2 * Wmapro compatible decoder
3 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 * Copyright (c) 2008 - 2009 Sascha Sommer, Benjamin Larsson
6 * This file is part of FFmpeg.
8 * FFmpeg 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 * FFmpeg 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 FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * @file libavcodec/wmaprodec.c
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"
97 /** current decoder limitations */
98 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
99 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
100 #define MAX_BANDS 29 ///< max number of scale factor bands
101 #define MAX_FRAMESIZE 16384 ///< maximum compressed frame size
103 #define WMAPRO_BLOCK_MAX_BITS 12 ///< log2 of max block size
104 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
105 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - BLOCK_MIN_BITS + 1) ///< possible block sizes
109 #define SCALEVLCBITS 8
110 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
111 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
112 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
113 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
114 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
116 static VLC sf_vlc; ///< scale factor DPCM vlc
117 static VLC sf_rl_vlc; ///< scale factor run length vlc
118 static VLC vec4_vlc; ///< 4 coefficients per symbol
119 static VLC vec2_vlc; ///< 2 coefficients per symbol
120 static VLC vec1_vlc; ///< 1 coefficient per symbol
121 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
122 static float sin64[33]; ///< sinus table for decorrelation
125 * @brief frame specific decoder context for a single channel
128 int16_t prev_block_len; ///< length of the previous block
129 uint8_t transmit_coefs;
130 uint8_t num_subframes;
131 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
132 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
133 uint8_t cur_subframe; ///< current subframe number
134 uint16_t decoded_samples; ///< number of already processed samples
135 uint8_t grouped; ///< channel is part of a group
136 int quant_step; ///< quantization step for the current subframe
137 int8_t reuse_sf; ///< share scale factors between subframes
138 int8_t scale_factor_step; ///< scaling step for the current subframe
139 int max_scale_factor; ///< maximum scale factor for the current subframe
140 int scale_factors[MAX_BANDS]; ///< scale factor values for the current subframe
141 int saved_scale_factors[MAX_BANDS]; ///< scale factors from a previous subframe
142 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
143 float* coeffs; ///< pointer to the subframe decode buffer
144 DECLARE_ALIGNED_16(float, out[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]); ///< output buffer
148 * @brief channel group for channel transformations
151 uint8_t num_channels; ///< number of channels in the group
152 int8_t transform; ///< transform on / off
153 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
154 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
155 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
159 * @brief main decoder context
161 typedef struct WMAProDecodeCtx {
162 /* generic decoder variables */
163 AVCodecContext* avctx; ///< codec context for av_log
164 DSPContext dsp; ///< accelerated DSP functions
165 uint8_t frame_data[MAX_FRAMESIZE +
166 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
167 PutBitContext pb; ///< context for filling the frame_data buffer
168 MDCTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
169 DECLARE_ALIGNED_16(float, tmp[WMAPRO_BLOCK_MAX_SIZE]); ///< IMDCT output buffer
170 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
172 /* frame size dependent frame information (set during initialization) */
173 uint32_t decode_flags; ///< used compression features
174 uint8_t len_prefix; ///< frame is prefixed with its length
175 uint8_t dynamic_range_compression; ///< frame contains DRC data
176 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
177 uint16_t samples_per_frame; ///< number of samples to output
178 uint16_t log2_frame_size;
179 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
180 int8_t lfe_channel; ///< lfe channel index
181 uint8_t max_num_subframes;
182 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
183 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
184 uint16_t min_samples_per_subframe;
185 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
186 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
187 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
188 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
190 /* packet decode state */
191 uint8_t packet_sequence_number; ///< current packet number
192 int num_saved_bits; ///< saved number of bits
193 int frame_offset; ///< frame offset in the bit reservoir
194 int subframe_offset; ///< subframe offset in the bit reservoir
195 uint8_t packet_loss; ///< set in case of bitstream error
197 /* frame decode state */
198 uint32_t frame_num; ///< current frame number (not used for decoding)
199 GetBitContext gb; ///< bitstream reader context
200 int buf_bit_size; ///< buffer size in bits
201 float* samples; ///< current samplebuffer pointer
202 float* samples_end; ///< maximum samplebuffer pointer
203 uint8_t drc_gain; ///< gain for the DRC tool
204 int8_t skip_frame; ///< skip output step
205 int8_t parsed_all_subframes; ///< all subframes decoded?
207 /* subframe/block decode state */
208 int16_t subframe_len; ///< current subframe length
209 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
210 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
211 int8_t num_bands; ///< number of scale factor bands
212 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
213 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
214 int8_t esc_len; ///< length of escaped coefficients
216 uint8_t num_chgroups; ///< number of channel groups
217 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
219 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
224 *@brief helper function to print the most important members of the context
227 static void av_cold dump_context(WMAProDecodeCtx *s)
229 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
230 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
232 PRINT("ed sample bit depth", s->bits_per_sample);
233 PRINT_HEX("ed decode flags", s->decode_flags);
234 PRINT("samples per frame", s->samples_per_frame);
235 PRINT("log2 frame size", s->log2_frame_size);
236 PRINT("max num subframes", s->max_num_subframes);
237 PRINT("len prefix", s->len_prefix);
238 PRINT("num channels", s->num_channels);
242 *@brief Uninitialize the decoder and free all resources.
