2 * Wmapro compatible decoder
3 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson
6 * This file is part of Libav.
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * @brief wmapro decoder implementation
26 * Wmapro is an MDCT based codec comparable to wma standard or AAC.
27 * The decoding therefore consists of the following steps:
28 * - bitstream decoding
29 * - reconstruction of per-channel data
30 * - rescaling and inverse quantization
32 * - windowing and overlapp-add
34 * The compressed wmapro bitstream is split into individual packets.
35 * Every such packet contains one or more wma frames.
36 * The compressed frames may have a variable length and frames may
37 * cross packet boundaries.
38 * Common to all wmapro frames is the number of samples that are stored in
40 * The number of samples and a few other decode flags are stored
41 * as extradata that has to be passed to the decoder.
43 * The wmapro frames themselves are again split into a variable number of
44 * subframes. Every subframe contains the data for 2^N time domain samples
45 * where N varies between 7 and 12.
47 * Example wmapro bitstream (in samples):
49 * || packet 0 || packet 1 || packet 2 packets
50 * ---------------------------------------------------
51 * || frame 0 || frame 1 || frame 2 || frames
52 * ---------------------------------------------------
53 * || | | || | | | || || subframes of channel 0
54 * ---------------------------------------------------
55 * || | | || | | | || || subframes of channel 1
56 * ---------------------------------------------------
58 * The frame layouts for the individual channels of a wma frame does not need
61 * However, if the offsets and lengths of several subframes of a frame are the
62 * same, the subframes of the channels can be grouped.
63 * Every group may then use special coding techniques like M/S stereo coding
64 * to improve the compression ratio. These channel transformations do not
65 * need to be applied to a whole subframe. Instead, they can also work on
66 * individual scale factor bands (see below).
67 * The coefficients that carry the audio signal in the frequency domain
68 * are transmitted as huffman-coded vectors with 4, 2 and 1 elements.
69 * In addition to that, the encoder can switch to a runlevel coding scheme
70 * by transmitting subframe_length / 128 zero coefficients.
72 * Before the audio signal can be converted to the time domain, the
73 * coefficients have to be rescaled and inverse quantized.
74 * A subframe is therefore split into several scale factor bands that get
75 * scaled individually.
76 * Scale factors are submitted for every frame but they might be shared
77 * between the subframes of a channel. Scale factors are initially DPCM-coded.
78 * Once scale factors are shared, the differences are transmitted as runlevel
80 * Every subframe length and offset combination in the frame layout shares a
81 * common quantization factor that can be adjusted for every channel by a
83 * After the inverse quantization, the coefficients get processed by an IMDCT.
84 * The resulting values are then windowed with a sine window and the first half
85 * of the values are added to the second half of the output from the previous
86 * subframe in order to reconstruct the output samples.
89 #include "libavutil/float_dsp.h"
90 #include "libavutil/intfloat.h"
91 #include "libavutil/intreadwrite.h"
96 #include "wmaprodata.h"
100 #include "wma_common.h"
102 /** current decoder limitations */
103 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
104 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
105 #define MAX_BANDS 29 ///< max number of scale factor bands
106 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
108 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
109 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
110 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
111 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
115 #define SCALEVLCBITS 8
116 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
117 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
118 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
119 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
120 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
122 static VLC sf_vlc; ///< scale factor DPCM vlc
123 static VLC sf_rl_vlc; ///< scale factor run length vlc
124 static VLC vec4_vlc; ///< 4 coefficients per symbol
125 static VLC vec2_vlc; ///< 2 coefficients per symbol
126 static VLC vec1_vlc; ///< 1 coefficient per symbol
127 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
128 static float sin64[33]; ///< sinus table for decorrelation
131 * @brief frame specific decoder context for a single channel
134 int16_t prev_block_len; ///< length of the previous block
135 uint8_t transmit_coefs;
136 uint8_t num_subframes;
137 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
138 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
139 uint8_t cur_subframe; ///< current subframe number
140 uint16_t decoded_samples; ///< number of already processed samples
141 uint8_t grouped; ///< channel is part of a group
142 int quant_step; ///< quantization step for the current subframe
143 int8_t reuse_sf; ///< share scale factors between subframes
144 int8_t scale_factor_step; ///< scaling step for the current subframe
145 int max_scale_factor; ///< maximum scale factor for the current subframe
146 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
147 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
148 int* scale_factors; ///< pointer to the scale factor values used for decoding
149 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
150 float* coeffs; ///< pointer to the subframe decode buffer
151 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
152 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
156 * @brief channel group for channel transformations
159 uint8_t num_channels; ///< number of channels in the group
160 int8_t transform; ///< transform on / off
161 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
162 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
163 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
167 * @brief main decoder context
169 typedef struct WMAProDecodeCtx {
170 /* generic decoder variables */
171 AVCodecContext* avctx; ///< codec context for av_log
172 AVFloatDSPContext fdsp;
173 uint8_t frame_data[MAX_FRAMESIZE +
174 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
175 PutBitContext pb; ///< context for filling the frame_data buffer
176 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
177 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
178 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
180 /* frame size dependent frame information (set during initialization) */
181 uint32_t decode_flags; ///< used compression features
182 uint8_t len_prefix; ///< frame is prefixed with its length
183 uint8_t dynamic_range_compression; ///< frame contains DRC data
184 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
185 uint16_t samples_per_frame; ///< number of samples to output
186 uint16_t log2_frame_size;
187 int8_t lfe_channel; ///< lfe channel index
188 uint8_t max_num_subframes;
189 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
190 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
191 uint16_t min_samples_per_subframe;
192 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
193 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
194 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
195 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
197 /* packet decode state */
198 GetBitContext pgb; ///< bitstream reader context for the packet
199 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
200 uint8_t packet_offset; ///< frame offset in the packet
201 uint8_t packet_sequence_number; ///< current packet number
202 int num_saved_bits; ///< saved number of bits
203 int frame_offset; ///< frame offset in the bit reservoir
204 int subframe_offset; ///< subframe offset in the bit reservoir
205 uint8_t packet_loss; ///< set in case of bitstream error
206 uint8_t packet_done; ///< set when a packet is fully decoded
208 /* frame decode state */
209 uint32_t frame_num; ///< current frame number (not used for decoding)
210 GetBitContext gb; ///< bitstream reader context
211 int buf_bit_size; ///< buffer size in bits
212 uint8_t drc_gain; ///< gain for the DRC tool
213 int8_t skip_frame; ///< skip output step
214 int8_t parsed_all_subframes; ///< all subframes decoded?
