ac0cce16bdcb8ff33acaed507c81554567921996
[libav.git] / libavcodec / wmaprodec.c
1 /*
2 * Wmapro compatible decoder
3 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson
5 *
6 * This file is part of Libav.
7 *
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.
12 *
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.
17 *
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
21 */
22
23 /**
24 * @file
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
31 * - IMDCT
32 * - windowing and overlapp-add
33 *
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
39 * a frame.
40 * The number of samples and a few other decode flags are stored
41 * as extradata that has to be passed to the decoder.
42 *
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.
46 *
47 * Example wmapro bitstream (in samples):
48 *
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 * ---------------------------------------------------
57 *
58 * The frame layouts for the individual channels of a wma frame does not need
59 * to be the same.
60 *
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.
71 *
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
79 * codes.
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
82 * modifier.
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.
87 */
88
89 #include "libavutil/float_dsp.h"
90 #include "libavutil/intfloat.h"
91 #include "libavutil/intreadwrite.h"
92 #include "avcodec.h"
93 #include "internal.h"
94 #include "get_bits.h"
95 #include "put_bits.h"
96 #include "wmaprodata.h"
97 #include "dsputil.h"
98 #include "sinewin.h"
99 #include "wma.h"
100 #include "wma_common.h"
101
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
107
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
112
113
114 #define VLCBITS 9
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)
121
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
129
130 /**
131 * @brief frame specific decoder context for a single channel
132 */
133 typedef struct {
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
153 } WMAProChannelCtx;
154
155 /**
156 * @brief channel group for channel transformations
157 */
158 typedef struct {
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
164 } WMAProChannelGrp;
165
166 /**
167 * @brief main decoder context
168 */
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 AVFloatDSPContext fdsp;
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
181
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 lfe_channel; ///< lfe channel index
190 uint8_t max_num_subframes;
191 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
192 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
193 uint16_t min_samples_per_subframe;
194 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
195 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
196 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
197 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
198
199 /* packet decode state */
200 GetBitContext pgb; ///< bitstream reader context for the packet
201 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
202 uint8_t packet_offset; ///< frame offset in the packet
203 uint8_t packet_sequence_number; ///< current packet number
204 int num_saved_bits; ///< saved number of bits
205 int frame_offset; ///< frame offset in the bit reservoir
206 int subframe_offset; ///< subframe offset in the bit reservoir
207 uint8_t packet_loss; ///< set in case of bitstream error
208 uint8_t packet_done; ///< set when a packet is fully decoded
209
210 /* frame decode state */
211 uint32_t frame_num; ///< current frame number (not used for decoding)
212 GetBitContext gb; ///< bitstream reader context
213 int buf_bit_size; ///< buffer size in bits
214 uint8_t drc_gain; ///< gain for the DRC tool
215 int8_t skip_frame; ///< skip output step
216 int8_t parsed_all_subframes; ///< all subframes decoded?
217
218 /* subframe/block decode state */
219 int16_t subframe_len; ///< current subframe length
220 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
221 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
222 int8_t num_bands; ///< number of scale factor bands
223 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
224 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
225 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
226 int8_t esc_len; ///< length of escaped coefficients
227
228 uint8_t num_chgroups; ///< number of channel groups
229 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
230
231 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
232 } WMAProDecodeCtx;
233
234
235 /**
236 *@brief helper function to print the most important members of the context
237 *@param s context
238 */
239 static av_cold void dump_context(WMAProDecodeCtx *s)
240 {
241 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
242 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
243
244 PRINT("ed sample bit depth", s->bits_per_sample);
245 PRINT_HEX("ed decode flags", s->decode_flags);
246 PRINT("samples per frame", s->samples_per_frame);
247 PRINT("log2 frame size", s->log2_frame_size);
248 PRINT("max num subframes", s->max_num_subframes);
249 PRINT("len prefix", s->len_prefix);
250 PRINT("num channels", s->avctx->channels);
251 }
252
253 /**
254 *@brief Uninitialize the decoder and free all resources.
255 *@param avctx codec context
256 *@return 0 on success, < 0 otherwise
257 */
258 static av_cold int decode_end(AVCodecContext *avctx)
259 {
260 WMAProDecodeCtx *s = avctx->priv_data;
261 int i;
262
263 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
264 ff_mdct_end(&s->mdct_ctx[i]);
265
266 return 0;
267 }
268
269 /**
270 *@brief Initialize the decoder.
