext *avctx)
331 WmallDecodeCtx *s = avctx->priv_data;
334 for (i = 0; i < WMALL_BLOCK_SIZES; i++)
335 ff_mdct_end(&s->mdct_ctx[i]);
341 *@brief Initialize the decoder.
342 *@param avctx codec context
343 *@return 0 on success, -1 otherwise
345 static av_cold int decode_init(AVCodecContext *avctx)
347 WmallDecodeCtx *s = avctx->priv_data;
348 uint8_t *edata_ptr = avctx->extradata;
349 unsigned int channel_mask;
351 int log2_max_num_subframes;
352 int num_possible_block_sizes;
355 dsputil_init(&s->dsp, avctx);
356 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
358 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
360 if (avctx->extradata_size >= 18) {
361 s->decode_flags = AV_RL16(edata_ptr+14);
362 channel_mask = AV_RL32(edata_ptr+2);
363 s->bits_per_sample = AV_RL16(edata_ptr);
364 /** dump the extradata */
365 for (i = 0; i < avctx->extradata_size; i++)
366 dprintf(avctx, "[%x] ", avctx->extradata[i]);
367 dprintf(avctx, "\n");
370 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
371 return AVERROR_INVALIDDATA;
375 s->log2_frame_size = av_log2(avctx->block_align) + 4;
378 s->skip_frame = 1; /* skip first frame */
380 s->len_prefix = (s->decode_flags & 0x40);
383 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
386 /** init previous block len */
387 for (i = 0; i < avctx->channels; i++)
388 s->channel[i].prev_block_len = s->samples_per_frame;
391 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
392 s->max_num_subframes = 1 << log2_max_num_subframes;
393 s->max_subframe_len_bit = 0;
394 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
396 num_possible_block_sizes = log2_max_num_subframes + 1;
397 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
398 s->dynamic_range_compression = (s->decode_flags & 0x80);
400 s->bV3RTM = s->decode_flags & 0x100;
402 if (s->max_num_subframes > MAX_SUBFRAMES) {
403 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
404 s->max_num_subframes);
405 return AVERROR_INVALIDDATA;
408 s->num_channels = avctx->channels;
410 /** extract lfe channel position */
413 if (channel_mask & 8) {
415 for (mask = 1; mask < 16; mask <<= 1) {
416 if (channel_mask & mask)
421 if (s->num_channels < 0) {
422 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
423 return AVERROR_INVALIDDATA;
424 } else if (s->num_channels > WMALL_MAX_CHANNELS) {
425 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
426 return AVERROR_PATCHWELCOME;
429 avctx->channel_layout = channel_mask;
434 *@brief Decode the subframe length.
436 *@param offset sample offset in the frame
437 *@return decoded subframe length on success, < 0 in case of an error
439 static int decode_subframe_length(WmallDecodeCtx *s, int offset)
442 int subframe_len, len;
444 /** no need to read from the bitstream when only one length is possible */
445 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
446 return s->min_samples_per_subframe;
448 len = av_log2(s->max_num_subframes - 1) + 1;
449 frame_len_ratio = get_bits(&s->gb, len);
451 subframe_len = s->min_samples_per_subframe * (frame_len_ratio + 1);
453 /** sanity check the length */
454 if (subframe_len < s->min_samples_per_subframe ||
455 subframe_len > s->samples_per_frame) {
456 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
458 return AVERROR_INVALIDDATA;
464 *@brief Decode how the data in the frame is split into subframes.
465 * Every WMA frame contains the encoded data for a fixed number of
466 * samples per channel. The data for every channel might be split
467 * into several subframes. This function will reconstruct the list of
468 * subframes for every channel.
470 * If the subframes are not evenly split, the algorithm estimates the
471 * channels with the lowest number of total samples.
472 * Afterwards, for each of these channels a bit is read from the
473 * bitstream that indicates if the channel contains a subframe with the
474 * next subframe size that is going to be read from the bitstream or not.
475 * If a channel contains such a subframe, the subframe size gets added to
476 * the channel's subframe list.
477 * The algorithm repeats these steps until the frame is properly divided
478 * between the individual channels.
