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 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
25 * @brief wmapro decoder implementation
26 * Wmapro is an MDCT based codec comparable to wma standard or AAC.
27 * The decoding therefore consists of the following steps:
28 * - bitstream decoding
29 * - reconstruction of per-channel data
30 * - rescaling and inverse quantization
32 * - windowing and overlapp-add
34 * The compressed wmapro bitstream is split into individual packets.
35 * Every such packet contains one or more wma frames.
36 * The compressed frames may have a variable length and frames may
37 * cross packet boundaries.
38 * Common to all wmapro frames is the number of samples that are stored in
40 * The number of samples and a few other decode flags are stored
41 * as extradata that has to be passed to the decoder.
43 * The wmapro frames themselves are again split into a variable number of
44 * subframes. Every subframe contains the data for 2^N time domain samples
45 * where N varies between 7 and 12.
47 * Example wmapro bitstream (in samples):
49 * || packet 0 || packet 1 || packet 2 packets
50 * ---------------------------------------------------
51 * || frame 0 || frame 1 || frame 2 || frames
52 * ---------------------------------------------------
53 * || | | || | | | || || subframes of channel 0
54 * ---------------------------------------------------
55 * || | | || | | | || || subframes of channel 1
56 * ---------------------------------------------------
58 * The frame layouts for the individual channels of a wma frame does not need
61 * However, if the offsets and lengths of several subframes of a frame are the
62 * same, the subframes of the channels can be grouped.
63 * Every group may then use special coding techniques like M/S stereo coding
64 * to improve the compression ratio. These channel transformations do not
65 * need to be applied to a whole subframe. Instead, they can also work on
66 * individual scale factor bands (see below).
67 * The coefficients that carry the audio signal in the frequency domain
68 * are transmitted as huffman-coded vectors with 4, 2 and 1 elements.
69 * In addition to that, the encoder can switch to a runlevel coding scheme
70 * by transmitting subframe_length / 128 zero coefficients.
72 * Before the audio signal can be converted to the time domain, the
73 * coefficients have to be rescaled and inverse quantized.
74 * A subframe is therefore split into several scale factor bands that get
75 * scaled individually.
76 * Scale factors are submitted for every frame but they might be shared
77 * between the subframes of a channel. Scale factors are initially DPCM-coded.
78 * Once scale factors are shared, the differences are transmitted as runlevel
80 * Every subframe length and offset combination in the frame layout shares a
81 * common quantization factor that can be adjusted for every channel by a
83 * After the inverse quantization, the coefficients get processed by an IMDCT.
84 * The resulting values are then windowed with a sine window and the first half
85 * of the values are added to the second half of the output from the previous
86 * subframe in order to reconstruct the output samples.
89 #include "libavutil/intfloat.h"
90 #include "libavutil/intreadwrite.h"
95 #include "wmaprodata.h"
99 #include "wma_common.h"
101 /** current decoder limitations */
102 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
103 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
104 #define MAX_BANDS 29 ///< max number of scale factor bands
105 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
107 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
108 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
109 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
110 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
114 #define SCALEVLCBITS 8
115 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
116 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
117 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
118 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
119 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
121 static VLC sf_vlc; ///< scale factor DPCM vlc
122 static VLC sf_rl_vlc; ///< scale factor run length vlc
123 static VLC vec4_vlc; ///< 4 coefficients per symbol
124 static VLC vec2_vlc; ///< 2 coefficients per symbol
125 static VLC vec1_vlc; ///< 1 coefficient per symbol
126 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
127 static float sin64[33]; ///< sinus table for decorrelation
130 * @brief frame specific decoder context for a single channel
133 int16_t prev_block_len; ///< length of the previous block
134 uint8_t transmit_coefs;
135 uint8_t num_subframes;
136 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
137 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
138 uint8_t cur_subframe; ///< current subframe number
139 uint16_t decoded_samples; ///< number of already processed samples
140 uint8_t grouped; ///< channel is part of a group
141 int quant_step; ///< quantization step for the current subframe
142 int8_t reuse_sf; ///< share scale factors between subframes
143 int8_t scale_factor_step; ///< scaling step for the current subframe
144 int max_scale_factor; ///< maximum scale factor for the current subframe
145 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
146 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
147 int* scale_factors; ///< pointer to the scale factor values used for decoding
148 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
149 float* coeffs; ///< pointer to the subframe decode buffer
150 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
151 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
155 * @brief channel group for channel transformations
158 uint8_t num_channels; ///< number of channels in the group
159 int8_t transform; ///< transform on / off
160 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
161 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
162 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
166 * @brief main decoder context
168 typedef struct WMAProDecodeCtx {
169 /* generic decoder variables */
170 AVCodecContext* avctx; ///< codec context for av_log
171 AVFrame frame; ///< AVFrame for decoded output
172 DSPContext dsp; ///< accelerated DSP functions
173 uint8_t frame_data[MAX_FRAMESIZE +
174 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
175 PutBitContext pb; ///< context for filling the frame_data buffer
176 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
177 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
178 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
180 /* frame size dependent frame information (set during initialization) */
181 uint32_t decode_flags; ///< used compression features
182 uint8_t len_prefix; ///< frame is prefixed with its length
183 uint8_t dynamic_range_compression; ///< frame contains DRC data
184 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
185 uint16_t samples_per_frame; ///< number of samples to output
186 uint16_t log2_frame_size;
187 int8_t lfe_channel; ///< lfe channel index
188 uint8_t max_num_subframes;
189 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
190 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
191 uint16_t min_samples_per_subframe;
192 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
193 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
194 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
195 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
197 /* packet decode state */
198 GetBitContext pgb; ///< bitstream reader context for the packet
199 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
200 uint8_t packet_offset; ///< frame offset in the packet
201 uint8_t packet_sequence_number; ///< current packet number
202 int num_saved_bits; ///< saved number of bits
203 int frame_offset; ///< frame offset in the bit reservoir
204 int subframe_offset; ///< subframe offset in the bit reservoir
205 uint8_t packet_loss; ///< set in case of bitstream error
206 uint8_t packet_done; ///< set when a packet is fully decoded
208 /* frame decode state */
209 uint32_t frame_num; ///< current frame number (not used for decoding)
210 GetBitContext gb; ///< bitstream reader context
211 int buf_bit_size; ///< buffer size in bits
212 uint8_t drc_gain; ///< gain for the DRC tool
213 int8_t skip_frame; ///< skip output step
214 int8_t parsed_all_subframes; ///< all subframes decoded?
