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.
93 #include "wmaprodata.h"
98 /** current decoder limitations */
99 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
100 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
101 #define MAX_BANDS 29 ///< max number of scale factor bands
102 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
104 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
105 #define WMAPRO_BLOCK_MAX_BITS 12 ///< log2 of max block size
106 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
107 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
111 #define SCALEVLCBITS 8
112 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
113 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
114 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
115 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
116 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
118 static VLC sf_vlc; ///< scale factor DPCM vlc
119 static VLC sf_rl_vlc; ///< scale factor run length vlc
120 static VLC vec4_vlc; ///< 4 coefficients per symbol
121 static VLC vec2_vlc; ///< 2 coefficients per symbol
122 static VLC vec1_vlc; ///< 1 coefficient per symbol
123 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
124 static float sin64[33]; ///< sinus table for decorrelation
127 * @brief frame specific decoder context for a single channel
130 int16_t prev_block_len; ///< length of the previous block
131 uint8_t transmit_coefs;
132 uint8_t num_subframes;
133 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
134 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
135 uint8_t cur_subframe; ///< current subframe number
136 uint16_t decoded_samples; ///< number of already processed samples
137 uint8_t grouped; ///< channel is part of a group
138 int quant_step; ///< quantization step for the current subframe
139 int8_t reuse_sf; ///< share scale factors between subframes
140 int8_t scale_factor_step; ///< scaling step for the current subframe
141 int max_scale_factor; ///< maximum scale factor for the current subframe
142 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
143 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
144 int* scale_factors; ///< pointer to the scale factor values used for decoding
145 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
146 float* coeffs; ///< pointer to the subframe decode buffer
147 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
148 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
152 * @brief channel group for channel transformations
155 uint8_t num_channels; ///< number of channels in the group
156 int8_t transform; ///< transform on / off
157 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
158 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
159 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
163 * @brief main decoder context
165 typedef struct WMAProDecodeCtx {
166 /* generic decoder variables */
167 AVCodecContext* avctx; ///< codec context for av_log
168 DSPContext dsp; ///< accelerated DSP functions
169 uint8_t frame_data[MAX_FRAMESIZE +
170 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
171 PutBitContext pb; ///< context for filling the frame_data buffer
172 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
173 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
174 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
176 /* frame size dependent frame information (set during initialization) */
177 uint32_t decode_flags; ///< used compression features
178 uint8_t len_prefix; ///< frame is prefixed with its length
179 uint8_t dynamic_range_compression; ///< frame contains DRC data
180 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
181 uint16_t samples_per_frame; ///< number of samples to output
182 uint16_t log2_frame_size;
183 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
184 int8_t lfe_channel; ///< lfe channel index
185 uint8_t max_num_subframes;
186 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
187 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
188 uint16_t min_samples_per_subframe;
189 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
190 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
191 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
192 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
194 /* packet decode state */
195 GetBitContext pgb; ///< bitstream reader context for the packet
196 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
197 uint8_t packet_offset; ///< frame offset in the packet
198 uint8_t packet_sequence_number; ///< current packet number
199 int num_saved_bits; ///< saved number of bits
200 int frame_offset; ///< frame offset in the bit reservoir
201 int subframe_offset; ///< subframe offset in the bit reservoir
202 uint8_t packet_loss; ///< set in case of bitstream error
203 uint8_t packet_done; ///< set when a packet is fully decoded
205 /* frame decode state */
206 uint32_t frame_num; ///< current frame number (not used for decoding)
207 GetBitContext gb; ///< bitstream reader context
208 int buf_bit_size; ///< buffer size in bits
209 float* samples; ///< current samplebuffer pointer
210 float* samples_end; ///< maximum samplebuffer pointer
211 uint8_t drc_gain; ///< gain for the DRC tool
212 int8_t skip_frame; ///< skip output step
213 int8_t parsed_all_subframes; ///< all subframes decoded?
215 /* subframe/block decode state */
216 int16_t subframe_len; ///< current subframe length
217 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
218 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
219 int8_t num_bands; ///< number of scale factor bands
220 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
221 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
222 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
223 int8_t esc_len; ///< length of escaped coefficients
225 uint8_t num_chgroups; ///< number of channel groups
226 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
228 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
233 *@brief helper function to print the most important members of the context
236 static void av_cold dump_context(WMAProDecodeCtx *s)
238 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
239 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
241 PRINT("ed sample bit depth", s->bits_per_sample);
242 PRINT_HEX("ed decode flags", s->decode_flags);
243 PRINT("samples per frame", s->samples_per_frame);
244 PRINT("log2 frame size", s->log2_frame_size);
245 PRINT("max num subframes", s->max_num_subframes);
246 PRINT("len prefix", s->len_prefix);
247 PRINT("num channels", s->num_channels);
251 *@brief Uninitialize the decoder and free all resources.
252 *@param avctx codec context
253 *@return 0 on success, < 0 otherwise
255 static av_cold int decode_end(AVCodecContext *avctx)
257 WMAProDecodeCtx *s = avctx->priv_data;
260 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
261 ff_mdct_end(&s->mdct_ctx[i]);
267 *@brief Initialize the decoder.