216 /* subframe/block decode state */
217 int16_t subframe_len; ///< current subframe length
218 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
219 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
220 int8_t num_bands; ///< number of scale factor bands
221 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
222 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
223 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
224 int8_t esc_len; ///< length of escaped coefficients
226 uint8_t num_chgroups; ///< number of channel groups
227 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
229 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
234 *@brief helper function to print the most important members of the context
237 static av_cold void dump_context(WMAProDecodeCtx *s)
239 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
240 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
242 PRINT("ed sample bit depth", s->bits_per_sample);
243 PRINT_HEX("ed decode flags", s->decode_flags);
244 PRINT("samples per frame", s->samples_per_frame);
245 PRINT("log2 frame size", s->log2_frame_size);
246 PRINT("max num subframes", s->max_num_subframes);
247 PRINT("len prefix", s->len_prefix);
248 PRINT("num channels", s->avctx->channels);
252 *@brief Uninitialize the decoder and free all resources.
253 *@param avctx codec context
254 *@return 0 on success, < 0 otherwise
256 static av_cold int decode_end(AVCodecContext *avctx)
258 WMAProDecodeCtx *s = avctx->priv_data;
261 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
262 ff_mdct_end(&s->mdct_ctx[i]);
268 *@brief Initialize the decoder.
269 *@param avctx codec context
270 *@return 0 on success, -1 otherwise
272 static av_cold int decode_init(AVCodecContext *avctx)
274 WMAProDecodeCtx *s = avctx->priv_data;
275 uint8_t *edata_ptr = avctx->extradata;
276 unsigned int channel_mask;
278 int log2_max_num_subframes;
279 int num_possible_block_sizes;
282 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
284 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
286 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
288 if (avctx->extradata_size >= 18) {
289 s->decode_flags = AV_RL16(edata_ptr+14);
290 channel_mask = AV_RL32(edata_ptr+2);
291 s->bits_per_sample = AV_RL16(edata_ptr);
292 /** dump the extradata */
293 for (i = 0; i < avctx->extradata_size; i++)
294 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
295 av_dlog(avctx, "\n");
298 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
299 return AVERROR_PATCHWELCOME;
303 s->log2_frame_size = av_log2(avctx->block_align) + 4;
306 s->skip_frame = 1; /* skip first frame */
308 s->len_prefix = (s->decode_flags & 0x40);
311 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
312 if (bits > WMAPRO_BLOCK_MAX_BITS) {
313 av_log_missing_feature(avctx, "14-bits block sizes", 1);
314 return AVERROR_PATCHWELCOME;
316 s->samples_per_frame = 1 << bits;
319 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
320 s->max_num_subframes = 1 << log2_max_num_subframes;
321 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
322 s->max_subframe_len_bit = 1;
323 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
325 num_possible_block_sizes = log2_max_num_subframes + 1;
326 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
327 s->dynamic_range_compression = (s->decode_flags & 0x80);
329 if (s->max_num_subframes > MAX_SUBFRAMES) {
330 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
331 s->max_num_subframes);
332 return AVERROR_INVALIDDATA;
335 if (s->avctx->sample_rate <= 0) {
336 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
337 return AVERROR_INVALIDDATA;
340 if (avctx->channels < 0) {
341 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
343 return AVERROR_INVALIDDATA;
344 } else if (avctx->channels > WMAPRO_MAX_CHANNELS) {
345 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
346 return AVERROR_PATCHWELCOME;
349 /** init previous block len */
350 for (i = 0; i < avctx->channels; i++)
351 s->channel[i].prev_block_len = s->samples_per_frame;
353 /** extract lfe channel position */
356 if (channel_mask & 8) {
358 for (mask = 1; mask < 16; mask <<= 1) {
359 if (channel_mask & mask)
364 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
365 scale_huffbits, 1, 1,
366 scale_huffcodes, 2, 2, 616);
368 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
369 scale_rl_huffbits, 1, 1,
370 scale_rl_huffcodes, 4, 4, 1406);
372 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
373 coef0_huffbits, 1, 1,
374 coef0_huffcodes, 4, 4, 2108);
376 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
377 coef1_huffbits, 1, 1,
378 coef1_huffcodes, 4, 4, 3912);
380 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
382 vec4_huffcodes, 2, 2, 604);
384 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
386 vec2_huffcodes, 2, 2, 562);
388 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
390 vec1_huffcodes, 2, 2, 562);
392 /** calculate number of scale factor bands and their offsets
393 for every possible block size */
394 for (i = 0; i < num_possible_block_sizes; i++) {
395 int subframe_len = s->samples_per_frame >> i;
399 s->sfb_offsets[i][0] = 0;
401 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
402 int offset = (subframe_len * 2 * critical_freq[x])
403 / s->avctx->sample_rate + 2;
405 if (offset > s->sfb_offsets[i][band - 1])
406 s->sfb_offsets[i][band++] = offset;
408 s->sfb_offsets[i][band - 1] = subframe_len;
409 s->num_sfb[i] = band - 1;
413 /** Scale factors can be shared between blocks of different size
414 as every block has a different scale factor band layout.