216 /* subframe/block decode state */
217 int16_t subframe_len; ///< current subframe length
218 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
219 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
220 int8_t num_bands; ///< number of scale factor bands
221 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
222 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
223 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
224 int8_t esc_len; ///< length of escaped coefficients
226 uint8_t num_chgroups; ///< number of channel groups
227 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
229 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
234 *@brief helper function to print the most important members of the context
237 static av_cold void dump_context(WMAProDecodeCtx *s)
239 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
240 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
242 PRINT("ed sample bit depth", s->bits_per_sample);
243 PRINT_HEX("ed decode flags", s->decode_flags);
244 PRINT("samples per frame", s->samples_per_frame);
245 PRINT("log2 frame size", s->log2_frame_size);
246 PRINT("max num subframes", s->max_num_subframes);
247 PRINT("len prefix", s->len_prefix);
248 PRINT("num channels", s->avctx->channels);
252 *@brief Uninitialize the decoder and free all resources.
253 *@param avctx codec context
254 *@return 0 on success, < 0 otherwise
256 static av_cold int decode_end(AVCodecContext *avctx)
258 WMAProDecodeCtx *s = avctx->priv_data;
261 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
262 ff_mdct_end(&s->mdct_ctx[i]);
268 *@brief Initialize the decoder.
269 *@param avctx codec context
270 *@return 0 on success, -1 otherwise
272 static av_cold int decode_init(AVCodecContext *avctx)
274 WMAProDecodeCtx *s = avctx->priv_data;
275 uint8_t *edata_ptr = avctx->extradata;
276 unsigned int channel_mask;
278 int log2_max_num_subframes;
279 int num_possible_block_sizes;
282 ff_dsputil_init(&s->dsp, avctx);
283 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
285 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
287 if (avctx->extradata_size >= 18) {
288 s->decode_flags = AV_RL16(edata_ptr+14);
289 channel_mask = AV_RL32(edata_ptr+2);
290 s->bits_per_sample = AV_RL16(edata_ptr);
291 /** dump the extradata */
292 for (i = 0; i < avctx->extradata_size; i++)
293 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
294 av_dlog(avctx, "\n");
297 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
298 return AVERROR_INVALIDDATA;
302 s->log2_frame_size = av_log2(avctx->block_align) + 4;
305 s->skip_frame = 1; /* skip first frame */
307 s->len_prefix = (s->decode_flags & 0x40);
310 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
311 if (bits > WMAPRO_BLOCK_MAX_BITS) {
312 av_log_missing_feature(avctx, "14-bits block sizes", 1);
313 return AVERROR_PATCHWELCOME;
315 s->samples_per_frame = 1 << bits;
318 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
319 s->max_num_subframes = 1 << log2_max_num_subframes;
320 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
321 s->max_subframe_len_bit = 1;
322 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
324 num_possible_block_sizes = log2_max_num_subframes + 1;
325 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
326 s->dynamic_range_compression = (s->decode_flags & 0x80);
328 if (s->max_num_subframes > MAX_SUBFRAMES) {
329 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
330 s->max_num_subframes);
331 return AVERROR_INVALIDDATA;
334 if (s->min_samples_per_subframe < (1<<WMAPRO_BLOCK_MIN_BITS)) {
335 av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
336 s->min_samples_per_subframe);
337 return AVERROR_INVALIDDATA;
340 if (s->avctx->sample_rate <= 0) {
341 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
342 return AVERROR_INVALIDDATA;
345 if (avctx->channels < 0) {
346 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
348 return AVERROR_INVALIDDATA;
349 } else if (avctx->channels > WMAPRO_MAX_CHANNELS) {
350 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
351 return AVERROR_PATCHWELCOME;
354 /** init previous block len */
355 for (i = 0; i < avctx->channels; i++)
356 s->channel[i].prev_block_len = s->samples_per_frame;
358 /** extract lfe channel position */
361 if (channel_mask & 8) {
363 for (mask = 1; mask < 16; mask <<= 1) {
364 if (channel_mask & mask)
369 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
370 scale_huffbits, 1, 1,
371 scale_huffcodes, 2, 2, 616);
373 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
374 scale_rl_huffbits, 1, 1,
375 scale_rl_huffcodes, 4, 4, 1406);
377 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
378 coef0_huffbits, 1, 1,
379 coef0_huffcodes, 4, 4, 2108);
381 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
382 coef1_huffbits, 1, 1,
383 coef1_huffcodes, 4, 4, 3912);
385 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
387 vec4_huffcodes, 2, 2, 604);
389 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
391 vec2_huffcodes, 2, 2, 562);
393 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
395 vec1_huffcodes, 2, 2, 562);
397 /** calculate number of scale factor bands and their offsets
398 for every possible block size */
399 for (i = 0; i < num_possible_block_sizes; i++) {
400 int subframe_len = s->samples_per_frame >> i;
404 s->sfb_offsets[i][0] = 0;
406 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
407 int offset = (subframe_len * 2 * critical_freq[x])
408 / s->avctx->sample_rate + 2;
410 if (offset > s->sfb_offsets[i][band - 1])
411 s->sfb_offsets[i][band++] = offset;
413 s->sfb_offsets[i][band - 1] = subframe_len;
414 s->num_sfb[i] = band - 1;
418 /** Scale factors can be shared between blocks of different size
419 as every block has a different scale factor band layout.
