2 * Wmapro compatible decoder
3 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson
6 * This file is part of Libav.
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * @brief wmapro decoder implementation
26 * Wmapro is an MDCT based codec comparable to wma standard or AAC.
27 * The decoding therefore consists of the following steps:
28 * - bitstream decoding
29 * - reconstruction of per-channel data
30 * - rescaling and inverse quantization
32 * - windowing and overlapp-add
34 * The compressed wmapro bitstream is split into individual packets.
35 * Every such packet contains one or more wma frames.
36 * The compressed frames may have a variable length and frames may
37 * cross packet boundaries.
38 * Common to all wmapro frames is the number of samples that are stored in
40 * The number of samples and a few other decode flags are stored
41 * as extradata that has to be passed to the decoder.
43 * The wmapro frames themselves are again split into a variable number of
44 * subframes. Every subframe contains the data for 2^N time domain samples
45 * where N varies between 7 and 12.
47 * Example wmapro bitstream (in samples):
49 * || packet 0 || packet 1 || packet 2 packets
50 * ---------------------------------------------------
51 * || frame 0 || frame 1 || frame 2 || frames
52 * ---------------------------------------------------
53 * || | | || | | | || || subframes of channel 0
54 * ---------------------------------------------------
55 * || | | || | | | || || subframes of channel 1
56 * ---------------------------------------------------
58 * The frame layouts for the individual channels of a wma frame does not need
61 * However, if the offsets and lengths of several subframes of a frame are the
62 * same, the subframes of the channels can be grouped.
63 * Every group may then use special coding techniques like M/S stereo coding
64 * to improve the compression ratio. These channel transformations do not
65 * need to be applied to a whole subframe. Instead, they can also work on
66 * individual scale factor bands (see below).
67 * The coefficients that carry the audio signal in the frequency domain
68 * are transmitted as huffman-coded vectors with 4, 2 and 1 elements.
69 * In addition to that, the encoder can switch to a runlevel coding scheme
70 * by transmitting subframe_length / 128 zero coefficients.
72 * Before the audio signal can be converted to the time domain, the
73 * coefficients have to be rescaled and inverse quantized.
74 * A subframe is therefore split into several scale factor bands that get
75 * scaled individually.
76 * Scale factors are submitted for every frame but they might be shared
77 * between the subframes of a channel. Scale factors are initially DPCM-coded.
78 * Once scale factors are shared, the differences are transmitted as runlevel
80 * Every subframe length and offset combination in the frame layout shares a
81 * common quantization factor that can be adjusted for every channel by a
83 * After the inverse quantization, the coefficients get processed by an IMDCT.
84 * The resulting values are then windowed with a sine window and the first half
85 * of the values are added to the second half of the output from the previous
86 * subframe in order to reconstruct the output samples.
89 #include "libavutil/intreadwrite.h"
94 #include "wmaprodata.h"
96 #include "fmtconvert.h"
100 /** current decoder limitations */
101 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
102 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
103 #define MAX_BANDS 29 ///< max number of scale factor bands
104 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
106 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
107 #define WMAPRO_BLOCK_MAX_BITS 12 ///< log2 of max block size
108 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
109 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
113 #define SCALEVLCBITS 8
114 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
115 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
116 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
117 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
118 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
120 static VLC sf_vlc; ///< scale factor DPCM vlc
121 static VLC sf_rl_vlc; ///< scale factor run length vlc
122 static VLC vec4_vlc; ///< 4 coefficients per symbol
123 static VLC vec2_vlc; ///< 2 coefficients per symbol
124 static VLC vec1_vlc; ///< 1 coefficient per symbol
125 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
126 static float sin64[33]; ///< sinus table for decorrelation
129 * @brief frame specific decoder context for a single channel
132 int16_t prev_block_len; ///< length of the previous block
133 uint8_t transmit_coefs;
134 uint8_t num_subframes;
135 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
136 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
137 uint8_t cur_subframe; ///< current subframe number
138 uint16_t decoded_samples; ///< number of already processed samples
139 uint8_t grouped; ///< channel is part of a group
140 int quant_step; ///< quantization step for the current subframe
141 int8_t reuse_sf; ///< share scale factors between subframes
142 int8_t scale_factor_step; ///< scaling step for the current subframe
143 int max_scale_factor; ///< maximum scale factor for the current subframe
144 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
145 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
146 int* scale_factors; ///< pointer to the scale factor values used for decoding
147 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
148 float* coeffs; ///< pointer to the subframe decode buffer
149 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
150 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
154 * @brief channel group for channel transformations
157 uint8_t num_channels; ///< number of channels in the group
158 int8_t transform; ///< transform on / off
159 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
160 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
161 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
165 * @brief main decoder context
167 typedef struct WMAProDecodeCtx {
168 /* generic decoder variables */
169 AVCodecContext* avctx; ///< codec context for av_log
170 AVFrame frame; ///< AVFrame for decoded output
171 DSPContext dsp; ///< accelerated DSP functions
172 FmtConvertContext fmt_conv;
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 num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
188 int8_t lfe_channel; ///< lfe channel index
189 uint8_t max_num_subframes;
190 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
191 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
192 uint16_t min_samples_per_subframe;
193 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
194 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
195 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
196 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
198 /* packet decode state */
199 GetBitContext pgb; ///< bitstream reader context for the packet
200 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
201 uint8_t packet_offset; ///< frame offset in the packet
202 uint8_t packet_sequence_number; ///< current packet number
203 int num_saved_bits; ///< saved number of bits
204 int frame_offset; ///< frame offset in the bit reservoir
205 int subframe_offset; ///< subframe offset in the bit reservoir
206 uint8_t packet_loss; ///< set in case of bitstream error
207 uint8_t packet_done; ///< set when a packet is fully decoded
209 /* frame decode state */
210 uint32_t frame_num; ///< current frame number (not used for decoding)
211 GetBitContext gb; ///< bitstream reader context
212 int buf_bit_size; ///< buffer size in bits
213 uint8_t drc_gain; ///< gain for the DRC tool
214 int8_t skip_frame; ///< skip output step
215 int8_t parsed_all_subframes; ///< all subframes decoded?