415 The matrix sf_offsets is needed to find the correct scale factor.
418 for (i = 0; i < num_possible_block_sizes; i++) {
420 for (b = 0; b < s->num_sfb[i]; b++) {
422 int offset = ((s->sfb_offsets[i][b]
423 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
424 for (x = 0; x < num_possible_block_sizes; x++) {
426 while (s->sfb_offsets[x][v + 1] << x < offset)
428 s->sf_offsets[i][x][b] = v;
433 /** init MDCT, FIXME: only init needed sizes */
434 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
435 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
436 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
437 / (1 << (s->bits_per_sample - 1)));
439 /** init MDCT windows: simple sinus window */
440 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
441 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
442 ff_init_ff_sine_windows(win_idx);
443 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
446 /** calculate subwoofer cutoff values */
447 for (i = 0; i < num_possible_block_sizes; i++) {
448 int block_size = s->samples_per_frame >> i;
449 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
450 / s->avctx->sample_rate;
451 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
454 /** calculate sine values for the decorrelation matrix */
455 for (i = 0; i < 33; i++)
456 sin64[i] = sin(i*M_PI / 64.0);
458 if (avctx->debug & FF_DEBUG_BITSTREAM)
461 avctx->channel_layout = channel_mask;
467 *@brief Decode the subframe length.
469 *@param offset sample offset in the frame
470 *@return decoded subframe length on success, < 0 in case of an error
472 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
474 int frame_len_shift = 0;
477 /** no need to read from the bitstream when only one length is possible */
478 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
479 return s->min_samples_per_subframe;
481 /** 1 bit indicates if the subframe is of maximum length */
482 if (s->max_subframe_len_bit) {
483 if (get_bits1(&s->gb))
484 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
486 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
488 subframe_len = s->samples_per_frame >> frame_len_shift;
490 /** sanity check the length */
491 if (subframe_len < s->min_samples_per_subframe ||
492 subframe_len > s->samples_per_frame) {
493 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
495 return AVERROR_INVALIDDATA;
501 *@brief Decode how the data in the frame is split into subframes.
502 * Every WMA frame contains the encoded data for a fixed number of
503 * samples per channel. The data for every channel might be split
504 * into several subframes. This function will reconstruct the list of
505 * subframes for every channel.
507 * If the subframes are not evenly split, the algorithm estimates the
508 * channels with the lowest number of total samples.
509 * Afterwards, for each of these channels a bit is read from the
510 * bitstream that indicates if the channel contains a subframe with the
511 * next subframe size that is going to be read from the bitstream or not.
512 * If a channel contains such a subframe, the subframe size gets added to
513 * the channel's subframe list.
514 * The algorithm repeats these steps until the frame is properly divided
515 * between the individual channels.
518 *@return 0 on success, < 0 in case of an error
520 static int decode_tilehdr(WMAProDecodeCtx *s)
522 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
523 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
524 int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */
525 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
526 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
529 /* Should never consume more than 3073 bits (256 iterations for the
530 * while loop when always the minimum amount of 128 samples is subtracted
531 * from missing samples in the 8 channel case).
532 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
535 /** reset tiling information */
536 for (c = 0; c < s->avctx->channels; c++)
537 s->channel[c].num_subframes = 0;
539 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
540 fixed_channel_layout = 1;
542 /** loop until the frame data is split between the subframes */
546 /** check which channels contain the subframe */
547 for (c = 0; c < s->avctx->channels; c++) {
548 if (num_samples[c] == min_channel_len) {
549 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
550 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
551 contains_subframe[c] = 1;
553 contains_subframe[c] = get_bits1(&s->gb);
555 contains_subframe[c] = 0;
558 /** get subframe length, subframe_len == 0 is not allowed */
559 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
560 return AVERROR_INVALIDDATA;
562 /** add subframes to the individual channels and find new min_channel_len */
563 min_channel_len += subframe_len;
564 for (c = 0; c < s->avctx->channels; c++) {
565 WMAProChannelCtx* chan = &s->channel[c];
567 if (contains_subframe[c]) {
568 if (chan->num_subframes >= MAX_SUBFRAMES) {
569 av_log(s->avctx, AV_LOG_ERROR,
570 "broken frame: num subframes > 31\n");
571 return AVERROR_INVALIDDATA;
573 chan->subframe_len[chan->num_subframes] = subframe_len;
574 num_samples[c] += subframe_len;
575 ++chan->num_subframes;
576 if (num_samples[c] > s->samples_per_frame) {
577 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
578 "channel len > samples_per_frame\n");
579 return AVERROR_INVALIDDATA;
581 } else if (num_samples[c] <= min_channel_len) {
582 if (num_samples[c] < min_channel_len) {
583 channels_for_cur_subframe = 0;
584 min_channel_len = num_samples[c];
586 ++channels_for_cur_subframe;
589 } while (min_channel_len < s->samples_per_frame);
591 for (c = 0; c < s->avctx->channels; c++) {
594 for (i = 0; i < s->channel[c].num_subframes; i++) {
595 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
596 " len %i\n", s->frame_num, c, i,
597 s->channel[c].subframe_len[i]);
598 s->channel[c].subframe_offset[i] = offset;
599 offset += s->channel[c].subframe_len[i];
607 *@brief Calculate a decorrelation matrix from the bitstream parameters.