420 The matrix sf_offsets is needed to find the correct scale factor.
423 for (i = 0; i < num_possible_block_sizes; i++) {
425 for (b = 0; b < s->num_sfb[i]; b++) {
427 int offset = ((s->sfb_offsets[i][b]
428 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
429 for (x = 0; x < num_possible_block_sizes; x++) {
431 while (s->sfb_offsets[x][v + 1] << x < offset)
433 s->sf_offsets[i][x][b] = v;
438 /** init MDCT, FIXME: only init needed sizes */
439 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
440 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
441 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
442 / (1 << (s->bits_per_sample - 1)));
444 /** init MDCT windows: simple sinus window */
445 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
446 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
447 ff_init_ff_sine_windows(win_idx);
448 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
451 /** calculate subwoofer cutoff values */
452 for (i = 0; i < num_possible_block_sizes; i++) {
453 int block_size = s->samples_per_frame >> i;
454 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
455 / s->avctx->sample_rate;
456 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
459 /** calculate sine values for the decorrelation matrix */
460 for (i = 0; i < 33; i++)
461 sin64[i] = sin(i*M_PI / 64.0);
463 if (avctx->debug & FF_DEBUG_BITSTREAM)
466 avctx->channel_layout = channel_mask;
468 avcodec_get_frame_defaults(&s->frame);
469 avctx->coded_frame = &s->frame;
475 *@brief Decode the subframe length.
477 *@param offset sample offset in the frame
478 *@return decoded subframe length on success, < 0 in case of an error
480 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
482 int frame_len_shift = 0;
485 /** no need to read from the bitstream when only one length is possible */
486 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
487 return s->min_samples_per_subframe;
489 /** 1 bit indicates if the subframe is of maximum length */
490 if (s->max_subframe_len_bit) {
491 if (get_bits1(&s->gb))
492 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
494 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
496 subframe_len = s->samples_per_frame >> frame_len_shift;
498 /** sanity check the length */
499 if (subframe_len < s->min_samples_per_subframe ||
500 subframe_len > s->samples_per_frame) {
501 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
503 return AVERROR_INVALIDDATA;
509 *@brief Decode how the data in the frame is split into subframes.
510 * Every WMA frame contains the encoded data for a fixed number of
511 * samples per channel. The data for every channel might be split
512 * into several subframes. This function will reconstruct the list of
513 * subframes for every channel.
515 * If the subframes are not evenly split, the algorithm estimates the
516 * channels with the lowest number of total samples.
517 * Afterwards, for each of these channels a bit is read from the
518 * bitstream that indicates if the channel contains a subframe with the
519 * next subframe size that is going to be read from the bitstream or not.
520 * If a channel contains such a subframe, the subframe size gets added to
521 * the channel's subframe list.
522 * The algorithm repeats these steps until the frame is properly divided
523 * between the individual channels.
526 *@return 0 on success, < 0 in case of an error
528 static int decode_tilehdr(WMAProDecodeCtx *s)
530 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
531 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
532 int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */
533 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
534 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
537 /* Should never consume more than 3073 bits (256 iterations for the
538 * while loop when always the minimum amount of 128 samples is subtracted
539 * from missing samples in the 8 channel case).
540 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
543 /** reset tiling information */
544 for (c = 0; c < s->avctx->channels; c++)
545 s->channel[c].num_subframes = 0;
547 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
548 fixed_channel_layout = 1;
550 /** loop until the frame data is split between the subframes */
554 /** check which channels contain the subframe */
555 for (c = 0; c < s->avctx->channels; c++) {
556 if (num_samples[c] == min_channel_len) {
557 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
558 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
559 contains_subframe[c] = 1;
561 contains_subframe[c] = get_bits1(&s->gb);
563 contains_subframe[c] = 0;
566 /** get subframe length, subframe_len == 0 is not allowed */
567 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
568 return AVERROR_INVALIDDATA;
570 /** add subframes to the individual channels and find new min_channel_len */
571 min_channel_len += subframe_len;
572 for (c = 0; c < s->avctx->channels; c++) {
573 WMAProChannelCtx* chan = &s->channel[c];
575 if (contains_subframe[c]) {
576 if (chan->num_subframes >= MAX_SUBFRAMES) {
577 av_log(s->avctx, AV_LOG_ERROR,
578 "broken frame: num subframes > 31\n");
579 return AVERROR_INVALIDDATA;
581 chan->subframe_len[chan->num_subframes] = subframe_len;
582 num_samples[c] += subframe_len;
583 ++chan->num_subframes;
584 if (num_samples[c] > s->samples_per_frame) {
585 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
586 "channel len > samples_per_frame\n");
587 return AVERROR_INVALIDDATA;
589 } else if (num_samples[c] <= min_channel_len) {
590 if (num_samples[c] < min_channel_len) {
591 channels_for_cur_subframe = 0;
592 min_channel_len = num_samples[c];
594 ++channels_for_cur_subframe;
597 } while (min_channel_len < s->samples_per_frame);
599 for (c = 0; c < s->avctx->channels; c++) {
602 for (i = 0; i < s->channel[c].num_subframes; i++) {
603 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
604 " len %i\n", s->frame_num, c, i,
605 s->channel[c].subframe_len[i]);
606 s->channel[c].subframe_offset[i] = offset;
607 offset += s->channel[c].subframe_len[i];
615 *@brief Calculate a decorrelation matrix from the bitstream parameters.
