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 FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
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 <inttypes.h>
90
91 #include "libavutil/ffmath.h"
92 #include "libavutil/float_dsp.h"
93 #include "libavutil/intfloat.h"
94 #include "libavutil/intreadwrite.h"
95 #include "avcodec.h"
96 #include "internal.h"
97 #include "get_bits.h"
98 #include "put_bits.h"
99 #include "wmaprodata.h"
100 #include "sinewin.h"
101 #include "wma.h"
102 #include "wma_common.h"
103
104 /** current decoder limitations */
105 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
106 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
107 #define MAX_BANDS 29 ///< max number of scale factor bands
108 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
109 #define XMA_MAX_STREAMS 8
110 #define XMA_MAX_CHANNELS_STREAM 2
111 #define XMA_MAX_CHANNELS (XMA_MAX_STREAMS * XMA_MAX_CHANNELS_STREAM)
112
113 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
114 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
115 #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size
116 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
117 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
118
119
120 #define VLCBITS 9
121 #define SCALEVLCBITS 8
122 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
123 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
124 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
125 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
126 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
127
128 static VLC sf_vlc; ///< scale factor DPCM vlc
129 static VLC sf_rl_vlc; ///< scale factor run length vlc
130 static VLC vec4_vlc; ///< 4 coefficients per symbol
131 static VLC vec2_vlc; ///< 2 coefficients per symbol
132 static VLC vec1_vlc; ///< 1 coefficient per symbol
133 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
134 static float sin64[33]; ///< sine table for decorrelation
135
136 /**
137 * @brief frame specific decoder context for a single channel
138 */
139 typedef struct WMAProChannelCtx {
140 int16_t prev_block_len; ///< length of the previous block
141 uint8_t transmit_coefs;
142 uint8_t num_subframes;
143 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
144 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
145 uint8_t cur_subframe; ///< current subframe number
146 uint16_t decoded_samples; ///< number of already processed samples
147 uint8_t grouped; ///< channel is part of a group
148 int quant_step; ///< quantization step for the current subframe
149 int8_t reuse_sf; ///< share scale factors between subframes
150 int8_t scale_factor_step; ///< scaling step for the current subframe
151 int max_scale_factor; ///< maximum scale factor for the current subframe
152 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
153 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
154 int* scale_factors; ///< pointer to the scale factor values used for decoding
155 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
156 float* coeffs; ///< pointer to the subframe decode buffer
157 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
158 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
159 } WMAProChannelCtx;
160
161 /**
162 * @brief channel group for channel transformations
163 */
164 typedef struct WMAProChannelGrp {
165 uint8_t num_channels; ///< number of channels in the group
166 int8_t transform; ///< transform on / off
167 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
168 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
169 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
170 } WMAProChannelGrp;
171
172 /**
173 * @brief main decoder context
174 */
175 typedef struct WMAProDecodeCtx {
176 /* generic decoder variables */
177 AVCodecContext* avctx; ///< codec context for av_log
178 AVFloatDSPContext *fdsp;
179 uint8_t frame_data[MAX_FRAMESIZE +
180 AV_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
181 PutBitContext pb; ///< context for filling the frame_data buffer
182 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
183 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
184 const float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
185
186 /* frame size dependent frame information (set during initialization) */
187 uint32_t decode_flags; ///< used compression features
188 uint8_t len_prefix; ///< frame is prefixed with its length
189 uint8_t dynamic_range_compression; ///< frame contains DRC data
190 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
191 uint16_t samples_per_frame; ///< number of samples to output
192 uint16_t log2_frame_size;
193 int8_t lfe_channel; ///< lfe channel index
194 uint8_t max_num_subframes;
195 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
196 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
197 uint16_t min_samples_per_subframe;
198 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
199 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
200 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
201 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
202
203 /* packet decode state */
204 GetBitContext pgb; ///< bitstream reader context for the packet
205 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
206 uint8_t packet_offset; ///< frame offset in the packet
207 uint8_t packet_sequence_number; ///< current packet number
208 int num_saved_bits; ///< saved number of bits
209 int frame_offset; ///< frame offset in the bit reservoir
210 int subframe_offset; ///< subframe offset in the bit reservoir
211 uint8_t packet_loss; ///< set in case of bitstream error
212 uint8_t packet_done; ///< set when a packet is fully decoded
213 uint8_t eof_done; ///< set when EOF reached and extra subframe is written (XMA1/2)
214
215 /* frame decode state */
216 uint32_t frame_num; ///< current frame number (not used for decoding)
217 GetBitContext gb; ///< bitstream reader context
218 int buf_bit_size; ///< buffer size in bits
219 uint8_t drc_gain; ///< gain for the DRC tool
220 int8_t skip_frame; ///< skip output step
221 int8_t parsed_all_subframes; ///< all subframes decoded?
222 uint8_t skip_packets; ///< packets to skip to find next packet in a stream (XMA1/2)
223
224 /* subframe/block decode state */
225 int16_t subframe_len; ///< current subframe length
226 int8_t nb_channels; ///< number of channels in stream (XMA1/2)
227 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
228 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
229 int8_t num_bands; ///< number of scale factor bands
230 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
231 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
232 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
233 int8_t esc_len; ///< length of escaped coefficients
234
235 uint8_t num_chgroups; ///< number of channel groups
236 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
237
238 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
239 } WMAProDecodeCtx;
240
241 typedef struct XMADecodeCtx {
242 WMAProDecodeCtx xma[XMA_MAX_STREAMS];
243 AVFrame *frames[XMA_MAX_STREAMS];
244 int current_stream;
245 int num_streams;
246 float samples[XMA_MAX_CHANNELS][512 * 64];
247 int offset[XMA_MAX_STREAMS];
248 int start_channel[XMA_MAX_STREAMS];
249 } XMADecodeCtx;
250
251 /**
252 *@brief helper function to print the most important members of the context
253 *@param s context
254 */
dump_context(WMAProDecodeCtx * s)255 static av_cold void dump_context(WMAProDecodeCtx *s)
256 {
257 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
258 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b);
259
260 PRINT("ed sample bit depth", s->bits_per_sample);
261 PRINT_HEX("ed decode flags", s->decode_flags);
262 PRINT("samples per frame", s->samples_per_frame);
263 PRINT("log2 frame size", s->log2_frame_size);
264 PRINT("max num subframes", s->max_num_subframes);
265 PRINT("len prefix", s->len_prefix);
266 PRINT("num channels", s->nb_channels);
267 }
268
269 /**
270 *@brief Uninitialize the decoder and free all resources.
271 *@param avctx codec context
272 *@return 0 on success, < 0 otherwise
273 */
decode_end(WMAProDecodeCtx * s)274 static av_cold int decode_end(WMAProDecodeCtx *s)
275 {
276 int i;
277
278 av_freep(&s->fdsp);
279
280 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
281 ff_mdct_end(&s->mdct_ctx[i]);
282
283 return 0;
284 }
285
wmapro_decode_end(AVCodecContext * avctx)286 static av_cold int wmapro_decode_end(AVCodecContext *avctx)
287 {
288 WMAProDecodeCtx *s = avctx->priv_data;
289
290 decode_end(s);
291
292 return 0;
293 }
294
get_rate(AVCodecContext * avctx)295 static av_cold int get_rate(AVCodecContext *avctx)
296 {
297 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) { // XXX: is this really only for XMA?