608 *@param s codec context
609 *@param chgroup channel group for which the matrix needs to be calculated
611 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
612 WMAProChannelGrp *chgroup)
616 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
617 memset(chgroup->decorrelation_matrix, 0, s->avctx->channels *
618 s->avctx->channels * sizeof(*chgroup->decorrelation_matrix));
620 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
621 rotation_offset[i] = get_bits(&s->gb, 6);
623 for (i = 0; i < chgroup->num_channels; i++)
624 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
625 get_bits1(&s->gb) ? 1.0 : -1.0;
627 for (i = 1; i < chgroup->num_channels; i++) {
629 for (x = 0; x < i; x++) {
631 for (y = 0; y < i + 1; y++) {
632 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
633 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
634 int n = rotation_offset[offset + x];
640 cosv = sin64[32 - n];
642 sinv = sin64[64 - n];
643 cosv = -sin64[n - 32];
646 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
647 (v1 * sinv) - (v2 * cosv);
648 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
649 (v1 * cosv) + (v2 * sinv);
657 *@brief Decode channel transformation parameters
658 *@param s codec context
659 *@return 0 in case of success, < 0 in case of bitstream errors
661 static int decode_channel_transform(WMAProDecodeCtx* s)
664 /* should never consume more than 1921 bits for the 8 channel case
665 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
666 * + MAX_CHANNELS + MAX_BANDS + 1)
669 /** in the one channel case channel transforms are pointless */
671 if (s->avctx->channels > 1) {
672 int remaining_channels = s->channels_for_cur_subframe;
674 if (get_bits1(&s->gb)) {
675 av_log_ask_for_sample(s->avctx,
676 "unsupported channel transform bit\n");
677 return AVERROR_PATCHWELCOME;
680 for (s->num_chgroups = 0; remaining_channels &&
681 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
682 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
683 float** channel_data = chgroup->channel_data;
684 chgroup->num_channels = 0;
685 chgroup->transform = 0;
687 /** decode channel mask */
688 if (remaining_channels > 2) {
689 for (i = 0; i < s->channels_for_cur_subframe; i++) {
690 int channel_idx = s->channel_indexes_for_cur_subframe[i];
691 if (!s->channel[channel_idx].grouped
692 && get_bits1(&s->gb)) {
693 ++chgroup->num_channels;
694 s->channel[channel_idx].grouped = 1;
695 *channel_data++ = s->channel[channel_idx].coeffs;
699 chgroup->num_channels = remaining_channels;
700 for (i = 0; i < s->channels_for_cur_subframe; i++) {
701 int channel_idx = s->channel_indexes_for_cur_subframe[i];
702 if (!s->channel[channel_idx].grouped)
703 *channel_data++ = s->channel[channel_idx].coeffs;
704 s->channel[channel_idx].grouped = 1;
708 /** decode transform type */
709 if (chgroup->num_channels == 2) {
710 if (get_bits1(&s->gb)) {
711 if (get_bits1(&s->gb)) {
712 av_log_ask_for_sample(s->avctx,
713 "unsupported channel transform type\n");
716 chgroup->transform = 1;
717 if (s->avctx->channels == 2) {
718 chgroup->decorrelation_matrix[0] = 1.0;
719 chgroup->decorrelation_matrix[1] = -1.0;
720 chgroup->decorrelation_matrix[2] = 1.0;
721 chgroup->decorrelation_matrix[3] = 1.0;
724 chgroup->decorrelation_matrix[0] = 0.70703125;
725 chgroup->decorrelation_matrix[1] = -0.70703125;
726 chgroup->decorrelation_matrix[2] = 0.70703125;
727 chgroup->decorrelation_matrix[3] = 0.70703125;
730 } else if (chgroup->num_channels > 2) {
731 if (get_bits1(&s->gb)) {
732 chgroup->transform = 1;
733 if (get_bits1(&s->gb)) {
734 decode_decorrelation_matrix(s, chgroup);
736 /** FIXME: more than 6 coupled channels not supported */
737 if (chgroup->num_channels > 6) {
738 av_log_ask_for_sample(s->avctx,
739 "coupled channels > 6\n");
741 memcpy(chgroup->decorrelation_matrix,
742 default_decorrelation[chgroup->num_channels],
743 chgroup->num_channels * chgroup->num_channels *
744 sizeof(*chgroup->decorrelation_matrix));
750 /** decode transform on / off */
751 if (chgroup->transform) {
752 if (!get_bits1(&s->gb)) {
754 /** transform can be enabled for individual bands */
755 for (i = 0; i < s->num_bands; i++) {
756 chgroup->transform_band[i] = get_bits1(&s->gb);
759 memset(chgroup->transform_band, 1, s->num_bands);
762 remaining_channels -= chgroup->num_channels;
769 *@brief Extract the coefficients from the bitstream.