616 *@param s codec context
617 *@param chgroup channel group for which the matrix needs to be calculated
619 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
620 WMAProChannelGrp *chgroup)
624 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
625 memset(chgroup->decorrelation_matrix, 0, s->avctx->channels *
626 s->avctx->channels * sizeof(*chgroup->decorrelation_matrix));
628 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
629 rotation_offset[i] = get_bits(&s->gb, 6);
631 for (i = 0; i < chgroup->num_channels; i++)
632 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
633 get_bits1(&s->gb) ? 1.0 : -1.0;
635 for (i = 1; i < chgroup->num_channels; i++) {
637 for (x = 0; x < i; x++) {
639 for (y = 0; y < i + 1; y++) {
640 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
641 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
642 int n = rotation_offset[offset + x];
648 cosv = sin64[32 - n];
650 sinv = sin64[64 - n];
651 cosv = -sin64[n - 32];
654 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
655 (v1 * sinv) - (v2 * cosv);
656 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
657 (v1 * cosv) + (v2 * sinv);
665 *@brief Decode channel transformation parameters
666 *@param s codec context
667 *@return 0 in case of success, < 0 in case of bitstream errors
669 static int decode_channel_transform(WMAProDecodeCtx* s)
672 /* should never consume more than 1921 bits for the 8 channel case
673 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
674 * + MAX_CHANNELS + MAX_BANDS + 1)
677 /** in the one channel case channel transforms are pointless */
679 if (s->avctx->channels > 1) {
680 int remaining_channels = s->channels_for_cur_subframe;
682 if (get_bits1(&s->gb)) {
683 av_log_ask_for_sample(s->avctx,
684 "unsupported channel transform bit\n");
685 return AVERROR_INVALIDDATA;
688 for (s->num_chgroups = 0; remaining_channels &&
689 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
690 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
691 float** channel_data = chgroup->channel_data;
692 chgroup->num_channels = 0;
693 chgroup->transform = 0;
695 /** decode channel mask */
696 if (remaining_channels > 2) {
697 for (i = 0; i < s->channels_for_cur_subframe; i++) {
698 int channel_idx = s->channel_indexes_for_cur_subframe[i];
699 if (!s->channel[channel_idx].grouped
700 && get_bits1(&s->gb)) {
701 ++chgroup->num_channels;
702 s->channel[channel_idx].grouped = 1;
703 *channel_data++ = s->channel[channel_idx].coeffs;
707 chgroup->num_channels = remaining_channels;
708 for (i = 0; i < s->channels_for_cur_subframe; i++) {
709 int channel_idx = s->channel_indexes_for_cur_subframe[i];
710 if (!s->channel[channel_idx].grouped)
711 *channel_data++ = s->channel[channel_idx].coeffs;
712 s->channel[channel_idx].grouped = 1;
716 /** decode transform type */
717 if (chgroup->num_channels == 2) {
718 if (get_bits1(&s->gb)) {
719 if (get_bits1(&s->gb)) {
720 av_log_ask_for_sample(s->avctx,
721 "unsupported channel transform type\n");
724 chgroup->transform = 1;
725 if (s->avctx->channels == 2) {
726 chgroup->decorrelation_matrix[0] = 1.0;
727 chgroup->decorrelation_matrix[1] = -1.0;
728 chgroup->decorrelation_matrix[2] = 1.0;
729 chgroup->decorrelation_matrix[3] = 1.0;
732 chgroup->decorrelation_matrix[0] = 0.70703125;
733 chgroup->decorrelation_matrix[1] = -0.70703125;
734 chgroup->decorrelation_matrix[2] = 0.70703125;
735 chgroup->decorrelation_matrix[3] = 0.70703125;
738 } else if (chgroup->num_channels > 2) {
739 if (get_bits1(&s->gb)) {
740 chgroup->transform = 1;
741 if (get_bits1(&s->gb)) {
742 decode_decorrelation_matrix(s, chgroup);
744 /** FIXME: more than 6 coupled channels not supported */
745 if (chgroup->num_channels > 6) {
746 av_log_ask_for_sample(s->avctx,
747 "coupled channels > 6\n");
749 memcpy(chgroup->decorrelation_matrix,
750 default_decorrelation[chgroup->num_channels],
751 chgroup->num_channels * chgroup->num_channels *
752 sizeof(*chgroup->decorrelation_matrix));
758 /** decode transform on / off */
759 if (chgroup->transform) {
760 if (!get_bits1(&s->gb)) {
762 /** transform can be enabled for individual bands */
763 for (i = 0; i < s->num_bands; i++) {
764 chgroup->transform_band[i] = get_bits1(&s->gb);
767 memset(chgroup->transform_band, 1, s->num_bands);
770 remaining_channels -= chgroup->num_channels;
777 *@brief Extract the coefficients from the bitstream.