268 *@param avctx codec context
269 *@return 0 on success, -1 otherwise
271 static av_cold int decode_init(AVCodecContext *avctx)
273 WMAProDecodeCtx *s = avctx->priv_data;
274 uint8_t *edata_ptr = avctx->extradata;
275 unsigned int channel_mask;
277 int log2_max_num_subframes;
278 int num_possible_block_sizes;
281 dsputil_init(&s->dsp, avctx);
282 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
284 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
286 if (avctx->extradata_size >= 18) {
287 s->decode_flags = AV_RL16(edata_ptr+14);
288 channel_mask = AV_RL32(edata_ptr+2);
289 s->bits_per_sample = AV_RL16(edata_ptr);
290 /** dump the extradata */
291 for (i = 0; i < avctx->extradata_size; i++)
292 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
293 av_dlog(avctx, "\n");
296 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
297 return AVERROR_INVALIDDATA;
301 s->log2_frame_size = av_log2(avctx->block_align) + 4;
304 s->skip_frame = 1; /* skip first frame */
306 s->len_prefix = (s->decode_flags & 0x40);
309 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
312 /** init previous block len */
313 for (i = 0; i < avctx->channels; i++)
314 s->channel[i].prev_block_len = s->samples_per_frame;
317 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
318 s->max_num_subframes = 1 << log2_max_num_subframes;
319 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
320 s->max_subframe_len_bit = 1;
321 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
323 num_possible_block_sizes = log2_max_num_subframes + 1;
324 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
325 s->dynamic_range_compression = (s->decode_flags & 0x80);
327 if (s->max_num_subframes > MAX_SUBFRAMES) {
328 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
329 s->max_num_subframes);
330 return AVERROR_INVALIDDATA;
333 s->num_channels = avctx->channels;
335 /** extract lfe channel position */
338 if (channel_mask & 8) {
340 for (mask = 1; mask < 16; mask <<= 1) {
341 if (channel_mask & mask)
346 if (s->num_channels < 0) {
347 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
348 return AVERROR_INVALIDDATA;
349 } else if (s->num_channels > WMAPRO_MAX_CHANNELS) {
350 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
351 return AVERROR_PATCHWELCOME;
354 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
355 scale_huffbits, 1, 1,
356 scale_huffcodes, 2, 2, 616);
358 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
359 scale_rl_huffbits, 1, 1,
360 scale_rl_huffcodes, 4, 4, 1406);
362 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
363 coef0_huffbits, 1, 1,
364 coef0_huffcodes, 4, 4, 2108);
366 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
367 coef1_huffbits, 1, 1,
368 coef1_huffcodes, 4, 4, 3912);
370 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
372 vec4_huffcodes, 2, 2, 604);
374 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
376 vec2_huffcodes, 2, 2, 562);
378 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
380 vec1_huffcodes, 2, 2, 562);
382 /** calculate number of scale factor bands and their offsets
383 for every possible block size */
384 for (i = 0; i < num_possible_block_sizes; i++) {
385 int subframe_len = s->samples_per_frame >> i;
389 s->sfb_offsets[i][0] = 0;
391 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
392 int offset = (subframe_len * 2 * critical_freq[x])
393 / s->avctx->sample_rate + 2;
395 if (offset > s->sfb_offsets[i][band - 1])
396 s->sfb_offsets[i][band++] = offset;
398 s->sfb_offsets[i][band - 1] = subframe_len;
399 s->num_sfb[i] = band - 1;
403 /** Scale factors can be shared between blocks of different size
404 as every block has a different scale factor band layout.
405 The matrix sf_offsets is needed to find the correct scale factor.
408 for (i = 0; i < num_possible_block_sizes; i++) {
410 for (b = 0; b < s->num_sfb[i]; b++) {
412 int offset = ((s->sfb_offsets[i][b]
413 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
414 for (x = 0; x < num_possible_block_sizes; x++) {
416 while (s->sfb_offsets[x][v + 1] << x < offset)
418 s->sf_offsets[i][x][b] = v;
423 /** init MDCT, FIXME: only init needed sizes */
424 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
425 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
426 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
427 / (1 << (s->bits_per_sample - 1)));
429 /** init MDCT windows: simple sinus window */
430 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
431 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
432 ff_init_ff_sine_windows(win_idx);
433 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
436 /** calculate subwoofer cutoff values */
437 for (i = 0; i < num_possible_block_sizes; i++) {
438 int block_size = s->samples_per_frame >> i;
439 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
440 / s->avctx->sample_rate;
441 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
444 /** calculate sine values for the decorrelation matrix */
445 for (i = 0; i < 33; i++)
446 sin64[i] = sin(i*M_PI / 64.0);
448 if (avctx->debug & FF_DEBUG_BITSTREAM)
451 avctx->channel_layout = channel_mask;
456 *@brief Decode the subframe length.
