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/intfloat.h"
90 #include "libavutil/intreadwrite.h"
95 #include "wmaprodata.h"
97 #include "fmtconvert.h"
100 #include "wma_common.h"
102 /** current decoder limitations */
103 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
104 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
105 #define MAX_BANDS 29 ///< max number of scale factor bands
106 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
108 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
109 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
110 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
111 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
115 #define SCALEVLCBITS 8
116 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
117 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
118 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
119 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
120 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
122 static VLC sf_vlc; ///< scale factor DPCM vlc
123 static VLC sf_rl_vlc; ///< scale factor run length vlc
124 static VLC vec4_vlc; ///< 4 coefficients per symbol
125 static VLC vec2_vlc; ///< 2 coefficients per symbol
126 static VLC vec1_vlc; ///< 1 coefficient per symbol
127 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
128 static float sin64[33]; ///< sinus table for decorrelation
131 * @brief frame specific decoder context for a single channel
134 int16_t prev_block_len; ///< length of the previous block
135 uint8_t transmit_coefs;
136 uint8_t num_subframes;
137 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
138 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
139 uint8_t cur_subframe; ///< current subframe number
140 uint16_t decoded_samples; ///< number of already processed samples
141 uint8_t grouped; ///< channel is part of a group
142 int quant_step; ///< quantization step for the current subframe
143 int8_t reuse_sf; ///< share scale factors between subframes
144 int8_t scale_factor_step; ///< scaling step for the current subframe
145 int max_scale_factor; ///< maximum scale factor for the current subframe
146 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
147 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
148 int* scale_factors; ///< pointer to the scale factor values used for decoding
149 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
150 float* coeffs; ///< pointer to the subframe decode buffer
151 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
152 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
156 * @brief channel group for channel transformations
159 uint8_t num_channels; ///< number of channels in the group
160 int8_t transform; ///< transform on / off
161 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
162 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
163 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
167 * @brief main decoder context
169 typedef struct WMAProDecodeCtx {
170 /* generic decoder variables */
171 AVCodecContext* avctx; ///< codec context for av_log
172 AVFrame frame; ///< AVFrame for decoded output
173 DSPContext dsp; ///< accelerated DSP functions
174 FmtConvertContext fmt_conv;
175 uint8_t frame_data[MAX_FRAMESIZE +
176 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
177 PutBitContext pb; ///< context for filling the frame_data buffer
178 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
179 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
180 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
182 /* frame size dependent frame information (set during initialization) */
183 uint32_t decode_flags; ///< used compression features
184 uint8_t len_prefix; ///< frame is prefixed with its length
185 uint8_t dynamic_range_compression; ///< frame contains DRC data
186 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
187 uint16_t samples_per_frame; ///< number of samples to output
188 uint16_t log2_frame_size;
189 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
190 int8_t lfe_channel; ///< lfe channel index
191 uint8_t max_num_subframes;
192 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
193 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
194 uint16_t min_samples_per_subframe;
195 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
196 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
197 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
198 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
200 /* packet decode state */
201 GetBitContext pgb; ///< bitstream reader context for the packet
202 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
203 uint8_t packet_offset; ///< frame offset in the packet
204 uint8_t packet_sequence_number; ///< current packet number
205 int num_saved_bits; ///< saved number of bits
206 int frame_offset; ///< frame offset in the bit reservoir
207 int subframe_offset; ///< subframe offset in the bit reservoir
208 uint8_t packet_loss; ///< set in case of bitstream error
209 uint8_t packet_done; ///< set when a packet is fully decoded
211 /* frame decode state */
212 uint32_t frame_num; ///< current frame number (not used for decoding)
213 GetBitContext gb; ///< bitstream reader context
214 int buf_bit_size; ///< buffer size in bits
215 uint8_t drc_gain; ///< gain for the DRC tool
216 int8_t skip_frame; ///< skip output step
217 int8_t parsed_all_subframes; ///< all subframes decoded?
219 /* subframe/block decode state */
220 int16_t subframe_len; ///< current subframe length
221 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
222 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
223 int8_t num_bands; ///< number of scale factor bands
224 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
225 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
226 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
227 int8_t esc_len; ///< length of escaped coefficients
229 uint8_t num_chgroups; ///< number of channel groups
230 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
232 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
237 *@brief helper function to print the most important members of the context
240 static av_cold void dump_context(WMAProDecodeCtx *s)
242 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
243 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
245 PRINT("ed sample bit depth", s->bits_per_sample);
246 PRINT_HEX("ed decode flags", s->decode_flags);
247 PRINT("samples per frame", s->samples_per_frame);
248 PRINT("log2 frame size", s->log2_frame_size);
249 PRINT("max num subframes", s->max_num_subframes);
250 PRINT("len prefix", s->len_prefix);
251 PRINT("num channels", s->num_channels);
255 *@brief Uninitialize the decoder and free all resources.
256 *@param avctx codec context
257 *@return 0 on success, < 0 otherwise
259 static av_cold int decode_end(AVCodecContext *avctx)
261 WMAProDecodeCtx *s = avctx->priv_data;
264 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
265 ff_mdct_end(&s->mdct_ctx[i]);
271 *@brief Initialize the decoder.