243 *@param avctx codec context
244 *@return 0 on success, < 0 otherwise
246 static av_cold int decode_end(AVCodecContext *avctx)
248 WMAProDecodeCtx *s = avctx->priv_data;
251 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
252 ff_mdct_end(&s->mdct_ctx[i]);
258 *@brief Initialize the decoder.
259 *@param avctx codec context
260 *@return 0 on success, -1 otherwise
262 static av_cold int decode_init(AVCodecContext *avctx)
264 WMAProDecodeCtx *s = avctx->priv_data;
265 uint8_t *edata_ptr = avctx->extradata;
266 unsigned int channel_mask;
268 int log2_max_num_subframes;
269 int num_possible_block_sizes;
272 dsputil_init(&s->dsp, avctx);
273 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
275 avctx->sample_fmt = SAMPLE_FMT_FLT;
277 if (avctx->extradata_size >= 18) {
278 s->decode_flags = AV_RL16(edata_ptr+14);
279 channel_mask = AV_RL32(edata_ptr+2);
280 s->bits_per_sample = AV_RL16(edata_ptr);
281 /** dump the extradata */
282 for (i = 0; i < avctx->extradata_size; i++)
283 dprintf(avctx, "[%x] ", avctx->extradata[i]);
284 dprintf(avctx, "\n");
287 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
288 return AVERROR_INVALIDDATA;
292 s->log2_frame_size = av_log2(avctx->block_align) + 4;
295 s->skip_frame = 1; /** skip first frame */
297 s->len_prefix = (s->decode_flags & 0x40);
299 if (!s->len_prefix) {
300 av_log_ask_for_sample(avctx, "no length prefix\n");
301 return AVERROR_INVALIDDATA;
305 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
308 /** init previous block len */
309 for (i = 0; i < avctx->channels; i++)
310 s->channel[i].prev_block_len = s->samples_per_frame;
313 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
314 s->max_num_subframes = 1 << log2_max_num_subframes;
315 if (s->max_num_subframes == 16)
316 s->max_subframe_len_bit = 1;
317 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
319 num_possible_block_sizes = log2_max_num_subframes + 1;
320 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
321 s->dynamic_range_compression = (s->decode_flags & 0x80);
323 if (s->max_num_subframes > MAX_SUBFRAMES) {
324 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
325 s->max_num_subframes);
326 return AVERROR_INVALIDDATA;
329 s->num_channels = avctx->channels;
331 /** extract lfe channel position */
334 if (channel_mask & 8) {
336 for (mask = 1; mask < 16; mask <<= 1) {
337 if (channel_mask & mask)
342 if (s->num_channels < 0 || s->num_channels > WMAPRO_MAX_CHANNELS) {
343 av_log_ask_for_sample(avctx, "invalid number of channels\n");
344 return AVERROR_NOTSUPP;
347 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
348 scale_huffbits, 1, 1,
349 scale_huffcodes, 2, 2, 616);
351 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
352 scale_rl_huffbits, 1, 1,
353 scale_rl_huffcodes, 4, 4, 1406);
355 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
356 coef0_huffbits, 1, 1,
357 coef0_huffcodes, 4, 4, 2108);
359 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
360 coef1_huffbits, 1, 1,
361 coef1_huffcodes, 4, 4, 3912);
363 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
365 vec4_huffcodes, 2, 2, 604);
367 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
369 vec2_huffcodes, 2, 2, 562);
371 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
373 vec1_huffcodes, 2, 2, 562);
375 /** calculate number of scale factor bands and their offsets
376 for every possible block size */
377 for (i = 0; i < num_possible_block_sizes; i++) {
378 int subframe_len = s->samples_per_frame >> i;
382 s->sfb_offsets[i][0] = 0;
384 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band-1] < subframe_len; x++) {
385 int offset = (subframe_len * 2 * critical_freq[x])
386 / s->avctx->sample_rate + 2;
388 if ( offset > s->sfb_offsets[i][band - 1] )
389 s->sfb_offsets[i][band++] = offset;
391 s->sfb_offsets[i][band - 1] = subframe_len;
392 s->num_sfb[i] = band - 1;
396 /** Scale factors can be shared between blocks of different size
397 as every block has a different scale factor band layout.
398 The matrix sf_offsets is needed to find the correct scale factor.
401 for (i = 0; i < num_possible_block_sizes; i++) {
403 for (b = 0; b < s->num_sfb[i]; b++) {
405 int offset = ((s->sfb_offsets[i][b]
406 + s->sfb_offsets[i][b + 1] - 1)<<i) >> 1;
407 for (x = 0; x < num_possible_block_sizes; x++) {
409 while (s->sfb_offsets[x][v + 1] << x < offset)
411 s->sf_offsets[i][x][b] = v;
416 /** init MDCT, FIXME: only init needed sizes */
417 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
418 ff_mdct_init(&s->mdct_ctx[i], BLOCK_MIN_BITS+1+i, 1,
419 1.0 / (1 <<(BLOCK_MIN_BITS + i - 1))
420 / (1 << (s->bits_per_sample - 1)));
422 /** init MDCT windows: simple sinus window */
423 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
424 const int n = 1 << (WMAPRO_BLOCK_MAX_BITS - i);
425 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i - 7;
426 ff_sine_window_init(ff_sine_windows[win_idx], n);
427 s->windows[WMAPRO_BLOCK_SIZES-i-1] = ff_sine_windows[win_idx];
430 /** calculate subwoofer cutoff values */
431 for (i = 0; i < num_possible_block_sizes; i++) {
432 int block_size = s->samples_per_frame >> i;
433 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
434 / s->avctx->sample_rate;
435 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
438 /** calculate sine values for the decorrelation matrix */
439 for (i = 0; i < 33; i++)
440 sin64[i] = sin(i*M_PI / 64.0);
442 if (avctx->debug & FF_DEBUG_BITSTREAM)
445 avctx->channel_layout = channel_mask;
450 *@brief Decode the subframe length.