458 *@param offset sample offset in the frame
459 *@return decoded subframe length on success, < 0 in case of an error
461 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
463 int frame_len_shift = 0;
466 /** no need to read from the bitstream when only one length is possible */
467 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
468 return s->min_samples_per_subframe;
470 /** 1 bit indicates if the subframe is of maximum length */
471 if (s->max_subframe_len_bit) {
472 if (get_bits1(&s->gb))
473 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
475 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
477 subframe_len = s->samples_per_frame >> frame_len_shift;
479 /** sanity check the length */
480 if (subframe_len < s->min_samples_per_subframe ||
481 subframe_len > s->samples_per_frame) {
482 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
484 return AVERROR_INVALIDDATA;
490 *@brief Decode how the data in the frame is split into subframes.
491 * Every WMA frame contains the encoded data for a fixed number of
492 * samples per channel. The data for every channel might be split
493 * into several subframes. This function will reconstruct the list of
494 * subframes for every channel.
496 * If the subframes are not evenly split, the algorithm estimates the
497 * channels with the lowest number of total samples.
498 * Afterwards, for each of these channels a bit is read from the
499 * bitstream that indicates if the channel contains a subframe with the
500 * next subframe size that is going to be read from the bitstream or not.
501 * If a channel contains such a subframe, the subframe size gets added to
502 * the channel's subframe list.
503 * The algorithm repeats these steps until the frame is properly divided
504 * between the individual channels.
507 *@return 0 on success, < 0 in case of an error
509 static int decode_tilehdr(WMAProDecodeCtx *s)
511 uint16_t num_samples[WMAPRO_MAX_CHANNELS]; /**< sum of samples for all currently known subframes of a channel */
512 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
513 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
514 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
515 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
518 /* Should never consume more than 3073 bits (256 iterations for the
519 * while loop when always the minimum amount of 128 samples is substracted
520 * from missing samples in the 8 channel case).
521 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
524 /** reset tiling information */
525 for (c = 0; c < s->num_channels; c++)
526 s->channel[c].num_subframes = 0;
528 memset(num_samples, 0, sizeof(num_samples));
530 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
531 fixed_channel_layout = 1;
533 /** loop until the frame data is split between the subframes */
537 /** check which channels contain the subframe */
538 for (c = 0; c < s->num_channels; c++) {
539 if (num_samples[c] == min_channel_len) {
540 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
541 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
542 contains_subframe[c] = 1;
544 contains_subframe[c] = get_bits1(&s->gb);
546 contains_subframe[c] = 0;
549 /** get subframe length, subframe_len == 0 is not allowed */
550 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
551 return AVERROR_INVALIDDATA;
553 /** add subframes to the individual channels and find new min_channel_len */
554 min_channel_len += subframe_len;
555 for (c = 0; c < s->num_channels; c++) {
556 WMAProChannelCtx* chan = &s->channel[c];
558 if (contains_subframe[c]) {
559 if (chan->num_subframes >= MAX_SUBFRAMES) {
560 av_log(s->avctx, AV_LOG_ERROR,
561 "broken frame: num subframes > 31\n");
562 return AVERROR_INVALIDDATA;
564 chan->subframe_len[chan->num_subframes] = subframe_len;
565 num_samples[c] += subframe_len;
566 ++chan->num_subframes;
567 if (num_samples[c] > s->samples_per_frame) {
568 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
569 "channel len > samples_per_frame\n");
570 return AVERROR_INVALIDDATA;
572 } else if (num_samples[c] <= min_channel_len) {
573 if (num_samples[c] < min_channel_len) {
574 channels_for_cur_subframe = 0;
575 min_channel_len = num_samples[c];
577 ++channels_for_cur_subframe;
580 } while (min_channel_len < s->samples_per_frame);
582 for (c = 0; c < s->num_channels; c++) {
585 for (i = 0; i < s->channel[c].num_subframes; i++) {
586 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
587 " len %i\n", s->frame_num, c, i,
588 s->channel[c].subframe_len[i]);
589 s->channel[c].subframe_offset[i] = offset;
590 offset += s->channel[c].subframe_len[i];
598 *@brief Calculate a decorrelation matrix from the bitstream parameters.