272 *@param avctx codec context
273 *@return 0 on success, -1 otherwise
275 static av_cold int decode_init(AVCodecContext *avctx)
277 WMAProDecodeCtx *s = avctx->priv_data;
278 uint8_t *edata_ptr = avctx->extradata;
279 unsigned int channel_mask;
281 int log2_max_num_subframes;
282 int num_possible_block_sizes;
285 ff_dsputil_init(&s->dsp, avctx);
286 ff_fmt_convert_init(&s->fmt_conv, avctx);
287 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
289 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
291 if (avctx->extradata_size >= 18) {
292 s->decode_flags = AV_RL16(edata_ptr+14);
293 channel_mask = AV_RL32(edata_ptr+2);
294 s->bits_per_sample = AV_RL16(edata_ptr);
295 /** dump the extradata */
296 for (i = 0; i < avctx->extradata_size; i++)
297 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
298 av_dlog(avctx, "\n");
301 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
302 return AVERROR_INVALIDDATA;
306 s->log2_frame_size = av_log2(avctx->block_align) + 4;
309 s->skip_frame = 1; /* skip first frame */
311 s->len_prefix = (s->decode_flags & 0x40);
314 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
315 if (bits > WMAPRO_BLOCK_MAX_BITS) {
316 av_log_missing_feature(avctx, "14-bits block sizes", 1);
317 return AVERROR_INVALIDDATA;
319 s->samples_per_frame = 1 << bits;
322 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
323 s->max_num_subframes = 1 << log2_max_num_subframes;
324 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
325 s->max_subframe_len_bit = 1;
326 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
328 num_possible_block_sizes = log2_max_num_subframes + 1;
329 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
330 s->dynamic_range_compression = (s->decode_flags & 0x80);
332 if (s->max_num_subframes > MAX_SUBFRAMES) {
333 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
334 s->max_num_subframes);
335 return AVERROR_INVALIDDATA;
338 if (s->avctx->sample_rate <= 0) {
339 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
340 return AVERROR_INVALIDDATA;
343 s->num_channels = avctx->channels;
345 if (s->num_channels < 0) {
346 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
347 return AVERROR_INVALIDDATA;
348 } else if (s->num_channels > WMAPRO_MAX_CHANNELS) {
349 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
350 return AVERROR_PATCHWELCOME;
353 /** init previous block len */
354 for (i = 0; i < s->num_channels; i++)
355 s->channel[i].prev_block_len = s->samples_per_frame;
357 /** extract lfe channel position */
360 if (channel_mask & 8) {
362 for (mask = 1; mask < 16; mask <<= 1) {
363 if (channel_mask & mask)
368 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
369 scale_huffbits, 1, 1,
370 scale_huffcodes, 2, 2, 616);
372 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
373 scale_rl_huffbits, 1, 1,
374 scale_rl_huffcodes, 4, 4, 1406);
376 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
377 coef0_huffbits, 1, 1,
378 coef0_huffcodes, 4, 4, 2108);
380 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
381 coef1_huffbits, 1, 1,
382 coef1_huffcodes, 4, 4, 3912);
384 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
386 vec4_huffcodes, 2, 2, 604);
388 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
390 vec2_huffcodes, 2, 2, 562);
392 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
394 vec1_huffcodes, 2, 2, 562);
396 /** calculate number of scale factor bands and their offsets
397 for every possible block size */
398 for (i = 0; i < num_possible_block_sizes; i++) {
399 int subframe_len = s->samples_per_frame >> i;
403 s->sfb_offsets[i][0] = 0;
405 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
406 int offset = (subframe_len * 2 * critical_freq[x])
407 / s->avctx->sample_rate + 2;
409 if (offset > s->sfb_offsets[i][band - 1])
410 s->sfb_offsets[i][band++] = offset;
412 s->sfb_offsets[i][band - 1] = subframe_len;
413 s->num_sfb[i] = band - 1;
417 /** Scale factors can be shared between blocks of different size
418 as every block has a different scale factor band layout.
419 The matrix sf_offsets is needed to find the correct scale factor.
422 for (i = 0; i < num_possible_block_sizes; i++) {
424 for (b = 0; b < s->num_sfb[i]; b++) {
426 int offset = ((s->sfb_offsets[i][b]
427 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
428 for (x = 0; x < num_possible_block_sizes; x++) {
430 while (s->sfb_offsets[x][v + 1] << x < offset)
432 s->sf_offsets[i][x][b] = v;
437 /** init MDCT, FIXME: only init needed sizes */
438 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
439 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
440 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
441 / (1 << (s->bits_per_sample - 1)));
443 /** init MDCT windows: simple sinus window */
444 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
445 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
446 ff_init_ff_sine_windows(win_idx);
447 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
450 /** calculate subwoofer cutoff values */
451 for (i = 0; i < num_possible_block_sizes; i++) {
452 int block_size = s->samples_per_frame >> i;
453 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
454 / s->avctx->sample_rate;
455 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
458 /** calculate sine values for the decorrelation matrix */
459 for (i = 0; i < 33; i++)
460 sin64[i] = sin(i*M_PI / 64.0);
462 if (avctx->debug & FF_DEBUG_BITSTREAM)
465 avctx->channel_layout = channel_mask;
467 avcodec_get_frame_defaults(&s->frame);
468 avctx->coded_frame = &s->frame;
474 *@brief Decode the subframe length.