770 *@param s codec context
771 *@param c current channel number
772 *@return 0 on success, < 0 in case of bitstream errors
774 static int decode_coeffs(WMAProDecodeCtx *s, int c)
776 /* Integers 0..15 as single-precision floats. The table saves a
777 costly int to float conversion, and storing the values as
778 integers allows fast sign-flipping. */
779 static const uint32_t fval_tab[16] = {
780 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
781 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
782 0x41000000, 0x41100000, 0x41200000, 0x41300000,
783 0x41400000, 0x41500000, 0x41600000, 0x41700000,
787 WMAProChannelCtx* ci = &s->channel[c];
794 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
796 vlctable = get_bits1(&s->gb);
797 vlc = &coef_vlc[vlctable];
807 /** decode vector coefficients (consumes up to 167 bits per iteration for
808 4 vector coded large values) */
809 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
810 (cur_coeff + 3 < ci->num_vec_coeffs)) {
815 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
817 if (idx == HUFF_VEC4_SIZE - 1) {
818 for (i = 0; i < 4; i += 2) {
819 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
820 if (idx == HUFF_VEC2_SIZE - 1) {
822 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
823 if (v0 == HUFF_VEC1_SIZE - 1)
824 v0 += ff_wma_get_large_val(&s->gb);
825 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
826 if (v1 == HUFF_VEC1_SIZE - 1)
827 v1 += ff_wma_get_large_val(&s->gb);
828 vals[i ] = av_float2int(v0);
829 vals[i+1] = av_float2int(v1);
831 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
832 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
836 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
837 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
838 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
839 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
843 for (i = 0; i < 4; i++) {
845 uint32_t sign = get_bits1(&s->gb) - 1;
846 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
849 ci->coeffs[cur_coeff] = 0;
850 /** switch to run level mode when subframe_len / 128 zeros
851 were found in a row */
852 rl_mode |= (++num_zeros > s->subframe_len >> 8);
858 /** decode run level coded coefficients */
859 if (cur_coeff < s->subframe_len) {
860 memset(&ci->coeffs[cur_coeff], 0,
861 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
862 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
863 level, run, 1, ci->coeffs,
864 cur_coeff, s->subframe_len,
865 s->subframe_len, s->esc_len, 0))
866 return AVERROR_INVALIDDATA;
873 *@brief Extract scale factors from the bitstream.
874 *@param s codec context
875 *@return 0 on success, < 0 in case of bitstream errors
877 static int decode_scale_factors(WMAProDecodeCtx* s)
881 /** should never consume more than 5344 bits
882 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
885 for (i = 0; i < s->channels_for_cur_subframe; i++) {
886 int c = s->channel_indexes_for_cur_subframe[i];
889 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
890 sf_end = s->channel[c].scale_factors + s->num_bands;
892 /** resample scale factors for the new block size
893 * as the scale factors might need to be resampled several times
894 * before some new values are transmitted, a backup of the last
895 * transmitted scale factors is kept in saved_scale_factors
897 if (s->channel[c].reuse_sf) {
898 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
900 for (b = 0; b < s->num_bands; b++)
901 s->channel[c].scale_factors[b] =
902 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
905 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
907 if (!s->channel[c].reuse_sf) {
909 /** decode DPCM coded scale factors */
910 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
911 val = 45 / s->channel[c].scale_factor_step;
912 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
913 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
918 /** run level decode differences to the resampled factors */
919 for (i = 0; i < s->num_bands; i++) {
925 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
928 uint32_t code = get_bits(&s->gb, 14);
930 sign = (code & 1) - 1;
931 skip = (code & 0x3f) >> 1;
932 } else if (idx == 1) {
935 skip = scale_rl_run[idx];
936 val = scale_rl_level[idx];
937 sign = get_bits1(&s->gb)-1;
941 if (i >= s->num_bands) {
942 av_log(s->avctx, AV_LOG_ERROR,
943 "invalid scale factor coding\n");
944 return AVERROR_INVALIDDATA;
946 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
950 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
951 s->channel[c].table_idx = s->table_idx;
952 s->channel[c].reuse_sf = 1;
955 /** calculate new scale factor maximum */
956 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
957 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
958 s->channel[c].max_scale_factor =
959 FFMAX(s->channel[c].max_scale_factor, *sf);
967 *@brief Reconstruct the individual channel data.
968 *@param s codec context
970 static void inverse_channel_transform(WMAProDecodeCtx *s)
974 for (i = 0; i < s->num_chgroups; i++) {
975 if (s->chgroup[i].transform) {
976 float data[WMAPRO_MAX_CHANNELS];
977 const int num_channels = s->chgroup[i].num_channels;
978 float** ch_data = s->chgroup[i].channel_data;
979 float** ch_end = ch_data + num_channels;
980 const int8_t* tb = s->chgroup[i].transform_band;
983 /** multichannel decorrelation */
984 for (sfb = s->cur_sfb_offsets;
985 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
988 /** multiply values with the decorrelation_matrix */
989 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
990 const float* mat = s->chgroup[i].decorrelation_matrix;
991 const float* data_end = data + num_channels;
992 float* data_ptr = data;
995 for (ch = ch_data; ch < ch_end; ch++)
996 *data_ptr++ = (*ch)[y];
998 for (ch = ch_data; ch < ch_end; ch++) {
1001 while (data_ptr < data_end)
1002 sum += *data_ptr++ * *mat++;
1007 } else if (s->avctx->channels == 2) {
1008 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1009 s->fdsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1010 ch_data[0] + sfb[0],
1012 s->fdsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1013 ch_data[1] + sfb[0],
1022 *@brief Apply sine window and reconstruct the output buffer.