599 *@param s codec context
600 *@param chgroup channel group for which the matrix needs to be calculated
602 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
603 WMAProChannelGrp *chgroup)
607 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
608 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
609 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
611 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
612 rotation_offset[i] = get_bits(&s->gb, 6);
614 for (i = 0; i < chgroup->num_channels; i++)
615 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
616 get_bits1(&s->gb) ? 1.0 : -1.0;
618 for (i = 1; i < chgroup->num_channels; i++) {
620 for (x = 0; x < i; x++) {
622 for (y = 0; y < i + 1; y++) {
623 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
624 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
625 int n = rotation_offset[offset + x];
631 cosv = sin64[32 - n];
633 sinv = sin64[64 - n];
634 cosv = -sin64[n - 32];
637 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
638 (v1 * sinv) - (v2 * cosv);
639 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
640 (v1 * cosv) + (v2 * sinv);
648 *@brief Decode channel transformation parameters
649 *@param s codec context
650 *@return 0 in case of success, < 0 in case of bitstream errors
652 static int decode_channel_transform(WMAProDecodeCtx* s)
655 /* should never consume more than 1921 bits for the 8 channel case
656 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
657 * + MAX_CHANNELS + MAX_BANDS + 1)
660 /** in the one channel case channel transforms are pointless */
662 if (s->num_channels > 1) {
663 int remaining_channels = s->channels_for_cur_subframe;
665 if (get_bits1(&s->gb)) {
666 av_log_ask_for_sample(s->avctx,
667 "unsupported channel transform bit\n");
668 return AVERROR_INVALIDDATA;
671 for (s->num_chgroups = 0; remaining_channels &&
672 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
673 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
674 float** channel_data = chgroup->channel_data;
675 chgroup->num_channels = 0;
676 chgroup->transform = 0;
678 /** decode channel mask */
679 if (remaining_channels > 2) {
680 for (i = 0; i < s->channels_for_cur_subframe; i++) {
681 int channel_idx = s->channel_indexes_for_cur_subframe[i];
682 if (!s->channel[channel_idx].grouped
683 && get_bits1(&s->gb)) {
684 ++chgroup->num_channels;
685 s->channel[channel_idx].grouped = 1;
686 *channel_data++ = s->channel[channel_idx].coeffs;
690 chgroup->num_channels = remaining_channels;
691 for (i = 0; i < s->channels_for_cur_subframe; i++) {
692 int channel_idx = s->channel_indexes_for_cur_subframe[i];
693 if (!s->channel[channel_idx].grouped)
694 *channel_data++ = s->channel[channel_idx].coeffs;
695 s->channel[channel_idx].grouped = 1;
699 /** decode transform type */
700 if (chgroup->num_channels == 2) {
701 if (get_bits1(&s->gb)) {
702 if (get_bits1(&s->gb)) {
703 av_log_ask_for_sample(s->avctx,
704 "unsupported channel transform type\n");
707 chgroup->transform = 1;
708 if (s->num_channels == 2) {
709 chgroup->decorrelation_matrix[0] = 1.0;
710 chgroup->decorrelation_matrix[1] = -1.0;
711 chgroup->decorrelation_matrix[2] = 1.0;
712 chgroup->decorrelation_matrix[3] = 1.0;
715 chgroup->decorrelation_matrix[0] = 0.70703125;
716 chgroup->decorrelation_matrix[1] = -0.70703125;
717 chgroup->decorrelation_matrix[2] = 0.70703125;
718 chgroup->decorrelation_matrix[3] = 0.70703125;
721 } else if (chgroup->num_channels > 2) {
722 if (get_bits1(&s->gb)) {
723 chgroup->transform = 1;
724 if (get_bits1(&s->gb)) {
725 decode_decorrelation_matrix(s, chgroup);
727 /** FIXME: more than 6 coupled channels not supported */
728 if (chgroup->num_channels > 6) {
729 av_log_ask_for_sample(s->avctx,
730 "coupled channels > 6\n");
732 memcpy(chgroup->decorrelation_matrix,
733 default_decorrelation[chgroup->num_channels],
734 chgroup->num_channels * chgroup->num_channels *
735 sizeof(*chgroup->decorrelation_matrix));
741 /** decode transform on / off */
742 if (chgroup->transform) {
743 if (!get_bits1(&s->gb)) {
745 /** transform can be enabled for individual bands */
746 for (i = 0; i < s->num_bands; i++) {
747 chgroup->transform_band[i] = get_bits1(&s->gb);
750 memset(chgroup->transform_band, 1, s->num_bands);
753 remaining_channels -= chgroup->num_channels;
760 *@brief Extract the coefficients from the bitstream.
761 *@param s codec context
762 *@param c current channel number
763 *@return 0 on success, < 0 in case of bitstream errors
765 static int decode_coeffs(WMAProDecodeCtx *s, int c)
767 /* Integers 0..15 as single-precision floats. The table saves a
768 costly int to float conversion, and storing the values as
769 integers allows fast sign-flipping. */
770 static const int fval_tab[16] = {
771 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
772 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
773 0x41000000, 0x41100000, 0x41200000, 0x41300000,
774 0x41400000, 0x41500000, 0x41600000, 0x41700000,
778 WMAProChannelCtx* ci = &s->channel[c];
785 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
787 vlctable = get_bits1(&s->gb);
788 vlc = &coef_vlc[vlctable];
798 /** decode vector coefficients (consumes up to 167 bits per iteration for
799 4 vector coded large values) */
800 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
801 (cur_coeff + 3 < ci->num_vec_coeffs)) {
806 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
808 if (idx == HUFF_VEC4_SIZE - 1) {
809 for (i = 0; i < 4; i += 2) {
810 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
811 if (idx == HUFF_VEC2_SIZE - 1) {
813 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
814 if (v0 == HUFF_VEC1_SIZE - 1)
815 v0 += ff_wma_get_large_val(&s->gb);
816 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
817 if (v1 == HUFF_VEC1_SIZE - 1)
818 v1 += ff_wma_get_large_val(&s->gb);
819 ((float*)vals)[i ] = v0;
820 ((float*)vals)[i+1] = v1;
822 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
823 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
827 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
828 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
829 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
830 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
834 for (i = 0; i < 4; i++) {
836 int sign = get_bits1(&s->gb) - 1;
837 *(uint32_t*)&ci->coeffs[cur_coeff] = vals[i] ^ sign<<31;
840 ci->coeffs[cur_coeff] = 0;
841 /** switch to run level mode when subframe_len / 128 zeros
842 were found in a row */
843 rl_mode |= (++num_zeros > s->subframe_len >> 8);
849 /** decode run level coded coefficients */
850 if (cur_coeff < s->subframe_len) {
851 memset(&ci->coeffs[cur_coeff], 0,
852 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
853 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
854 level, run, 1, ci->coeffs,
855 cur_coeff, s->subframe_len,
856 s->subframe_len, s->esc_len, 0))
857 return AVERROR_INVALIDDATA;
864 *@brief Extract scale factors from the bitstream.