298 if (avctx->sample_rate > 44100)
299 return 48000;
300 else if (avctx->sample_rate > 32000)
301 return 44100;
302 else if (avctx->sample_rate > 24000)
303 return 32000;
304 return 24000;
305 }
306
307 return avctx->sample_rate;
308 }
309
310 /**
311 *@brief Initialize the decoder.
312 *@param avctx codec context
313 *@return 0 on success, -1 otherwise
314 */
decode_init(WMAProDecodeCtx * s,AVCodecContext * avctx,int num_stream)315 static av_cold int decode_init(WMAProDecodeCtx *s, AVCodecContext *avctx, int num_stream)
316 {
317 uint8_t *edata_ptr = avctx->extradata;
318 unsigned int channel_mask;
319 int i, bits;
320 int log2_max_num_subframes;
321 int num_possible_block_sizes;
322
323 if (avctx->codec_id == AV_CODEC_ID_XMA1 || avctx->codec_id == AV_CODEC_ID_XMA2)
324 avctx->block_align = 2048;
325
326 if (!avctx->block_align) {
327 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
328 return AVERROR(EINVAL);
329 }
330
331 s->avctx = avctx;
332
333 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
334
335 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
336
337 /** dump the extradata */
338 av_log(avctx, AV_LOG_DEBUG, "extradata:\n");
339 for (i = 0; i < avctx->extradata_size; i++)
340 av_log(avctx, AV_LOG_DEBUG, "[%x] ", avctx->extradata[i]);
341 av_log(avctx, AV_LOG_DEBUG, "\n");
342
343 if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size == 34) { /* XMA2WAVEFORMATEX */
344 s->decode_flags = 0x10d6;
345 s->bits_per_sample = 16;
346 channel_mask = 0; //AV_RL32(edata_ptr+2); /* not always in expected order */
347 if ((num_stream+1) * XMA_MAX_CHANNELS_STREAM > avctx->channels) /* stream config is 2ch + 2ch + ... + 1/2ch */
348 s->nb_channels = 1;
349 else
350 s->nb_channels = 2;
351 } else if (avctx->codec_id == AV_CODEC_ID_XMA2) { /* XMA2WAVEFORMAT */
352 s->decode_flags = 0x10d6;
353 s->bits_per_sample = 16;
354 channel_mask = 0; /* would need to aggregate from all streams */
355 s->nb_channels = edata_ptr[32 + ((edata_ptr[0]==3)?0:8) + 4*num_stream + 0]; /* nth stream config */
356 } else if (avctx->codec_id == AV_CODEC_ID_XMA1) { /* XMAWAVEFORMAT */
357 s->decode_flags = 0x10d6;
358 s->bits_per_sample = 16;
359 channel_mask = 0; /* would need to aggregate from all streams */
360 s->nb_channels = edata_ptr[8 + 20*num_stream + 17]; /* nth stream config */
361 } else if (avctx->codec_id == AV_CODEC_ID_WMAPRO && avctx->extradata_size >= 18) {
362 s->decode_flags = AV_RL16(edata_ptr+14);
363 channel_mask = AV_RL32(edata_ptr+2);
364 s->bits_per_sample = AV_RL16(edata_ptr);
365 s->nb_channels = avctx->channels;
366
367 if (s->bits_per_sample > 32 || s->bits_per_sample < 1) {
368 avpriv_request_sample(avctx, "bits per sample is %d", s->bits_per_sample);
369 return AVERROR_PATCHWELCOME;
370 }
371 } else {
372 avpriv_request_sample(avctx, "Unknown extradata size");
373 return AVERROR_PATCHWELCOME;
374 }
375
376 /** generic init */
377 s->log2_frame_size = av_log2(avctx->block_align) + 4;
378 if (s->log2_frame_size > 25) {
379 avpriv_request_sample(avctx, "Large block align");
380 return AVERROR_PATCHWELCOME;
381 }
382
383 /** frame info */
384 if (avctx->codec_id != AV_CODEC_ID_WMAPRO)
385 s->skip_frame = 0;
386 else
387 s->skip_frame = 1; /* skip first frame */
388
389 s->packet_loss = 1;
390 s->len_prefix = (s->decode_flags & 0x40);
391
392 /** get frame len */
393 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
394 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
395 if (bits > WMAPRO_BLOCK_MAX_BITS) {
396 avpriv_request_sample(avctx, "14-bit block sizes");
397 return AVERROR_PATCHWELCOME;
398 }
399 s->samples_per_frame = 1 << bits;
400 } else {
401 s->samples_per_frame = 512;
402 }
403
404 /** subframe info */
405 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
406 s->max_num_subframes = 1 << log2_max_num_subframes;
407 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
408 s->max_subframe_len_bit = 1;
409 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
410
411 num_possible_block_sizes = log2_max_num_subframes + 1;
412 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
413 s->dynamic_range_compression = (s->decode_flags & 0x80);
414
415 if (s->max_num_subframes > MAX_SUBFRAMES) {
416 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n",
417 s->max_num_subframes);
418 return AVERROR_INVALIDDATA;
419 }
420
421 if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) {
422 av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
423 s->min_samples_per_subframe);
424 return AVERROR_INVALIDDATA;
425 }
426
427 if (s->avctx->sample_rate <= 0) {
428 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
429 return AVERROR_INVALIDDATA;
430 }
431
432 if (s->nb_channels <= 0) {
433 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
434 s->nb_channels);
435 return AVERROR_INVALIDDATA;
436 } else if (avctx->codec_id != AV_CODEC_ID_WMAPRO && s->nb_channels > XMA_MAX_CHANNELS_STREAM) {
437 av_log(avctx, AV_LOG_ERROR, "invalid number of channels per XMA stream %d\n",
438 s->nb_channels);
439 return AVERROR_INVALIDDATA;
440 } else if (s->nb_channels > WMAPRO_MAX_CHANNELS || s->nb_channels > avctx->channels) {
441 avpriv_request_sample(avctx,
442 "More than %d channels", WMAPRO_MAX_CHANNELS);
443 return AVERROR_PATCHWELCOME;
444 }
445
446 /** init previous block len */
447 for (i = 0; i < s->nb_channels; i++)
448 s->channel[i].prev_block_len = s->samples_per_frame;
449
450 /** extract lfe channel position */
451 s->lfe_channel = -1;
452
453 if (channel_mask & 8) {
454 unsigned int mask;
455 for (mask = 1; mask < 16; mask <<= 1) {
456 if (channel_mask & mask)
457 ++s->lfe_channel;
458 }
459 }
460
461 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
462 scale_huffbits, 1, 1,
463 scale_huffcodes, 2, 2, 616);
464
465 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
466 scale_rl_huffbits, 1, 1,
467 scale_rl_huffcodes, 4, 4, 1406);
468
469 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
470 coef0_huffbits, 1, 1,
471 coef0_huffcodes, 4, 4, 2108);
472
473 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
474 coef1_huffbits, 1, 1,
475 coef1_huffcodes, 4, 4, 3912);
476
477 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
478 vec4_huffbits, 1, 1,
479 vec4_huffcodes, 2, 2, 604);
480
481 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
482 vec2_huffbits, 1, 1,
483 vec2_huffcodes, 2, 2, 562);
484
485 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
486 vec1_huffbits, 1, 1,
487 vec1_huffcodes, 2, 2, 562);
488
489 /** calculate number of scale factor bands and their offsets
490 for every possible block size */
491 for (i = 0; i < num_possible_block_sizes; i++) {
492 int subframe_len = s->samples_per_frame >> i;
493 int x;
494 int band = 1;
495 int rate = get_rate(avctx);
496
497 s->sfb_offsets[i][0] = 0;
498
499 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
500 int offset = (subframe_len * 2 * critical_freq[x]) / rate + 2;
501 offset &= ~3;
502 if (offset > s->sfb_offsets[i][band - 1])
503 s->sfb_offsets[i][band++] = offset;
504
505 if (offset >= subframe_len)
506 break;
507 }
508 s->sfb_offsets[i][band - 1] = subframe_len;
509 s->num_sfb[i] = band - 1;
510 if (s->num_sfb[i] <= 0) {
511 av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n");
512 return AVERROR_INVALIDDATA;
513 }
514 }
515
516
517 /** Scale factors can be shared between blocks of different size
518 as every block has a different scale factor band layout.