778 *@param s codec context
779 *@param c current channel number
780 *@return 0 on success, < 0 in case of bitstream errors
782 static int decode_coeffs(WMAProDecodeCtx *s, int c)
784 /* Integers 0..15 as single-precision floats. The table saves a
785 costly int to float conversion, and storing the values as
786 integers allows fast sign-flipping. */
787 static const uint32_t fval_tab[16] = {
788 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
789 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
790 0x41000000, 0x41100000, 0x41200000, 0x41300000,
791 0x41400000, 0x41500000, 0x41600000, 0x41700000,
795 WMAProChannelCtx* ci = &s->channel[c];
802 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
804 vlctable = get_bits1(&s->gb);
805 vlc = &coef_vlc[vlctable];
815 /** decode vector coefficients (consumes up to 167 bits per iteration for
816 4 vector coded large values) */
817 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
818 (cur_coeff + 3 < ci->num_vec_coeffs)) {
823 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
825 if (idx == HUFF_VEC4_SIZE - 1) {
826 for (i = 0; i < 4; i += 2) {
827 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
828 if (idx == HUFF_VEC2_SIZE - 1) {
830 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
831 if (v0 == HUFF_VEC1_SIZE - 1)
832 v0 += ff_wma_get_large_val(&s->gb);
833 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
834 if (v1 == HUFF_VEC1_SIZE - 1)
835 v1 += ff_wma_get_large_val(&s->gb);
836 vals[i ] = av_float2int(v0);
837 vals[i+1] = av_float2int(v1);
839 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
840 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
844 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
845 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
846 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
847 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
851 for (i = 0; i < 4; i++) {
853 uint32_t sign = get_bits1(&s->gb) - 1;
854 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
857 ci->coeffs[cur_coeff] = 0;
858 /** switch to run level mode when subframe_len / 128 zeros
859 were found in a row */
860 rl_mode |= (++num_zeros > s->subframe_len >> 8);
866 /** decode run level coded coefficients */
867 if (cur_coeff < s->subframe_len) {
868 memset(&ci->coeffs[cur_coeff], 0,
869 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
870 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
871 level, run, 1, ci->coeffs,
872 cur_coeff, s->subframe_len,
873 s->subframe_len, s->esc_len, 0))
874 return AVERROR_INVALIDDATA;
881 *@brief Extract scale factors from the bitstream.
882 *@param s codec context
883 *@return 0 on success, < 0 in case of bitstream errors
885 static int decode_scale_factors(WMAProDecodeCtx* s)
889 /** should never consume more than 5344 bits
890 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
893 for (i = 0; i < s->channels_for_cur_subframe; i++) {
894 int c = s->channel_indexes_for_cur_subframe[i];
897 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
898 sf_end = s->channel[c].scale_factors + s->num_bands;
900 /** resample scale factors for the new block size
901 * as the scale factors might need to be resampled several times
902 * before some new values are transmitted, a backup of the last
903 * transmitted scale factors is kept in saved_scale_factors
905 if (s->channel[c].reuse_sf) {
906 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
908 for (b = 0; b < s->num_bands; b++)
909 s->channel[c].scale_factors[b] =
910 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
913 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
915 if (!s->channel[c].reuse_sf) {
917 /** decode DPCM coded scale factors */
918 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
919 val = 45 / s->channel[c].scale_factor_step;
920 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
921 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
926 /** run level decode differences to the resampled factors */
927 for (i = 0; i < s->num_bands; i++) {
933 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
936 uint32_t code = get_bits(&s->gb, 14);
938 sign = (code & 1) - 1;
939 skip = (code & 0x3f) >> 1;
940 } else if (idx == 1) {
943 skip = scale_rl_run[idx];
944 val = scale_rl_level[idx];
945 sign = get_bits1(&s->gb)-1;
949 if (i >= s->num_bands) {
950 av_log(s->avctx, AV_LOG_ERROR,
951 "invalid scale factor coding\n");
952 return AVERROR_INVALIDDATA;
954 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
958 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
959 s->channel[c].table_idx = s->table_idx;
960 s->channel[c].reuse_sf = 1;
963 /** calculate new scale factor maximum */
964 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
965 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
966 s->channel[c].max_scale_factor =
967 FFMAX(s->channel[c].max_scale_factor, *sf);
975 *@brief Reconstruct the individual channel data.
976 *@param s codec context
978 static void inverse_channel_transform(WMAProDecodeCtx *s)
982 for (i = 0; i < s->num_chgroups; i++) {
983 if (s->chgroup[i].transform) {
984 float data[WMAPRO_MAX_CHANNELS];
985 const int num_channels = s->chgroup[i].num_channels;
986 float** ch_data = s->chgroup[i].channel_data;
987 float** ch_end = ch_data + num_channels;
988 const int8_t* tb = s->chgroup[i].transform_band;
991 /** multichannel decorrelation */
992 for (sfb = s->cur_sfb_offsets;
993 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
996 /** multiply values with the decorrelation_matrix */
997 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
998 const float* mat = s->chgroup[i].decorrelation_matrix;
999 const float* data_end = data + num_channels;
1000 float* data_ptr = data;
1003 for (ch = ch_data; ch < ch_end; ch++)
1004 *data_ptr++ = (*ch)[y];
1006 for (ch = ch_data; ch < ch_end; ch++) {
1009 while (data_ptr < data_end)
1010 sum += *data_ptr++ * *mat++;
1015 } else if (s->avctx->channels == 2) {
1016 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1017 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1018 ch_data[0] + sfb[0],
1020 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1021 ch_data[1] + sfb[0],
1030 *@brief Apply sine window and reconstruct the output buffer.