476 *@param offset sample offset in the frame
477 *@return decoded subframe length on success, < 0 in case of an error
479 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
481 int frame_len_shift = 0;
484 /** no need to read from the bitstream when only one length is possible */
485 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
486 return s->min_samples_per_subframe;
488 /** 1 bit indicates if the subframe is of maximum length */
489 if (s->max_subframe_len_bit) {
490 if (get_bits1(&s->gb))
491 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
493 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
495 subframe_len = s->samples_per_frame >> frame_len_shift;
497 /** sanity check the length */
498 if (subframe_len < s->min_samples_per_subframe ||
499 subframe_len > s->samples_per_frame) {
500 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
502 return AVERROR_INVALIDDATA;
508 *@brief Decode how the data in the frame is split into subframes.
509 * Every WMA frame contains the encoded data for a fixed number of
510 * samples per channel. The data for every channel might be split
511 * into several subframes. This function will reconstruct the list of
512 * subframes for every channel.
514 * If the subframes are not evenly split, the algorithm estimates the
515 * channels with the lowest number of total samples.
516 * Afterwards, for each of these channels a bit is read from the
517 * bitstream that indicates if the channel contains a subframe with the
518 * next subframe size that is going to be read from the bitstream or not.
519 * If a channel contains such a subframe, the subframe size gets added to
520 * the channel's subframe list.
521 * The algorithm repeats these steps until the frame is properly divided
522 * between the individual channels.
525 *@return 0 on success, < 0 in case of an error
527 static int decode_tilehdr(WMAProDecodeCtx *s)
529 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
530 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
531 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
532 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
533 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
536 /* Should never consume more than 3073 bits (256 iterations for the
537 * while loop when always the minimum amount of 128 samples is substracted
538 * from missing samples in the 8 channel case).
539 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
542 /** reset tiling information */
543 for (c = 0; c < s->num_channels; c++)
544 s->channel[c].num_subframes = 0;
546 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
547 fixed_channel_layout = 1;
549 /** loop until the frame data is split between the subframes */
553 /** check which channels contain the subframe */
554 for (c = 0; c < s->num_channels; c++) {
555 if (num_samples[c] == min_channel_len) {
556 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
557 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
558 contains_subframe[c] = 1;
560 contains_subframe[c] = get_bits1(&s->gb);
562 contains_subframe[c] = 0;
565 /** get subframe length, subframe_len == 0 is not allowed */
566 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
567 return AVERROR_INVALIDDATA;
569 /** add subframes to the individual channels and find new min_channel_len */
570 min_channel_len += subframe_len;
571 for (c = 0; c < s->num_channels; c++) {
572 WMAProChannelCtx* chan = &s->channel[c];
574 if (contains_subframe[c]) {
575 if (chan->num_subframes >= MAX_SUBFRAMES) {
576 av_log(s->avctx, AV_LOG_ERROR,
577 "broken frame: num subframes > 31\n");
578 return AVERROR_INVALIDDATA;
580 chan->subframe_len[chan->num_subframes] = subframe_len;
581 num_samples[c] += subframe_len;
582 ++chan->num_subframes;
583 if (num_samples[c] > s->samples_per_frame) {
584 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
585 "channel len > samples_per_frame\n");
586 return AVERROR_INVALIDDATA;
588 } else if (num_samples[c] <= min_channel_len) {
589 if (num_samples[c] < min_channel_len) {
590 channels_for_cur_subframe = 0;
591 min_channel_len = num_samples[c];
593 ++channels_for_cur_subframe;
596 } while (min_channel_len < s->samples_per_frame);
598 for (c = 0; c < s->num_channels; c++) {
601 for (i = 0; i < s->channel[c].num_subframes; i++) {
602 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
603 " len %i\n", s->frame_num, c, i,
604 s->channel[c].subframe_len[i]);
605 s->channel[c].subframe_offset[i] = offset;
606 offset += s->channel[c].subframe_len[i];
614 *@brief Calculate a decorrelation matrix from the bitstream parameters.