519 The matrix sf_offsets is needed to find the correct scale factor.
520 */
521
522 for (i = 0; i < num_possible_block_sizes; i++) {
523 int b;
524 for (b = 0; b < s->num_sfb[i]; b++) {
525 int x;
526 int offset = ((s->sfb_offsets[i][b]
527 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
528 for (x = 0; x < num_possible_block_sizes; x++) {
529 int v = 0;
530 while (s->sfb_offsets[x][v + 1] << x < offset) {
531 v++;
532 av_assert0(v < MAX_BANDS);
533 }
534 s->sf_offsets[i][x][b] = v;
535 }
536 }
537 }
538
539 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
540 if (!s->fdsp)
541 return AVERROR(ENOMEM);
542
543 /** init MDCT, FIXME: only init needed sizes */
544 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
545 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
546 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
547 / (1ll << (s->bits_per_sample - 1)));
548
549 /** init MDCT windows: simple sine window */
550 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
551 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
552 ff_init_ff_sine_windows(win_idx);
553 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
554 }
555
556 /** calculate subwoofer cutoff values */
557 for (i = 0; i < num_possible_block_sizes; i++) {
558 int block_size = s->samples_per_frame >> i;
559 int cutoff = (440*block_size + 3LL * (s->avctx->sample_rate >> 1) - 1)
560 / s->avctx->sample_rate;
561 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
562 }
563
564 /** calculate sine values for the decorrelation matrix */
565 for (i = 0; i < 33; i++)
566 sin64[i] = sin(i*M_PI / 64.0);
567
568 if (avctx->debug & FF_DEBUG_BITSTREAM)
569 dump_context(s);
570
571 avctx->channel_layout = channel_mask;
572
573 return 0;
574 }
575
576 /**
577 *@brief Initialize the decoder.
578 *@param avctx codec context
579 *@return 0 on success, -1 otherwise
580 */
wmapro_decode_init(AVCodecContext * avctx)581 static av_cold int wmapro_decode_init(AVCodecContext *avctx)
582 {
583 WMAProDecodeCtx *s = avctx->priv_data;
584
585 return decode_init(s, avctx, 0);
586 }
587
588 /**
589 *@brief Decode the subframe length.
590 *@param s context
591 *@param offset sample offset in the frame
592 *@return decoded subframe length on success, < 0 in case of an error
593 */
decode_subframe_length(WMAProDecodeCtx * s,int offset)594 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
595 {
596 int frame_len_shift = 0;
597 int subframe_len;
598
599 /** no need to read from the bitstream when only one length is possible */
600 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
601 return s->min_samples_per_subframe;
602
603 if (get_bits_left(&s->gb) < 1)
604 return AVERROR_INVALIDDATA;
605
606 /** 1 bit indicates if the subframe is of maximum length */
607 if (s->max_subframe_len_bit) {
608 if (get_bits1(&s->gb))
609 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
610 } else
611 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
612
613 subframe_len = s->samples_per_frame >> frame_len_shift;
614
615 /** sanity check the length */
616 if (subframe_len < s->min_samples_per_subframe ||
617 subframe_len > s->samples_per_frame) {
618 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
619 subframe_len);
620 return AVERROR_INVALIDDATA;
621 }
622 return subframe_len;
623 }
624
625 /**
626 *@brief Decode how the data in the frame is split into subframes.
627 * Every WMA frame contains the encoded data for a fixed number of
628 * samples per channel. The data for every channel might be split
629 * into several subframes. This function will reconstruct the list of
630 * subframes for every channel.
631 *
632 * If the subframes are not evenly split, the algorithm estimates the
633 * channels with the lowest number of total samples.
634 * Afterwards, for each of these channels a bit is read from the
635 * bitstream that indicates if the channel contains a subframe with the
636 * next subframe size that is going to be read from the bitstream or not.
637 * If a channel contains such a subframe, the subframe size gets added to
638 * the channel's subframe list.
639 * The algorithm repeats these steps until the frame is properly divided
640 * between the individual channels.
641 *
642 *@param s context
643 *@return 0 on success, < 0 in case of an error
644 */
decode_tilehdr(WMAProDecodeCtx * s)645 static int decode_tilehdr(WMAProDecodeCtx *s)
646 {
647 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
648 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
649 int channels_for_cur_subframe = s->nb_channels; /**< number of channels that contain the current subframe */
650 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
651 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
652 int c;
653
654 /* Should never consume more than 3073 bits (256 iterations for the
655 * while loop when always the minimum amount of 128 samples is subtracted
656 * from missing samples in the 8 channel case).
657 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
658 */
659
660 /** reset tiling information */
661 for (c = 0; c < s->nb_channels; c++)
662 s->channel[c].num_subframes = 0;
663
664 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
665 fixed_channel_layout = 1;
666
667 /** loop until the frame data is split between the subframes */
668 do {
669 int subframe_len;
670
671 /** check which channels contain the subframe */
672 for (c = 0; c < s->nb_channels; c++) {
673 if (num_samples[c] == min_channel_len) {
674 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
675 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
676 contains_subframe[c] = 1;
677 else
678 contains_subframe[c] = get_bits1(&s->gb);
679 } else
680 contains_subframe[c] = 0;
681 }
682
683 /** get subframe length, subframe_len == 0 is not allowed */
684 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
685 return AVERROR_INVALIDDATA;
686
687 /** add subframes to the individual channels and find new min_channel_len */
688 min_channel_len += subframe_len;
689 for (c = 0; c < s->nb_channels; c++) {
690 WMAProChannelCtx* chan = &s->channel[c];
691
692 if (contains_subframe[c]) {
693 if (chan->num_subframes >= MAX_SUBFRAMES) {
694 av_log(s->avctx, AV_LOG_ERROR,
695 "broken frame: num subframes > 31\n");
696 return AVERROR_INVALIDDATA;
697 }
698 chan->subframe_len[chan->num_subframes] = subframe_len;
699 num_samples[c] += subframe_len;
700 ++chan->num_subframes;
701 if (num_samples[c] > s->samples_per_frame) {
702 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
703 "channel len > samples_per_frame\n");
704 return AVERROR_INVALIDDATA;
705 }
706 } else if (num_samples[c] <= min_channel_len) {
707 if (num_samples[c] < min_channel_len) {
708 channels_for_cur_subframe = 0;
709 min_channel_len = num_samples[c];
710 }
711 ++channels_for_cur_subframe;
712 }
713 }
714 } while (min_channel_len < s->samples_per_frame);
715
716 for (c = 0; c < s->nb_channels; c++) {
717 int i;
718 int offset = 0;
719 for (i = 0; i < s->channel[c].num_subframes; i++) {
720 ff_dlog(s->avctx, "frame[%"PRIu32"] channel[%i] subframe[%i]"
721 " len %i\n", s->frame_num, c, i,
722 s->channel[c].subframe_len[i]);
723 s->channel[c].subframe_offset[i] = offset;
724 offset += s->channel[c].subframe_len[i];
725 }
726 }
727
728 return 0;
729 }
730
731 /**
732 *@brief Calculate a decorrelation matrix from the bitstream parameters.