217 /* subframe/block decode state */
218 int16_t subframe_len; ///< current subframe length
219 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
220 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
221 int8_t num_bands; ///< number of scale factor bands
222 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
223 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
224 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
225 int8_t esc_len; ///< length of escaped coefficients
227 uint8_t num_chgroups; ///< number of channel groups
228 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
230 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
235 *@brief helper function to print the most important members of the context
238 static void av_cold dump_context(WMAProDecodeCtx *s)
240 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
241 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
243 PRINT("ed sample bit depth", s->bits_per_sample);
244 PRINT_HEX("ed decode flags", s->decode_flags);
245 PRINT("samples per frame", s->samples_per_frame);
246 PRINT("log2 frame size", s->log2_frame_size);
247 PRINT("max num subframes", s->max_num_subframes);
248 PRINT("len prefix", s->len_prefix);
249 PRINT("num channels", s->num_channels);
253 *@brief Uninitialize the decoder and free all resources.
254 *@param avctx codec context
255 *@return 0 on success, < 0 otherwise
257 static av_cold int decode_end(AVCodecContext *avctx)
259 WMAProDecodeCtx *s = avctx->priv_data;
262 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
263 ff_mdct_end(&s->mdct_ctx[i]);
269 *@brief Initialize the decoder.
270 *@param avctx codec context
271 *@return 0 on success, -1 otherwise
273 static av_cold int decode_init(AVCodecContext *avctx)
275 WMAProDecodeCtx *s = avctx->priv_data;
276 uint8_t *edata_ptr = avctx->extradata;
277 unsigned int channel_mask;
279 int log2_max_num_subframes;
280 int num_possible_block_sizes;
283 dsputil_init(&s->dsp, avctx);
284 ff_fmt_convert_init(&s->fmt_conv, avctx);
285 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
287 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
289 if (avctx->extradata_size >= 18) {
290 s->decode_flags = AV_RL16(edata_ptr+14);
291 channel_mask = AV_RL32(edata_ptr+2);
292 s->bits_per_sample = AV_RL16(edata_ptr);
293 /** dump the extradata */
294 for (i = 0; i < avctx->extradata_size; i++)
295 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
296 av_dlog(avctx, "\n");
299 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
300 return AVERROR_INVALIDDATA;
304 s->log2_frame_size = av_log2(avctx->block_align) + 4;
307 s->skip_frame = 1; /* skip first frame */
309 s->len_prefix = (s->decode_flags & 0x40);
312 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
316 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
317 s->max_num_subframes = 1 << log2_max_num_subframes;
318 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
319 s->max_subframe_len_bit = 1;
320 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
322 num_possible_block_sizes = log2_max_num_subframes + 1;
323 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
324 s->dynamic_range_compression = (s->decode_flags & 0x80);
326 if (s->max_num_subframes > MAX_SUBFRAMES) {
327 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
328 s->max_num_subframes);
329 return AVERROR_INVALIDDATA;
332 s->num_channels = avctx->channels;
334 if (s->num_channels < 0) {
335 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
336 return AVERROR_INVALIDDATA;
337 } else if (s->num_channels > WMAPRO_MAX_CHANNELS) {
338 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
339 return AVERROR_PATCHWELCOME;
342 /** init previous block len */
343 for (i = 0; i < s->num_channels; i++)
344 s->channel[i].prev_block_len = s->samples_per_frame;
346 /** extract lfe channel position */
349 if (channel_mask & 8) {
351 for (mask = 1; mask < 16; mask <<= 1) {
352 if (channel_mask & mask)
357 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
358 scale_huffbits, 1, 1,
359 scale_huffcodes, 2, 2, 616);
361 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
362 scale_rl_huffbits, 1, 1,
363 scale_rl_huffcodes, 4, 4, 1406);
365 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
366 coef0_huffbits, 1, 1,
367 coef0_huffcodes, 4, 4, 2108);
369 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
370 coef1_huffbits, 1, 1,
371 coef1_huffcodes, 4, 4, 3912);
373 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
375 vec4_huffcodes, 2, 2, 604);
377 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
379 vec2_huffcodes, 2, 2, 562);
381 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
383 vec1_huffcodes, 2, 2, 562);
385 /** calculate number of scale factor bands and their offsets
386 for every possible block size */
387 for (i = 0; i < num_possible_block_sizes; i++) {
388 int subframe_len = s->samples_per_frame >> i;
392 s->sfb_offsets[i][0] = 0;
394 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
395 int offset = (subframe_len * 2 * critical_freq[x])
396 / s->avctx->sample_rate + 2;
398 if (offset > s->sfb_offsets[i][band - 1])
399 s->sfb_offsets[i][band++] = offset;
401 s->sfb_offsets[i][band - 1] = subframe_len;
402 s->num_sfb[i] = band - 1;
406 /** Scale factors can be shared between blocks of different size
407 as every block has a different scale factor band layout.