271 *@param avctx codec context
272 *@return 0 on success, -1 otherwise
273 */
274 static av_cold int decode_init(AVCodecContext *avctx)
275 {
276 WMAProDecodeCtx *s = avctx->priv_data;
277 uint8_t *edata_ptr = avctx->extradata;
278 unsigned int channel_mask;
279 int i, bits;
280 int log2_max_num_subframes;
281 int num_possible_block_sizes;
282
283 s->avctx = avctx;
284 ff_dsputil_init(&s->dsp, avctx);
285 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
286
287 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
288
289 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
290
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");
299
300 } else {
301 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
302 return AVERROR_INVALIDDATA;
303 }
304
305 /** generic init */
306 s->log2_frame_size = av_log2(avctx->block_align) + 4;
307
308 /** frame info */
309 s->skip_frame = 1; /* skip first frame */
310 s->packet_loss = 1;
311 s->len_prefix = (s->decode_flags & 0x40);
312
313 /** get frame len */
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_PATCHWELCOME;
318 }
319 s->samples_per_frame = 1 << bits;
320
321 /** subframe info */
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;
327
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);
331
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;
336 }
337
338 if (s->avctx->sample_rate <= 0) {
339 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
340 return AVERROR_INVALIDDATA;
341 }
342
343 if (avctx->channels < 0) {
344 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
345 avctx->channels);
346 return AVERROR_INVALIDDATA;
347 } else if (avctx->channels > WMAPRO_MAX_CHANNELS) {
348 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
349 return AVERROR_PATCHWELCOME;
350 }
351
352 /** init previous block len */
353 for (i = 0; i < avctx->channels; i++)
354 s->channel[i].prev_block_len = s->samples_per_frame;
355
356 /** extract lfe channel position */
357 s->lfe_channel = -1;
358
359 if (channel_mask & 8) {
360 unsigned int mask;
361 for (mask = 1; mask < 16; mask <<= 1) {
362 if (channel_mask & mask)
363 ++s->lfe_channel;
364 }
365 }
366
367 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
368 scale_huffbits, 1, 1,
369 scale_huffcodes, 2, 2, 616);
370
371 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
372 scale_rl_huffbits, 1, 1,
373 scale_rl_huffcodes, 4, 4, 1406);
374
375 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
376 coef0_huffbits, 1, 1,
377 coef0_huffcodes, 4, 4, 2108);
378
379 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
380 coef1_huffbits, 1, 1,
381 coef1_huffcodes, 4, 4, 3912);
382
383 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
384 vec4_huffbits, 1, 1,
385 vec4_huffcodes, 2, 2, 604);
386
387 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
388 vec2_huffbits, 1, 1,
389 vec2_huffcodes, 2, 2, 562);
390
391 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
392 vec1_huffbits, 1, 1,
393 vec1_huffcodes, 2, 2, 562);
394
395 /** calculate number of scale factor bands and their offsets
396 for every possible block size */
397 for (i = 0; i < num_possible_block_sizes; i++) {
398 int subframe_len = s->samples_per_frame >> i;
399 int x;
400 int band = 1;
401
402 s->sfb_offsets[i][0] = 0;
403
404 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
405 int offset = (subframe_len * 2 * critical_freq[x])
406 / s->avctx->sample_rate + 2;
407 offset &= ~3;
408 if (offset > s->sfb_offsets[i][band - 1])
409 s->sfb_offsets[i][band++] = offset;
410 }
411 s->sfb_offsets[i][band - 1] = subframe_len;
412 s->num_sfb[i] = band - 1;
413 }
414
415
416 /** Scale factors can be shared between blocks of different size
417 as every block has a different scale factor band layout.
418 The matrix sf_offsets is needed to find the correct scale factor.
419 */
420
421 for (i = 0; i < num_possible_block_sizes; i++) {
422 int b;
423 for (b = 0; b < s->num_sfb[i]; b++) {
424 int x;
425 int offset = ((s->sfb_offsets[i][b]
426 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
427 for (x = 0; x < num_possible_block_sizes; x++) {
428 int v = 0;
429 while (s->sfb_offsets[x][v + 1] << x < offset)
430 ++v;
431 s->sf_offsets[i][x][b] = v;
432 }
433 }
434 }
435
436 /** init MDCT, FIXME: only init needed sizes */
437 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
438 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
439 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
440 / (1 << (s->bits_per_sample - 1)));
441
442 /** init MDCT windows: simple sinus window */
443 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
444 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
445 ff_init_ff_sine_windows(win_idx);
446 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
447 }
448
449 /** calculate subwoofer cutoff values */
450 for (i = 0; i < num_possible_block_sizes; i++) {
451 int block_size = s->samples_per_frame >> i;
452 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
453 / s->avctx->sample_rate;
454 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
455 }
456
457 /** calculate sine values for the decorrelation matrix */
458 for (i = 0; i < 33; i++)
459 sin64[i] = sin(i*M_PI / 64.0);
460
461 if (avctx->debug & FF_DEBUG_BITSTREAM)
462 dump_context(s);
463
464 avctx->channel_layout = channel_mask;
465
466 avcodec_get_frame_defaults(&s->frame);
467 avctx->coded_frame = &s->frame;
468
469 return 0;
470 }
471
472 /**
473 *@brief Decode the subframe length.
474 *@param s context
475 *@param offset sample offset in the frame
476 *@return decoded subframe length on success, < 0 in case of an error
477 */
478 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
479 {
480 int frame_len_shift = 0;
481 int subframe_len;
482
483 /** no need to read from the bitstream when only one length is possible */
484 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
485 return s->min_samples_per_subframe;
486
487 /** 1 bit indicates if the subframe is of maximum length */
488 if (s->max_subframe_len_bit) {
489 if (get_bits1(&s->gb))
490 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
491 } else
492 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
493
494 subframe_len = s->samples_per_frame >> frame_len_shift;
495
496 /** sanity check the length */
497 if (subframe_len < s->min_samples_per_subframe ||
498 subframe_len > s->samples_per_frame) {
499 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
500 subframe_len);
501 return AVERROR_INVALIDDATA;
502 }
503 return subframe_len;
504 }
505
506 /**
507 *@brief Decode how the data in the frame is split into subframes.
508 * Every WMA frame contains the encoded data for a fixed number of
509 * samples per channel. The data for every channel might be split
510 * into several subframes. This function will reconstruct the list of
511 * subframes for every channel.