733 *@param s codec context
734 *@param chgroup channel group for which the matrix needs to be calculated
735 */
decode_decorrelation_matrix(WMAProDecodeCtx * s,WMAProChannelGrp * chgroup)736 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
737 WMAProChannelGrp *chgroup)
738 {
739 int i;
740 int offset = 0;
741 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
742 memset(chgroup->decorrelation_matrix, 0, s->nb_channels *
743 s->nb_channels * sizeof(*chgroup->decorrelation_matrix));
744
745 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
746 rotation_offset[i] = get_bits(&s->gb, 6);
747
748 for (i = 0; i < chgroup->num_channels; i++)
749 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
750 get_bits1(&s->gb) ? 1.0 : -1.0;
751
752 for (i = 1; i < chgroup->num_channels; i++) {
753 int x;
754 for (x = 0; x < i; x++) {
755 int y;
756 for (y = 0; y < i + 1; y++) {
757 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
758 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
759 int n = rotation_offset[offset + x];
760 float sinv;
761 float cosv;
762
763 if (n < 32) {
764 sinv = sin64[n];
765 cosv = sin64[32 - n];
766 } else {
767 sinv = sin64[64 - n];
768 cosv = -sin64[n - 32];
769 }
770
771 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
772 (v1 * sinv) - (v2 * cosv);
773 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
774 (v1 * cosv) + (v2 * sinv);
775 }
776 }
777 offset += i;
778 }
779 }
780
781 /**
782 *@brief Decode channel transformation parameters
783 *@param s codec context
784 *@return >= 0 in case of success, < 0 in case of bitstream errors
785 */
decode_channel_transform(WMAProDecodeCtx * s)786 static int decode_channel_transform(WMAProDecodeCtx* s)
787 {
788 int i;
789 /* should never consume more than 1921 bits for the 8 channel case
790 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
791 * + MAX_CHANNELS + MAX_BANDS + 1)
792 */
793
794 /** in the one channel case channel transforms are pointless */
795 s->num_chgroups = 0;
796 if (s->nb_channels > 1) {
797 int remaining_channels = s->channels_for_cur_subframe;
798
799 if (get_bits1(&s->gb)) {
800 avpriv_request_sample(s->avctx,
801 "Channel transform bit");
802 return AVERROR_PATCHWELCOME;
803 }
804
805 for (s->num_chgroups = 0; remaining_channels &&
806 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
807 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
808 float** channel_data = chgroup->channel_data;
809 chgroup->num_channels = 0;
810 chgroup->transform = 0;
811
812 /** decode channel mask */
813 if (remaining_channels > 2) {
814 for (i = 0; i < s->channels_for_cur_subframe; i++) {
815 int channel_idx = s->channel_indexes_for_cur_subframe[i];
816 if (!s->channel[channel_idx].grouped
817 && get_bits1(&s->gb)) {
818 ++chgroup->num_channels;
819 s->channel[channel_idx].grouped = 1;
820 *channel_data++ = s->channel[channel_idx].coeffs;
821 }
822 }
823 } else {
824 chgroup->num_channels = remaining_channels;
825 for (i = 0; i < s->channels_for_cur_subframe; i++) {
826 int channel_idx = s->channel_indexes_for_cur_subframe[i];
827 if (!s->channel[channel_idx].grouped)
828 *channel_data++ = s->channel[channel_idx].coeffs;
829 s->channel[channel_idx].grouped = 1;
830 }
831 }
832
833 /** decode transform type */
834 if (chgroup->num_channels == 2) {
835 if (get_bits1(&s->gb)) {
836 if (get_bits1(&s->gb)) {
837 avpriv_request_sample(s->avctx,
838 "Unknown channel transform type");
839 return AVERROR_PATCHWELCOME;
840 }
841 } else {
842 chgroup->transform = 1;
843 if (s->nb_channels == 2) {
844 chgroup->decorrelation_matrix[0] = 1.0;
845 chgroup->decorrelation_matrix[1] = -1.0;
846 chgroup->decorrelation_matrix[2] = 1.0;
847 chgroup->decorrelation_matrix[3] = 1.0;
848 } else {
849 /** cos(pi/4) */
850 chgroup->decorrelation_matrix[0] = 0.70703125;
851 chgroup->decorrelation_matrix[1] = -0.70703125;
852 chgroup->decorrelation_matrix[2] = 0.70703125;
853 chgroup->decorrelation_matrix[3] = 0.70703125;
854 }
855 }
856 } else if (chgroup->num_channels > 2) {
857 if (get_bits1(&s->gb)) {
858 chgroup->transform = 1;
859 if (get_bits1(&s->gb)) {
860 decode_decorrelation_matrix(s, chgroup);
861 } else {
862 /** FIXME: more than 6 coupled channels not supported */
863 if (chgroup->num_channels > 6) {
864 avpriv_request_sample(s->avctx,
865 "Coupled channels > 6");
866 } else {
867 memcpy(chgroup->decorrelation_matrix,
868 default_decorrelation[chgroup->num_channels],
869 chgroup->num_channels * chgroup->num_channels *
870 sizeof(*chgroup->decorrelation_matrix));
871 }
872 }
873 }
874 }
875
876 /** decode transform on / off */
877 if (chgroup->transform) {
878 if (!get_bits1(&s->gb)) {
879 int i;
880 /** transform can be enabled for individual bands */
881 for (i = 0; i < s->num_bands; i++) {
882 chgroup->transform_band[i] = get_bits1(&s->gb);
883 }
884 } else {
885 memset(chgroup->transform_band, 1, s->num_bands);
886 }
887 }
888 remaining_channels -= chgroup->num_channels;
889 }
890 }
891 return 0;
892 }
893
894 /**
895 *@brief Extract the coefficients from the bitstream.