481 *@return 0 on success, < 0 in case of an error
483 static int decode_tilehdr(WmallDecodeCtx *s)
485 uint16_t num_samples[WMALL_MAX_CHANNELS]; /**< sum of samples for all currently known subframes of a channel */
486 uint8_t contains_subframe[WMALL_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
487 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
488 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subfra2me offsets and sizes */
489 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
492 /* Should never consume more than 3073 bits (256 iterations for the
493 * while loop when always the minimum amount of 128 samples is substracted
494 * from missing samples in the 8 channel case).
495 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
498 /** reset tiling information */
499 for (c = 0; c < s->num_channels; c++)
500 s->channel[c].num_subframes = 0;
502 memset(num_samples, 0, sizeof(num_samples));
504 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
505 fixed_channel_layout = 1;
507 /** loop until the frame data is split between the subframes */
511 /** check which channels contain the subframe */
512 for (c = 0; c < s->num_channels; c++) {
513 if (num_samples[c] == min_channel_len) {
514 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
515 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) {
516 contains_subframe[c] = 1;
519 contains_subframe[c] = get_bits1(&s->gb);
522 contains_subframe[c] = 0;
525 /** get subframe length, subframe_len == 0 is not allowed */
526 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
527 return AVERROR_INVALIDDATA;
528 /** add subframes to the individual channels and find new min_channel_len */
529 min_channel_len += subframe_len;
530 for (c = 0; c < s->num_channels; c++) {
531 WmallChannelCtx* chan = &s->channel[c];
533 if (contains_subframe[c]) {
534 if (chan->num_subframes >= MAX_SUBFRAMES) {
535 av_log(s->avctx, AV_LOG_ERROR,
536 "broken frame: num subframes > 31\n");
537 return AVERROR_INVALIDDATA;
539 chan->subframe_len[chan->num_subframes] = subframe_len;
540 num_samples[c] += subframe_len;
541 ++chan->num_subframes;
542 if (num_samples[c] > s->samples_per_frame) {
543 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
544 "channel len(%d) > samples_per_frame(%d)\n",
545 num_samples[c], s->samples_per_frame);
546 return AVERROR_INVALIDDATA;
548 } else if (num_samples[c] <= min_channel_len) {
549 if (num_samples[c] < min_channel_len) {
550 channels_for_cur_subframe = 0;
551 min_channel_len = num_samples[c];
553 ++channels_for_cur_subframe;
556 } while (min_channel_len < s->samples_per_frame);
558 for (c = 0; c < s->num_channels; c++) {
561 for (i = 0; i < s->channel[c].num_subframes; i++) {
562 s->channel[c].subframe_offset[i] = offset;
563 offset += s->channel[c].subframe_len[i];
571 static int my_log2(unsigned int i)
573 unsigned int iLog2 = 0;
574 while ((i >> iLog2) > 1)
583 static void decode_ac_filter(WmallDecodeCtx *s)
586 s->acfilter_order = get_bits(&s->gb, 4) + 1;
587 s->acfilter_scaling = get_bits(&s->gb, 4);
589 for(i = 0; i < s->acfilter_order; i++) {
590 s->acfilter_coeffs[i] = get_bits(&s->gb, s->acfilter_scaling) + 1;
598 static void decode_mclms(WmallDecodeCtx *s)
600 s->mclms_order = (get_bits(&s->gb, 4) + 1) * 2;
601 s->mclms_scaling = get_bits(&s->gb, 4);
602 if(get_bits1(&s->gb)) {
606 int cbits = av_log2(s->mclms_scaling + 1);
607 assert(cbits == my_log2(s->mclms_scaling + 1));
608 if(1 << cbits < s->mclms_scaling + 1)
611 send_coef_bits = (cbits ? get_bits(&s->gb, cbits) : 0) + 2;
613 for(i = 0; i < s->mclms_order * s->num_channels * s->num_channels; i++) {
614 s->mclms_coeffs[i] = get_bits(&s->gb, send_coef_bits);
617 for(i = 0; i < s->num_channels; i++) {
619 for(c = 0; c < i; c++) {
620 s->mclms_coeffs_cur[i * s->num_channels + c] = get_bits(&s->gb, send_coef_bits);