512 *
513 * If the subframes are not evenly split, the algorithm estimates the
514 * channels with the lowest number of total samples.
515 * Afterwards, for each of these channels a bit is read from the
516 * bitstream that indicates if the channel contains a subframe with the
517 * next subframe size that is going to be read from the bitstream or not.
518 * If a channel contains such a subframe, the subframe size gets added to
519 * the channel's subframe list.
520 * The algorithm repeats these steps until the frame is properly divided
521 * between the individual channels.
522 *
523 *@param s context
524 *@return 0 on success, < 0 in case of an error
525 */
526 static int decode_tilehdr(WMAProDecodeCtx *s)
527 {
528 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
529 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
530 int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */
531 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
532 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
533 int c;
534
535 /* Should never consume more than 3073 bits (256 iterations for the
536 * while loop when always the minimum amount of 128 samples is substracted
537 * from missing samples in the 8 channel case).
538 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
539 */
540
541 /** reset tiling information */
542 for (c = 0; c < s->avctx->channels; c++)
543 s->channel[c].num_subframes = 0;
544
545 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
546 fixed_channel_layout = 1;
547
548 /** loop until the frame data is split between the subframes */
549 do {
550 int subframe_len;
551
552 /** check which channels contain the subframe */
553 for (c = 0; c < s->avctx->channels; c++) {
554 if (num_samples[c] == min_channel_len) {
555 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
556 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
557 contains_subframe[c] = 1;
558 else
559 contains_subframe[c] = get_bits1(&s->gb);
560 } else
561 contains_subframe[c] = 0;
562 }
563
564 /** get subframe length, subframe_len == 0 is not allowed */
565 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
566 return AVERROR_INVALIDDATA;
567
568 /** add subframes to the individual channels and find new min_channel_len */
569 min_channel_len += subframe_len;
570 for (c = 0; c < s->avctx->channels; c++) {
571 WMAProChannelCtx* chan = &s->channel[c];
572
573 if (contains_subframe[c]) {
574 if (chan->num_subframes >= MAX_SUBFRAMES) {
575 av_log(s->avctx, AV_LOG_ERROR,
576 "broken frame: num subframes > 31\n");
577 return AVERROR_INVALIDDATA;
578 }
579 chan->subframe_len[chan->num_subframes] = subframe_len;
580 num_samples[c] += subframe_len;
581 ++chan->num_subframes;
582 if (num_samples[c] > s->samples_per_frame) {
583 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
584 "channel len > samples_per_frame\n");
585 return AVERROR_INVALIDDATA;
586 }
587 } else if (num_samples[c] <= min_channel_len) {
588 if (num_samples[c] < min_channel_len) {
589 channels_for_cur_subframe = 0;
590 min_channel_len = num_samples[c];
591 }
592 ++channels_for_cur_subframe;
593 }
594 }
595 } while (min_channel_len < s->samples_per_frame);
596
597 for (c = 0; c < s->avctx->channels; c++) {
598 int i;
599 int offset = 0;
600 for (i = 0; i < s->channel[c].num_subframes; i++) {
601 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
602 " len %i\n", s->frame_num, c, i,
603 s->channel[c].subframe_len[i]);
604 s->channel[c].subframe_offset[i] = offset;
605 offset += s->channel[c].subframe_len[i];
606 }
607 }
608
609 return 0;
610 }
611
612 /**
613 *@brief Calculate a decorrelation matrix from the bitstream parameters.
614 *@param s codec context
615 *@param chgroup channel group for which the matrix needs to be calculated
616 */
617 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
618 WMAProChannelGrp *chgroup)
619 {
620 int i;
621 int offset = 0;
622 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
623 memset(chgroup->decorrelation_matrix, 0, s->avctx->channels *
624 s->avctx->channels * sizeof(*chgroup->decorrelation_matrix));
625
626 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
627 rotation_offset[i] = get_bits(&s->gb, 6);
628
629 for (i = 0; i < chgroup->num_channels; i++)
630 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
631 get_bits1(&s->gb) ? 1.0 : -1.0;
632
633 for (i = 1; i < chgroup->num_channels; i++) {
634 int x;
635 for (x = 0; x < i; x++) {
636 int y;
637 for (y = 0; y < i + 1; y++) {
638 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
639 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
640 int n = rotation_offset[offset + x];
641 float sinv;
642 float cosv;
643
644 if (n < 32) {
645 sinv = sin64[n];
646 cosv = sin64[32 - n];
647 } else {
648 sinv = sin64[64 - n];
649 cosv = -sin64[n - 32];
650 }
651
652 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
653 (v1 * sinv) - (v2 * cosv);
654 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
655 (v1 * cosv) + (v2 * sinv);
656 }
657 }
658 offset += i;
659 }
660 }
661
662 /**
663 *@brief Decode channel transformation parameters
664 *@param s codec context
665 *@return 0 in case of success, < 0 in case of bitstream errors
666 */
667 static int decode_channel_transform(WMAProDecodeCtx* s)
668 {
669 int i;
670 /* should never consume more than 1921 bits for the 8 channel case
671 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
672 * + MAX_CHANNELS + MAX_BANDS + 1)
673 */
674
675 /** in the one channel case channel transforms are pointless */
676 s->num_chgroups = 0;
677 if (s->avctx->channels > 1) {
678 int remaining_channels = s->channels_for_cur_subframe;
679
680 if (get_bits1(&s->gb)) {
681 av_log_ask_for_sample(s->avctx,
682 "unsupported channel transform bit\n");
683 return AVERROR_INVALIDDATA;
684 }
685
686 for (s->num_chgroups = 0; remaining_channels &&
687 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