452 *@param offset sample offset in the frame
453 *@return decoded subframe length on success, < 0 in case of an error
455 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
457 int frame_len_shift = 0;
460 /** no need to read from the bitstream when only one length is possible */
461 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
462 return s->min_samples_per_subframe;
464 /** 1 bit indicates if the subframe is of maximum length */
465 if (s->max_subframe_len_bit) {
466 if (get_bits1(&s->gb))
467 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
469 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
471 subframe_len = s->samples_per_frame >> frame_len_shift;
473 /** sanity check the length */
474 if (subframe_len < s->min_samples_per_subframe
475 || subframe_len > s->samples_per_frame) {
476 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
478 return AVERROR_INVALIDDATA;
484 *@brief Decode how the data in the frame is split into subframes.
485 * Every WMA frame contains the encoded data for a fixed number of
486 * samples per channel. The data for every channel might be split
487 * into several subframes. This function will reconstruct the list of
488 * subframes for every channel.
490 * If the subframes are not evenly split, the algorithm estimates the
491 * channels with the lowest number of total samples.
492 * Afterwards, for each of these channels a bit is read from the
493 * bitstream that indicates if the channel contains a subframe with the
494 * next subframe size that is going to be read from the bitstream or not.
495 * If a channel contains such a subframe, the subframe size gets added to
496 * the channel's subframe list.
497 * The algorithm repeats these steps until the frame is properly divided
498 * between the individual channels.
501 *@return 0 on success, < 0 in case of an error
503 static int decode_tilehdr(WMAProDecodeCtx *s)
505 uint16_t num_samples[WMAPRO_MAX_CHANNELS]; /** sum of samples for all currently known subframes of a channel */
506 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /** flag indicating if a channel contains the current subframe */
507 int channels_for_cur_subframe = s->num_channels; /** number of channels that contain the current subframe */
508 int fixed_channel_layout = 0; /** flag indicating that all channels use the same subframe offsets and sizes */
509 int min_channel_len = 0; /** smallest sum of samples (channels with this length will be processed first) */
512 /* Should never consume more than 3073 bits (256 iterations for the
513 * while loop when always the minimum amount of 128 samples is substracted
514 * from missing samples in the 8 channel case).
515 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
518 /** reset tiling information */
519 for (c = 0; c < s->num_channels; c++)
520 s->channel[c].num_subframes = 0;
522 memset(num_samples, 0, sizeof(num_samples));
524 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
525 fixed_channel_layout = 1;
527 /** loop until the frame data is split between the subframes */
531 /** check which channels contain the subframe */
532 for (c = 0; c < s->num_channels; c++) {
533 if (num_samples[c] == min_channel_len) {
534 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
535 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
536 contains_subframe[c] = 1;
538 contains_subframe[c] = get_bits1(&s->gb);
540 contains_subframe[c] = 0;
543 /** get subframe length, subframe_len == 0 is not allowed */
544 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
545 return AVERROR_INVALIDDATA;
547 /** add subframes to the individual channels and find new min_channel_len */
548 min_channel_len += subframe_len;
549 for (c = 0; c < s->num_channels; c++) {
550 WMAProChannelCtx* chan = &s->channel[c];
552 if (contains_subframe[c]) {
553 if (chan->num_subframes >= MAX_SUBFRAMES) {
554 av_log(s->avctx, AV_LOG_ERROR,
555 "broken frame: num subframes > 31\n");
556 return AVERROR_INVALIDDATA;
558 chan->subframe_len[chan->num_subframes] = subframe_len;
559 num_samples[c] += subframe_len;
560 ++chan->num_subframes;
561 if (num_samples[c] > s->samples_per_frame) {
562 av_log(s->avctx, AV_LOG_ERROR,"broken frame: "
563 "channel len > samples_per_frame\n");
564 return AVERROR_INVALIDDATA;
566 } else if(num_samples[c] <= min_channel_len) {
567 if (num_samples[c] < min_channel_len) {
568 channels_for_cur_subframe = 0;
569 min_channel_len = num_samples[c];
571 ++channels_for_cur_subframe;
574 } while (min_channel_len < s->samples_per_frame);
576 for (c = 0; c < s->num_channels; c++) {
579 for (i = 0; i < s->channel[c].num_subframes; i++) {
580 dprintf(s->avctx, "frame[%i] channel[%i] subframe[%i]"
581 " len %i\n", s->frame_num, c, i, s->channel[c].subframe_len[i]);
582 s->channel[c].subframe_offset[i] = offset;
583 offset += s->channel[c].subframe_len[i];
591 *@brief Calculate a decorrelation matrix from the bitstream parameters.