408 The matrix sf_offsets is needed to find the correct scale factor.
411 for (i = 0; i < num_possible_block_sizes; i++) {
413 for (b = 0; b < s->num_sfb[i]; b++) {
415 int offset = ((s->sfb_offsets[i][b]
416 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
417 for (x = 0; x < num_possible_block_sizes; x++) {
419 while (s->sfb_offsets[x][v + 1] << x < offset)
421 s->sf_offsets[i][x][b] = v;
426 /** init MDCT, FIXME: only init needed sizes */
427 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
428 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
429 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
430 / (1 << (s->bits_per_sample - 1)));
432 /** init MDCT windows: simple sinus window */
433 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
434 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
435 ff_init_ff_sine_windows(win_idx);
436 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
439 /** calculate subwoofer cutoff values */
440 for (i = 0; i < num_possible_block_sizes; i++) {
441 int block_size = s->samples_per_frame >> i;
442 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
443 / s->avctx->sample_rate;
444 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
447 /** calculate sine values for the decorrelation matrix */
448 for (i = 0; i < 33; i++)
449 sin64[i] = sin(i*M_PI / 64.0);
451 if (avctx->debug & FF_DEBUG_BITSTREAM)
454 avctx->channel_layout = channel_mask;
456 avcodec_get_frame_defaults(&s->frame);
457 avctx->coded_frame = &s->frame;
463 *@brief Decode the subframe length.
465 *@param offset sample offset in the frame
466 *@return decoded subframe length on success, < 0 in case of an error
468 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
470 int frame_len_shift = 0;
473 /** no need to read from the bitstream when only one length is possible */
474 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
475 return s->min_samples_per_subframe;
477 /** 1 bit indicates if the subframe is of maximum length */
478 if (s->max_subframe_len_bit) {
479 if (get_bits1(&s->gb))
480 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
482 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
484 subframe_len = s->samples_per_frame >> frame_len_shift;
486 /** sanity check the length */
487 if (subframe_len < s->min_samples_per_subframe ||
488 subframe_len > s->samples_per_frame) {
489 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
491 return AVERROR_INVALIDDATA;
497 *@brief Decode how the data in the frame is split into subframes.
498 * Every WMA frame contains the encoded data for a fixed number of
499 * samples per channel. The data for every channel might be split
500 * into several subframes. This function will reconstruct the list of
501 * subframes for every channel.
503 * If the subframes are not evenly split, the algorithm estimates the
504 * channels with the lowest number of total samples.
505 * Afterwards, for each of these channels a bit is read from the
506 * bitstream that indicates if the channel contains a subframe with the
507 * next subframe size that is going to be read from the bitstream or not.
508 * If a channel contains such a subframe, the subframe size gets added to
509 * the channel's subframe list.
510 * The algorithm repeats these steps until the frame is properly divided
511 * between the individual channels.
514 *@return 0 on success, < 0 in case of an error
516 static int decode_tilehdr(WMAProDecodeCtx *s)
518 uint16_t num_samples[WMAPRO_MAX_CHANNELS]; /**< sum of samples for all currently known subframes of a channel */
519 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
520 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
521 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
522 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
525 /* Should never consume more than 3073 bits (256 iterations for the
526 * while loop when always the minimum amount of 128 samples is substracted
527 * from missing samples in the 8 channel case).
528 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
531 /** reset tiling information */
532 for (c = 0; c < s->num_channels; c++)
533 s->channel[c].num_subframes = 0;
535 memset(num_samples, 0, sizeof(num_samples));
537 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
538 fixed_channel_layout = 1;
540 /** loop until the frame data is split between the subframes */
544 /** check which channels contain the subframe */
545 for (c = 0; c < s->num_channels; c++) {
546 if (num_samples[c] == min_channel_len) {
547 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
548 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
549 contains_subframe[c] = 1;
551 contains_subframe[c] = get_bits1(&s->gb);
553 contains_subframe[c] = 0;
556 /** get subframe length, subframe_len == 0 is not allowed */
557 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
558 return AVERROR_INVALIDDATA;
560 /** add subframes to the individual channels and find new min_channel_len */
561 min_channel_len += subframe_len;
562 for (c = 0; c < s->num_channels; c++) {
563 WMAProChannelCtx* chan = &s->channel[c];
565 if (contains_subframe[c]) {
566 if (chan->num_subframes >= MAX_SUBFRAMES) {
567 av_log(s->avctx, AV_LOG_ERROR,
568 "broken frame: num subframes > 31\n");
569 return AVERROR_INVALIDDATA;
571 chan->subframe_len[chan->num_subframes] = subframe_len;
572 num_samples[c] += subframe_len;
573 ++chan->num_subframes;
574 if (num_samples[c] > s->samples_per_frame) {
575 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
576 "channel len > samples_per_frame\n");
577 return AVERROR_INVALIDDATA;
579 } else if (num_samples[c] <= min_channel_len) {
580 if (num_samples[c] < min_channel_len) {
581 channels_for_cur_subframe = 0;
582 min_channel_len = num_samples[c];
584 ++channels_for_cur_subframe;
587 } while (min_channel_len < s->samples_per_frame);
589 for (c = 0; c < s->num_channels; c++) {
592 for (i = 0; i < s->channel[c].num_subframes; i++) {
593 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
594 " len %i\n", s->frame_num, c, i,
595 s->channel[c].subframe_len[i]);
596 s->channel[c].subframe_offset[i] = offset;
597 offset += s->channel[c].subframe_len[i];
605 *@brief Calculate a decorrelation matrix from the bitstream parameters.