688 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
689 float** channel_data = chgroup->channel_data;
690 chgroup->num_channels = 0;
691 chgroup->transform = 0;
692
693 /** decode channel mask */
694 if (remaining_channels > 2) {
695 for (i = 0; i < s->channels_for_cur_subframe; i++) {
696 int channel_idx = s->channel_indexes_for_cur_subframe[i];
697 if (!s->channel[channel_idx].grouped
698 && get_bits1(&s->gb)) {
699 ++chgroup->num_channels;
700 s->channel[channel_idx].grouped = 1;
701 *channel_data++ = s->channel[channel_idx].coeffs;
702 }
703 }
704 } else {
705 chgroup->num_channels = remaining_channels;
706 for (i = 0; i < s->channels_for_cur_subframe; i++) {
707 int channel_idx = s->channel_indexes_for_cur_subframe[i];
708 if (!s->channel[channel_idx].grouped)
709 *channel_data++ = s->channel[channel_idx].coeffs;
710 s->channel[channel_idx].grouped = 1;
711 }
712 }
713
714 /** decode transform type */
715 if (chgroup->num_channels == 2) {
716 if (get_bits1(&s->gb)) {
717 if (get_bits1(&s->gb)) {
718 av_log_ask_for_sample(s->avctx,
719 "unsupported channel transform type\n");
720 }
721 } else {
722 chgroup->transform = 1;
723 if (s->avctx->channels == 2) {
724 chgroup->decorrelation_matrix[0] = 1.0;
725 chgroup->decorrelation_matrix[1] = -1.0;
726 chgroup->decorrelation_matrix[2] = 1.0;
727 chgroup->decorrelation_matrix[3] = 1.0;
728 } else {
729 /** cos(pi/4) */
730 chgroup->decorrelation_matrix[0] = 0.70703125;
731 chgroup->decorrelation_matrix[1] = -0.70703125;
732 chgroup->decorrelation_matrix[2] = 0.70703125;
733 chgroup->decorrelation_matrix[3] = 0.70703125;
734 }
735 }
736 } else if (chgroup->num_channels > 2) {
737 if (get_bits1(&s->gb)) {
738 chgroup->transform = 1;
739 if (get_bits1(&s->gb)) {
740 decode_decorrelation_matrix(s, chgroup);
741 } else {
742 /** FIXME: more than 6 coupled channels not supported */
743 if (chgroup->num_channels > 6) {
744 av_log_ask_for_sample(s->avctx,
745 "coupled channels > 6\n");
746 } else {
747 memcpy(chgroup->decorrelation_matrix,
748 default_decorrelation[chgroup->num_channels],
749 chgroup->num_channels * chgroup->num_channels *
750 sizeof(*chgroup->decorrelation_matrix));
751 }
752 }
753 }
754 }
755
756 /** decode transform on / off */
757 if (chgroup->transform) {
758 if (!get_bits1(&s->gb)) {
759 int i;
760 /** transform can be enabled for individual bands */
761 for (i = 0; i < s->num_bands; i++) {
762 chgroup->transform_band[i] = get_bits1(&s->gb);
763 }
764 } else {
765 memset(chgroup->transform_band, 1, s->num_bands);
766 }
767 }
768 remaining_channels -= chgroup->num_channels;
769 }
770 }
771 return 0;
772 }
773
774 /**
775 *@brief Extract the coefficients from the bitstream.
776 *@param s codec context
777 *@param c current channel number
778 *@return 0 on success, < 0 in case of bitstream errors
779 */
780 static int decode_coeffs(WMAProDecodeCtx *s, int c)
781 {
782 /* Integers 0..15 as single-precision floats. The table saves a
783 costly int to float conversion, and storing the values as
784 integers allows fast sign-flipping. */
785 static const uint32_t fval_tab[16] = {
786 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
787 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
788 0x41000000, 0x41100000, 0x41200000, 0x41300000,
789 0x41400000, 0x41500000, 0x41600000, 0x41700000,
790 };
791 int vlctable;
792 VLC* vlc;
793 WMAProChannelCtx* ci = &s->channel[c];
794 int rl_mode = 0;
795 int cur_coeff = 0;
796 int num_zeros = 0;
797 const uint16_t* run;
798 const float* level;
799
800 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
801
802 vlctable = get_bits1(&s->gb);
803 vlc = &coef_vlc[vlctable];
804
805 if (vlctable) {
806 run = coef1_run;
807 level = coef1_level;
808 } else {
809 run = coef0_run;
810 level = coef0_level;
811 }
812
813 /** decode vector coefficients (consumes up to 167 bits per iteration for
814 4 vector coded large values) */
815 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
816 (cur_coeff + 3 < ci->num_vec_coeffs)) {
817 uint32_t vals[4];
818 int i;
819 unsigned int idx;
820
821 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
822
823 if (idx == HUFF_VEC4_SIZE - 1) {
824 for (i = 0; i < 4; i += 2) {
825 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
826 if (idx == HUFF_VEC2_SIZE - 1) {
827 uint32_t v0, v1;
828 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
829 if (v0 == HUFF_VEC1_SIZE - 1)
830 v0 += ff_wma_get_large_val(&s->gb);
831 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
832 if (v1 == HUFF_VEC1_SIZE - 1)
833 v1 += ff_wma_get_large_val(&s->gb);
834 vals[i ] = av_float2int(v0);
835 vals[i+1] = av_float2int(v1);
836 } else {
837 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
838 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
839 }
840 }
841 } else {
842 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
843 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
844 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
845 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
846 }
847
848 /** decode sign */
849 for (i = 0; i < 4; i++) {
850 if (vals[i]) {
851 uint32_t sign = get_bits1(&s->gb) - 1;
852 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
853 num_zeros = 0;
854 } else {
855 ci->coeffs[cur_coeff] = 0;
856 /** switch to run level mode when subframe_len / 128 zeros
857 were found in a row */
858 rl_mode |= (++num_zeros > s->subframe_len >> 8);
859 }
860 ++cur_coeff;
861 }
862 }
863
864 /** decode run level coded coefficients */
865 if (cur_coeff < s->subframe_len) {
866 memset(&ci->coeffs[cur_coeff], 0,
867 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
868 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
869 level, run, 1, ci->coeffs,
870 cur_coeff, s->subframe_len,
871 s->subframe_len, s->esc_len, 0))
872 return AVERROR_INVALIDDATA;
873 }
874
875 return 0;
876 }
877
878 /**
879 *@brief Extract scale factors from the bitstream.