592 *@param s codec context
593 *@param chgroup channel group for which the matrix needs to be calculated
595 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
596 WMAProChannelGrp *chgroup)
600 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
601 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
602 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
604 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
605 rotation_offset[i] = get_bits(&s->gb, 6);
607 for (i = 0; i < chgroup->num_channels; i++)
608 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
609 get_bits1(&s->gb) ? 1.0 : -1.0;
611 for (i = 1; i < chgroup->num_channels; i++) {
613 for (x = 0; x < i; x++) {
615 for (y = 0; y < i + 1; y++) {
616 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
617 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
618 int n = rotation_offset[offset + x];
630 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
631 (v1 * sinv) - (v2 * cosv);
632 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
633 (v1 * cosv) + (v2 * sinv);
641 *@brief Decode channel transformation parameters
642 *@param s codec context
643 *@return 0 in case of success, < 0 in case of bitstream errors
645 static int decode_channel_transform(WMAProDecodeCtx* s)
648 /* should never consume more than 1921 bits for the 8 channel case
649 * 1 + MAX_CHANNELS * ( MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
650 * + MAX_CHANNELS + MAX_BANDS + 1)
653 /** in the one channel case channel transforms are pointless */
655 if (s->num_channels > 1) {
656 int remaining_channels = s->channels_for_cur_subframe;
658 if (get_bits1(&s->gb)) {
659 av_log_ask_for_sample(s->avctx,
660 "unsupported channel transform bit\n");
661 return AVERROR_INVALIDDATA;
664 for (s->num_chgroups = 0; remaining_channels &&
665 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
666 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
667 float** channel_data = chgroup->channel_data;
668 chgroup->num_channels = 0;
669 chgroup->transform = 0;
671 /** decode channel mask */
672 if (remaining_channels > 2) {
673 for (i = 0; i < s->channels_for_cur_subframe; i++) {
674 int channel_idx = s->channel_indexes_for_cur_subframe[i];
675 if (!s->channel[channel_idx].grouped
676 && get_bits1(&s->gb)) {
677 ++chgroup->num_channels;
678 s->channel[channel_idx].grouped = 1;
679 *channel_data++ = s->channel[channel_idx].coeffs;
683 chgroup->num_channels = remaining_channels;
684 for (i = 0; i < s->channels_for_cur_subframe; i++) {
685 int channel_idx = s->channel_indexes_for_cur_subframe[i];
686 if (!s->channel[channel_idx].grouped)
687 *channel_data++ = s->channel[channel_idx].coeffs;
688 s->channel[channel_idx].grouped = 1;
692 /** decode transform type */
693 if (chgroup->num_channels == 2) {
694 if (get_bits1(&s->gb)) {
695 if (get_bits1(&s->gb)) {
696 av_log_ask_for_sample(s->avctx,
697 "unsupported channel transform type\n");
700 chgroup->transform = 1;
701 if (s->num_channels == 2) {
702 chgroup->decorrelation_matrix[0] = 1.0;
703 chgroup->decorrelation_matrix[1] = -1.0;
704 chgroup->decorrelation_matrix[2] = 1.0;
705 chgroup->decorrelation_matrix[3] = 1.0;
708 chgroup->decorrelation_matrix[0] = 0.70703125;
709 chgroup->decorrelation_matrix[1] = -0.70703125;
710 chgroup->decorrelation_matrix[2] = 0.70703125;
711 chgroup->decorrelation_matrix[3] = 0.70703125;
714 } else if (chgroup->num_channels > 2) {
715 if (get_bits1(&s->gb)) {
716 chgroup->transform = 1;
717 if (get_bits1(&s->gb)) {
718 decode_decorrelation_matrix(s, chgroup);
720 /** FIXME: more than 6 coupled channels not supported */
721 if (chgroup->num_channels > 6) {
722 av_log_ask_for_sample(s->avctx,
723 "coupled channels > 6\n");
725 memcpy(chgroup->decorrelation_matrix,
726 default_decorrelation[chgroup->num_channels],
727 chgroup->num_channels * chgroup->num_channels *
728 sizeof(*chgroup->decorrelation_matrix));
734 /** decode transform on / off */
735 if (chgroup->transform) {
736 if (!get_bits1(&s->gb)) {
738 /** transform can be enabled for individual bands */
739 for (i = 0; i < s->num_bands; i++) {
740 chgroup->transform_band[i] = get_bits1(&s->gb);
743 memset(chgroup->transform_band, 1, s->num_bands);
746 remaining_channels -= chgroup->num_channels;
753 *@brief Extract the coefficients from the bitstream.