896 *@param s codec context
897 *@param c current channel number
898 *@return 0 on success, < 0 in case of bitstream errors
899 */
decode_coeffs(WMAProDecodeCtx * s,int c)900 static int decode_coeffs(WMAProDecodeCtx *s, int c)
901 {
902 /* Integers 0..15 as single-precision floats. The table saves a
903 costly int to float conversion, and storing the values as
904 integers allows fast sign-flipping. */
905 static const uint32_t fval_tab[16] = {
906 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
907 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
908 0x41000000, 0x41100000, 0x41200000, 0x41300000,
909 0x41400000, 0x41500000, 0x41600000, 0x41700000,
910 };
911 int vlctable;
912 VLC* vlc;
913 WMAProChannelCtx* ci = &s->channel[c];
914 int rl_mode = 0;
915 int cur_coeff = 0;
916 int num_zeros = 0;
917 const uint16_t* run;
918 const float* level;
919
920 ff_dlog(s->avctx, "decode coefficients for channel %i\n", c);
921
922 vlctable = get_bits1(&s->gb);
923 vlc = &coef_vlc[vlctable];
924
925 if (vlctable) {
926 run = coef1_run;
927 level = coef1_level;
928 } else {
929 run = coef0_run;
930 level = coef0_level;
931 }
932
933 /** decode vector coefficients (consumes up to 167 bits per iteration for
934 4 vector coded large values) */
935 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
936 (cur_coeff + 3 < ci->num_vec_coeffs)) {
937 uint32_t vals[4];
938 int i;
939 unsigned int idx;
940
941 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
942
943 if (idx == HUFF_VEC4_SIZE - 1) {
944 for (i = 0; i < 4; i += 2) {
945 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
946 if (idx == HUFF_VEC2_SIZE - 1) {
947 uint32_t v0, v1;
948 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
949 if (v0 == HUFF_VEC1_SIZE - 1)
950 v0 += ff_wma_get_large_val(&s->gb);
951 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
952 if (v1 == HUFF_VEC1_SIZE - 1)
953 v1 += ff_wma_get_large_val(&s->gb);
954 vals[i ] = av_float2int(v0);
955 vals[i+1] = av_float2int(v1);
956 } else {
957 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
958 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
959 }
960 }
961 } else {
962 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
963 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
964 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
965 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
966 }
967
968 /** decode sign */
969 for (i = 0; i < 4; i++) {
970 if (vals[i]) {
971 uint32_t sign = get_bits1(&s->gb) - 1;
972 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
973 num_zeros = 0;
974 } else {
975 ci->coeffs[cur_coeff] = 0;
976 /** switch to run level mode when subframe_len / 128 zeros
977 were found in a row */
978 rl_mode |= (++num_zeros > s->subframe_len >> 8);
979 }
980 ++cur_coeff;
981 }
982 }
983
984 /** decode run level coded coefficients */
985 if (cur_coeff < s->subframe_len) {
986 memset(&ci->coeffs[cur_coeff], 0,
987 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
988 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
989 level, run, 1, ci->coeffs,
990 cur_coeff, s->subframe_len,
991 s->subframe_len, s->esc_len, 0))
992 return AVERROR_INVALIDDATA;
993 }
994
995 return 0;
996 }
997
998 /**
999 *@brief Extract scale factors from the bitstream.
1000 *@param s codec context
1001 *@return 0 on success, < 0 in case of bitstream errors
1002 */
decode_scale_factors(WMAProDecodeCtx * s)1003 static int decode_scale_factors(WMAProDecodeCtx* s)
1004 {
1005 int i;
1006
1007 /** should never consume more than 5344 bits
1008 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
1009 */
1010
1011 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1012 int c = s->channel_indexes_for_cur_subframe[i];
1013 int* sf;
1014 int* sf_end;
1015 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
1016 sf_end = s->channel[c].scale_factors + s->num_bands;
1017
1018 /** resample scale factors for the new block size
1019 * as the scale factors might need to be resampled several times
1020 * before some new values are transmitted, a backup of the last
1021 * transmitted scale factors is kept in saved_scale_factors
1022 */
1023 if (s->channel[c].reuse_sf) {
1024 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
1025 int b;
1026 for (b = 0; b < s->num_bands; b++)
1027 s->channel[c].scale_factors[b] =
1028 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
1029 }
1030
1031 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
1032
1033 if (!s->channel[c].reuse_sf) {
1034 int val;
1035 /** decode DPCM coded scale factors */
1036 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
1037 val = 45 / s->channel[c].scale_factor_step;
1038 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
1039 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
1040 *sf = val;
1041 }
1042 } else {
1043 int i;
1044 /** run level decode differences to the resampled factors */
1045 for (i = 0; i < s->num_bands; i++) {
1046 int idx;
1047 int skip;
1048 int val;
1049 int sign;
1050
1051 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
1052
1053 if (!idx) {
1054 uint32_t code = get_bits(&s->gb, 14);
1055 val = code >> 6;
1056 sign = (code & 1) - 1;
1057 skip = (code & 0x3f) >> 1;
1058 } else if (idx == 1) {
1059 break;
1060 } else {
1061 skip = scale_rl_run[idx];
1062 val = scale_rl_level[idx];
1063 sign = get_bits1(&s->gb)-1;
1064 }
1065
1066 i += skip;
1067 if (i >= s->num_bands) {
1068 av_log(s->avctx, AV_LOG_ERROR,
1069 "invalid scale factor coding\n");
1070 return AVERROR_INVALIDDATA;
1071 }
1072 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
1073 }
1074 }
1075 /** swap buffers */
1076 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
1077 s->channel[c].table_idx = s->table_idx;
1078 s->channel[c].reuse_sf = 1;
1079 }
1080
1081 /** calculate new scale factor maximum */
1082 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
1083 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
1084 s->channel[c].max_scale_factor =
1085 FFMAX(s->channel[c].max_scale_factor, *sf);
1086 }
1087
1088 }
1089 return 0;
1090 }
1091
1092 /**
1093 *@brief Reconstruct the individual channel data.
1094 *@param s codec context
1095 */
inverse_channel_transform(WMAProDecodeCtx * s)1096 static void inverse_channel_transform(WMAProDecodeCtx *s)
1097 {
1098 int i;
1099
1100 for (i = 0; i < s->num_chgroups; i++) {
1101 if (s->chgroup[i].transform) {
1102 float data[WMAPRO_MAX_CHANNELS];
1103 const int num_channels = s->chgroup[i].num_channels;
1104 float** ch_data = s->chgroup[i].channel_data;
1105 float** ch_end = ch_data + num_channels;
1106 const int8_t* tb = s->chgroup[i].transform_band;
1107 int16_t* sfb;
1108
1109 /** multichannel decorrelation */
1110 for (sfb = s->cur_sfb_offsets;
1111 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1112 int y;
1113 if (*tb++ == 1) {
1114 /** multiply values with the decorrelation_matrix */
1115 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1116 const float* mat = s->chgroup[i].decorrelation_matrix;
1117 const float* data_end = data + num_channels;
1118 float* data_ptr = data;
1119 float** ch;
1120
1121 for (ch = ch_data; ch < ch_end; ch++)
1122 *data_ptr++ = (*ch)[y];
1123
1124 for (ch = ch_data; ch < ch_end; ch++) {
1125 float sum = 0;
1126 data_ptr = data;
1127 while (data_ptr < data_end)
1128 sum += *data_ptr++ * *mat++;
1129
1130 (*ch)[y] = sum;
1131 }
1132 }
1133 } else if (s->nb_channels == 2) {
1134 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1135 s->fdsp->vector_fmul_scalar(ch_data[0] + sfb[0],
1136 ch_data[0] + sfb[0],
1137 181.0 / 128, len);
1138 s->fdsp->vector_fmul_scalar(ch_data[1] + sfb[0],
1139 ch_data[1] + sfb[0],
1140 181.0 / 128, len);
1141 }
1142 }
1143 }
1144 }
1145 }
1146
1147 /**
1148 *@brief Apply sine window and reconstruct the output buffer.
1149 *@param s codec context
1150 */
wmapro_window(WMAProDecodeCtx * s)1151 static void wmapro_window(WMAProDecodeCtx *s)
1152 {
1153 int i;
1154 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1155 int c = s->channel_indexes_for_cur_subframe[i];
1156 const float* window;
1157 int winlen = s->channel[c].prev_block_len;
1158 float* start = s->channel[c].coeffs - (winlen >> 1);
1159
1160 if (s->subframe_len < winlen) {
1161 start += (winlen - s->subframe_len) >> 1;
1162 winlen = s->subframe_len;
1163 }
1164
1165 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1166
1167 winlen >>= 1;
1168
1169 s->fdsp->vector_fmul_window(start, start, start + winlen,
1170 window, winlen);
1171
1172 s->channel[c].prev_block_len = s->subframe_len;
1173 }
1174 }
1175
1176 /**
1177 *@brief Decode a single subframe (block).