606 *@param s codec context
607 *@param chgroup channel group for which the matrix needs to be calculated
609 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
610 WMAProChannelGrp *chgroup)
614 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
615 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
616 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
618 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
619 rotation_offset[i] = get_bits(&s->gb, 6);
621 for (i = 0; i < chgroup->num_channels; i++)
622 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
623 get_bits1(&s->gb) ? 1.0 : -1.0;
625 for (i = 1; i < chgroup->num_channels; i++) {
627 for (x = 0; x < i; x++) {
629 for (y = 0; y < i + 1; y++) {
630 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
631 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
632 int n = rotation_offset[offset + x];
638 cosv = sin64[32 - n];
640 sinv = sin64[64 - n];
641 cosv = -sin64[n - 32];
644 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
645 (v1 * sinv) - (v2 * cosv);
646 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
647 (v1 * cosv) + (v2 * sinv);
655 *@brief Decode channel transformation parameters
656 *@param s codec context
657 *@return 0 in case of success, < 0 in case of bitstream errors
659 static int decode_channel_transform(WMAProDecodeCtx* s)
662 /* should never consume more than 1921 bits for the 8 channel case
663 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
664 * + MAX_CHANNELS + MAX_BANDS + 1)
667 /** in the one channel case channel transforms are pointless */
669 if (s->num_channels > 1) {
670 int remaining_channels = s->channels_for_cur_subframe;
672 if (get_bits1(&s->gb)) {
673 av_log_ask_for_sample(s->avctx,
674 "unsupported channel transform bit\n");
675 return AVERROR_INVALIDDATA;
678 for (s->num_chgroups = 0; remaining_channels &&
679 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
680 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
681 float** channel_data = chgroup->channel_data;
682 chgroup->num_channels = 0;
683 chgroup->transform = 0;
685 /** decode channel mask */
686 if (remaining_channels > 2) {
687 for (i = 0; i < s->channels_for_cur_subframe; i++) {
688 int channel_idx = s->channel_indexes_for_cur_subframe[i];
689 if (!s->channel[channel_idx].grouped
690 && get_bits1(&s->gb)) {
691 ++chgroup->num_channels;
692 s->channel[channel_idx].grouped = 1;
693 *channel_data++ = s->channel[channel_idx].coeffs;
697 chgroup->num_channels = remaining_channels;
698 for (i = 0; i < s->channels_for_cur_subframe; i++) {
699 int channel_idx = s->channel_indexes_for_cur_subframe[i];
700 if (!s->channel[channel_idx].grouped)
701 *channel_data++ = s->channel[channel_idx].coeffs;
702 s->channel[channel_idx].grouped = 1;
706 /** decode transform type */
707 if (chgroup->num_channels == 2) {
708 if (get_bits1(&s->gb)) {
709 if (get_bits1(&s->gb)) {
710 av_log_ask_for_sample(s->avctx,
711 "unsupported channel transform type\n");
714 chgroup->transform = 1;
715 if (s->num_channels == 2) {
716 chgroup->decorrelation_matrix[0] = 1.0;
717 chgroup->decorrelation_matrix[1] = -1.0;
718 chgroup->decorrelation_matrix[2] = 1.0;
719 chgroup->decorrelation_matrix[3] = 1.0;
722 chgroup->decorrelation_matrix[0] = 0.70703125;
723 chgroup->decorrelation_matrix[1] = -0.70703125;
724 chgroup->decorrelation_matrix[2] = 0.70703125;
725 chgroup->decorrelation_matrix[3] = 0.70703125;
728 } else if (chgroup->num_channels > 2) {
729 if (get_bits1(&s->gb)) {
730 chgroup->transform = 1;
731 if (get_bits1(&s->gb)) {
732 decode_decorrelation_matrix(s, chgroup);
734 /** FIXME: more than 6 coupled channels not supported */
735 if (chgroup->num_channels > 6) {
736 av_log_ask_for_sample(s->avctx,
737 "coupled channels > 6\n");
739 memcpy(chgroup->decorrelation_matrix,
740 default_decorrelation[chgroup->num_channels],
741 chgroup->num_channels * chgroup->num_channels *
742 sizeof(*chgroup->decorrelation_matrix));
748 /** decode transform on / off */
749 if (chgroup->transform) {
750 if (!get_bits1(&s->gb)) {
752 /** transform can be enabled for individual bands */
753 for (i = 0; i < s->num_bands; i++) {
754 chgroup->transform_band[i] = get_bits1(&s->gb);
757 memset(chgroup->transform_band, 1, s->num_bands);
760 remaining_channels -= chgroup->num_channels;
767 *@brief Extract the coefficients from the bitstream.