754 *@param s codec context
755 *@param c current channel number
756 *@return 0 on success, < 0 in case of bitstream errors
758 static int decode_coeffs(WMAProDecodeCtx *s, int c)
762 WMAProChannelCtx* ci = &s->channel[c];
767 const uint16_t* level;
769 dprintf(s->avctx, "decode coefficients for channel %i\n", c);
771 vlctable = get_bits1(&s->gb);
772 vlc = &coef_vlc[vlctable];
782 /** decode vector coefficients (consumes up to 167 bits per iteration for
783 4 vector coded large values) */
784 while (!rl_mode && cur_coeff + 3 < s->subframe_len) {
789 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
791 if (idx == HUFF_VEC4_SIZE - 1) {
792 for (i = 0; i < 4; i += 2) {
793 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
794 if (idx == HUFF_VEC2_SIZE - 1) {
795 vals[i] = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
796 if (vals[i] == HUFF_VEC1_SIZE - 1)
797 vals[i] += ff_wma_get_large_val(&s->gb);
798 vals[i+1] = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
799 if (vals[i+1] == HUFF_VEC1_SIZE - 1)
800 vals[i+1] += ff_wma_get_large_val(&s->gb);
802 vals[i] = symbol_to_vec2[idx] >> 4;
803 vals[i+1] = symbol_to_vec2[idx] & 0xF;
807 vals[0] = symbol_to_vec4[idx] >> 12;
808 vals[1] = (symbol_to_vec4[idx] >> 8) & 0xF;
809 vals[2] = (symbol_to_vec4[idx] >> 4) & 0xF;
810 vals[3] = symbol_to_vec4[idx] & 0xF;
814 for (i = 0; i < 4; i++) {
816 int sign = get_bits1(&s->gb) - 1;
817 ci->coeffs[cur_coeff] = (vals[i]^sign) - sign;
820 ci->coeffs[cur_coeff] = 0;
821 /** switch to run level mode when subframe_len / 128 zeros
822 were found in a row */
823 rl_mode |= (++num_zeros > s->subframe_len>>8);
829 /** decode run level coded coefficients */
831 memset(&ci->coeffs[cur_coeff], 0,
832 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
833 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
834 level, run, 1, ci->coeffs,
835 cur_coeff, s->subframe_len,
836 s->subframe_len, s->esc_len, 0))
837 return AVERROR_INVALIDDATA;
844 *@brief Extract scale factors from the bitstream.
845 *@param s codec context
846 *@return 0 on success, < 0 in case of bitstream errors
848 static int decode_scale_factors(WMAProDecodeCtx* s)
852 /** should never consume more than 5344 bits
853 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
856 for (i = 0; i < s->channels_for_cur_subframe; i++) {
857 int c = s->channel_indexes_for_cur_subframe[i];
859 int* sf_end = s->channel[c].scale_factors + s->num_bands;
861 /** resample scale factors for the new block size
862 * as the scale factors might need to be resampled several times
863 * before some new values are transmitted, a backup of the last
864 * transmitted scale factors is kept in saved_scale_factors
866 if (s->channel[c].reuse_sf) {
867 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
869 for (b = 0; b < s->num_bands; b++)
870 s->channel[c].scale_factors[b] =
871 s->channel[c].saved_scale_factors[*sf_offsets++];
874 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
876 if (!s->channel[c].reuse_sf) {
878 /** decode DPCM coded scale factors */
879 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
880 val = 45 / s->channel[c].scale_factor_step;
881 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
882 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
887 /** run level decode differences to the resampled factors */
888 for (i = 0; i < s->num_bands; i++) {
894 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
897 uint32_t code = get_bits(&s->gb, 14);
899 sign = (code & 1) - 1;
900 skip = (code & 0x3f) >> 1;
901 } else if (idx == 1) {
904 skip = scale_rl_run[idx];
905 val = scale_rl_level[idx];
906 sign = get_bits1(&s->gb)-1;
910 if (i >= s->num_bands) {
911 av_log(s->avctx,AV_LOG_ERROR,
912 "invalid scale factor coding\n");
913 return AVERROR_INVALIDDATA;
915 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
919 /** save transmitted scale factors so that they can be reused for
921 memcpy(s->channel[c].saved_scale_factors,
922 s->channel[c].scale_factors, s->num_bands *
923 sizeof(*s->channel[c].saved_scale_factors));
924 s->channel[c].table_idx = s->table_idx;
925 s->channel[c].reuse_sf = 1;
928 /** calculate new scale factor maximum */
929 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
930 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
931 s->channel[c].max_scale_factor =
932 FFMAX(s->channel[c].max_scale_factor, *sf);
940 *@brief Reconstruct the individual channel data.
941 *@param s codec context
943 static void inverse_channel_transform(WMAProDecodeCtx *s)
947 for (i = 0; i < s->num_chgroups; i++) {
948 if (s->chgroup[i].transform) {
949 float data[WMAPRO_MAX_CHANNELS];
950 const int num_channels = s->chgroup[i].num_channels;
951 float** ch_data = s->chgroup[i].channel_data;
952 float** ch_end = ch_data + num_channels;
953 const int8_t* tb = s->chgroup[i].transform_band;
956 /** multichannel decorrelation */
957 for (sfb = s->cur_sfb_offsets;
958 sfb < s->cur_sfb_offsets + s->num_bands;sfb++) {
961 /** multiply values with the decorrelation_matrix */
962 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
963 const float* mat = s->chgroup[i].decorrelation_matrix;
964 const float* data_end = data + num_channels;
965 float* data_ptr = data;
968 for (ch = ch_data; ch < ch_end; ch++)
969 *data_ptr++ = (*ch)[y];
971 for (ch = ch_data; ch < ch_end; ch++) {
974 while (data_ptr < data_end)
975 sum += *data_ptr++ * *mat++;
980 } else if (s->num_channels == 2) {
981 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
982 ch_data[0][y] *= 181.0 / 128;
983 ch_data[1][y] *= 181.0 / 128;
992 *@brief Apply sine window and reconstruct the output buffer.