630 static void decode_cdlms(WmallDecodeCtx *s)
633 int cdlms_send_coef = get_bits1(&s->gb);
635 for(c = 0; c < s->num_channels; c++) {
636 s->cdlms_ttl[c] = get_bits(&s->gb, 3) + 1;
637 for(i = 0; i < s->cdlms_ttl[c]; i++) {
638 s->cdlms[c][i].order = (get_bits(&s->gb, 7) + 1) * 8;
641 for(i = 0; i < s->cdlms_ttl[c]; i++) {
642 s->cdlms[c][i].scaling = get_bits(&s->gb, 4);
645 if(cdlms_send_coef) {
646 for(i = 0; i < s->cdlms_ttl[c]; i++) {
647 int cbits, shift_l, shift_r, j;
648 cbits = av_log2(s->cdlms[c][i].order);
649 if(1 << cbits < s->cdlms[c][i].order)
651 s->cdlms[c][i].coefsend = get_bits(&s->gb, cbits) + 1;
653 cbits = av_log2(s->cdlms[c][i].scaling + 1);
654 if(1 << cbits < s->cdlms[c][i].scaling + 1)
657 s->cdlms[c][i].bitsend = get_bits(&s->gb, cbits) + 2;
658 shift_l = 32 - s->cdlms[c][i].bitsend;
659 shift_r = 32 - 2 - s->cdlms[c][i].scaling;
660 for(j = 0; j < s->cdlms[c][i].coefsend; j++) {
661 s->cdlms[c][i].coefs[j] =
662 (get_bits(&s->gb, s->cdlms[c][i].bitsend) << shift_l) >> shift_r;
672 static int decode_channel_residues(WmallDecodeCtx *s, int ch, int tile_size)
675 unsigned int ave_mean;
676 s->transient[ch] = get_bits1(&s->gb);
677 if(s->transient[ch]) {
678 s->transient_pos[ch] = get_bits(&s->gb, av_log2(tile_size));
679 if (s->transient_pos[ch])
680 s->transient[ch] = 0;
681 s->channel[ch].transient_counter =
682 FFMAX(s->channel[ch].transient_counter, s->samples_per_frame / 2);
683 } else if (s->channel[ch].transient_counter)
684 s->transient[ch] = 1;
686 if(s->seekable_tile) {
687 ave_mean = get_bits(&s->gb, s->bits_per_sample);
688 s->ave_sum[ch] = ave_mean << (s->movave_scaling + 1);
689 // s->ave_sum[ch] *= 2;
692 if(s->seekable_tile) {
693 if(s->do_inter_ch_decorr)
694 s->channel_residues[ch][0] = get_sbits(&s->gb, s->bits_per_sample + 1);
696 s->channel_residues[ch][0] = get_sbits(&s->gb, s->bits_per_sample);
699 //av_log(0, 0, "%8d: ", num_logged_tiles++);
700 for(; i < tile_size; i++) {
701 int quo = 0, rem, rem_bits, residue;
702 while(get_bits1(&s->gb))
705 quo += get_bits_long(&s->gb, get_bits(&s->gb, 5) + 1);
707 ave_mean = (s->ave_sum[ch] + (1 << s->movave_scaling)) >> (s->movave_scaling + 1);
708 rem_bits = av_ceil_log2(ave_mean);
709 rem = rem_bits ? get_bits(&s->gb, rem_bits) : 0;
710 residue = (quo << rem_bits) + rem;
712 s->ave_sum[ch] = residue + s->ave_sum[ch] - (s->ave_sum[ch] >> s->movave_scaling);
715 residue = -(residue >> 1) - 1;
717 residue = residue >> 1;
718 s->channel_residues[ch][i] = residue;
720 /*if (num_logged_tiles < 1)
721 av_log(0, 0, "%4d ", residue); */
723 dump_int_buffer(s->channel_residues[ch], 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(int));
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;
790 static int lms_predict(WmallDecodeCtx *s, int ich, int ilms)
792 int32_t pred = 0, icoef;
793 int recent = s->cdlms[ich][ilms].recent;
795 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
796 pred += s->cdlms[ich][ilms].coefs[icoef] *
797 s->cdlms[ich][ilms].lms_prevvalues[icoef + recent];
799 pred += (1 << (s->cdlms[ich][ilms].scaling - 1));
800 /* XXX: Table 29 has:
801 iPred >= cdlms[iCh][ilms].scaling;
802 seems to me like a missing > */
803 pred >>= s->cdlms[ich][ilms].scaling;
807 static void lms_update(WmallDecodeCtx *s, int ich, int ilms, int32_t input, int32_t pred)
810 int recent = s->cdlms[ich][ilms].recent;
811 int range = 1 << (s->bits_per_sample - 1);
812 int bps = s->bits_per_sample > 16 ? 4 : 2; // bytes per sample
815 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
816 s->cdlms[ich][ilms].coefs[icoef] +=
817 s->cdlms[ich][ilms].lms_updates[icoef + recent];
819 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
820 s->cdlms[ich][ilms].coefs[icoef] -=
821 s->cdlms[ich][ilms].lms_updates[icoef]; // XXX: [icoef + recent] ?