1023 *@param s codec context
1025 static void wmapro_window(WMAProDecodeCtx *s)
1028 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1029 int c = s->channel_indexes_for_cur_subframe[i];
1031 int winlen = s->channel[c].prev_block_len;
1032 float* start = s->channel[c].coeffs - (winlen >> 1);
1034 if (s->subframe_len < winlen) {
1035 start += (winlen - s->subframe_len) >> 1;
1036 winlen = s->subframe_len;
1039 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1043 s->fdsp.vector_fmul_window(start, start, start + winlen,
1046 s->channel[c].prev_block_len = s->subframe_len;
1051 *@brief Decode a single subframe (block).
1052 *@param s codec context
1053 *@return 0 on success, < 0 when decoding failed
1055 static int decode_subframe(WMAProDecodeCtx *s)
1057 int offset = s->samples_per_frame;
1058 int subframe_len = s->samples_per_frame;
1060 int total_samples = s->samples_per_frame * s->avctx->channels;
1061 int transmit_coeffs = 0;
1062 int cur_subwoofer_cutoff;
1064 s->subframe_offset = get_bits_count(&s->gb);
1066 /** reset channel context and find the next block offset and size
1067 == the next block of the channel with the smallest number of
1070 for (i = 0; i < s->avctx->channels; i++) {
1071 s->channel[i].grouped = 0;
1072 if (offset > s->channel[i].decoded_samples) {
1073 offset = s->channel[i].decoded_samples;
1075 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1080 "processing subframe with offset %i len %i\n", offset, subframe_len);
1082 /** get a list of all channels that contain the estimated block */
1083 s->channels_for_cur_subframe = 0;
1084 for (i = 0; i < s->avctx->channels; i++) {
1085 const int cur_subframe = s->channel[i].cur_subframe;
1086 /** subtract already processed samples */
1087 total_samples -= s->channel[i].decoded_samples;
1089 /** and count if there are multiple subframes that match our profile */
1090 if (offset == s->channel[i].decoded_samples &&
1091 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1092 total_samples -= s->channel[i].subframe_len[cur_subframe];
1093 s->channel[i].decoded_samples +=
1094 s->channel[i].subframe_len[cur_subframe];
1095 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1096 ++s->channels_for_cur_subframe;
1100 /** check if the frame will be complete after processing the
1103 s->parsed_all_subframes = 1;
1106 av_dlog(s->avctx, "subframe is part of %i channels\n",
1107 s->channels_for_cur_subframe);
1109 /** calculate number of scale factor bands and their offsets */
1110 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1111 s->num_bands = s->num_sfb[s->table_idx];
1112 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1113 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1115 /** configure the decoder for the current subframe */
1116 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1117 int c = s->channel_indexes_for_cur_subframe[i];
1119 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1123 s->subframe_len = subframe_len;
1124 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1126 /** skip extended header if any */
1127 if (get_bits1(&s->gb)) {
1129 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1130 int len = get_bits(&s->gb, 4);
1131 num_fill_bits = get_bits(&s->gb, len) + 1;
1134 if (num_fill_bits >= 0) {
1135 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1136 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1137 return AVERROR_INVALIDDATA;
1140 skip_bits_long(&s->gb, num_fill_bits);
1144 /** no idea for what the following bit is used */
1145 if (get_bits1(&s->gb)) {
1146 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1147 return AVERROR_PATCHWELCOME;
1151 if (decode_channel_transform(s) < 0)
1152 return AVERROR_INVALIDDATA;
1155 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1156 int c = s->channel_indexes_for_cur_subframe[i];
1157 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1158 transmit_coeffs = 1;
1161 if (transmit_coeffs) {
1163 int quant_step = 90 * s->bits_per_sample >> 4;
1165 /** decode number of vector coded coefficients */
1166 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1167 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1168 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1169 int c = s->channel_indexes_for_cur_subframe[i];
1170 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1171 if (num_vec_coeffs > WMAPRO_BLOCK_MAX_SIZE) {
1172 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1173 return AVERROR_INVALIDDATA;
1175 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1178 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1179 int c = s->channel_indexes_for_cur_subframe[i];
1180 s->channel[c].num_vec_coeffs = s->subframe_len;
1183 /** decode quantization step */
1184 step = get_sbits(&s->gb, 6);
1186 if (step == -32 || step == 31) {
1187 const int sign = (step == 31) - 1;
1189 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1190 (step = get_bits(&s->gb, 5)) == 31) {
1193 quant_step += ((quant + step) ^ sign) - sign;
1195 if (quant_step < 0) {
1196 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1199 /** decode quantization step modifiers for every channel */
1201 if (s->channels_for_cur_subframe == 1) {
1202 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1204 int modifier_len = get_bits(&s->gb, 3);
1205 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1206 int c = s->channel_indexes_for_cur_subframe[i];
1207 s->channel[c].quant_step = quant_step;
1208 if (get_bits1(&s->gb)) {