993 *@param s codec context
995 static void wmapro_window(WMAProDecodeCtx *s)
998 for (i = 0; i< s->channels_for_cur_subframe; i++) {
999 int c = s->channel_indexes_for_cur_subframe[i];
1001 int winlen = s->channel[c].prev_block_len;
1002 float* start = s->channel[c].coeffs - (winlen >> 1);
1004 if (s->subframe_len < winlen) {
1005 start += (winlen - s->subframe_len)>>1;
1006 winlen = s->subframe_len;
1009 window = s->windows[av_log2(winlen)-BLOCK_MIN_BITS];
1013 s->dsp.vector_fmul_window(start, start, start + winlen,
1016 s->channel[c].prev_block_len = s->subframe_len;
1021 *@brief Decode a single subframe (block).
1022 *@param s codec context
1023 *@return 0 on success, < 0 when decoding failed
1025 static int decode_subframe(WMAProDecodeCtx *s)
1027 int offset = s->samples_per_frame;
1028 int subframe_len = s->samples_per_frame;
1030 int total_samples = s->samples_per_frame * s->num_channels;
1031 int transmit_coeffs = 0;
1032 int cur_subwoofer_cutoff;
1034 s->subframe_offset = get_bits_count(&s->gb);
1036 /** reset channel context and find the next block offset and size
1037 == the next block of the channel with the smallest number of
1040 for (i = 0; i < s->num_channels; i++) {
1041 s->channel[i].grouped = 0;
1042 if (offset > s->channel[i].decoded_samples) {
1043 offset = s->channel[i].decoded_samples;
1045 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1050 "processing subframe with offset %i len %i\n", offset, subframe_len);
1052 /** get a list of all channels that contain the estimated block */
1053 s->channels_for_cur_subframe = 0;
1054 for (i = 0; i < s->num_channels; i++) {
1055 const int cur_subframe = s->channel[i].cur_subframe;
1056 /** substract already processed samples */
1057 total_samples -= s->channel[i].decoded_samples;
1059 /** and count if there are multiple subframes that match our profile */
1060 if (offset == s->channel[i].decoded_samples &&
1061 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1062 total_samples -= s->channel[i].subframe_len[cur_subframe];
1063 s->channel[i].decoded_samples +=
1064 s->channel[i].subframe_len[cur_subframe];
1065 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1066 ++s->channels_for_cur_subframe;
1070 /** check if the frame will be complete after processing the
1073 s->parsed_all_subframes = 1;
1076 dprintf(s->avctx, "subframe is part of %i channels\n",
1077 s->channels_for_cur_subframe);
1079 /** calculate number of scale factor bands and their offsets */
1080 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1081 s->num_bands = s->num_sfb[s->table_idx];
1082 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1083 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1085 /** configure the decoder for the current subframe */
1086 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1087 int c = s->channel_indexes_for_cur_subframe[i];
1089 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame>>1)
1093 s->subframe_len = subframe_len;
1094 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1096 /** skip extended header if any */
1097 if (get_bits1(&s->gb)) {
1099 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1100 int len = get_bits(&s->gb, 4);
1101 num_fill_bits = get_bits(&s->gb, len) + 1;
1104 if (num_fill_bits >= 0) {
1105 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1106 av_log(s->avctx,AV_LOG_ERROR,"invalid number of fill bits\n");
1107 return AVERROR_INVALIDDATA;
1110 skip_bits_long(&s->gb, num_fill_bits);
1114 /** no idea for what the following bit is used */
1115 if (get_bits1(&s->gb)) {
1116 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1117 return AVERROR_INVALIDDATA;
1121 if (decode_channel_transform(s) < 0)
1122 return AVERROR_INVALIDDATA;
1125 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1126 int c = s->channel_indexes_for_cur_subframe[i];
1127 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1128 transmit_coeffs = 1;
1131 if (transmit_coeffs) {
1133 int quant_step = 90 * s->bits_per_sample >> 4;
1134 if ((get_bits1(&s->gb))) {
1135 /** FIXME: might change run level mode decision */
1136 av_log_ask_for_sample(s->avctx, "unsupported quant step coding\n");
1137 return AVERROR_INVALIDDATA;
1139 /** decode quantization step */
1140 step = get_sbits(&s->gb, 6);
1142 if (step == -32 || step == 31) {
1143 const int sign = (step == 31) - 1;
1145 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1146 (step = get_bits(&s->gb, 5)) == 31 ) {
1149 quant_step += ((quant + step) ^ sign) - sign;
1151 if (quant_step < 0) {
1152 av_log(s->avctx,AV_LOG_DEBUG,"negative quant step\n");
1155 /** decode quantization step modifiers for every channel */
1157 if (s->channels_for_cur_subframe == 1) {
1158 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1160 int modifier_len = get_bits(&s->gb, 3);
1161 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1162 int c = s->channel_indexes_for_cur_subframe[i];
1163 s->channel[c].quant_step = quant_step;
1164 if (get_bits1(&s->gb)) {
1166 s->channel[c].quant_step +=
1167 get_bits(&s->gb, modifier_len) + 1;
1169 ++s->channel[c].quant_step;
1174 /** decode scale factors */