615 *@param s codec context
616 *@param chgroup channel group for which the matrix needs to be calculated
618 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
619 WMAProChannelGrp *chgroup)
623 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
624 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
625 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
627 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
628 rotation_offset[i] = get_bits(&s->gb, 6);
630 for (i = 0; i < chgroup->num_channels; i++)
631 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
632 get_bits1(&s->gb) ? 1.0 : -1.0;
634 for (i = 1; i < chgroup->num_channels; i++) {
636 for (x = 0; x < i; x++) {
638 for (y = 0; y < i + 1; y++) {
639 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
640 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
641 int n = rotation_offset[offset + x];
647 cosv = sin64[32 - n];
649 sinv = sin64[64 - n];
650 cosv = -sin64[n - 32];
653 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
654 (v1 * sinv) - (v2 * cosv);
655 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
656 (v1 * cosv) + (v2 * sinv);
664 *@brief Decode channel transformation parameters
665 *@param s codec context
666 *@return 0 in case of success, < 0 in case of bitstream errors
668 static int decode_channel_transform(WMAProDecodeCtx* s)
671 /* should never consume more than 1921 bits for the 8 channel case
672 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
673 * + MAX_CHANNELS + MAX_BANDS + 1)
676 /** in the one channel case channel transforms are pointless */
678 if (s->num_channels > 1) {
679 int remaining_channels = s->channels_for_cur_subframe;
681 if (get_bits1(&s->gb)) {
682 av_log_ask_for_sample(s->avctx,
683 "unsupported channel transform bit\n");
684 return AVERROR_INVALIDDATA;
687 for (s->num_chgroups = 0; remaining_channels &&
688 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
689 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
690 float** channel_data = chgroup->channel_data;
691 chgroup->num_channels = 0;
692 chgroup->transform = 0;
694 /** decode channel mask */
695 if (remaining_channels > 2) {
696 for (i = 0; i < s->channels_for_cur_subframe; i++) {
697 int channel_idx = s->channel_indexes_for_cur_subframe[i];
698 if (!s->channel[channel_idx].grouped
699 && get_bits1(&s->gb)) {
700 ++chgroup->num_channels;
701 s->channel[channel_idx].grouped = 1;
702 *channel_data++ = s->channel[channel_idx].coeffs;
706 chgroup->num_channels = remaining_channels;
707 for (i = 0; i < s->channels_for_cur_subframe; i++) {
708 int channel_idx = s->channel_indexes_for_cur_subframe[i];
709 if (!s->channel[channel_idx].grouped)
710 *channel_data++ = s->channel[channel_idx].coeffs;
711 s->channel[channel_idx].grouped = 1;
715 /** decode transform type */
716 if (chgroup->num_channels == 2) {
717 if (get_bits1(&s->gb)) {
718 if (get_bits1(&s->gb)) {
719 av_log_ask_for_sample(s->avctx,
720 "unsupported channel transform type\n");
723 chgroup->transform = 1;
724 if (s->num_channels == 2) {
725 chgroup->decorrelation_matrix[0] = 1.0;
726 chgroup->decorrelation_matrix[1] = -1.0;
727 chgroup->decorrelation_matrix[2] = 1.0;
728 chgroup->decorrelation_matrix[3] = 1.0;
731 chgroup->decorrelation_matrix[0] = 0.70703125;
732 chgroup->decorrelation_matrix[1] = -0.70703125;
733 chgroup->decorrelation_matrix[2] = 0.70703125;
734 chgroup->decorrelation_matrix[3] = 0.70703125;
737 } else if (chgroup->num_channels > 2) {
738 if (get_bits1(&s->gb)) {
739 chgroup->transform = 1;
740 if (get_bits1(&s->gb)) {
741 decode_decorrelation_matrix(s, chgroup);
743 /** FIXME: more than 6 coupled channels not supported */
744 if (chgroup->num_channels > 6) {
745 av_log_ask_for_sample(s->avctx,
746 "coupled channels > 6\n");
748 memcpy(chgroup->decorrelation_matrix,
749 default_decorrelation[chgroup->num_channels],
750 chgroup->num_channels * chgroup->num_channels *
751 sizeof(*chgroup->decorrelation_matrix));
757 /** decode transform on / off */
758 if (chgroup->transform) {
759 if (!get_bits1(&s->gb)) {
761 /** transform can be enabled for individual bands */
762 for (i = 0; i < s->num_bands; i++) {
763 chgroup->transform_band[i] = get_bits1(&s->gb);
766 memset(chgroup->transform_band, 1, s->num_bands);
769 remaining_channels -= chgroup->num_channels;
776 *@brief Extract the coefficients from the bitstream.