1210 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1212 ++s->channel[c].quant_step;
1217 /** decode scale factors */
1218 if (decode_scale_factors(s) < 0)
1219 return AVERROR_INVALIDDATA;
1222 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1223 get_bits_count(&s->gb) - s->subframe_offset);
1225 /** parse coefficients */
1226 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1227 int c = s->channel_indexes_for_cur_subframe[i];
1228 if (s->channel[c].transmit_coefs &&
1229 get_bits_count(&s->gb) < s->num_saved_bits) {
1230 decode_coeffs(s, c);
1232 memset(s->channel[c].coeffs, 0,
1233 sizeof(*s->channel[c].coeffs) * subframe_len);
1236 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1237 get_bits_count(&s->gb) - s->subframe_offset);
1239 if (transmit_coeffs) {
1240 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1241 /** reconstruct the per channel data */
1242 inverse_channel_transform(s);
1243 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1244 int c = s->channel_indexes_for_cur_subframe[i];
1245 const int* sf = s->channel[c].scale_factors;
1248 if (c == s->lfe_channel)
1249 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1250 (subframe_len - cur_subwoofer_cutoff));
1252 /** inverse quantization and rescaling */
1253 for (b = 0; b < s->num_bands; b++) {
1254 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1255 const int exp = s->channel[c].quant_step -
1256 (s->channel[c].max_scale_factor - *sf++) *
1257 s->channel[c].scale_factor_step;
1258 const float quant = pow(10.0, exp / 20.0);
1259 int start = s->cur_sfb_offsets[b];
1260 s->fdsp.vector_fmul_scalar(s->tmp + start,
1261 s->channel[c].coeffs + start,
1262 quant, end - start);
1265 /** apply imdct (imdct_half == DCTIV with reverse) */
1266 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1270 /** window and overlapp-add */
1273 /** handled one subframe */
1274 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1275 int c = s->channel_indexes_for_cur_subframe[i];
1276 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1277 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1278 return AVERROR_INVALIDDATA;
1280 ++s->channel[c].cur_subframe;
1287 *@brief Decode one WMA frame.
1288 *@param s codec context
1289 *@return 0 if the trailer bit indicates that this is the last frame,
1290 * 1 if there are additional frames
1292 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1294 AVCodecContext *avctx = s->avctx;
1295 GetBitContext* gb = &s->gb;
1296 int more_frames = 0;
1300 /** get frame length */
1302 len = get_bits(gb, s->log2_frame_size);
1304 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1306 /** decode tile information */
1307 if (decode_tilehdr(s)) {
1312 /** read postproc transform */
1313 if (s->avctx->channels > 1 && get_bits1(gb)) {
1314 if (get_bits1(gb)) {
1315 for (i = 0; i < avctx->channels * avctx->channels; i++)
1320 /** read drc info */
1321 if (s->dynamic_range_compression) {
1322 s->drc_gain = get_bits(gb, 8);
1323 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1326 /** no idea what these are for, might be the number of samples
1327 that need to be skipped at the beginning or end of a stream */
1328 if (get_bits1(gb)) {
1331 /** usually true for the first frame */
1332 if (get_bits1(gb)) {
1333 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1334 av_dlog(s->avctx, "start skip: %i\n", skip);
1337 /** sometimes true for the last frame */
1338 if (get_bits1(gb)) {
1339 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1340 av_dlog(s->avctx, "end skip: %i\n", skip);
1345 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1346 get_bits_count(gb) - s->frame_offset);
1348 /** reset subframe states */
1349 s->parsed_all_subframes = 0;
1350 for (i = 0; i < avctx->channels; i++) {
1351 s->channel[i].decoded_samples = 0;
1352 s->channel[i].cur_subframe = 0;
1353 s->channel[i].reuse_sf = 0;
1356 /** decode all subframes */
1357 while (!s->parsed_all_subframes) {
1358 if (decode_subframe(s) < 0) {
1364 /* get output buffer */
1365 frame->nb_samples = s->samples_per_frame;
1366 if ((ret = ff_get_buffer(avctx, frame)) < 0) {
1367 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1372 /** copy samples to the output buffer */
1373 for (i = 0; i < avctx->channels; i++)
1374 memcpy(frame->extended_data[i], s->channel[i].out,
1375 s->samples_per_frame * sizeof(*s->channel[i].out));
1377 for (i = 0; i < avctx->channels; i++) {
1378 /** reuse second half of the IMDCT output for the next frame */
1379 memcpy(&s->channel[i].out[0],
1380 &s->channel[i].out[s->samples_per_frame],
1381 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1384 if (s->skip_frame) {
1391 if (s->len_prefix) {
1392 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1393 /** FIXME: not sure if this is always an error */
1394 av_log(s->avctx, AV_LOG_ERROR,
1395 "frame[%i] would have to skip %i bits\n", s->frame_num,
1396 len - (get_bits_count(gb) - s->frame_offset) - 1);
1401 /** skip the rest of the frame data */
1402 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1404 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1408 /** decode trailer bit */
1409 more_frames = get_bits1(gb);
1416 *@brief Calculate remaining input buffer length.
1417 *@param s codec context
1418 *@param gb bitstream reader context
1419 *@return remaining size in bits
1421 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1423 return s->buf_bit_size - get_bits_count(gb);
1427 *@brief Fill the bit reservoir with a (partial) frame.