880 *@param s codec context
881 *@return 0 on success, < 0 in case of bitstream errors
882 */
883 static int decode_scale_factors(WMAProDecodeCtx* s)
884 {
885 int i;
886
887 /** should never consume more than 5344 bits
888 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
889 */
890
891 for (i = 0; i < s->channels_for_cur_subframe; i++) {
892 int c = s->channel_indexes_for_cur_subframe[i];
893 int* sf;
894 int* sf_end;
895 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
896 sf_end = s->channel[c].scale_factors + s->num_bands;
897
898 /** resample scale factors for the new block size
899 * as the scale factors might need to be resampled several times
900 * before some new values are transmitted, a backup of the last
901 * transmitted scale factors is kept in saved_scale_factors
902 */
903 if (s->channel[c].reuse_sf) {
904 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
905 int b;
906 for (b = 0; b < s->num_bands; b++)
907 s->channel[c].scale_factors[b] =
908 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
909 }
910
911 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
912
913 if (!s->channel[c].reuse_sf) {
914 int val;
915 /** decode DPCM coded scale factors */
916 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
917 val = 45 / s->channel[c].scale_factor_step;
918 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
919 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
920 *sf = val;
921 }
922 } else {
923 int i;
924 /** run level decode differences to the resampled factors */
925 for (i = 0; i < s->num_bands; i++) {
926 int idx;
927 int skip;
928 int val;
929 int sign;
930
931 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
932
933 if (!idx) {
934 uint32_t code = get_bits(&s->gb, 14);
935 val = code >> 6;
936 sign = (code & 1) - 1;
937 skip = (code & 0x3f) >> 1;
938 } else if (idx == 1) {
939 break;
940 } else {
941 skip = scale_rl_run[idx];
942 val = scale_rl_level[idx];
943 sign = get_bits1(&s->gb)-1;
944 }
945
946 i += skip;
947 if (i >= s->num_bands) {
948 av_log(s->avctx, AV_LOG_ERROR,
949 "invalid scale factor coding\n");
950 return AVERROR_INVALIDDATA;
951 }
952 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
953 }
954 }
955 /** swap buffers */
956 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
957 s->channel[c].table_idx = s->table_idx;
958 s->channel[c].reuse_sf = 1;
959 }
960
961 /** calculate new scale factor maximum */
962 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
963 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
964 s->channel[c].max_scale_factor =
965 FFMAX(s->channel[c].max_scale_factor, *sf);
966 }
967
968 }
969 return 0;
970 }
971
972 /**
973 *@brief Reconstruct the individual channel data.
974 *@param s codec context
975 */
976 static void inverse_channel_transform(WMAProDecodeCtx *s)
977 {
978 int i;
979
980 for (i = 0; i < s->num_chgroups; i++) {
981 if (s->chgroup[i].transform) {
982 float data[WMAPRO_MAX_CHANNELS];
983 const int num_channels = s->chgroup[i].num_channels;
984 float** ch_data = s->chgroup[i].channel_data;
985 float** ch_end = ch_data + num_channels;
986 const int8_t* tb = s->chgroup[i].transform_band;
987 int16_t* sfb;
988
989 /** multichannel decorrelation */
990 for (sfb = s->cur_sfb_offsets;
991 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
992 int y;
993 if (*tb++ == 1) {
994 /** multiply values with the decorrelation_matrix */
995 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
996 const float* mat = s->chgroup[i].decorrelation_matrix;
997 const float* data_end = data + num_channels;
998 float* data_ptr = data;
999 float** ch;
1000
1001 for (ch = ch_data; ch < ch_end; ch++)
1002 *data_ptr++ = (*ch)[y];
1003
1004 for (ch = ch_data; ch < ch_end; ch++) {
1005 float sum = 0;
1006 data_ptr = data;
1007 while (data_ptr < data_end)
1008 sum += *data_ptr++ * *mat++;
1009
1010 (*ch)[y] = sum;
1011 }
1012 }
1013 } else if (s->avctx->channels == 2) {
1014 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1015 s->fdsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1016 ch_data[0] + sfb[0],
1017 181.0 / 128, len);
1018 s->fdsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1019 ch_data[1] + sfb[0],
1020 181.0 / 128, len);
1021 }
1022 }
1023 }
1024 }
1025 }
1026
1027 /**
1028 *@brief Apply sine window and reconstruct the output buffer.