823 s->cdlms[ich][ilms].recent--;
824 s->cdlms[ich][ilms].lms_prevvalues[recent] = av_clip(input, -range, range - 1);
827 s->cdlms[ich][ilms].lms_updates[recent] = s->update_speed[ich];
828 else if (input < pred)
829 s->cdlms[ich][ilms].lms_updates[recent] = -s->update_speed[ich];
832 cdlms[iCh][ilms].updates[iRecent + cdlms[iCh][ilms].order >> 4] >>= 2;
833 lms_updates[iCh][ilms][iRecent + cdlms[iCh][ilms].order >> 3] >>= 1;
835 Questions is - are cdlms[iCh][ilms].updates[] and lms_updates[][][] two
836 seperate buffers? Here I've assumed that the two are same which makes
839 s->cdlms[ich][ilms].lms_updates[recent + s->cdlms[ich][ilms].order >> 4] >>= 2;
840 s->cdlms[ich][ilms].lms_updates[recent + s->cdlms[ich][ilms].order >> 3] >>= 1;
841 /* XXX: recent + (s->cdlms[ich][ilms].order >> 4) ? */
843 if (s->cdlms[ich][ilms].recent == 0) {
844 /* XXX: This memcpy()s will probably fail if a fixed 32-bit buffer is used.
845 follow kshishkov's suggestion of using a union. */
846 memcpy(s->cdlms[ich][ilms].lms_prevvalues + s->cdlms[ich][ilms].order,
847 s->cdlms[ich][ilms].lms_prevvalues,
848 bps * s->cdlms[ich][ilms].order);
849 memcpy(s->cdlms[ich][ilms].lms_updates + s->cdlms[ich][ilms].order,
850 s->cdlms[ich][ilms].lms_updates,
851 bps * s->cdlms[ich][ilms].order);
852 s->cdlms[ich][ilms].recent = s->cdlms[ich][ilms].order;
856 static void use_high_update_speed(WmallDecodeCtx *s, int ich)
858 int ilms, recent, icoef;
859 s->update_speed[ich] = 16;
860 for (ilms = s->cdlms_ttl[ich]; ilms >= 0; ilms--) {
861 recent = s->cdlms[ich][ilms].recent;
863 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
864 s->cdlms[ich][ilms].lms_updates[icoef + recent] *= 2;
866 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
867 s->cdlms[ich][ilms].lms_updates[icoef] *= 2;
872 static void use_normal_update_speed(WmallDecodeCtx *s, int ich)
874 int ilms, recent, icoef;
875 s->update_speed[ich] = 8;
876 for (ilms = s->cdlms_ttl[ich]; ilms >= 0; ilms--) {
877 recent = s->cdlms[ich][ilms].recent;
879 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
880 s->cdlms[ich][ilms].lms_updates[icoef + recent] /= 2;
882 for (icoef = 0; icoef < s->cdlms[ich][ilms].order; icoef++)
883 s->cdlms[ich][ilms].lms_updates[icoef] /= 2;
888 static void revert_cdlms(WmallDecodeCtx *s, int tile_size)
891 int32_t pred, channel_coeff;
894 for (ich = 0; ich < s->num_channels; ich++) {
895 if (!s->is_channel_coded[ich])
897 for (icoef = 0; icoef < tile_size; icoef++) {
898 num_lms = s->cdlms_ttl[ich];
899 channel_coeff = s->channel_residues[ich][icoef];
900 if (icoef == s->transient_pos[ich]) {
901 s->transient[ich] = 1;
902 use_high_update_speed(s, ich);
904 for (ilms = num_lms; ilms >= 0; ilms--) {
905 pred = lms_predict(s, ich, ilms);
906 channel_coeff += pred;
907 lms_update(s, ich, ilms, channel_coeff, pred);
909 if (s->transient[ich]) {
910 --s->channel[ich].transient_counter;
911 if(!s->channel[ich].transient_counter)
912 use_normal_update_speed(s, ich);
914 s->channel_coeffs[ich][icoef] = channel_coeff;
922 *@brief Decode a single subframe (block).