1428 *@param s codec context
1429 *@param gb bitstream reader context
1430 *@param len length of the partial frame
1431 *@param append decides whether to reset the buffer or not
1433 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1438 /** when the frame data does not need to be concatenated, the input buffer
1439 is resetted and additional bits from the previous frame are copyed
1440 and skipped later so that a fast byte copy is possible */
1443 s->frame_offset = get_bits_count(gb) & 7;
1444 s->num_saved_bits = s->frame_offset;
1445 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1448 buflen = (s->num_saved_bits + len + 8) >> 3;
1450 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1451 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1456 s->num_saved_bits += len;
1458 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1461 int align = 8 - (get_bits_count(gb) & 7);
1462 align = FFMIN(align, len);
1463 put_bits(&s->pb, align, get_bits(gb, align));
1465 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1467 skip_bits_long(gb, len);
1470 PutBitContext tmp = s->pb;
1471 flush_put_bits(&tmp);
1474 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1475 skip_bits(&s->gb, s->frame_offset);
1479 *@brief Decode a single WMA packet.
1480 *@param avctx codec context
1481 *@param data the output buffer
1482 *@param avpkt input packet
1483 *@return number of bytes that were read from the input buffer
1485 static int decode_packet(AVCodecContext *avctx, void *data,
1486 int *got_frame_ptr, AVPacket* avpkt)
1488 WMAProDecodeCtx *s = avctx->priv_data;
1489 GetBitContext* gb = &s->pgb;
1490 const uint8_t* buf = avpkt->data;
1491 int buf_size = avpkt->size;
1492 int num_bits_prev_frame;
1493 int packet_sequence_number;
1497 if (s->packet_done || s->packet_loss) {
1500 /** sanity check for the buffer length */
1501 if (buf_size < avctx->block_align)
1504 s->next_packet_start = buf_size - avctx->block_align;
1505 buf_size = avctx->block_align;
1506 s->buf_bit_size = buf_size << 3;
1508 /** parse packet header */
1509 init_get_bits(gb, buf, s->buf_bit_size);
1510 packet_sequence_number = get_bits(gb, 4);
1513 /** get number of bits that need to be added to the previous frame */
1514 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1515 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1516 num_bits_prev_frame);
1518 /** check for packet loss */
1519 if (!s->packet_loss &&
1520 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1522 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1523 s->packet_sequence_number, packet_sequence_number);
1525 s->packet_sequence_number = packet_sequence_number;
1527 if (num_bits_prev_frame > 0) {
1528 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1529 if (num_bits_prev_frame >= remaining_packet_bits) {
1530 num_bits_prev_frame = remaining_packet_bits;
1534 /** append the previous frame data to the remaining data from the
1535 previous packet to create a full frame */
1536 save_bits(s, gb, num_bits_prev_frame, 1);
1537 av_dlog(avctx, "accumulated %x bits of frame data\n",
1538 s->num_saved_bits - s->frame_offset);
1540 /** decode the cross packet frame if it is valid */
1541 if (!s->packet_loss)
1542 decode_frame(s, data, got_frame_ptr);
1543 } else if (s->num_saved_bits - s->frame_offset) {
1544 av_dlog(avctx, "ignoring %x previously saved bits\n",
1545 s->num_saved_bits - s->frame_offset);
1548 if (s->packet_loss) {
1549 /** reset number of saved bits so that the decoder
1550 does not start to decode incomplete frames in the
1551 s->len_prefix == 0 case */
1552 s->num_saved_bits = 0;
1558 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1559 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1560 skip_bits(gb, s->packet_offset);
1561 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1562 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1563 frame_size <= remaining_bits(s, gb)) {
1564 save_bits(s, gb, frame_size, 0);
1565 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1566 } else if (!s->len_prefix
1567 && s->num_saved_bits > get_bits_count(&s->gb)) {
1568 /** when the frames do not have a length prefix, we don't know
1569 the compressed length of the individual frames
1570 however, we know what part of a new packet belongs to the
1572 therefore we save the incoming packet first, then we append
1573 the "previous frame" data from the next packet so that
1574 we get a buffer that only contains full frames */
1575 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1580 if (s->packet_done && !s->packet_loss &&
1581 remaining_bits(s, gb) > 0) {
1582 /** save the rest of the data so that it can be decoded
1583 with the next packet */
1584 save_bits(s, gb, remaining_bits(s, gb), 0);
1587 s->packet_offset = get_bits_count(gb) & 7;
1589 return AVERROR_INVALIDDATA;
1591 return get_bits_count(gb) >> 3;
1595 *@brief Clear decoder buffers (for seeking).
1596 *@param avctx codec context
1598 static void flush(AVCodecContext *avctx)
1600 WMAProDecodeCtx *s = avctx->priv_data;
1602 /** reset output buffer as a part of it is used during the windowing of a
1604 for (i = 0; i < avctx->channels; i++)
1605 memset(s->channel[i].out, 0, s->samples_per_frame *
1606 sizeof(*s->channel[i].out));
1612 *@brief wmapro decoder
1614 AVCodec ff_wmapro_decoder = {
1616 .type = AVMEDIA_TYPE_AUDIO,
1617 .id = AV_CODEC_ID_WMAPRO,
1618 .priv_data_size = sizeof(WMAProDecodeCtx),
1619 .init = decode_init,
1620 .close = decode_end,
1621 .decode = decode_packet,
1622 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1624 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1625 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1626 AV_SAMPLE_FMT_NONE },