1029 *@param s codec context
1030 */
1031 static void wmapro_window(WMAProDecodeCtx *s)
1032 {
1033 int i;
1034 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1035 int c = s->channel_indexes_for_cur_subframe[i];
1036 float* window;
1037 int winlen = s->channel[c].prev_block_len;
1038 float* start = s->channel[c].coeffs - (winlen >> 1);
1039
1040 if (s->subframe_len < winlen) {
1041 start += (winlen - s->subframe_len) >> 1;
1042 winlen = s->subframe_len;
1043 }
1044
1045 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1046
1047 winlen >>= 1;
1048
1049 s->dsp.vector_fmul_window(start, start, start + winlen,
1050 window, winlen);
1051
1052 s->channel[c].prev_block_len = s->subframe_len;
1053 }
1054 }
1055
1056 /**
1057 *@brief Decode a single subframe (block).
1058 *@param s codec context
1059 *@return 0 on success, < 0 when decoding failed
1060 */
1061 static int decode_subframe(WMAProDecodeCtx *s)
1062 {
1063 int offset = s->samples_per_frame;
1064 int subframe_len = s->samples_per_frame;
1065 int i;
1066 int total_samples = s->samples_per_frame * s->avctx->channels;
1067 int transmit_coeffs = 0;
1068 int cur_subwoofer_cutoff;
1069
1070 s->subframe_offset = get_bits_count(&s->gb);
1071
1072 /** reset channel context and find the next block offset and size
1073 == the next block of the channel with the smallest number of
1074 decoded samples
1075 */
1076 for (i = 0; i < s->avctx->channels; i++) {
1077 s->channel[i].grouped = 0;
1078 if (offset > s->channel[i].decoded_samples) {
1079 offset = s->channel[i].decoded_samples;
1080 subframe_len =
1081 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1082 }
1083 }
1084
1085 av_dlog(s->avctx,
1086 "processing subframe with offset %i len %i\n", offset, subframe_len);
1087
1088 /** get a list of all channels that contain the estimated block */
1089 s->channels_for_cur_subframe = 0;
1090 for (i = 0; i < s->avctx->channels; i++) {
1091 const int cur_subframe = s->channel[i].cur_subframe;
1092 /** substract already processed samples */
1093 total_samples -= s->channel[i].decoded_samples;
1094
1095 /** and count if there are multiple subframes that match our profile */
1096 if (offset == s->channel[i].decoded_samples &&
1097 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1098 total_samples -= s->channel[i].subframe_len[cur_subframe];
1099 s->channel[i].decoded_samples +=
1100 s->channel[i].subframe_len[cur_subframe];
1101 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1102 ++s->channels_for_cur_subframe;
1103 }
1104 }
1105
1106 /** check if the frame will be complete after processing the
1107 estimated block */
1108 if (!total_samples)
1109 s->parsed_all_subframes = 1;
1110
1111
1112 av_dlog(s->avctx, "subframe is part of %i channels\n",
1113 s->channels_for_cur_subframe);
1114
1115 /** calculate number of scale factor bands and their offsets */
1116 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1117 s->num_bands = s->num_sfb[s->table_idx];
1118 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1119 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1120
1121 /** configure the decoder for the current subframe */
1122 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1123 int c = s->channel_indexes_for_cur_subframe[i];
1124
1125 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1126 + offset];
1127 }
1128
1129 s->subframe_len = subframe_len;
1130 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1131
1132 /** skip extended header if any */
1133 if (get_bits1(&s->gb)) {
1134 int num_fill_bits;
1135 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1136 int len = get_bits(&s->gb, 4);
1137 num_fill_bits = get_bits(&s->gb, len) + 1;
1138 }
1139
1140 if (num_fill_bits >= 0) {
1141 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1142 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1143 return AVERROR_INVALIDDATA;
1144 }
1145
1146 skip_bits_long(&s->gb, num_fill_bits);
1147 }
1148 }
1149
1150 /** no idea for what the following bit is used */
1151 if (get_bits1(&s->gb)) {
1152 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1153 return AVERROR_INVALIDDATA;
1154 }
1155
1156
1157 if (decode_channel_transform(s) < 0)
1158 return AVERROR_INVALIDDATA;
1159
1160
1161 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1162 int c = s->channel_indexes_for_cur_subframe[i];
1163 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1164 transmit_coeffs = 1;
1165 }
1166
1167 if (transmit_coeffs) {
1168 int step;
1169 int quant_step = 90 * s->bits_per_sample >> 4;
1170
1171 /** decode number of vector coded coefficients */
1172 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1173 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1174 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1175 int c = s->channel_indexes_for_cur_subframe[i];
1176 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1177 if (num_vec_coeffs > WMAPRO_BLOCK_MAX_SIZE) {
1178 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1179 return AVERROR_INVALIDDATA;
1180 }
1181 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1182 }
1183 } else {
1184 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1185 int c = s->channel_indexes_for_cur_subframe[i];
1186 s->channel[c].num_vec_coeffs = s->subframe_len;
1187 }
1188 }
1189 /** decode quantization step */
1190 step = get_sbits(&s->gb, 6);
1191 quant_step += step;
1192 if (step == -32 || step == 31) {
1193 const int sign = (step == 31) - 1;
1194 int quant = 0;
1195 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1196 (step = get_bits(&s->gb, 5)) == 31) {
1197 quant += 31;
1198 }
1199 quant_step += ((quant + step) ^ sign) - sign;
1200 }
1201 if (quant_step < 0) {
1202 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1203 }
1204
1205 /** decode quantization step modifiers for every channel */
1206
1207 if (s->channels_for_cur_subframe == 1) {
1208 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1209 } else {
1210 int modifier_len = get_bits(&s->gb, 3);