923 *@param s codec context
924 *@return 0 on success, < 0 when decoding failed
926 static int decode_subframe(WmallDecodeCtx *s)
928 int offset = s->samples_per_frame;
929 int subframe_len = s->samples_per_frame;
931 int total_samples = s->samples_per_frame * s->num_channels;
935 s->subframe_offset = get_bits_count(&s->gb);
937 /** reset channel context and find the next block offset and size
938 == the next block of the channel with the smallest number of
941 for (i = 0; i < s->num_channels; i++) {
942 s->channel[i].grouped = 0;
943 if (offset > s->channel[i].decoded_samples) {
944 offset = s->channel[i].decoded_samples;
946 s->channel[i].subframe_len[s->channel[i].cur_subframe];
950 /** get a list of all channels that contain the estimated block */
951 s->channels_for_cur_subframe = 0;
952 for (i = 0; i < s->num_channels; i++) {
953 const int cur_subframe = s->channel[i].cur_subframe;
954 /** substract already processed samples */
955 total_samples -= s->channel[i].decoded_samples;
957 /** and count if there are multiple subframes that match our profile */
958 if (offset == s->channel[i].decoded_samples &&
959 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
960 total_samples -= s->channel[i].subframe_len[cur_subframe];
961 s->channel[i].decoded_samples +=
962 s->channel[i].subframe_len[cur_subframe];
963 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
964 ++s->channels_for_cur_subframe;
968 /** check if the frame will be complete after processing the
971 s->parsed_all_subframes = 1;
974 s->seekable_tile = get_bits1(&s->gb);
975 if(s->seekable_tile) {
976 clear_codec_buffers(s);
978 s->do_arith_coding = get_bits1(&s->gb);
979 if(s->do_arith_coding) {
980 dprintf(s->avctx, "do_arith_coding == 1");
983 s->do_ac_filter = get_bits1(&s->gb);
984 s->do_inter_ch_decorr = get_bits1(&s->gb);
985 s->do_mclms = get_bits1(&s->gb);
994 s->movave_scaling = get_bits(&s->gb, 3);
995 s->quant_stepsize = get_bits(&s->gb, 8) + 1;
1000 rawpcm_tile = get_bits1(&s->gb);
1002 for(i = 0; i < s->num_channels; i++) {
1003 s->is_channel_coded[i] = 1;
1008 for(i = 0; i < s->num_channels; i++) {
1009 s->is_channel_coded[i] = get_bits1(&s->gb);
1014 s->do_lpc = get_bits1(&s->gb);
1024 if(get_bits1(&s->gb)) {
1025 padding_zeroes = get_bits(&s->gb, 5);
1032 int bits = s->bits_per_sample - padding_zeroes;
1034 dprintf(s->avctx, "RAWPCM %d bits per sample. total %d bits, remain=%d\n", bits,
1035 bits * s->num_channels * subframe_len, get_bits_count(&s->gb));
1036 for(i = 0; i < s->num_channels; i++) {
1037 for(j = 0; j < subframe_len; j++) {
1038 s->channel_coeffs[i][j] = get_sbits(&s->gb, bits);
1039 // dprintf(s->avctx, "PCM[%d][%d] = 0x%04x\n", i, j, s->channel_coeffs[i][j]);
1043 for(i = 0; i < s->num_channels; i++)
1044 if(s->is_channel_coded[i])
1045 decode_channel_residues(s, i, subframe_len);
1048 /** handled one subframe */
1050 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1051 int c = s->channel_indexes_for_cur_subframe[i];
1052 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1053 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1054 return AVERROR_INVALIDDATA;
1056 ++s->channel[c].cur_subframe;
1062 *@brief Decode one WMA frame.