768 *@param s codec context
769 *@param c current channel number
770 *@return 0 on success, < 0 in case of bitstream errors
772 static int decode_coeffs(WMAProDecodeCtx *s, int c)
774 /* Integers 0..15 as single-precision floats. The table saves a
775 costly int to float conversion, and storing the values as
776 integers allows fast sign-flipping. */
777 static const uint32_t fval_tab[16] = {
778 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
779 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
780 0x41000000, 0x41100000, 0x41200000, 0x41300000,
781 0x41400000, 0x41500000, 0x41600000, 0x41700000,
785 WMAProChannelCtx* ci = &s->channel[c];
792 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
794 vlctable = get_bits1(&s->gb);
795 vlc = &coef_vlc[vlctable];
805 /** decode vector coefficients (consumes up to 167 bits per iteration for
806 4 vector coded large values) */
807 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
808 (cur_coeff + 3 < ci->num_vec_coeffs)) {
813 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
815 if (idx == HUFF_VEC4_SIZE - 1) {
816 for (i = 0; i < 4; i += 2) {
817 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
818 if (idx == HUFF_VEC2_SIZE - 1) {
820 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
821 if (v0 == HUFF_VEC1_SIZE - 1)
822 v0 += ff_wma_get_large_val(&s->gb);
823 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
824 if (v1 == HUFF_VEC1_SIZE - 1)
825 v1 += ff_wma_get_large_val(&s->gb);
826 vals[i ] = ((av_alias32){ .f32 = v0 }).u32;
827 vals[i+1] = ((av_alias32){ .f32 = v1 }).u32;
829 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
830 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
834 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
835 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
836 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
837 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
841 for (i = 0; i < 4; i++) {
843 uint32_t sign = get_bits1(&s->gb) - 1;
844 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
847 ci->coeffs[cur_coeff] = 0;
848 /** switch to run level mode when subframe_len / 128 zeros
849 were found in a row */
850 rl_mode |= (++num_zeros > s->subframe_len >> 8);
856 /** decode run level coded coefficients */
857 if (cur_coeff < s->subframe_len) {
858 memset(&ci->coeffs[cur_coeff], 0,
859 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
860 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
861 level, run, 1, ci->coeffs,
862 cur_coeff, s->subframe_len,
863 s->subframe_len, s->esc_len, 0))
864 return AVERROR_INVALIDDATA;
871 *@brief Extract scale factors from the bitstream.
872 *@param s codec context
873 *@return 0 on success, < 0 in case of bitstream errors
875 static int decode_scale_factors(WMAProDecodeCtx* s)
879 /** should never consume more than 5344 bits
880 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
883 for (i = 0; i < s->channels_for_cur_subframe; i++) {
884 int c = s->channel_indexes_for_cur_subframe[i];
887 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
888 sf_end = s->channel[c].scale_factors + s->num_bands;
890 /** resample scale factors for the new block size
891 * as the scale factors might need to be resampled several times
892 * before some new values are transmitted, a backup of the last
893 * transmitted scale factors is kept in saved_scale_factors
895 if (s->channel[c].reuse_sf) {
896 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
898 for (b = 0; b < s->num_bands; b++)
899 s->channel[c].scale_factors[b] =
900 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
903 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
905 if (!s->channel[c].reuse_sf) {
907 /** decode DPCM coded scale factors */
908 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
909 val = 45 / s->channel[c].scale_factor_step;
910 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
911 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
916 /** run level decode differences to the resampled factors */
917 for (i = 0; i < s->num_bands; i++) {
923 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
926 uint32_t code = get_bits(&s->gb, 14);
928 sign = (code & 1) - 1;
929 skip = (code & 0x3f) >> 1;
930 } else if (idx == 1) {
933 skip = scale_rl_run[idx];
934 val = scale_rl_level[idx];
935 sign = get_bits1(&s->gb)-1;
939 if (i >= s->num_bands) {
940 av_log(s->avctx, AV_LOG_ERROR,
941 "invalid scale factor coding\n");
942 return AVERROR_INVALIDDATA;
944 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
948 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
949 s->channel[c].table_idx = s->table_idx;
950 s->channel[c].reuse_sf = 1;
953 /** calculate new scale factor maximum */
954 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
955 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
956 s->channel[c].max_scale_factor =
957 FFMAX(s->channel[c].max_scale_factor, *sf);
965 *@brief Reconstruct the individual channel data.
966 *@param s codec context
968 static void inverse_channel_transform(WMAProDecodeCtx *s)
972 for (i = 0; i < s->num_chgroups; i++) {
973 if (s->chgroup[i].transform) {
974 float data[WMAPRO_MAX_CHANNELS];
975 const int num_channels = s->chgroup[i].num_channels;
976 float** ch_data = s->chgroup[i].channel_data;
977 float** ch_end = ch_data + num_channels;
978 const int8_t* tb = s->chgroup[i].transform_band;
981 /** multichannel decorrelation */
982 for (sfb = s->cur_sfb_offsets;
983 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
986 /** multiply values with the decorrelation_matrix */
987 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
988 const float* mat = s->chgroup[i].decorrelation_matrix;
989 const float* data_end = data + num_channels;
990 float* data_ptr = data;
993 for (ch = ch_data; ch < ch_end; ch++)
994 *data_ptr++ = (*ch)[y];
996 for (ch = ch_data; ch < ch_end; ch++) {
999 while (data_ptr < data_end)
1000 sum += *data_ptr++ * *mat++;
1005 } else if (s->num_channels == 2) {
1006 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1007 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1008 ch_data[0] + sfb[0],
1010 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1011 ch_data[1] + sfb[0],
1020 *@brief Apply sine window and reconstruct the output buffer.