777 *@param s codec context
778 *@param c current channel number
779 *@return 0 on success, < 0 in case of bitstream errors
781 static int decode_coeffs(WMAProDecodeCtx *s, int c)
783 /* Integers 0..15 as single-precision floats. The table saves a
784 costly int to float conversion, and storing the values as
785 integers allows fast sign-flipping. */
786 static const uint32_t fval_tab[16] = {
787 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
788 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
789 0x41000000, 0x41100000, 0x41200000, 0x41300000,
790 0x41400000, 0x41500000, 0x41600000, 0x41700000,
794 WMAProChannelCtx* ci = &s->channel[c];
801 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
803 vlctable = get_bits1(&s->gb);
804 vlc = &coef_vlc[vlctable];
814 /** decode vector coefficients (consumes up to 167 bits per iteration for
815 4 vector coded large values) */
816 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
817 (cur_coeff + 3 < ci->num_vec_coeffs)) {
822 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
824 if (idx == HUFF_VEC4_SIZE - 1) {
825 for (i = 0; i < 4; i += 2) {
826 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
827 if (idx == HUFF_VEC2_SIZE - 1) {
829 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
830 if (v0 == HUFF_VEC1_SIZE - 1)
831 v0 += ff_wma_get_large_val(&s->gb);
832 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
833 if (v1 == HUFF_VEC1_SIZE - 1)
834 v1 += ff_wma_get_large_val(&s->gb);
835 vals[i ] = av_float2int(v0);
836 vals[i+1] = av_float2int(v1);
838 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
839 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
843 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
844 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
845 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
846 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
850 for (i = 0; i < 4; i++) {
852 uint32_t sign = get_bits1(&s->gb) - 1;
853 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
856 ci->coeffs[cur_coeff] = 0;
857 /** switch to run level mode when subframe_len / 128 zeros
858 were found in a row */
859 rl_mode |= (++num_zeros > s->subframe_len >> 8);
865 /** decode run level coded coefficients */
866 if (cur_coeff < s->subframe_len) {
867 memset(&ci->coeffs[cur_coeff], 0,
868 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
869 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
870 level, run, 1, ci->coeffs,
871 cur_coeff, s->subframe_len,
872 s->subframe_len, s->esc_len, 0))
873 return AVERROR_INVALIDDATA;
880 *@brief Extract scale factors from the bitstream.
881 *@param s codec context
882 *@return 0 on success, < 0 in case of bitstream errors
884 static int decode_scale_factors(WMAProDecodeCtx* s)
888 /** should never consume more than 5344 bits
889 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
892 for (i = 0; i < s->channels_for_cur_subframe; i++) {
893 int c = s->channel_indexes_for_cur_subframe[i];
896 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
897 sf_end = s->channel[c].scale_factors + s->num_bands;
899 /** resample scale factors for the new block size
900 * as the scale factors might need to be resampled several times
901 * before some new values are transmitted, a backup of the last
902 * transmitted scale factors is kept in saved_scale_factors
904 if (s->channel[c].reuse_sf) {
905 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
907 for (b = 0; b < s->num_bands; b++)
908 s->channel[c].scale_factors[b] =
909 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
912 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
914 if (!s->channel[c].reuse_sf) {
916 /** decode DPCM coded scale factors */
917 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
918 val = 45 / s->channel[c].scale_factor_step;
919 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
920 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
925 /** run level decode differences to the resampled factors */
926 for (i = 0; i < s->num_bands; i++) {
932 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
935 uint32_t code = get_bits(&s->gb, 14);
937 sign = (code & 1) - 1;
938 skip = (code & 0x3f) >> 1;
939 } else if (idx == 1) {
942 skip = scale_rl_run[idx];
943 val = scale_rl_level[idx];
944 sign = get_bits1(&s->gb)-1;
948 if (i >= s->num_bands) {
949 av_log(s->avctx, AV_LOG_ERROR,
950 "invalid scale factor coding\n");
951 return AVERROR_INVALIDDATA;
953 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
957 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
958 s->channel[c].table_idx = s->table_idx;
959 s->channel[c].reuse_sf = 1;
962 /** calculate new scale factor maximum */
963 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
964 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
965 s->channel[c].max_scale_factor =
966 FFMAX(s->channel[c].max_scale_factor, *sf);
974 *@brief Reconstruct the individual channel data.
975 *@param s codec context
977 static void inverse_channel_transform(WMAProDecodeCtx *s)
981 for (i = 0; i < s->num_chgroups; i++) {
982 if (s->chgroup[i].transform) {
983 float data[WMAPRO_MAX_CHANNELS];
984 const int num_channels = s->chgroup[i].num_channels;
985 float** ch_data = s->chgroup[i].channel_data;
986 float** ch_end = ch_data + num_channels;
987 const int8_t* tb = s->chgroup[i].transform_band;
990 /** multichannel decorrelation */
991 for (sfb = s->cur_sfb_offsets;
992 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
995 /** multiply values with the decorrelation_matrix */
996 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
997 const float* mat = s->chgroup[i].decorrelation_matrix;
998 const float* data_end = data + num_channels;
999 float* data_ptr = data;
1002 for (ch = ch_data; ch < ch_end; ch++)
1003 *data_ptr++ = (*ch)[y];
1005 for (ch = ch_data; ch < ch_end; ch++) {
1008 while (data_ptr < data_end)
1009 sum += *data_ptr++ * *mat++;
1014 } else if (s->num_channels == 2) {
1015 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1016 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1017 ch_data[0] + sfb[0],
1019 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1020 ch_data[1] + sfb[0],
1029 *@brief Apply sine window and reconstruct the output buffer.
1030 *@param s codec context
1032 static void wmapro_window(WMAProDecodeCtx *s)
1035 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1036 int c = s->channel_indexes_for_cur_subframe[i];
1038 int winlen = s->channel[c].prev_block_len;
1039 float* start = s->channel[c].coeffs - (winlen >> 1);
1041 if (s->subframe_len < winlen) {
1042 start += (winlen - s->subframe_len) >> 1;
1043 winlen = s->subframe_len;
1046 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1050 s->dsp.vector_fmul_window(start, start, start + winlen,
1053 s->channel[c].prev_block_len = s->subframe_len;
1058 *@brief Decode a single subframe (block).