1063 *@param s codec context
1064 *@return 0 if the trailer bit indicates that this is the last frame,
1065 * 1 if there are additional frames
1067 static int decode_frame(WmallDecodeCtx *s)
1069 GetBitContext* gb = &s->gb;
1070 int more_frames = 0;
1074 /** check for potential output buffer overflow */
1075 if (s->num_channels * s->samples_per_frame > s->samples_end - s->samples) {
1076 /** return an error if no frame could be decoded at all */
1077 av_log(s->avctx, AV_LOG_ERROR,
1078 "not enough space for the output samples\n");
1083 /** get frame length */
1085 len = get_bits(gb, s->log2_frame_size);
1087 /** decode tile information */
1088 if (decode_tilehdr(s)) {
1093 /** read drc info */
1094 if (s->dynamic_range_compression) {
1095 s->drc_gain = get_bits(gb, 8);
1098 /** no idea what these are for, might be the number of samples
1099 that need to be skipped at the beginning or end of a stream */
1100 if (get_bits1(gb)) {
1103 /** usually true for the first frame */
1104 if (get_bits1(gb)) {
1105 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1106 dprintf(s->avctx, "start skip: %i\n", skip);
1109 /** sometimes true for the last frame */
1110 if (get_bits1(gb)) {
1111 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1112 dprintf(s->avctx, "end skip: %i\n", skip);
1117 /** reset subframe states */
1118 s->parsed_all_subframes = 0;
1119 for (i = 0; i < s->num_channels; i++) {
1120 s->channel[i].decoded_samples = 0;
1121 s->channel[i].cur_subframe = 0;
1122 s->channel[i].reuse_sf = 0;
1125 /** decode all subframes */
1126 while (!s->parsed_all_subframes) {
1127 if (decode_subframe(s) < 0) {
1133 dprintf(s->avctx, "Frame done\n");
1135 if (s->skip_frame) {
1138 s->samples += s->num_channels * s->samples_per_frame;
1140 if (s->len_prefix) {
1141 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1142 /** FIXME: not sure if this is always an error */
1143 av_log(s->avctx, AV_LOG_ERROR,
1144 "frame[%i] would have to skip %i bits\n", s->frame_num,
1145 len - (get_bits_count(gb) - s->frame_offset) - 1);
1150 /** skip the rest of the frame data */
1151 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1154 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1155 dprintf(s->avctx, "skip1\n");
1160 /** decode trailer bit */
1161 more_frames = get_bits1(gb);
1167 *@brief Calculate remaining input buffer length.
1168 *@param s codec context
1169 *@param gb bitstream reader context
1170 *@return remaining size in bits
1172 static int remaining_bits(WmallDecodeCtx *s, GetBitContext *gb)
1174 return s->buf_bit_size - get_bits_count(gb);
1178 *@brief Fill the bit reservoir with a (partial) frame.
1179 *@param s codec context
1180 *@param gb bitstream reader context
1181 *@param len length of the partial frame
1182 *@param append decides wether to reset the buffer or not
1184 static void save_bits(WmallDecodeCtx *s, GetBitContext* gb, int len,
1189 /** when the frame data does not need to be concatenated, the input buffer
1190 is resetted and additional bits from the previous frame are copyed
1191 and skipped later so that a fast byte copy is possible */
1194 s->frame_offset = get_bits_count(gb) & 7;
1195 s->num_saved_bits = s->frame_offset;
1196 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1199 buflen = (s->num_saved_bits + len + 8) >> 3;
1201 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1202 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1207 s->num_saved_bits += len;
1209 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1212 int align = 8 - (get_bits_count(gb) & 7);
1213 align = FFMIN(align, len);
1214 put_bits(&s->pb, align, get_bits(gb, align));
1216 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1218 skip_bits_long(gb, len);
1221 PutBitContext tmp = s->pb;
1222 flush_put_bits(&tmp);
1225 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1226 skip_bits(&s->gb, s->frame_offset);
1230 *@brief Decode a single WMA packet.