1021 *@param s codec context
1023 static void wmapro_window(WMAProDecodeCtx *s)
1026 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1027 int c = s->channel_indexes_for_cur_subframe[i];
1029 int winlen = s->channel[c].prev_block_len;
1030 float* start = s->channel[c].coeffs - (winlen >> 1);
1032 if (s->subframe_len < winlen) {
1033 start += (winlen - s->subframe_len) >> 1;
1034 winlen = s->subframe_len;
1037 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1041 s->dsp.vector_fmul_window(start, start, start + winlen,
1044 s->channel[c].prev_block_len = s->subframe_len;
1049 *@brief Decode a single subframe (block).
1050 *@param s codec context
1051 *@return 0 on success, < 0 when decoding failed
1053 static int decode_subframe(WMAProDecodeCtx *s)
1055 int offset = s->samples_per_frame;
1056 int subframe_len = s->samples_per_frame;
1058 int total_samples = s->samples_per_frame * s->num_channels;
1059 int transmit_coeffs = 0;
1060 int cur_subwoofer_cutoff;
1062 s->subframe_offset = get_bits_count(&s->gb);
1064 /** reset channel context and find the next block offset and size
1065 == the next block of the channel with the smallest number of
1068 for (i = 0; i < s->num_channels; i++) {
1069 s->channel[i].grouped = 0;
1070 if (offset > s->channel[i].decoded_samples) {
1071 offset = s->channel[i].decoded_samples;
1073 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1078 "processing subframe with offset %i len %i\n", offset, subframe_len);
1080 /** get a list of all channels that contain the estimated block */
1081 s->channels_for_cur_subframe = 0;
1082 for (i = 0; i < s->num_channels; i++) {
1083 const int cur_subframe = s->channel[i].cur_subframe;
1084 /** substract already processed samples */
1085 total_samples -= s->channel[i].decoded_samples;
1087 /** and count if there are multiple subframes that match our profile */
1088 if (offset == s->channel[i].decoded_samples &&
1089 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1090 total_samples -= s->channel[i].subframe_len[cur_subframe];
1091 s->channel[i].decoded_samples +=
1092 s->channel[i].subframe_len[cur_subframe];
1093 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1094 ++s->channels_for_cur_subframe;
1098 /** check if the frame will be complete after processing the
1101 s->parsed_all_subframes = 1;
1104 av_dlog(s->avctx, "subframe is part of %i channels\n",
1105 s->channels_for_cur_subframe);
1107 /** calculate number of scale factor bands and their offsets */
1108 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1109 s->num_bands = s->num_sfb[s->table_idx];
1110 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1111 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1113 /** configure the decoder for the current subframe */
1114 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1115 int c = s->channel_indexes_for_cur_subframe[i];
1117 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1121 s->subframe_len = subframe_len;
1122 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1124 /** skip extended header if any */
1125 if (get_bits1(&s->gb)) {
1127 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1128 int len = get_bits(&s->gb, 4);
1129 num_fill_bits = get_bits(&s->gb, len) + 1;
1132 if (num_fill_bits >= 0) {
1133 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1134 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1135 return AVERROR_INVALIDDATA;
1138 skip_bits_long(&s->gb, num_fill_bits);
1142 /** no idea for what the following bit is used */
1143 if (get_bits1(&s->gb)) {
1144 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1145 return AVERROR_INVALIDDATA;
1149 if (decode_channel_transform(s) < 0)
1150 return AVERROR_INVALIDDATA;
1153 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1154 int c = s->channel_indexes_for_cur_subframe[i];
1155 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1156 transmit_coeffs = 1;
1159 if (transmit_coeffs) {
1161 int quant_step = 90 * s->bits_per_sample >> 4;
1163 /** decode number of vector coded coefficients */
1164 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1165 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1166 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1167 int c = s->channel_indexes_for_cur_subframe[i];
1168 s->channel[c].num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1171 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1172 int c = s->channel_indexes_for_cur_subframe[i];
1173 s->channel[c].num_vec_coeffs = s->subframe_len;
1176 /** decode quantization step */
1177 step = get_sbits(&s->gb, 6);
1179 if (step == -32 || step == 31) {
1180 const int sign = (step == 31) - 1;
1182 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1183 (step = get_bits(&s->gb, 5)) == 31) {
1186 quant_step += ((quant + step) ^ sign) - sign;
1188 if (quant_step < 0) {
1189 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1192 /** decode quantization step modifiers for every channel */
1194 if (s->channels_for_cur_subframe == 1) {
1195 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1197 int modifier_len = get_bits(&s->gb, 3);
1198 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1199 int c = s->channel_indexes_for_cur_subframe[i];
1200 s->channel[c].quant_step = quant_step;
1201 if (get_bits1(&s->gb)) {
1203 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1205 ++s->channel[c].quant_step;
1210 /** decode scale factors */
1211 if (decode_scale_factors(s) < 0)