1059 *@param s codec context
1060 *@return 0 on success, < 0 when decoding failed
1062 static int decode_subframe(WMAProDecodeCtx *s)
1064 int offset = s->samples_per_frame;
1065 int subframe_len = s->samples_per_frame;
1067 int total_samples = s->samples_per_frame * s->num_channels;
1068 int transmit_coeffs = 0;
1069 int cur_subwoofer_cutoff;
1071 s->subframe_offset = get_bits_count(&s->gb);
1073 /** reset channel context and find the next block offset and size
1074 == the next block of the channel with the smallest number of
1077 for (i = 0; i < s->num_channels; i++) {
1078 s->channel[i].grouped = 0;
1079 if (offset > s->channel[i].decoded_samples) {
1080 offset = s->channel[i].decoded_samples;
1082 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1087 "processing subframe with offset %i len %i\n", offset, subframe_len);
1089 /** get a list of all channels that contain the estimated block */
1090 s->channels_for_cur_subframe = 0;
1091 for (i = 0; i < s->num_channels; i++) {
1092 const int cur_subframe = s->channel[i].cur_subframe;
1093 /** substract already processed samples */
1094 total_samples -= s->channel[i].decoded_samples;
1096 /** and count if there are multiple subframes that match our profile */
1097 if (offset == s->channel[i].decoded_samples &&
1098 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1099 total_samples -= s->channel[i].subframe_len[cur_subframe];
1100 s->channel[i].decoded_samples +=
1101 s->channel[i].subframe_len[cur_subframe];
1102 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1103 ++s->channels_for_cur_subframe;
1107 /** check if the frame will be complete after processing the
1110 s->parsed_all_subframes = 1;
1113 av_dlog(s->avctx, "subframe is part of %i channels\n",
1114 s->channels_for_cur_subframe);
1116 /** calculate number of scale factor bands and their offsets */
1117 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1118 s->num_bands = s->num_sfb[s->table_idx];
1119 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1120 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1122 /** configure the decoder for the current subframe */
1123 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1124 int c = s->channel_indexes_for_cur_subframe[i];
1126 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1130 s->subframe_len = subframe_len;
1131 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1133 /** skip extended header if any */
1134 if (get_bits1(&s->gb)) {
1136 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1137 int len = get_bits(&s->gb, 4);
1138 num_fill_bits = get_bits(&s->gb, len) + 1;
1141 if (num_fill_bits >= 0) {
1142 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1143 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1144 return AVERROR_INVALIDDATA;
1147 skip_bits_long(&s->gb, num_fill_bits);
1151 /** no idea for what the following bit is used */
1152 if (get_bits1(&s->gb)) {
1153 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1154 return AVERROR_INVALIDDATA;
1158 if (decode_channel_transform(s) < 0)
1159 return AVERROR_INVALIDDATA;
1162 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1163 int c = s->channel_indexes_for_cur_subframe[i];
1164 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1165 transmit_coeffs = 1;
1168 if (transmit_coeffs) {
1170 int quant_step = 90 * s->bits_per_sample >> 4;
1172 /** decode number of vector coded coefficients */
1173 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1174 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1175 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1176 int c = s->channel_indexes_for_cur_subframe[i];
1177 s->channel[c].num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1180 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1181 int c = s->channel_indexes_for_cur_subframe[i];
1182 s->channel[c].num_vec_coeffs = s->subframe_len;
1185 /** decode quantization step */
1186 step = get_sbits(&s->gb, 6);
1188 if (step == -32 || step == 31) {
1189 const int sign = (step == 31) - 1;
1191 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1192 (step = get_bits(&s->gb, 5)) == 31) {
1195 quant_step += ((quant + step) ^ sign) - sign;
1197 if (quant_step < 0) {
1198 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1201 /** decode quantization step modifiers for every channel */
1203 if (s->channels_for_cur_subframe == 1) {
1204 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1206 int modifier_len = get_bits(&s->gb, 3);
1207 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1208 int c = s->channel_indexes_for_cur_subframe[i];
1209 s->channel[c].quant_step = quant_step;
1210 if (get_bits1(&s->gb)) {
1212 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1214 ++s->channel[c].quant_step;
1219 /** decode scale factors */
1220 if (decode_scale_factors(s) < 0)
1221 return AVERROR_INVALIDDATA;
1224 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1225 get_bits_count(&s->gb) - s->subframe_offset);
1227 /** parse coefficients */
1228 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1229 int c = s->channel_indexes_for_cur_subframe[i];
1230 if (s->channel[c].transmit_coefs &&
1231 get_bits_count(&s->gb) < s->num_saved_bits) {
1232 decode_coeffs(s, c);
1234 memset(s->channel[c].coeffs, 0,
1235 sizeof(*s->channel[c].coeffs) * subframe_len);
1238 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1239 get_bits_count(&s->gb) - s->subframe_offset);
1241 if (transmit_coeffs) {
1242 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1243 /** reconstruct the per channel data */
1244 inverse_channel_transform(s);
1245 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1246 int c = s->channel_indexes_for_cur_subframe[i];
1247 const int* sf = s->channel[c].scale_factors;
1250 if (c == s->lfe_channel)
1251 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1252 (subframe_len - cur_subwoofer_cutoff));
1254 /** inverse quantization and rescaling */
1255 for (b = 0; b < s->num_bands; b++) {
1256 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1257 const int exp = s->channel[c].quant_step -
1258 (s->channel[c].max_scale_factor - *sf++) *
1259 s->channel[c].scale_factor_step;
1260 const float quant = pow(10.0, exp / 20.0);
1261 int start = s->cur_sfb_offsets[b];
1262 s->dsp.vector_fmul_scalar(s->tmp + start,
1263 s->channel[c].coeffs + start,
1264 quant, end - start);
1267 /** apply imdct (imdct_half == DCTIV with reverse) */
1268 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1272 /** window and overlapp-add */
1275 /** handled one subframe */
1276 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1277 int c = s->channel_indexes_for_cur_subframe[i];
1278 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1279 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1280 return AVERROR_INVALIDDATA;
1282 ++s->channel[c].cur_subframe;
1289 *@brief Decode one WMA frame.