1031 *@param s codec context
1033 static void wmapro_window(WMAProDecodeCtx *s)
1036 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1037 int c = s->channel_indexes_for_cur_subframe[i];
1039 int winlen = s->channel[c].prev_block_len;
1040 float* start = s->channel[c].coeffs - (winlen >> 1);
1042 if (s->subframe_len < winlen) {
1043 start += (winlen - s->subframe_len) >> 1;
1044 winlen = s->subframe_len;
1047 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1051 s->dsp.vector_fmul_window(start, start, start + winlen,
1054 s->channel[c].prev_block_len = s->subframe_len;
1059 *@brief Decode a single subframe (block).
1060 *@param s codec context
1061 *@return 0 on success, < 0 when decoding failed
1063 static int decode_subframe(WMAProDecodeCtx *s)
1065 int offset = s->samples_per_frame;
1066 int subframe_len = s->samples_per_frame;
1068 int total_samples = s->samples_per_frame * s->avctx->channels;
1069 int transmit_coeffs = 0;
1070 int cur_subwoofer_cutoff;
1072 s->subframe_offset = get_bits_count(&s->gb);
1074 /** reset channel context and find the next block offset and size
1075 == the next block of the channel with the smallest number of
1078 for (i = 0; i < s->avctx->channels; i++) {
1079 s->channel[i].grouped = 0;
1080 if (offset > s->channel[i].decoded_samples) {
1081 offset = s->channel[i].decoded_samples;
1083 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1088 "processing subframe with offset %i len %i\n", offset, subframe_len);
1090 /** get a list of all channels that contain the estimated block */
1091 s->channels_for_cur_subframe = 0;
1092 for (i = 0; i < s->avctx->channels; i++) {
1093 const int cur_subframe = s->channel[i].cur_subframe;
1094 /** substract already processed samples */
1095 total_samples -= s->channel[i].decoded_samples;
1097 /** and count if there are multiple subframes that match our profile */
1098 if (offset == s->channel[i].decoded_samples &&
1099 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1100 total_samples -= s->channel[i].subframe_len[cur_subframe];
1101 s->channel[i].decoded_samples +=
1102 s->channel[i].subframe_len[cur_subframe];
1103 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1104 ++s->channels_for_cur_subframe;
1108 /** check if the frame will be complete after processing the
1111 s->parsed_all_subframes = 1;
1114 av_dlog(s->avctx, "subframe is part of %i channels\n",
1115 s->channels_for_cur_subframe);
1117 /** calculate number of scale factor bands and their offsets */
1118 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1119 s->num_bands = s->num_sfb[s->table_idx];
1120 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1121 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1123 /** configure the decoder for the current subframe */
1124 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1125 int c = s->channel_indexes_for_cur_subframe[i];
1127 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1131 s->subframe_len = subframe_len;
1132 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1134 /** skip extended header if any */
1135 if (get_bits1(&s->gb)) {
1137 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1138 int len = get_bits(&s->gb, 4);
1139 num_fill_bits = (len ? get_bits(&s->gb, len) : 0) + 1;
1142 if (num_fill_bits >= 0) {
1143 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1144 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1145 return AVERROR_INVALIDDATA;
1148 skip_bits_long(&s->gb, num_fill_bits);
1152 /** no idea for what the following bit is used */
1153 if (get_bits1(&s->gb)) {
1154 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1155 return AVERROR_INVALIDDATA;
1159 if (decode_channel_transform(s) < 0)
1160 return AVERROR_INVALIDDATA;
1163 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1164 int c = s->channel_indexes_for_cur_subframe[i];
1165 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1166 transmit_coeffs = 1;
1169 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1170 if (transmit_coeffs) {
1172 int quant_step = 90 * s->bits_per_sample >> 4;
1174 /** decode number of vector coded coefficients */
1175 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1176 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1177 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1178 int c = s->channel_indexes_for_cur_subframe[i];
1179 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1180 if (num_vec_coeffs > s->subframe_len) {
1181 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1182 return AVERROR_INVALIDDATA;
1184 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1187 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1188 int c = s->channel_indexes_for_cur_subframe[i];
1189 s->channel[c].num_vec_coeffs = s->subframe_len;
1192 /** decode quantization step */
1193 step = get_sbits(&s->gb, 6);
1195 if (step == -32 || step == 31) {
1196 const int sign = (step == 31) - 1;
1198 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1199 (step = get_bits(&s->gb, 5)) == 31) {
1202 quant_step += ((quant + step) ^ sign) - sign;
1204 if (quant_step < 0) {
1205 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1208 /** decode quantization step modifiers for every channel */
1210 if (s->channels_for_cur_subframe == 1) {
1211 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1213 int modifier_len = get_bits(&s->gb, 3);
1214 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1215 int c = s->channel_indexes_for_cur_subframe[i];
1216 s->channel[c].quant_step = quant_step;
1217 if (get_bits1(&s->gb)) {