1211 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1212 int c = s->channel_indexes_for_cur_subframe[i];
1213 s->channel[c].quant_step = quant_step;
1214 if (get_bits1(&s->gb)) {
1215 if (modifier_len) {
1216 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1217 } else
1218 ++s->channel[c].quant_step;
1219 }
1220 }
1221 }
1222
1223 /** decode scale factors */
1224 if (decode_scale_factors(s) < 0)
1225 return AVERROR_INVALIDDATA;
1226 }
1227
1228 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1229 get_bits_count(&s->gb) - s->subframe_offset);
1230
1231 /** parse coefficients */
1232 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1233 int c = s->channel_indexes_for_cur_subframe[i];
1234 if (s->channel[c].transmit_coefs &&
1235 get_bits_count(&s->gb) < s->num_saved_bits) {
1236 decode_coeffs(s, c);
1237 } else
1238 memset(s->channel[c].coeffs, 0,
1239 sizeof(*s->channel[c].coeffs) * subframe_len);
1240 }
1241
1242 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1243 get_bits_count(&s->gb) - s->subframe_offset);
1244
1245 if (transmit_coeffs) {
1246 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1247 /** reconstruct the per channel data */
1248 inverse_channel_transform(s);
1249 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1250 int c = s->channel_indexes_for_cur_subframe[i];
1251 const int* sf = s->channel[c].scale_factors;
1252 int b;
1253
1254 if (c == s->lfe_channel)
1255 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1256 (subframe_len - cur_subwoofer_cutoff));
1257
1258 /** inverse quantization and rescaling */
1259 for (b = 0; b < s->num_bands; b++) {
1260 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1261 const int exp = s->channel[c].quant_step -
1262 (s->channel[c].max_scale_factor - *sf++) *
1263 s->channel[c].scale_factor_step;
1264 const float quant = pow(10.0, exp / 20.0);
1265 int start = s->cur_sfb_offsets[b];
1266 s->fdsp.vector_fmul_scalar(s->tmp + start,
1267 s->channel[c].coeffs + start,
1268 quant, end - start);
1269 }
1270
1271 /** apply imdct (imdct_half == DCTIV with reverse) */
1272 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1273 }
1274 }
1275
1276 /** window and overlapp-add */
1277 wmapro_window(s);
1278
1279 /** handled one subframe */
1280 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1281 int c = s->channel_indexes_for_cur_subframe[i];
1282 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1283 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1284 return AVERROR_INVALIDDATA;
1285 }
1286 ++s->channel[c].cur_subframe;
1287 }
1288
1289 return 0;
1290 }
1291
1292 /**
1293 *@brief Decode one WMA frame.
1294 *@param s codec context
1295 *@return 0 if the trailer bit indicates that this is the last frame,
1296 * 1 if there are additional frames
1297 */
1298 static int decode_frame(WMAProDecodeCtx *s, int *got_frame_ptr)
1299 {
1300 AVCodecContext *avctx = s->avctx;
1301 GetBitContext* gb = &s->gb;
1302 int more_frames = 0;
1303 int len = 0;
1304 int i, ret;
1305
1306 /** get frame length */
1307 if (s->len_prefix)
1308 len = get_bits(gb, s->log2_frame_size);
1309
1310 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1311
1312 /** decode tile information */
1313 if (decode_tilehdr(s)) {
1314 s->packet_loss = 1;
1315 return 0;
1316 }
1317
1318 /** read postproc transform */
1319 if (s->avctx->channels > 1 && get_bits1(gb)) {
1320 if (get_bits1(gb)) {
1321 for (i = 0; i < avctx->channels * avctx->channels; i++)
1322 skip_bits(gb, 4);
1323 }
1324 }
1325
1326 /** read drc info */
1327 if (s->dynamic_range_compression) {
1328 s->drc_gain = get_bits(gb, 8);
1329 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1330 }
1331
1332 /** no idea what these are for, might be the number of samples
1333 that need to be skipped at the beginning or end of a stream */
1334 if (get_bits1(gb)) {
1335 int av_unused skip;
1336
1337 /** usually true for the first frame */
1338 if (get_bits1(gb)) {
1339 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1340 av_dlog(s->avctx, "start skip: %i\n", skip);
1341 }
1342
1343 /** sometimes true for the last frame */
1344 if (get_bits1(gb)) {
1345 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1346 av_dlog(s->avctx, "end skip: %i\n", skip);
1347 }
1348
1349 }
1350
1351 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1352 get_bits_count(gb) - s->frame_offset);
1353
1354 /** reset subframe states */
1355 s->parsed_all_subframes = 0;
1356 for (i = 0; i < avctx->channels; i++) {
1357 s->channel[i].decoded_samples = 0;
1358 s->channel[i].cur_subframe = 0;
1359 s->channel[i].reuse_sf = 0;
1360 }
1361
1362 /** decode all subframes */
1363 while (!s->parsed_all_subframes) {
1364 if (decode_subframe(s) < 0) {
1365 s->packet_loss = 1;
1366 return 0;
1367 }
1368 }
1369
1370 /* get output buffer */
1371 s->frame.nb_samples = s->samples_per_frame;
1372 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
1373 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1374 s->packet_loss = 1;
1375 return 0;
1376 }
1377
1378 /** copy samples to the output buffer */
1379 for (i = 0; i < avctx->channels; i++)
1380 memcpy(s->frame.extended_data[i], s->channel[i].out,
1381 s->samples_per_frame * sizeof(*s->channel[i].out));
1382
1383 for (i = 0; i < avctx->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);
1388 }
1389
1390 if (s->skip_frame) {
1391 s->skip_frame = 0;
1392 *got_frame_ptr = 0;
1393 } else {
1394 *got_frame_ptr = 1;
1395 }
1396
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);
1403 s->packet_loss = 1;
1404 return 0;
1405 }
1406
1407 /** skip the rest of the frame data */
1408 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1409 } else {
1410 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1411 }
1412 }
1413
1414 /** decode trailer bit */
1415 more_frames = get_bits1(gb);
1416
1417 ++s->frame_num;
1418 return more_frames;
1419 }
1420
1421 /**
1422 *@brief Calculate remaining input buffer length.