1178 *@param s codec context
1179 *@return 0 on success, < 0 when decoding failed
1180 */
decode_subframe(WMAProDecodeCtx * s)1181 static int decode_subframe(WMAProDecodeCtx *s)
1182 {
1183 int offset = s->samples_per_frame;
1184 int subframe_len = s->samples_per_frame;
1185 int i;
1186 int total_samples = s->samples_per_frame * s->nb_channels;
1187 int transmit_coeffs = 0;
1188 int cur_subwoofer_cutoff;
1189
1190 s->subframe_offset = get_bits_count(&s->gb);
1191
1192 /** reset channel context and find the next block offset and size
1193 == the next block of the channel with the smallest number of
1194 decoded samples
1195 */
1196 for (i = 0; i < s->nb_channels; i++) {
1197 s->channel[i].grouped = 0;
1198 if (offset > s->channel[i].decoded_samples) {
1199 offset = s->channel[i].decoded_samples;
1200 subframe_len =
1201 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1202 }
1203 }
1204
1205 ff_dlog(s->avctx,
1206 "processing subframe with offset %i len %i\n", offset, subframe_len);
1207
1208 /** get a list of all channels that contain the estimated block */
1209 s->channels_for_cur_subframe = 0;
1210 for (i = 0; i < s->nb_channels; i++) {
1211 const int cur_subframe = s->channel[i].cur_subframe;
1212 /** subtract already processed samples */
1213 total_samples -= s->channel[i].decoded_samples;
1214
1215 /** and count if there are multiple subframes that match our profile */
1216 if (offset == s->channel[i].decoded_samples &&
1217 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1218 total_samples -= s->channel[i].subframe_len[cur_subframe];
1219 s->channel[i].decoded_samples +=
1220 s->channel[i].subframe_len[cur_subframe];
1221 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1222 ++s->channels_for_cur_subframe;
1223 }
1224 }
1225
1226 /** check if the frame will be complete after processing the
1227 estimated block */
1228 if (!total_samples)
1229 s->parsed_all_subframes = 1;
1230
1231
1232 ff_dlog(s->avctx, "subframe is part of %i channels\n",
1233 s->channels_for_cur_subframe);
1234
1235 /** calculate number of scale factor bands and their offsets */
1236 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1237 s->num_bands = s->num_sfb[s->table_idx];
1238 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1239 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1240
1241 /** configure the decoder for the current subframe */
1242 offset += s->samples_per_frame >> 1;
1243
1244 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1245 int c = s->channel_indexes_for_cur_subframe[i];
1246
1247 s->channel[c].coeffs = &s->channel[c].out[offset];
1248 }
1249
1250 s->subframe_len = subframe_len;
1251 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1252
1253 /** skip extended header if any */
1254 if (get_bits1(&s->gb)) {
1255 int num_fill_bits;
1256 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1257 int len = get_bits(&s->gb, 4);
1258 num_fill_bits = get_bitsz(&s->gb, len) + 1;
1259 }
1260
1261 if (num_fill_bits >= 0) {
1262 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1263 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1264 return AVERROR_INVALIDDATA;
1265 }
1266
1267 skip_bits_long(&s->gb, num_fill_bits);
1268 }
1269 }
1270
1271 /** no idea for what the following bit is used */
1272 if (get_bits1(&s->gb)) {
1273 avpriv_request_sample(s->avctx, "Reserved bit");
1274 return AVERROR_PATCHWELCOME;
1275 }
1276
1277
1278 if (decode_channel_transform(s) < 0)
1279 return AVERROR_INVALIDDATA;
1280
1281
1282 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1283 int c = s->channel_indexes_for_cur_subframe[i];
1284 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1285 transmit_coeffs = 1;
1286 }
1287
1288 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1289 if (transmit_coeffs) {
1290 int step;
1291 int quant_step = 90 * s->bits_per_sample >> 4;
1292
1293 /** decode number of vector coded coefficients */
1294 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1295 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1296 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1297 int c = s->channel_indexes_for_cur_subframe[i];
1298 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1299 if (num_vec_coeffs > s->subframe_len) {
1300 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1301 return AVERROR_INVALIDDATA;
1302 }
1303 av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out));
1304 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1305 }
1306 } else {
1307 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1308 int c = s->channel_indexes_for_cur_subframe[i];
1309 s->channel[c].num_vec_coeffs = s->subframe_len;
1310 }
1311 }
1312 /** decode quantization step */
1313 step = get_sbits(&s->gb, 6);
1314 quant_step += step;
1315 if (step == -32 || step == 31) {
1316 const int sign = (step == 31) - 1;
1317 int quant = 0;
1318 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1319 (step = get_bits(&s->gb, 5)) == 31) {
1320 quant += 31;
1321 }
1322 quant_step += ((quant + step) ^ sign) - sign;
1323 }
1324 if (quant_step < 0) {
1325 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1326 }
1327
1328 /** decode quantization step modifiers for every channel */
1329
1330 if (s->channels_for_cur_subframe == 1) {
1331 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1332 } else {
1333 int modifier_len = get_bits(&s->gb, 3);
1334 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1335 int c = s->channel_indexes_for_cur_subframe[i];
1336 s->channel[c].quant_step = quant_step;
1337 if (get_bits1(&s->gb)) {
1338 if (modifier_len) {
1339 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1340 } else
1341 ++s->channel[c].quant_step;
1342 }
1343 }
1344 }
1345
1346 /** decode scale factors */
1347 if (decode_scale_factors(s) < 0)
1348 return AVERROR_INVALIDDATA;
1349 }
1350
1351 ff_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1352 get_bits_count(&s->gb) - s->subframe_offset);
1353
1354 /** parse coefficients */
1355 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1356 int c = s->channel_indexes_for_cur_subframe[i];
1357 if (s->channel[c].transmit_coefs &&
1358 get_bits_count(&s->gb) < s->num_saved_bits) {
1359 decode_coeffs(s, c);
1360 } else
1361 memset(s->channel[c].coeffs, 0,
1362 sizeof(*s->channel[c].coeffs) * subframe_len);
1363 }
1364
1365 ff_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1366 get_bits_count(&s->gb) - s->subframe_offset);
1367
1368 if (transmit_coeffs) {
1369 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1370 /** reconstruct the per channel data */
1371 inverse_channel_transform(s);
1372 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1373 int c = s->channel_indexes_for_cur_subframe[i];
1374 const int* sf = s->channel[c].scale_factors;
1375 int b;
1376
1377 if (c == s->lfe_channel)
1378 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1379 (subframe_len - cur_subwoofer_cutoff));
1380
1381 /** inverse quantization and rescaling */
1382 for (b = 0; b < s->num_bands; b++) {
1383 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1384 const int exp = s->channel[c].quant_step -
1385 (s->channel[c].max_scale_factor - *sf++) *
1386 s->channel[c].scale_factor_step;
1387 const float quant = ff_exp10(exp / 20.0);
1388 int start = s->cur_sfb_offsets[b];
1389 s->fdsp->vector_fmul_scalar(s->tmp + start,
1390 s->channel[c].coeffs + start,
1391 quant, end - start);
1392 }
1393
1394 /** apply imdct (imdct_half == DCTIV with reverse) */
1395 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1396 }
1397 }
1398
1399 /** window and overlapp-add */
1400 wmapro_window(s);
1401
1402 /** handled one subframe */
1403 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1404 int c = s->channel_indexes_for_cur_subframe[i];
1405 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1406 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1407 return AVERROR_INVALIDDATA;
1408 }
1409 ++s->channel[c].cur_subframe;
1410 }
1411
1412 return 0;
1413 }
1414
1415 /**
1416 *@brief Decode one WMA frame.