1219 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1221 ++s->channel[c].quant_step;
1226 /** decode scale factors */
1227 if (decode_scale_factors(s) < 0)
1228 return AVERROR_INVALIDDATA;
1231 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1232 get_bits_count(&s->gb) - s->subframe_offset);
1234 /** parse coefficients */
1235 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1236 int c = s->channel_indexes_for_cur_subframe[i];
1237 if (s->channel[c].transmit_coefs &&
1238 get_bits_count(&s->gb) < s->num_saved_bits) {
1239 decode_coeffs(s, c);
1241 memset(s->channel[c].coeffs, 0,
1242 sizeof(*s->channel[c].coeffs) * subframe_len);
1245 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1246 get_bits_count(&s->gb) - s->subframe_offset);
1248 if (transmit_coeffs) {
1249 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1250 /** reconstruct the per channel data */
1251 inverse_channel_transform(s);
1252 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1253 int c = s->channel_indexes_for_cur_subframe[i];
1254 const int* sf = s->channel[c].scale_factors;
1257 if (c == s->lfe_channel)
1258 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1259 (subframe_len - cur_subwoofer_cutoff));
1261 /** inverse quantization and rescaling */
1262 for (b = 0; b < s->num_bands; b++) {
1263 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1264 const int exp = s->channel[c].quant_step -
1265 (s->channel[c].max_scale_factor - *sf++) *
1266 s->channel[c].scale_factor_step;
1267 const float quant = pow(10.0, exp / 20.0);
1268 int start = s->cur_sfb_offsets[b];
1269 s->dsp.vector_fmul_scalar(s->tmp + start,
1270 s->channel[c].coeffs + start,
1271 quant, end - start);
1274 /** apply imdct (imdct_half == DCTIV with reverse) */
1275 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1279 /** window and overlapp-add */
1282 /** handled one subframe */
1283 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1284 int c = s->channel_indexes_for_cur_subframe[i];
1285 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1286 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1287 return AVERROR_INVALIDDATA;
1289 ++s->channel[c].cur_subframe;
1296 *@brief Decode one WMA frame.
1297 *@param s codec context
1298 *@return 0 if the trailer bit indicates that this is the last frame,
1299 * 1 if there are additional frames
1301 static int decode_frame(WMAProDecodeCtx *s, int *got_frame_ptr)
1303 AVCodecContext *avctx = s->avctx;
1304 GetBitContext* gb = &s->gb;
1305 int more_frames = 0;
1309 /** get frame length */
1311 len = get_bits(gb, s->log2_frame_size);
1313 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1315 /** decode tile information */
1316 if (decode_tilehdr(s)) {
1321 /** read postproc transform */
1322 if (s->avctx->channels > 1 && get_bits1(gb)) {
1323 if (get_bits1(gb)) {
1324 for (i = 0; i < avctx->channels * avctx->channels; i++)
1329 /** read drc info */
1330 if (s->dynamic_range_compression) {
1331 s->drc_gain = get_bits(gb, 8);
1332 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1335 /** no idea what these are for, might be the number of samples
1336 that need to be skipped at the beginning or end of a stream */
1337 if (get_bits1(gb)) {
1340 /** usually true for the first frame */
1341 if (get_bits1(gb)) {
1342 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1343 av_dlog(s->avctx, "start skip: %i\n", skip);
1346 /** sometimes true for the last frame */
1347 if (get_bits1(gb)) {
1348 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1349 av_dlog(s->avctx, "end skip: %i\n", skip);
1354 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1355 get_bits_count(gb) - s->frame_offset);
1357 /** reset subframe states */
1358 s->parsed_all_subframes = 0;
1359 for (i = 0; i < avctx->channels; i++) {
1360 s->channel[i].decoded_samples = 0;
1361 s->channel[i].cur_subframe = 0;
1362 s->channel[i].reuse_sf = 0;
1365 /** decode all subframes */
1366 while (!s->parsed_all_subframes) {
1367 if (decode_subframe(s) < 0) {
1373 /* get output buffer */
1374 s->frame.nb_samples = s->samples_per_frame;
1375 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
1376 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1381 /** copy samples to the output buffer */
1382 for (i = 0; i < avctx->channels; i++)
1383 memcpy(s->frame.extended_data[i], s->channel[i].out,
1384 s->samples_per_frame * sizeof(*s->channel[i].out));
1386 for (i = 0; i < avctx->channels; i++) {
1387 /** reuse second half of the IMDCT output for the next frame */
1388 memcpy(&s->channel[i].out[0],
1389 &s->channel[i].out[s->samples_per_frame],
1390 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1393 if (s->skip_frame) {
1400 if (s->len_prefix) {
1401 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1402 /** FIXME: not sure if this is always an error */
1403 av_log(s->avctx, AV_LOG_ERROR,
1404 "frame[%i] would have to skip %i bits\n", s->frame_num,
1405 len - (get_bits_count(gb) - s->frame_offset) - 1);
1410 /** skip the rest of the frame data */
1411 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1413 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1417 /** decode trailer bit */
1418 more_frames = get_bits1(gb);
1425 *@brief Calculate remaining input buffer length.
1426 *@param s codec context
1427 *@param gb bitstream reader context
1428 *@return remaining size in bits
1430 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1432 return s->buf_bit_size - get_bits_count(gb);
1436 *@brief Fill the bit reservoir with a (partial) frame.