865 *@param s codec context
866 *@return 0 on success, < 0 in case of bitstream errors
868 static int decode_scale_factors(WMAProDecodeCtx* s)
872 /** should never consume more than 5344 bits
873 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
876 for (i = 0; i < s->channels_for_cur_subframe; i++) {
877 int c = s->channel_indexes_for_cur_subframe[i];
880 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
881 sf_end = s->channel[c].scale_factors + s->num_bands;
883 /** resample scale factors for the new block size
884 * as the scale factors might need to be resampled several times
885 * before some new values are transmitted, a backup of the last
886 * transmitted scale factors is kept in saved_scale_factors
888 if (s->channel[c].reuse_sf) {
889 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
891 for (b = 0; b < s->num_bands; b++)
892 s->channel[c].scale_factors[b] =
893 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
896 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
898 if (!s->channel[c].reuse_sf) {
900 /** decode DPCM coded scale factors */
901 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
902 val = 45 / s->channel[c].scale_factor_step;
903 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
904 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
909 /** run level decode differences to the resampled factors */
910 for (i = 0; i < s->num_bands; i++) {
916 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
919 uint32_t code = get_bits(&s->gb, 14);
921 sign = (code & 1) - 1;
922 skip = (code & 0x3f) >> 1;
923 } else if (idx == 1) {
926 skip = scale_rl_run[idx];
927 val = scale_rl_level[idx];
928 sign = get_bits1(&s->gb)-1;
932 if (i >= s->num_bands) {
933 av_log(s->avctx, AV_LOG_ERROR,
934 "invalid scale factor coding\n");
935 return AVERROR_INVALIDDATA;
937 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
941 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
942 s->channel[c].table_idx = s->table_idx;
943 s->channel[c].reuse_sf = 1;
946 /** calculate new scale factor maximum */
947 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
948 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
949 s->channel[c].max_scale_factor =
950 FFMAX(s->channel[c].max_scale_factor, *sf);
958 *@brief Reconstruct the individual channel data.
959 *@param s codec context
961 static void inverse_channel_transform(WMAProDecodeCtx *s)
965 for (i = 0; i < s->num_chgroups; i++) {
966 if (s->chgroup[i].transform) {
967 float data[WMAPRO_MAX_CHANNELS];
968 const int num_channels = s->chgroup[i].num_channels;
969 float** ch_data = s->chgroup[i].channel_data;
970 float** ch_end = ch_data + num_channels;
971 const int8_t* tb = s->chgroup[i].transform_band;
974 /** multichannel decorrelation */
975 for (sfb = s->cur_sfb_offsets;
976 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
979 /** multiply values with the decorrelation_matrix */
980 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
981 const float* mat = s->chgroup[i].decorrelation_matrix;
982 const float* data_end = data + num_channels;
983 float* data_ptr = data;
986 for (ch = ch_data; ch < ch_end; ch++)
987 *data_ptr++ = (*ch)[y];
989 for (ch = ch_data; ch < ch_end; ch++) {
992 while (data_ptr < data_end)
993 sum += *data_ptr++ * *mat++;
998 } else if (s->num_channels == 2) {
999 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1000 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1001 ch_data[0] + sfb[0],
1003 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1004 ch_data[1] + sfb[0],
1013 *@brief Apply sine window and reconstruct the output buffer.
1014 *@param s codec context
1016 static void wmapro_window(WMAProDecodeCtx *s)
1019 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1020 int c = s->channel_indexes_for_cur_subframe[i];
1022 int winlen = s->channel[c].prev_block_len;
1023 float* start = s->channel[c].coeffs - (winlen >> 1);
1025 if (s->subframe_len < winlen) {
1026 start += (winlen - s->subframe_len) >> 1;
1027 winlen = s->subframe_len;
1030 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1034 s->dsp.vector_fmul_window(start, start, start + winlen,
1037 s->channel[c].prev_block_len = s->subframe_len;
1042 *@brief Decode a single subframe (block).