1290 *@param s codec context
1291 *@return 0 if the trailer bit indicates that this is the last frame,
1292 * 1 if there are additional frames
1294 static int decode_frame(WMAProDecodeCtx *s, int *got_frame_ptr)
1296 AVCodecContext *avctx = s->avctx;
1297 GetBitContext* gb = &s->gb;
1298 int more_frames = 0;
1301 const float *out_ptr[WMAPRO_MAX_CHANNELS];
1304 /** get frame length */
1306 len = get_bits(gb, s->log2_frame_size);
1308 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1310 /** decode tile information */
1311 if (decode_tilehdr(s)) {
1316 /** read postproc transform */
1317 if (s->num_channels > 1 && get_bits1(gb)) {
1318 if (get_bits1(gb)) {
1319 for (i = 0; i < s->num_channels * s->num_channels; i++)
1324 /** read drc info */
1325 if (s->dynamic_range_compression) {
1326 s->drc_gain = get_bits(gb, 8);
1327 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1330 /** no idea what these are for, might be the number of samples
1331 that need to be skipped at the beginning or end of a stream */
1332 if (get_bits1(gb)) {
1335 /** usually true for the first frame */
1336 if (get_bits1(gb)) {
1337 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1338 av_dlog(s->avctx, "start skip: %i\n", skip);
1341 /** sometimes true for the last frame */
1342 if (get_bits1(gb)) {
1343 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1344 av_dlog(s->avctx, "end skip: %i\n", skip);
1349 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1350 get_bits_count(gb) - s->frame_offset);
1352 /** reset subframe states */
1353 s->parsed_all_subframes = 0;
1354 for (i = 0; i < s->num_channels; i++) {
1355 s->channel[i].decoded_samples = 0;
1356 s->channel[i].cur_subframe = 0;
1357 s->channel[i].reuse_sf = 0;
1360 /** decode all subframes */
1361 while (!s->parsed_all_subframes) {
1362 if (decode_subframe(s) < 0) {
1368 /* get output buffer */
1369 s->frame.nb_samples = s->samples_per_frame;
1370 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
1371 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1375 samples = (float *)s->frame.data[0];
1377 /** interleave samples and write them to the output buffer */
1378 for (i = 0; i < s->num_channels; i++)
1379 out_ptr[i] = s->channel[i].out;
1380 s->fmt_conv.float_interleave(samples, out_ptr, s->samples_per_frame,
1383 for (i = 0; i < s->num_channels; i++) {
1384 /** reuse second half of the IMDCT output for the next frame */
1385 memcpy(&s->channel[i].out[0],
1386 &s->channel[i].out[s->samples_per_frame],
1387 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1390 if (s->skip_frame) {
1397 if (s->len_prefix) {
1398 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1399 /** FIXME: not sure if this is always an error */
1400 av_log(s->avctx, AV_LOG_ERROR,
1401 "frame[%i] would have to skip %i bits\n", s->frame_num,
1402 len - (get_bits_count(gb) - s->frame_offset) - 1);
1407 /** skip the rest of the frame data */
1408 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1410 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1414 /** decode trailer bit */
1415 more_frames = get_bits1(gb);
1422 *@brief Calculate remaining input buffer length.
1423 *@param s codec context
1424 *@param gb bitstream reader context
1425 *@return remaining size in bits
1427 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1429 return s->buf_bit_size - get_bits_count(gb);
1433 *@brief Fill the bit reservoir with a (partial) frame.