1417 *@param s codec context
1418 *@return 0 if the trailer bit indicates that this is the last frame,
1419 * 1 if there are additional frames
1420 */
decode_frame(WMAProDecodeCtx * s,AVFrame * frame,int * got_frame_ptr)1421 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1422 {
1423 GetBitContext* gb = &s->gb;
1424 int more_frames = 0;
1425 int len = 0;
1426 int i;
1427
1428 /** get frame length */
1429 if (s->len_prefix)
1430 len = get_bits(gb, s->log2_frame_size);
1431
1432 ff_dlog(s->avctx, "decoding frame with length %x\n", len);
1433
1434 /** decode tile information */
1435 if (decode_tilehdr(s)) {
1436 s->packet_loss = 1;
1437 return 0;
1438 }
1439
1440 /** read postproc transform */
1441 if (s->nb_channels > 1 && get_bits1(gb)) {
1442 if (get_bits1(gb)) {
1443 for (i = 0; i < s->nb_channels * s->nb_channels; i++)
1444 skip_bits(gb, 4);
1445 }
1446 }
1447
1448 /** read drc info */
1449 if (s->dynamic_range_compression) {
1450 s->drc_gain = get_bits(gb, 8);
1451 ff_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1452 }
1453
1454 /** no idea what these are for, might be the number of samples
1455 that need to be skipped at the beginning or end of a stream */
1456 if (get_bits1(gb)) {
1457 int av_unused skip;
1458
1459 /** usually true for the first frame */
1460 if (get_bits1(gb)) {
1461 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1462 ff_dlog(s->avctx, "start skip: %i\n", skip);
1463 }
1464
1465 /** sometimes true for the last frame */
1466 if (get_bits1(gb)) {
1467 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1468 ff_dlog(s->avctx, "end skip: %i\n", skip);
1469 }
1470
1471 }
1472
1473 ff_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1474 get_bits_count(gb) - s->frame_offset);
1475
1476 /** reset subframe states */
1477 s->parsed_all_subframes = 0;
1478 for (i = 0; i < s->nb_channels; i++) {
1479 s->channel[i].decoded_samples = 0;
1480 s->channel[i].cur_subframe = 0;
1481 s->channel[i].reuse_sf = 0;
1482 }
1483
1484 /** decode all subframes */
1485 while (!s->parsed_all_subframes) {
1486 if (decode_subframe(s) < 0) {
1487 s->packet_loss = 1;
1488 return 0;
1489 }
1490 }
1491
1492 /** copy samples to the output buffer */
1493 for (i = 0; i < s->nb_channels; i++)
1494 memcpy(frame->extended_data[i], s->channel[i].out,
1495 s->samples_per_frame * sizeof(*s->channel[i].out));
1496
1497 for (i = 0; i < s->nb_channels; i++) {
1498 /** reuse second half of the IMDCT output for the next frame */
1499 memcpy(&s->channel[i].out[0],
1500 &s->channel[i].out[s->samples_per_frame],
1501 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1502 }
1503
1504 if (s->skip_frame) {
1505 s->skip_frame = 0;
1506 *got_frame_ptr = 0;
1507 av_frame_unref(frame);
1508 } else {
1509 *got_frame_ptr = 1;
1510 }
1511
1512 if (s->len_prefix) {
1513 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1514 /** FIXME: not sure if this is always an error */
1515 av_log(s->avctx, AV_LOG_ERROR,
1516 "frame[%"PRIu32"] would have to skip %i bits\n",
1517 s->frame_num,
1518 len - (get_bits_count(gb) - s->frame_offset) - 1);
1519 s->packet_loss = 1;
1520 return 0;
1521 }
1522
1523 /** skip the rest of the frame data */
1524 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1525 } else {
1526 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1527 }
1528 }
1529
1530 /** decode trailer bit */
1531 more_frames = get_bits1(gb);
1532
1533 ++s->frame_num;
1534 return more_frames;
1535 }
1536
1537 /**
1538 *@brief Calculate remaining input buffer length.
1539 *@param s codec context
1540 *@param gb bitstream reader context
1541 *@return remaining size in bits
1542 */
remaining_bits(WMAProDecodeCtx * s,GetBitContext * gb)1543 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1544 {
1545 return s->buf_bit_size - get_bits_count(gb);
1546 }
1547
1548 /**
1549 *@brief Fill the bit reservoir with a (partial) frame.
1550 *@param s codec context
1551 *@param gb bitstream reader context
1552 *@param len length of the partial frame
1553 *@param append decides whether to reset the buffer or not
1554 */
save_bits(WMAProDecodeCtx * s,GetBitContext * gb,int len,int append)1555 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1556 int append)
1557 {
1558 int buflen;
1559
1560 /** when the frame data does not need to be concatenated, the input buffer
1561 is reset and additional bits from the previous frame are copied
1562 and skipped later so that a fast byte copy is possible */
1563
1564 if (!append) {
1565 s->frame_offset = get_bits_count(gb) & 7;
1566 s->num_saved_bits = s->frame_offset;
1567 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1568 buflen = (s->num_saved_bits + len + 7) >> 3;
1569 } else
1570 buflen = (put_bits_count(&s->pb) + len + 7) >> 3;
1571
1572 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1573 avpriv_request_sample(s->avctx, "Too small input buffer");
1574 s->packet_loss = 1;
1575 return;
1576 }
1577
1578 av_assert0(len <= put_bits_left(&s->pb));
1579
1580 s->num_saved_bits += len;
1581 if (!append) {
1582 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1583 s->num_saved_bits);
1584 } else {
1585 int align = 8 - (get_bits_count(gb) & 7);
1586 align = FFMIN(align, len);
1587 put_bits(&s->pb, align, get_bits(gb, align));
1588 len -= align;
1589 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1590 }
1591 skip_bits_long(gb, len);
1592
1593 {
1594 PutBitContext tmp = s->pb;
1595 flush_put_bits(&tmp);
1596 }
1597
1598 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1599 skip_bits(&s->gb, s->frame_offset);
1600 }
1601
decode_packet(AVCodecContext * avctx,WMAProDecodeCtx * s,void * data,int * got_frame_ptr,AVPacket * avpkt)1602 static int decode_packet(AVCodecContext *avctx, WMAProDecodeCtx *s,
1603 void *data, int *got_frame_ptr, AVPacket *avpkt)
1604 {
1605 GetBitContext* gb = &s->pgb;
1606 const uint8_t* buf = avpkt->data;
1607 int buf_size = avpkt->size;
1608 int num_bits_prev_frame;
1609 int packet_sequence_number;
1610
1611 *got_frame_ptr = 0;
1612
1613 if (!buf_size) {
1614 AVFrame *frame = data;
1615 int i;
1616
1617 /** Must output remaining samples after stream end. WMAPRO 5.1 created
1618 * by XWMA encoder don't though (maybe only 1/2ch streams need it). */
1619 s->packet_done = 0;
1620 if (s->eof_done)
1621 return 0;
1622
1623 /** clean output buffer and copy last IMDCT samples */
1624 for (i = 0; i < s->nb_channels; i++) {
1625 memset(frame->extended_data[i], 0,
1626 s->samples_per_frame * sizeof(*s->channel[i].out));
1627
1628 memcpy(frame->extended_data[i], s->channel[i].out,
1629 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1630 }