1231 *@param avctx codec context
1232 *@param data the output buffer
1233 *@param data_size number of bytes that were written to the output buffer
1234 *@param avpkt input packet
1235 *@return number of bytes that were read from the input buffer
1237 static int decode_packet(AVCodecContext *avctx,
1238 void *data, int *data_size, AVPacket* avpkt)
1240 WmallDecodeCtx *s = avctx->priv_data;
1241 GetBitContext* gb = &s->pgb;
1242 const uint8_t* buf = avpkt->data;
1243 int buf_size = avpkt->size;
1244 int num_bits_prev_frame;
1245 int packet_sequence_number;
1248 s->samples_end = (float*)((int8_t*)data + *data_size);
1251 if (s->packet_done || s->packet_loss) {
1254 /** sanity check for the buffer length */
1255 if (buf_size < avctx->block_align)
1258 s->next_packet_start = buf_size - avctx->block_align;
1259 buf_size = avctx->block_align;
1260 s->buf_bit_size = buf_size << 3;
1262 /** parse packet header */
1263 init_get_bits(gb, buf, s->buf_bit_size);
1264 packet_sequence_number = get_bits(gb, 4);
1265 int seekable_frame_in_packet = get_bits1(gb);
1266 int spliced_packet = get_bits1(gb);
1268 /** get number of bits that need to be added to the previous frame */
1269 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1271 /** check for packet loss */
1272 if (!s->packet_loss &&
1273 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1275 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1276 s->packet_sequence_number, packet_sequence_number);
1278 s->packet_sequence_number = packet_sequence_number;
1280 if (num_bits_prev_frame > 0) {
1281 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1282 if (num_bits_prev_frame >= remaining_packet_bits) {
1283 num_bits_prev_frame = remaining_packet_bits;
1287 /** append the previous frame data to the remaining data from the
1288 previous packet to create a full frame */
1289 save_bits(s, gb, num_bits_prev_frame, 1);
1291 /** decode the cross packet frame if it is valid */
1292 if (!s->packet_loss)
1294 } else if (s->num_saved_bits - s->frame_offset) {
1295 dprintf(avctx, "ignoring %x previously saved bits\n",
1296 s->num_saved_bits - s->frame_offset);
1299 if (s->packet_loss) {
1300 /** reset number of saved bits so that the decoder
1301 does not start to decode incomplete frames in the
1302 s->len_prefix == 0 case */
1303 s->num_saved_bits = 0;
1310 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1311 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1312 skip_bits(gb, s->packet_offset);
1314 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1315 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1316 frame_size <= remaining_bits(s, gb)) {
1317 save_bits(s, gb, frame_size, 0);
1318 s->packet_done = !decode_frame(s);
1319 } else if (!s->len_prefix
1320 && s->num_saved_bits > get_bits_count(&s->gb)) {
1321 /** when the frames do not have a length prefix, we don't know
1322 the compressed length of the individual frames
1323 however, we know what part of a new packet belongs to the
1325 therefore we save the incoming packet first, then we append
1326 the "previous frame" data from the next packet so that
1327 we get a buffer that only contains full frames */
1328 s->packet_done = !decode_frame(s);
1334 if (s->packet_done && !s->packet_loss &&
1335 remaining_bits(s, gb) > 0) {
1336 /** save the rest of the data so that it can be decoded
1337 with the next packet */
1338 save_bits(s, gb, remaining_bits(s, gb), 0);
1341 *data_size = 0; // (int8_t *)s->samples - (int8_t *)data;
1342 s->packet_offset = get_bits_count(gb) & 7;
1344 return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3;
1348 *@brief Clear decoder buffers (for seeking).
1349 *@param avctx codec context
1351 static void flush(AVCodecContext *avctx)
1353 WmallDecodeCtx *s = avctx->priv_data;
1355 /** reset output buffer as a part of it is used during the windowing of a
1357 for (i = 0; i < s->num_channels; i++)
1358 memset(s->channel[i].out, 0, s->samples_per_frame *
1359 sizeof(*s->channel[i].out));
1365 *@brief wmall decoder
1367 AVCodec ff_wmalossless_decoder = {
1370 CODEC_ID_WMALOSSLESS,
1371 sizeof(WmallDecodeCtx),
1376 .capabilities = CODEC_CAP_SUBFRAMES,
1378 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Lossless"),