1043 *@param s codec context
1044 *@return 0 on success, < 0 when decoding failed
1046 static int decode_subframe(WMAProDecodeCtx *s)
1048 int offset = s->samples_per_frame;
1049 int subframe_len = s->samples_per_frame;
1051 int total_samples = s->samples_per_frame * s->num_channels;
1052 int transmit_coeffs = 0;
1053 int cur_subwoofer_cutoff;
1055 s->subframe_offset = get_bits_count(&s->gb);
1057 /** reset channel context and find the next block offset and size
1058 == the next block of the channel with the smallest number of
1061 for (i = 0; i < s->num_channels; i++) {
1062 s->channel[i].grouped = 0;
1063 if (offset > s->channel[i].decoded_samples) {
1064 offset = s->channel[i].decoded_samples;
1066 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1071 "processing subframe with offset %i len %i\n", offset, subframe_len);
1073 /** get a list of all channels that contain the estimated block */
1074 s->channels_for_cur_subframe = 0;
1075 for (i = 0; i < s->num_channels; i++) {
1076 const int cur_subframe = s->channel[i].cur_subframe;
1077 /** substract already processed samples */
1078 total_samples -= s->channel[i].decoded_samples;
1080 /** and count if there are multiple subframes that match our profile */
1081 if (offset == s->channel[i].decoded_samples &&
1082 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1083 total_samples -= s->channel[i].subframe_len[cur_subframe];
1084 s->channel[i].decoded_samples +=
1085 s->channel[i].subframe_len[cur_subframe];
1086 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1087 ++s->channels_for_cur_subframe;
1091 /** check if the frame will be complete after processing the
1094 s->parsed_all_subframes = 1;
1097 av_dlog(s->avctx, "subframe is part of %i channels\n",
1098 s->channels_for_cur_subframe);
1100 /** calculate number of scale factor bands and their offsets */
1101 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1102 s->num_bands = s->num_sfb[s->table_idx];
1103 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1104 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1106 /** configure the decoder for the current subframe */
1107 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1108 int c = s->channel_indexes_for_cur_subframe[i];
1110 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1114 s->subframe_len = subframe_len;
1115 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1117 /** skip extended header if any */
1118 if (get_bits1(&s->gb)) {
1120 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1121 int len = get_bits(&s->gb, 4);
1122 num_fill_bits = get_bits(&s->gb, len) + 1;
1125 if (num_fill_bits >= 0) {
1126 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1127 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1128 return AVERROR_INVALIDDATA;
1131 skip_bits_long(&s->gb, num_fill_bits);
1135 /** no idea for what the following bit is used */
1136 if (get_bits1(&s->gb)) {
1137 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1138 return AVERROR_INVALIDDATA;
1142 if (decode_channel_transform(s) < 0)
1143 return AVERROR_INVALIDDATA;
1146 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1147 int c = s->channel_indexes_for_cur_subframe[i];
1148 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1149 transmit_coeffs = 1;
1152 if (transmit_coeffs) {
1154 int quant_step = 90 * s->bits_per_sample >> 4;
1156 /** decode number of vector coded coefficients */
1157 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1158 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1159 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1160 int c = s->channel_indexes_for_cur_subframe[i];
1161 s->channel[c].num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1164 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1165 int c = s->channel_indexes_for_cur_subframe[i];
1166 s->channel[c].num_vec_coeffs = s->subframe_len;
1169 /** decode quantization step */
1170 step = get_sbits(&s->gb, 6);
1172 if (step == -32 || step == 31) {
1173 const int sign = (step == 31) - 1;
1175 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1176 (step = get_bits(&s->gb, 5)) == 31) {
1179 quant_step += ((quant + step) ^ sign) - sign;
1181 if (quant_step < 0) {
1182 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1185 /** decode quantization step modifiers for every channel */
1187 if (s->channels_for_cur_subframe == 1) {
1188 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1190 int modifier_len = get_bits(&s->gb, 3);
1191 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1192 int c = s->channel_indexes_for_cur_subframe[i];
1193 s->channel[c].quant_step = quant_step;
1194 if (get_bits1(&s->gb)) {
1196 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1198 ++s->channel[c].quant_step;
1203 /** decode scale factors */
1204 if (decode_scale_factors(s) < 0)
1205 return AVERROR_INVALIDDATA;
1208 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1209 get_bits_count(&s->gb) - s->subframe_offset);
1211 /** parse coefficients */
1212 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1213 int c = s->channel_indexes_for_cur_subframe[i];
1214 if (s->channel[c].transmit_coefs &&
1215 get_bits_count(&s->gb) < s->num_saved_bits) {
1216 decode_coeffs(s, c);
1218 memset(s->channel[c].coeffs, 0,
1219 sizeof(*s->channel[c].coeffs) * subframe_len);
1222 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1223 get_bits_count(&s->gb) - s->subframe_offset);
1225 if (transmit_coeffs) {
1226 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1227 /** reconstruct the per channel data */
1228 inverse_channel_transform(s);
1229 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1230 int c = s->channel_indexes_for_cur_subframe[i];
1231 const int* sf = s->channel[c].scale_factors;
1234 if (c == s->lfe_channel)
1235 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1236 (subframe_len - cur_subwoofer_cutoff));
1238 /** inverse quantization and rescaling */
1239 for (b = 0; b < s->num_bands; b++) {
1240 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1241 const int exp = s->channel[c].quant_step -
1242 (s->channel[c].max_scale_factor - *sf++) *
1243 s->channel[c].scale_factor_step;
1244 const float quant = pow(10.0, exp / 20.0);
1245 int start = s->cur_sfb_offsets[b];
1246 s->dsp.vector_fmul_scalar(s->tmp + start,
1247 s->channel[c].coeffs + start,
1248 quant, end - start);
1251 /** apply imdct (imdct_half == DCTIV with reverse) */
1252 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1256 /** window and overlapp-add */
1259 /** handled one subframe */
1260 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1261 int c = s->channel_indexes_for_cur_subframe[i];
1262 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1263 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1264 return AVERROR_INVALIDDATA;
1266 ++s->channel[c].cur_subframe;
1273 *@brief Decode one WMA frame.