1212 return AVERROR_INVALIDDATA;
1215 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1216 get_bits_count(&s->gb) - s->subframe_offset);
1218 /** parse coefficients */
1219 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1220 int c = s->channel_indexes_for_cur_subframe[i];
1221 if (s->channel[c].transmit_coefs &&
1222 get_bits_count(&s->gb) < s->num_saved_bits) {
1223 decode_coeffs(s, c);
1225 memset(s->channel[c].coeffs, 0,
1226 sizeof(*s->channel[c].coeffs) * subframe_len);
1229 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1230 get_bits_count(&s->gb) - s->subframe_offset);
1232 if (transmit_coeffs) {
1233 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1234 /** reconstruct the per channel data */
1235 inverse_channel_transform(s);
1236 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1237 int c = s->channel_indexes_for_cur_subframe[i];
1238 const int* sf = s->channel[c].scale_factors;
1241 if (c == s->lfe_channel)
1242 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1243 (subframe_len - cur_subwoofer_cutoff));
1245 /** inverse quantization and rescaling */
1246 for (b = 0; b < s->num_bands; b++) {
1247 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1248 const int exp = s->channel[c].quant_step -
1249 (s->channel[c].max_scale_factor - *sf++) *
1250 s->channel[c].scale_factor_step;
1251 const float quant = pow(10.0, exp / 20.0);
1252 int start = s->cur_sfb_offsets[b];
1253 s->dsp.vector_fmul_scalar(s->tmp + start,
1254 s->channel[c].coeffs + start,
1255 quant, end - start);
1258 /** apply imdct (imdct_half == DCTIV with reverse) */
1259 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1263 /** window and overlapp-add */
1266 /** handled one subframe */
1267 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1268 int c = s->channel_indexes_for_cur_subframe[i];
1269 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1270 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1271 return AVERROR_INVALIDDATA;
1273 ++s->channel[c].cur_subframe;
1280 *@brief Decode one WMA frame.
1281 *@param s codec context
1282 *@return 0 if the trailer bit indicates that this is the last frame,
1283 * 1 if there are additional frames
1285 static int decode_frame(WMAProDecodeCtx *s, int *got_frame_ptr)
1287 AVCodecContext *avctx = s->avctx;
1288 GetBitContext* gb = &s->gb;
1289 int more_frames = 0;
1292 const float *out_ptr[WMAPRO_MAX_CHANNELS];
1295 /** get frame length */
1297 len = get_bits(gb, s->log2_frame_size);
1299 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1301 /** decode tile information */
1302 if (decode_tilehdr(s)) {
1307 /** read postproc transform */
1308 if (s->num_channels > 1 && get_bits1(gb)) {
1309 if (get_bits1(gb)) {
1310 for (i = 0; i < s->num_channels * s->num_channels; i++)
1315 /** read drc info */
1316 if (s->dynamic_range_compression) {
1317 s->drc_gain = get_bits(gb, 8);
1318 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1321 /** no idea what these are for, might be the number of samples
1322 that need to be skipped at the beginning or end of a stream */
1323 if (get_bits1(gb)) {
1326 /** usually true for the first frame */
1327 if (get_bits1(gb)) {
1328 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1329 av_dlog(s->avctx, "start skip: %i\n", skip);
1332 /** sometimes true for the last frame */
1333 if (get_bits1(gb)) {
1334 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1335 av_dlog(s->avctx, "end skip: %i\n", skip);
1340 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1341 get_bits_count(gb) - s->frame_offset);
1343 /** reset subframe states */
1344 s->parsed_all_subframes = 0;
1345 for (i = 0; i < s->num_channels; i++) {
1346 s->channel[i].decoded_samples = 0;
1347 s->channel[i].cur_subframe = 0;
1348 s->channel[i].reuse_sf = 0;
1351 /** decode all subframes */
1352 while (!s->parsed_all_subframes) {
1353 if (decode_subframe(s) < 0) {
1359 /* get output buffer */
1360 s->frame.nb_samples = s->samples_per_frame;
1361 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
1362 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1366 samples = (float *)s->frame.data[0];
1368 /** interleave samples and write them to the output buffer */
1369 for (i = 0; i < s->num_channels; i++)
1370 out_ptr[i] = s->channel[i].out;
1371 s->fmt_conv.float_interleave(samples, out_ptr, s->samples_per_frame,
1374 for (i = 0; i < s->num_channels; i++) {
1375 /** reuse second half of the IMDCT output for the next frame */
1376 memcpy(&s->channel[i].out[0],
1377 &s->channel[i].out[s->samples_per_frame],
1378 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1381 if (s->skip_frame) {
1388 if (s->len_prefix) {
1389 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1390 /** FIXME: not sure if this is always an error */
1391 av_log(s->avctx, AV_LOG_ERROR,
1392 "frame[%i] would have to skip %i bits\n", s->frame_num,
1393 len - (get_bits_count(gb) - s->frame_offset) - 1);
1398 /** skip the rest of the frame data */
1399 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1401 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1405 /** decode trailer bit */
1406 more_frames = get_bits1(gb);
1413 *@brief Calculate remaining input buffer length.
1414 *@param s codec context
1415 *@param gb bitstream reader context
1416 *@return remaining size in bits
1418 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1420 return s->buf_bit_size - get_bits_count(gb);
1424 *@brief Fill the bit reservoir with a (partial) frame.