1434 *@param s codec context
1435 *@param gb bitstream reader context
1436 *@param len length of the partial frame
1437 *@param append decides whether to reset the buffer or not
1439 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1444 /** when the frame data does not need to be concatenated, the input buffer
1445 is resetted and additional bits from the previous frame are copyed
1446 and skipped later so that a fast byte copy is possible */
1449 s->frame_offset = get_bits_count(gb) & 7;
1450 s->num_saved_bits = s->frame_offset;
1451 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1454 buflen = (s->num_saved_bits + len + 8) >> 3;
1456 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1457 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1462 s->num_saved_bits += len;
1464 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1467 int align = 8 - (get_bits_count(gb) & 7);
1468 align = FFMIN(align, len);
1469 put_bits(&s->pb, align, get_bits(gb, align));
1471 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1473 skip_bits_long(gb, len);
1476 PutBitContext tmp = s->pb;
1477 flush_put_bits(&tmp);
1480 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1481 skip_bits(&s->gb, s->frame_offset);
1485 *@brief Decode a single WMA packet.
1486 *@param avctx codec context
1487 *@param data the output buffer
1488 *@param avpkt input packet
1489 *@return number of bytes that were read from the input buffer
1491 static int decode_packet(AVCodecContext *avctx, void *data,
1492 int *got_frame_ptr, AVPacket* avpkt)
1494 WMAProDecodeCtx *s = avctx->priv_data;
1495 GetBitContext* gb = &s->pgb;
1496 const uint8_t* buf = avpkt->data;
1497 int buf_size = avpkt->size;
1498 int num_bits_prev_frame;
1499 int packet_sequence_number;
1503 if (s->packet_done || s->packet_loss) {
1506 /** sanity check for the buffer length */
1507 if (buf_size < avctx->block_align)
1510 s->next_packet_start = buf_size - avctx->block_align;
1511 buf_size = avctx->block_align;
1512 s->buf_bit_size = buf_size << 3;
1514 /** parse packet header */
1515 init_get_bits(gb, buf, s->buf_bit_size);
1516 packet_sequence_number = get_bits(gb, 4);
1519 /** get number of bits that need to be added to the previous frame */
1520 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1521 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1522 num_bits_prev_frame);
1524 /** check for packet loss */
1525 if (!s->packet_loss &&
1526 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1528 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1529 s->packet_sequence_number, packet_sequence_number);
1531 s->packet_sequence_number = packet_sequence_number;
1533 if (num_bits_prev_frame > 0) {
1534 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1535 if (num_bits_prev_frame >= remaining_packet_bits) {
1536 num_bits_prev_frame = remaining_packet_bits;
1540 /** append the previous frame data to the remaining data from the
1541 previous packet to create a full frame */
1542 save_bits(s, gb, num_bits_prev_frame, 1);
1543 av_dlog(avctx, "accumulated %x bits of frame data\n",
1544 s->num_saved_bits - s->frame_offset);
1546 /** decode the cross packet frame if it is valid */
1547 if (!s->packet_loss)
1548 decode_frame(s, got_frame_ptr);
1549 } else if (s->num_saved_bits - s->frame_offset) {
1550 av_dlog(avctx, "ignoring %x previously saved bits\n",
1551 s->num_saved_bits - s->frame_offset);
1554 if (s->packet_loss) {
1555 /** reset number of saved bits so that the decoder
1556 does not start to decode incomplete frames in the
1557 s->len_prefix == 0 case */
1558 s->num_saved_bits = 0;
1564 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1565 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1566 skip_bits(gb, s->packet_offset);
1567 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1568 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1569 frame_size <= remaining_bits(s, gb)) {
1570 save_bits(s, gb, frame_size, 0);
1571 s->packet_done = !decode_frame(s, got_frame_ptr);
1572 } else if (!s->len_prefix
1573 && s->num_saved_bits > get_bits_count(&s->gb)) {
1574 /** when the frames do not have a length prefix, we don't know
1575 the compressed length of the individual frames
1576 however, we know what part of a new packet belongs to the
1578 therefore we save the incoming packet first, then we append
1579 the "previous frame" data from the next packet so that
1580 we get a buffer that only contains full frames */
1581 s->packet_done = !decode_frame(s, got_frame_ptr);
1586 if (s->packet_done && !s->packet_loss &&
1587 remaining_bits(s, gb) > 0) {
1588 /** save the rest of the data so that it can be decoded
1589 with the next packet */
1590 save_bits(s, gb, remaining_bits(s, gb), 0);
1593 s->packet_offset = get_bits_count(gb) & 7;
1595 return AVERROR_INVALIDDATA;
1598 *(AVFrame *)data = s->frame;
1600 return get_bits_count(gb) >> 3;
1604 *@brief Clear decoder buffers (for seeking).
1605 *@param avctx codec context
1607 static void flush(AVCodecContext *avctx)
1609 WMAProDecodeCtx *s = avctx->priv_data;
1611 /** reset output buffer as a part of it is used during the windowing of a
1613 for (i = 0; i < s->num_channels; i++)
1614 memset(s->channel[i].out, 0, s->samples_per_frame *
1615 sizeof(*s->channel[i].out));
1621 *@brief wmapro decoder
1623 AVCodec ff_wmapro_decoder = {
1625 .type = AVMEDIA_TYPE_AUDIO,
1626 .id = AV_CODEC_ID_WMAPRO,
1627 .priv_data_size = sizeof(WMAProDecodeCtx),
1628 .init = decode_init,
1629 .close = decode_end,
1630 .decode = decode_packet,
1631 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1633 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),