1274 *@param s codec context
1275 *@return 0 if the trailer bit indicates that this is the last frame,
1276 * 1 if there are additional frames
1278 static int decode_frame(WMAProDecodeCtx *s)
1280 GetBitContext* gb = &s->gb;
1281 int more_frames = 0;
1285 /** check for potential output buffer overflow */
1286 if (s->num_channels * s->samples_per_frame > s->samples_end - s->samples) {
1287 /** return an error if no frame could be decoded at all */
1288 av_log(s->avctx, AV_LOG_ERROR,
1289 "not enough space for the output samples\n");
1294 /** get frame length */
1296 len = get_bits(gb, s->log2_frame_size);
1298 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1300 /** decode tile information */
1301 if (decode_tilehdr(s)) {
1306 /** read postproc transform */
1307 if (s->num_channels > 1 && get_bits1(gb)) {
1308 if (get_bits1(gb)) {
1309 for (i = 0; i < s->num_channels * s->num_channels; i++)
1314 /** read drc info */
1315 if (s->dynamic_range_compression) {
1316 s->drc_gain = get_bits(gb, 8);
1317 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1320 /** no idea what these are for, might be the number of samples
1321 that need to be skipped at the beginning or end of a stream */
1322 if (get_bits1(gb)) {
1325 /** usually true for the first frame */
1326 if (get_bits1(gb)) {
1327 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1328 av_dlog(s->avctx, "start skip: %i\n", skip);
1331 /** sometimes true for the last frame */
1332 if (get_bits1(gb)) {
1333 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1334 av_dlog(s->avctx, "end skip: %i\n", skip);
1339 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1340 get_bits_count(gb) - s->frame_offset);
1342 /** reset subframe states */
1343 s->parsed_all_subframes = 0;
1344 for (i = 0; i < s->num_channels; i++) {
1345 s->channel[i].decoded_samples = 0;
1346 s->channel[i].cur_subframe = 0;
1347 s->channel[i].reuse_sf = 0;
1350 /** decode all subframes */
1351 while (!s->parsed_all_subframes) {
1352 if (decode_subframe(s) < 0) {
1358 /** interleave samples and write them to the output buffer */
1359 for (i = 0; i < s->num_channels; i++) {
1360 float* ptr = s->samples + i;
1361 int incr = s->num_channels;
1362 float* iptr = s->channel[i].out;
1363 float* iend = iptr + s->samples_per_frame;
1365 // FIXME should create/use a DSP function here
1366 while (iptr < iend) {
1371 /** reuse second half of the IMDCT output for the next frame */
1372 memcpy(&s->channel[i].out[0],
1373 &s->channel[i].out[s->samples_per_frame],
1374 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1377 if (s->skip_frame) {
1380 s->samples += s->num_channels * s->samples_per_frame;
1382 if (s->len_prefix) {
1383 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1384 /** FIXME: not sure if this is always an error */
1385 av_log(s->avctx, AV_LOG_ERROR,
1386 "frame[%i] would have to skip %i bits\n", s->frame_num,
1387 len - (get_bits_count(gb) - s->frame_offset) - 1);
1392 /** skip the rest of the frame data */
1393 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1395 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1399 /** decode trailer bit */
1400 more_frames = get_bits1(gb);
1407 *@brief Calculate remaining input buffer length.
1408 *@param s codec context
1409 *@param gb bitstream reader context
1410 *@return remaining size in bits
1412 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1414 return s->buf_bit_size - get_bits_count(gb);
1418 *@brief Fill the bit reservoir with a (partial) frame.
1419 *@param s codec context
1420 *@param gb bitstream reader context
1421 *@param len length of the partial frame
1422 *@param append decides wether to reset the buffer or not
1424 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1429 /** when the frame data does not need to be concatenated, the input buffer
1430 is resetted and additional bits from the previous frame are copyed
1431 and skipped later so that a fast byte copy is possible */
1434 s->frame_offset = get_bits_count(gb) & 7;
1435 s->num_saved_bits = s->frame_offset;
1436 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1439 buflen = (s->num_saved_bits + len + 8) >> 3;
1441 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1442 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1447 s->num_saved_bits += len;
1449 ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1452 int align = 8 - (get_bits_count(gb) & 7);
1453 align = FFMIN(align, len);
1454 put_bits(&s->pb, align, get_bits(gb, align));
1456 ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1458 skip_bits_long(gb, len);
1461 PutBitContext tmp = s->pb;
1462 flush_put_bits(&tmp);
1465 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1466 skip_bits(&s->gb, s->frame_offset);
1470 *@brief Decode a single WMA packet.