1425 *@param s codec context
1426 *@param gb bitstream reader context
1427 *@param len length of the partial frame
1428 *@param append decides wether to reset the buffer or not
1430 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1435 /** when the frame data does not need to be concatenated, the input buffer
1436 is resetted and additional bits from the previous frame are copyed
1437 and skipped later so that a fast byte copy is possible */
1440 s->frame_offset = get_bits_count(gb) & 7;
1441 s->num_saved_bits = s->frame_offset;
1442 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1445 buflen = (s->num_saved_bits + len + 8) >> 3;
1447 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1448 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1453 s->num_saved_bits += len;
1455 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1458 int align = 8 - (get_bits_count(gb) & 7);
1459 align = FFMIN(align, len);
1460 put_bits(&s->pb, align, get_bits(gb, align));
1462 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1464 skip_bits_long(gb, len);
1467 PutBitContext tmp = s->pb;
1468 flush_put_bits(&tmp);
1471 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1472 skip_bits(&s->gb, s->frame_offset);
1476 *@brief Decode a single WMA packet.
1477 *@param avctx codec context
1478 *@param data the output buffer
1479 *@param data_size number of bytes that were written to the output buffer
1480 *@param avpkt input packet
1481 *@return number of bytes that were read from the input buffer
1483 static int decode_packet(AVCodecContext *avctx, void *data,
1484 int *got_frame_ptr, AVPacket* avpkt)
1486 WMAProDecodeCtx *s = avctx->priv_data;
1487 GetBitContext* gb = &s->pgb;
1488 const uint8_t* buf = avpkt->data;
1489 int buf_size = avpkt->size;
1490 int num_bits_prev_frame;
1491 int packet_sequence_number;
1495 if (s->packet_done || s->packet_loss) {
1498 /** sanity check for the buffer length */
1499 if (buf_size < avctx->block_align)
1502 s->next_packet_start = buf_size - avctx->block_align;
1503 buf_size = avctx->block_align;
1504 s->buf_bit_size = buf_size << 3;
1506 /** parse packet header */
1507 init_get_bits(gb, buf, s->buf_bit_size);
1508 packet_sequence_number = get_bits(gb, 4);
1511 /** get number of bits that need to be added to the previous frame */
1512 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1513 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1514 num_bits_prev_frame);
1516 /** check for packet loss */
1517 if (!s->packet_loss &&
1518 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1520 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1521 s->packet_sequence_number, packet_sequence_number);
1523 s->packet_sequence_number = packet_sequence_number;
1525 if (num_bits_prev_frame > 0) {
1526 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1527 if (num_bits_prev_frame >= remaining_packet_bits) {
1528 num_bits_prev_frame = remaining_packet_bits;
1532 /** append the previous frame data to the remaining data from the
1533 previous packet to create a full frame */
1534 save_bits(s, gb, num_bits_prev_frame, 1);
1535 av_dlog(avctx, "accumulated %x bits of frame data\n",
1536 s->num_saved_bits - s->frame_offset);
1538 /** decode the cross packet frame if it is valid */
1539 if (!s->packet_loss)
1540 decode_frame(s, got_frame_ptr);
1541 } else if (s->num_saved_bits - s->frame_offset) {
1542 av_dlog(avctx, "ignoring %x previously saved bits\n",
1543 s->num_saved_bits - s->frame_offset);
1546 if (s->packet_loss) {
1547 /** reset number of saved bits so that the decoder
1548 does not start to decode incomplete frames in the
1549 s->len_prefix == 0 case */
1550 s->num_saved_bits = 0;
1556 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1557 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1558 skip_bits(gb, s->packet_offset);
1559 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1560 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1561 frame_size <= remaining_bits(s, gb)) {
1562 save_bits(s, gb, frame_size, 0);
1563 s->packet_done = !decode_frame(s, got_frame_ptr);
1564 } else if (!s->len_prefix
1565 && s->num_saved_bits > get_bits_count(&s->gb)) {
1566 /** when the frames do not have a length prefix, we don't know
1567 the compressed length of the individual frames
1568 however, we know what part of a new packet belongs to the
1570 therefore we save the incoming packet first, then we append
1571 the "previous frame" data from the next packet so that
1572 we get a buffer that only contains full frames */
1573 s->packet_done = !decode_frame(s, got_frame_ptr);
1578 if (s->packet_done && !s->packet_loss &&
1579 remaining_bits(s, gb) > 0) {
1580 /** save the rest of the data so that it can be decoded
1581 with the next packet */
1582 save_bits(s, gb, remaining_bits(s, gb), 0);
1585 s->packet_offset = get_bits_count(gb) & 7;
1587 return AVERROR_INVALIDDATA;
1590 *(AVFrame *)data = s->frame;
1592 return get_bits_count(gb) >> 3;
1596 *@brief Clear decoder buffers (for seeking).
1597 *@param avctx codec context
1599 static void flush(AVCodecContext *avctx)
1601 WMAProDecodeCtx *s = avctx->priv_data;
1603 /** reset output buffer as a part of it is used during the windowing of a
1605 for (i = 0; i < s->num_channels; i++)
1606 memset(s->channel[i].out, 0, s->samples_per_frame *
1607 sizeof(*s->channel[i].out));
1613 *@brief wmapro decoder
1615 AVCodec ff_wmapro_decoder = {
1617 .type = AVMEDIA_TYPE_AUDIO,
1618 .id = CODEC_ID_WMAPRO,
1619 .priv_data_size = sizeof(WMAProDecodeCtx),
1620 .init = decode_init,
1621 .close = decode_end,
1622 .decode = decode_packet,
1623 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1625 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),