1437 *@param s codec context
1438 *@param gb bitstream reader context
1439 *@param len length of the partial frame
1440 *@param append decides whether to reset the buffer or not
1442 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1447 /** when the frame data does not need to be concatenated, the input buffer
1448 is reset and additional bits from the previous frame are copied
1449 and skipped later so that a fast byte copy is possible */
1452 s->frame_offset = get_bits_count(gb) & 7;
1453 s->num_saved_bits = s->frame_offset;
1454 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1457 buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1459 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1460 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1465 s->num_saved_bits += len;
1467 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1470 int align = 8 - (get_bits_count(gb) & 7);
1471 align = FFMIN(align, len);
1472 put_bits(&s->pb, align, get_bits(gb, align));
1474 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1476 skip_bits_long(gb, len);
1479 PutBitContext tmp = s->pb;
1480 flush_put_bits(&tmp);
1483 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1484 skip_bits(&s->gb, s->frame_offset);
1488 *@brief Decode a single WMA packet.
1489 *@param avctx codec context
1490 *@param data the output buffer
1491 *@param avpkt input packet
1492 *@return number of bytes that were read from the input buffer
1494 static int decode_packet(AVCodecContext *avctx, void *data,
1495 int *got_frame_ptr, AVPacket* avpkt)
1497 WMAProDecodeCtx *s = avctx->priv_data;
1498 GetBitContext* gb = &s->pgb;
1499 const uint8_t* buf = avpkt->data;
1500 int buf_size = avpkt->size;
1501 int num_bits_prev_frame;
1502 int packet_sequence_number;
1506 if (s->packet_done || s->packet_loss) {
1509 /** sanity check for the buffer length */
1510 if (buf_size < avctx->block_align)
1513 s->next_packet_start = buf_size - avctx->block_align;
1514 buf_size = avctx->block_align;
1515 s->buf_bit_size = buf_size << 3;
1517 /** parse packet header */
1518 init_get_bits(gb, buf, s->buf_bit_size);
1519 packet_sequence_number = get_bits(gb, 4);
1522 /** get number of bits that need to be added to the previous frame */
1523 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1524 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1525 num_bits_prev_frame);
1527 /** check for packet loss */
1528 if (!s->packet_loss &&
1529 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1531 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1532 s->packet_sequence_number, packet_sequence_number);
1534 s->packet_sequence_number = packet_sequence_number;
1536 if (num_bits_prev_frame > 0) {
1537 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1538 if (num_bits_prev_frame >= remaining_packet_bits) {
1539 num_bits_prev_frame = remaining_packet_bits;
1543 /** append the previous frame data to the remaining data from the
1544 previous packet to create a full frame */
1545 save_bits(s, gb, num_bits_prev_frame, 1);
1546 av_dlog(avctx, "accumulated %x bits of frame data\n",
1547 s->num_saved_bits - s->frame_offset);
1549 /** decode the cross packet frame if it is valid */
1550 if (!s->packet_loss)
1551 decode_frame(s, got_frame_ptr);
1552 } else if (s->num_saved_bits - s->frame_offset) {
1553 av_dlog(avctx, "ignoring %x previously saved bits\n",
1554 s->num_saved_bits - s->frame_offset);
1557 if (s->packet_loss) {
1558 /** reset number of saved bits so that the decoder
1559 does not start to decode incomplete frames in the
1560 s->len_prefix == 0 case */
1561 s->num_saved_bits = 0;
1567 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1568 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1569 skip_bits(gb, s->packet_offset);
1570 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1571 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1572 frame_size <= remaining_bits(s, gb)) {
1573 save_bits(s, gb, frame_size, 0);
1574 s->packet_done = !decode_frame(s, got_frame_ptr);
1575 } else if (!s->len_prefix
1576 && s->num_saved_bits > get_bits_count(&s->gb)) {
1577 /** when the frames do not have a length prefix, we don't know
1578 the compressed length of the individual frames
1579 however, we know what part of a new packet belongs to the
1581 therefore we save the incoming packet first, then we append
1582 the "previous frame" data from the next packet so that
1583 we get a buffer that only contains full frames */
1584 s->packet_done = !decode_frame(s, got_frame_ptr);
1589 if (s->packet_done && !s->packet_loss &&
1590 remaining_bits(s, gb) > 0) {
1591 /** save the rest of the data so that it can be decoded
1592 with the next packet */
1593 save_bits(s, gb, remaining_bits(s, gb), 0);
1596 s->packet_offset = get_bits_count(gb) & 7;
1598 return AVERROR_INVALIDDATA;
1601 *(AVFrame *)data = s->frame;
1603 return get_bits_count(gb) >> 3;
1607 *@brief Clear decoder buffers (for seeking).
1608 *@param avctx codec context
1610 static void flush(AVCodecContext *avctx)
1612 WMAProDecodeCtx *s = avctx->priv_data;
1614 /** reset output buffer as a part of it is used during the windowing of a
1616 for (i = 0; i < avctx->channels; i++)
1617 memset(s->channel[i].out, 0, s->samples_per_frame *
1618 sizeof(*s->channel[i].out));
1624 *@brief wmapro decoder
1626 AVCodec ff_wmapro_decoder = {
1628 .type = AVMEDIA_TYPE_AUDIO,
1629 .id = AV_CODEC_ID_WMAPRO,
1630 .priv_data_size = sizeof(WMAProDecodeCtx),
1631 .init = decode_init,
1632 .close = decode_end,
1633 .decode = decode_packet,
1634 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1636 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1637 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1638 AV_SAMPLE_FMT_NONE },