1423 *@param s codec context
1424 *@param gb bitstream reader context
1425 *@return remaining size in bits
1426 */
1427 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1428 {
1429 return s->buf_bit_size - get_bits_count(gb);
1430 }
1431
1432 /**
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
1438 */
1439 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1440 int append)
1441 {
1442 int buflen;
1443
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 */
1447
1448 if (!append) {
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);
1452 }
1453
1454 buflen = (s->num_saved_bits + len + 8) >> 3;
1455
1456 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1457 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1458 s->packet_loss = 1;
1459 return;
1460 }
1461
1462 s->num_saved_bits += len;
1463 if (!append) {
1464 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1465 s->num_saved_bits);
1466 } else {
1467 int align = 8 - (get_bits_count(gb) & 7);
1468 align = FFMIN(align, len);
1469 put_bits(&s->pb, align, get_bits(gb, align));
1470 len -= align;
1471 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1472 }
1473 skip_bits_long(gb, len);
1474
1475 {
1476 PutBitContext tmp = s->pb;
1477 flush_put_bits(&tmp);
1478 }
1479
1480 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1481 skip_bits(&s->gb, s->frame_offset);
1482 }
1483
1484 /**
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
1490 */
1491 static int decode_packet(AVCodecContext *avctx, void *data,
1492 int *got_frame_ptr, AVPacket* avpkt)
1493 {
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;
1500
1501 *got_frame_ptr = 0;
1502
1503 if (s->packet_done || s->packet_loss) {
1504 s->packet_done = 0;
1505
1506 /** sanity check for the buffer length */
1507 if (buf_size < avctx->block_align)
1508 return 0;
1509
1510 s->next_packet_start = buf_size - avctx->block_align;
1511 buf_size = avctx->block_align;
1512 s->buf_bit_size = buf_size << 3;
1513
1514 /** parse packet header */
1515 init_get_bits(gb, buf, s->buf_bit_size);
1516 packet_sequence_number = get_bits(gb, 4);
1517 skip_bits(gb, 2);
1518
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);
1523
1524 /** check for packet loss */
1525 if (!s->packet_loss &&
1526 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1527 s->packet_loss = 1;
1528 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1529 s->packet_sequence_number, packet_sequence_number);
1530 }
1531 s->packet_sequence_number = packet_sequence_number;
1532
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;
1537 s->packet_done = 1;
1538 }
1539
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);
1545
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);
1552 }
1553
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;
1559 s->packet_loss = 0;
1560 }
1561
1562 } else {
1563 int frame_size;
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
1577 previous frame
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);
1582 } else
1583 s->packet_done = 1;
1584 }
1585
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);
1591 }
1592
1593 s->packet_offset = get_bits_count(gb) & 7;
1594 if (s->packet_loss)
1595 return AVERROR_INVALIDDATA;
1596
1597 if (*got_frame_ptr)
1598 *(AVFrame *)data = s->frame;
1599
1600 return get_bits_count(gb) >> 3;
1601 }
1602
1603 /**
1604 *@brief Clear decoder buffers (for seeking).
1605 *@param avctx codec context
1606 */
1607 static void flush(AVCodecContext *avctx)
1608 {
1609 WMAProDecodeCtx *s = avctx->priv_data;
1610 int i;
1611 /** reset output buffer as a part of it is used during the windowing of a
1612 new frame */
1613 for (i = 0; i < avctx->channels; i++)
1614 memset(s->channel[i].out, 0, s->samples_per_frame *
1615 sizeof(*s->channel[i].out));
1616 s->packet_loss = 1;
1617 }
1618
1619
1620 /**
1621 *@brief wmapro decoder
1622 */
1623 AVCodec ff_wmapro_decoder = {
1624 .name = "wmapro",
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,
1632 .flush = flush,
1633 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1634 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1635 AV_SAMPLE_FMT_NONE },
1636 };