1471 *@param avctx codec context
1472 *@param data the output buffer
1473 *@param data_size number of bytes that were written to the output buffer
1474 *@param avpkt input packet
1475 *@return number of bytes that were read from the input buffer
1477 static int decode_packet(AVCodecContext *avctx,
1478 void *data, int *data_size, AVPacket* avpkt)
1480 WMAProDecodeCtx *s = avctx->priv_data;
1481 GetBitContext* gb = &s->pgb;
1482 const uint8_t* buf = avpkt->data;
1483 int buf_size = avpkt->size;
1484 int num_bits_prev_frame;
1485 int packet_sequence_number;
1488 s->samples_end = (float*)((int8_t*)data + *data_size);
1491 if (s->packet_done || s->packet_loss) {
1494 /** sanity check for the buffer length */
1495 if (buf_size < avctx->block_align)
1498 s->next_packet_start = buf_size - avctx->block_align;
1499 buf_size = avctx->block_align;
1500 s->buf_bit_size = buf_size << 3;
1502 /** parse packet header */
1503 init_get_bits(gb, buf, s->buf_bit_size);
1504 packet_sequence_number = get_bits(gb, 4);
1507 /** get number of bits that need to be added to the previous frame */
1508 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1509 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1510 num_bits_prev_frame);
1512 /** check for packet loss */
1513 if (!s->packet_loss &&
1514 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1516 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1517 s->packet_sequence_number, packet_sequence_number);
1519 s->packet_sequence_number = packet_sequence_number;
1521 if (num_bits_prev_frame > 0) {
1522 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1523 if (num_bits_prev_frame >= remaining_packet_bits) {
1524 num_bits_prev_frame = remaining_packet_bits;
1528 /** append the previous frame data to the remaining data from the
1529 previous packet to create a full frame */
1530 save_bits(s, gb, num_bits_prev_frame, 1);
1531 av_dlog(avctx, "accumulated %x bits of frame data\n",
1532 s->num_saved_bits - s->frame_offset);
1534 /** decode the cross packet frame if it is valid */
1535 if (!s->packet_loss)
1537 } else if (s->num_saved_bits - s->frame_offset) {
1538 av_dlog(avctx, "ignoring %x previously saved bits\n",
1539 s->num_saved_bits - s->frame_offset);
1542 if (s->packet_loss) {
1543 /** reset number of saved bits so that the decoder
1544 does not start to decode incomplete frames in the
1545 s->len_prefix == 0 case */
1546 s->num_saved_bits = 0;
1552 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1553 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1554 skip_bits(gb, s->packet_offset);
1555 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1556 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1557 frame_size <= remaining_bits(s, gb)) {
1558 save_bits(s, gb, frame_size, 0);
1559 s->packet_done = !decode_frame(s);
1560 } else if (!s->len_prefix
1561 && s->num_saved_bits > get_bits_count(&s->gb)) {
1562 /** when the frames do not have a length prefix, we don't know
1563 the compressed length of the individual frames
1564 however, we know what part of a new packet belongs to the
1566 therefore we save the incoming packet first, then we append
1567 the "previous frame" data from the next packet so that
1568 we get a buffer that only contains full frames */
1569 s->packet_done = !decode_frame(s);
1574 if (s->packet_done && !s->packet_loss &&
1575 remaining_bits(s, gb) > 0) {
1576 /** save the rest of the data so that it can be decoded
1577 with the next packet */
1578 save_bits(s, gb, remaining_bits(s, gb), 0);
1581 *data_size = (int8_t *)s->samples - (int8_t *)data;
1582 s->packet_offset = get_bits_count(gb) & 7;
1584 return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3;
1588 *@brief Clear decoder buffers (for seeking).
1589 *@param avctx codec context
1591 static void flush(AVCodecContext *avctx)
1593 WMAProDecodeCtx *s = avctx->priv_data;
1595 /** reset output buffer as a part of it is used during the windowing of a
1597 for (i = 0; i < s->num_channels; i++)
1598 memset(s->channel[i].out, 0, s->samples_per_frame *
1599 sizeof(*s->channel[i].out));
1605 *@brief wmapro decoder
1607 AVCodec ff_wmapro_decoder = {
1609 .type = AVMEDIA_TYPE_AUDIO,
1610 .id = CODEC_ID_WMAPRO,
1611 .priv_data_size = sizeof(WMAProDecodeCtx),
1612 .init = decode_init,
1613 .close = decode_end,
1614 .decode = decode_packet,
1615 .capabilities = CODEC_CAP_SUBFRAMES,
1617 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),