1175 if (decode_scale_factors(s) < 0)
1176 return AVERROR_INVALIDDATA;
1179 dprintf(s->avctx, "BITSTREAM: subframe header length was %i\n",
1180 get_bits_count(&s->gb) - s->subframe_offset);
1182 /** parse coefficients */
1183 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1184 int c = s->channel_indexes_for_cur_subframe[i];
1185 if (s->channel[c].transmit_coefs &&
1186 get_bits_count(&s->gb) < s->num_saved_bits) {
1187 decode_coeffs(s, c);
1189 memset(s->channel[c].coeffs, 0,
1190 sizeof(*s->channel[c].coeffs) * subframe_len);
1193 dprintf(s->avctx, "BITSTREAM: subframe length was %i\n",
1194 get_bits_count(&s->gb) - s->subframe_offset);
1196 if (transmit_coeffs) {
1197 /** reconstruct the per channel data */
1198 inverse_channel_transform(s);
1199 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1200 int c = s->channel_indexes_for_cur_subframe[i];
1201 const int* sf = s->channel[c].scale_factors;
1204 if (c == s->lfe_channel)
1205 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1206 (subframe_len - cur_subwoofer_cutoff));
1208 /** inverse quantization and rescaling */
1209 for (b = 0; b < s->num_bands; b++) {
1210 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1211 const int exp = s->channel[c].quant_step -
1212 (s->channel[c].max_scale_factor - *sf++) *
1213 s->channel[c].scale_factor_step;
1214 const float quant = pow(10.0, exp / 20.0);
1217 for (start = s->cur_sfb_offsets[b]; start < end; start++)
1218 s->tmp[start] = s->channel[c].coeffs[start] * quant;
1221 /** apply imdct (ff_imdct_half == DCTIV with reverse) */
1222 ff_imdct_half(&s->mdct_ctx[av_log2(subframe_len)-BLOCK_MIN_BITS],
1223 s->channel[c].coeffs, s->tmp);
1227 /** window and overlapp-add */
1230 /** handled one subframe */
1231 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1232 int c = s->channel_indexes_for_cur_subframe[i];
1233 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1234 av_log(s->avctx,AV_LOG_ERROR,"broken subframe\n");
1235 return AVERROR_INVALIDDATA;
1237 ++s->channel[c].cur_subframe;
1244 *@brief Decode one WMA frame.
1245 *@param s codec context
1246 *@return 0 if the trailer bit indicates that this is the last frame,
1247 * 1 if there are additional frames
1249 static int decode_frame(WMAProDecodeCtx *s)
1251 GetBitContext* gb = &s->gb;
1252 int more_frames = 0;
1256 /** check for potential output buffer overflow */
1257 if (s->num_channels * s->samples_per_frame > s->samples_end - s->samples) {
1258 av_log(s->avctx,AV_LOG_ERROR,
1259 "not enough space for the output samples\n");
1264 /** get frame length */
1266 len = get_bits(gb, s->log2_frame_size);
1268 dprintf(s->avctx, "decoding frame with length %x\n", len);
1270 /** decode tile information */
1271 if (decode_tilehdr(s)) {
1276 /** read postproc transform */
1277 if (s->num_channels > 1 && get_bits1(gb)) {
1278 av_log_ask_for_sample(s->avctx, "Unsupported postproc transform found\n");
1283 /** read drc info */
1284 if (s->dynamic_range_compression) {
1285 s->drc_gain = get_bits(gb, 8);
1286 dprintf(s->avctx, "drc_gain %i\n", s->drc_gain);
1289 /** no idea what these are for, might be the number of samples
1290 that need to be skipped at the beginning or end of a stream */
1291 if (get_bits1(gb)) {
1294 /** usually true for the first frame */
1295 if (get_bits1(gb)) {
1296 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1297 dprintf(s->avctx, "start skip: %i\n", skip);
1300 /** sometimes true for the last frame */
1301 if (get_bits1(gb)) {
1302 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1303 dprintf(s->avctx, "end skip: %i\n", skip);
1308 dprintf(s->avctx, "BITSTREAM: frame header length was %i\n",
1309 get_bits_count(gb) - s->frame_offset);
1311 /** reset subframe states */
1312 s->parsed_all_subframes = 0;
1313 for (i = 0; i < s->num_channels; i++) {
1314 s->channel[i].decoded_samples = 0;
1315 s->channel[i].cur_subframe = 0;
1316 s->channel[i].reuse_sf = 0;
1319 /** decode all subframes */
1320 while (!s->parsed_all_subframes) {
1321 if (decode_subframe(s) < 0) {
1327 /** interleave samples and write them to the output buffer */
1328 for (i = 0; i < s->num_channels; i++) {
1330 int incr = s->num_channels;
1331 float* iptr = s->channel[i].out;
1334 ptr = s->samples + i;
1336 for (x = 0; x < s->samples_per_frame; x++) {
1337 *ptr = av_clipf(*iptr++, -1.0, 32767.0 / 32768.0);
1341 /** reuse second half of the IMDCT output for the next frame */
1342 memcpy(&s->channel[i].out[0],
1343 &s->channel[i].out[s->samples_per_frame],
1344 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1347 if (s->skip_frame) {
1350 s->samples += s->num_channels * s->samples_per_frame;
1352 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1353 /** FIXME: not sure if this is always an error */
1354 av_log(s->avctx,AV_LOG_ERROR,"frame[%i] would have to skip %i bits\n",
1355 s->frame_num, len - (get_bits_count(gb) - s->frame_offset) - 1);
1360 /** skip the rest of the frame data */
1361 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1363 /** decode trailer bit */
1364 more_frames = get_bits1(gb);
1371 *@brief Calculate remaining input buffer length.
1372 *@param s codec context
1373 *@param gb bitstream reader context
1374 *@return remaining size in bits
1376 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext* gb)
1378 return s->buf_bit_size - get_bits_count(gb);
1382 *@brief Fill the bit reservoir with a (partial) frame.
1383 *@param s codec context
1384 *@param gb bitstream reader context
1385 *@param len length of the partial frame
1386 *@param append decides wether to reset the buffer or not
1388 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1393 /** when the frame data does not need to be concatenated, the input buffer
1394 is resetted and additional bits from the previous frame are copyed
1395 and skipped later so that a fast byte copy is possible */
1398 s->frame_offset = get_bits_count(gb) & 7;
1399 s->num_saved_bits = s->frame_offset;
1400 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1403 buflen = (s->num_saved_bits + len + 8) >> 3;
1405 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1406 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1411 s->num_saved_bits += len;
1413 ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), s->num_saved_bits);
1415 int align = 8 - (get_bits_count(gb) & 7);
1416 align = FFMIN(align, len);
1417 put_bits(&s->pb, align, get_bits(gb, align));
1419 ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1421 skip_bits_long(gb, len);
1424 PutBitContext tmp = s->pb;
1425 flush_put_bits(&tmp);
1428 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1429 skip_bits(&s->gb, s->frame_offset);
1433 *@brief Decode a single WMA packet.
1434 *@param avctx codec context
1435 *@param data the output buffer
1436 *@param data_size number of bytes that were written to the output buffer
1437 *@param avpkt input packet
1438 *@return number of bytes that were read from the input buffer
1440 static int decode_packet(AVCodecContext *avctx,
1441 void *data, int *data_size, AVPacket* avpkt)
1444 WMAProDecodeCtx *s = avctx->priv_data;
1445 const uint8_t* buf = avpkt->data;
1446 int buf_size = avpkt->size;
1447 int more_frames = 1;
1448 int num_bits_prev_frame;
1449 int packet_sequence_number;
1452 s->samples_end = (float*)((int8_t*)data + *data_size);
1453 s->buf_bit_size = buf_size << 3;
1458 /** sanity check for the buffer length */
1459 if (buf_size < avctx->block_align)
1462 buf_size = avctx->block_align;
1464 /** parse packet header */
1465 init_get_bits(&gb, buf, s->buf_bit_size);
1466 packet_sequence_number = get_bits(&gb, 4);
1469 /** get number of bits that need to be added to the previous frame */
1470 num_bits_prev_frame = get_bits(&gb, s->log2_frame_size);
1471 dprintf(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1472 num_bits_prev_frame);
1474 /** check for packet loss */
1475 if (!s->packet_loss &&
1476 ((s->packet_sequence_number + 1)&0xF) != packet_sequence_number) {
1478 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1479 s->packet_sequence_number, packet_sequence_number);
1481 s->packet_sequence_number = packet_sequence_number;
1483 if (num_bits_prev_frame > 0) {
1484 /** append the previous frame data to the remaining data from the
1485 previous packet to create a full frame */
1486 save_bits(s, &gb, num_bits_prev_frame, 1);
1487 dprintf(avctx, "accumulated %x bits of frame data\n",
1488 s->num_saved_bits - s->frame_offset);
1490 /** decode the cross packet frame if it is valid */
1491 if (!s->packet_loss)
1493 } else if (s->num_saved_bits - s->frame_offset) {
1494 dprintf(avctx, "ignoring %x previously saved bits\n",
1495 s->num_saved_bits - s->frame_offset);
1499 /** decode the rest of the packet */
1500 while (!s->packet_loss && more_frames &&
1501 remaining_bits(s, &gb) > s->log2_frame_size) {
1502 int frame_size = show_bits(&gb, s->log2_frame_size);
1504 /** there is enough data for a full frame */
1505 if (remaining_bits(s,&gb) >= frame_size && frame_size > 0) {
1506 save_bits(s, &gb, frame_size, 0);
1508 /** decode the frame */
1509 more_frames = decode_frame(s);
1512 dprintf(avctx, "no more frames\n");
1518 if (!s->packet_loss && remaining_bits(s,&gb) > 0) {
1519 /** save the rest of the data so that it can be decoded
1520 with the next packet */
1521 save_bits(s, &gb, remaining_bits(s,&gb), 0);
1524 *data_size = (int8_t *)s->samples - (int8_t *)data;
1526 return avctx->block_align;
1530 *@brief Clear decoder buffers (for seeking).
1531 *@param avctx codec context
1533 static void flush(AVCodecContext *avctx)
1535 WMAProDecodeCtx *s = avctx->priv_data;
1537 /** reset output buffer as a part of it is used during the windowing of a
1539 for (i = 0; i < s->num_channels; i++)
1540 memset(s->channel[i].out, 0, s->samples_per_frame *
1541 sizeof(*s->channel[i].out));
1547 *@brief wmapro decoder
1549 AVCodec wmapro_decoder = {
1553 sizeof(WMAProDecodeCtx),
1559 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),