1631
1632 /* TODO: XMA should output 128 samples only (instead of 512) and WMAPRO
1633 * maybe 768 (with 2048), XMA needs changes in multi-stream handling though. */
1634
1635 s->eof_done = 1;
1636 s->packet_done = 1;
1637 *got_frame_ptr = 1;
1638 return 0;
1639 }
1640 else if (s->packet_done || s->packet_loss) {
1641 s->packet_done = 0;
1642
1643 /** sanity check for the buffer length */
1644 if (avctx->codec_id == AV_CODEC_ID_WMAPRO && buf_size < avctx->block_align) {
1645 av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1646 buf_size, avctx->block_align);
1647 s->packet_loss = 1;
1648 return AVERROR_INVALIDDATA;
1649 }
1650
1651 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
1652 s->next_packet_start = buf_size - avctx->block_align;
1653 buf_size = avctx->block_align;
1654 } else {
1655 s->next_packet_start = buf_size - FFMIN(buf_size, avctx->block_align);
1656 buf_size = FFMIN(buf_size, avctx->block_align);
1657 }
1658 s->buf_bit_size = buf_size << 3;
1659
1660 /** parse packet header */
1661 init_get_bits(gb, buf, s->buf_bit_size);
1662 if (avctx->codec_id != AV_CODEC_ID_XMA2) {
1663 packet_sequence_number = get_bits(gb, 4);
1664 skip_bits(gb, 2);
1665 } else {
1666 int num_frames = get_bits(gb, 6);
1667 ff_dlog(avctx, "packet[%d]: number of frames %d\n", avctx->frame_number, num_frames);
1668 packet_sequence_number = 0;
1669 }
1670
1671 /** get number of bits that need to be added to the previous frame */
1672 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1673 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) {
1674 skip_bits(gb, 3);
1675 s->skip_packets = get_bits(gb, 8);
1676 ff_dlog(avctx, "packet[%d]: skip packets %d\n", avctx->frame_number, s->skip_packets);
1677 }
1678
1679 ff_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1680 num_bits_prev_frame);
1681
1682 /** check for packet loss */
1683 if (avctx->codec_id == AV_CODEC_ID_WMAPRO && !s->packet_loss &&
1684 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1685 s->packet_loss = 1;
1686 av_log(avctx, AV_LOG_ERROR,
1687 "Packet loss detected! seq %"PRIx8" vs %x\n",
1688 s->packet_sequence_number, packet_sequence_number);
1689 }
1690 s->packet_sequence_number = packet_sequence_number;
1691
1692 if (num_bits_prev_frame > 0) {
1693 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1694 if (num_bits_prev_frame >= remaining_packet_bits) {
1695 num_bits_prev_frame = remaining_packet_bits;
1696 s->packet_done = 1;
1697 }
1698
1699 /** append the previous frame data to the remaining data from the
1700 previous packet to create a full frame */
1701 save_bits(s, gb, num_bits_prev_frame, 1);
1702 ff_dlog(avctx, "accumulated %x bits of frame data\n",
1703 s->num_saved_bits - s->frame_offset);
1704
1705 /** decode the cross packet frame if it is valid */
1706 if (!s->packet_loss)
1707 decode_frame(s, data, got_frame_ptr);
1708 } else if (s->num_saved_bits - s->frame_offset) {
1709 ff_dlog(avctx, "ignoring %x previously saved bits\n",
1710 s->num_saved_bits - s->frame_offset);
1711 }
1712
1713 if (s->packet_loss) {
1714 /** reset number of saved bits so that the decoder
1715 does not start to decode incomplete frames in the
1716 s->len_prefix == 0 case */
1717 s->num_saved_bits = 0;
1718 s->packet_loss = 0;
1719 }
1720 } else {
1721 int frame_size;
1722
1723 if (avpkt->size < s->next_packet_start) {
1724 s->packet_loss = 1;
1725 return AVERROR_INVALIDDATA;
1726 }
1727
1728 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1729 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1730 skip_bits(gb, s->packet_offset);
1731 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1732 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1733 frame_size <= remaining_bits(s, gb)) {
1734 save_bits(s, gb, frame_size, 0);
1735 if (!s->packet_loss)
1736 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1737 } else if (!s->len_prefix
1738 && s->num_saved_bits > get_bits_count(&s->gb)) {
1739 /** when the frames do not have a length prefix, we don't know
1740 the compressed length of the individual frames
1741 however, we know what part of a new packet belongs to the
1742 previous frame
1743 therefore we save the incoming packet first, then we append
1744 the "previous frame" data from the next packet so that
1745 we get a buffer that only contains full frames */
1746 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1747 } else {
1748 s->packet_done = 1;
1749 }
1750 }
1751
1752 if (remaining_bits(s, gb) < 0) {
1753 av_log(avctx, AV_LOG_ERROR, "Overread %d\n", -remaining_bits(s, gb));
1754 s->packet_loss = 1;
1755 }
1756
1757 if (s->packet_done && !s->packet_loss &&
1758 remaining_bits(s, gb) > 0) {
1759 /** save the rest of the data so that it can be decoded
1760 with the next packet */
1761 save_bits(s, gb, remaining_bits(s, gb), 0);
1762 }
1763
1764 s->packet_offset = get_bits_count(gb) & 7;
1765 if (s->packet_loss)
1766 return AVERROR_INVALIDDATA;
1767
1768 return get_bits_count(gb) >> 3;
1769 }
1770
1771 /**
1772 *@brief Decode a single WMA packet.
1773 *@param avctx codec context
1774 *@param data the output buffer
1775 *@param avpkt input packet
1776 *@return number of bytes that were read from the input buffer
1777 */
wmapro_decode_packet(AVCodecContext * avctx,void * data,int * got_frame_ptr,AVPacket * avpkt)1778 static int wmapro_decode_packet(AVCodecContext *avctx, void *data,
1779 int *got_frame_ptr, AVPacket *avpkt)
1780 {
1781 WMAProDecodeCtx *s = avctx->priv_data;
1782 AVFrame *frame = data;
1783 int ret;
1784
1785 /* get output buffer */
1786 frame->nb_samples = s->samples_per_frame;
1787 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1788 s->packet_loss = 1;
1789 return 0;
1790 }
1791
1792 return decode_packet(avctx, s, data, got_frame_ptr, avpkt);
1793 }
1794
xma_decode_packet(AVCodecContext * avctx,void * data,int * got_frame_ptr,AVPacket * avpkt)1795 static int xma_decode_packet(AVCodecContext *avctx, void *data,
1796 int *got_frame_ptr, AVPacket *avpkt)
1797 {
1798 XMADecodeCtx *s = avctx->priv_data;
1799 int got_stream_frame_ptr = 0;
1800 AVFrame *frame = data;
1801 int i, ret, offset = INT_MAX;
1802
1803 if (!s->frames[s->current_stream]->data[0]) {
1804 s->frames[s->current_stream]->nb_samples = 512;
1805 if ((ret = ff_get_buffer(avctx, s->frames[s->current_stream], 0)) < 0) {
1806 return ret;
1807 }
1808 }
1809 /* decode current stream packet */
1810 ret = decode_packet(avctx, &s->xma[s->current_stream], s->frames[s->current_stream],
1811 &got_stream_frame_ptr, avpkt);
1812
1813 if (got_stream_frame_ptr && s->offset[s->current_stream] >= 64) {
1814 got_stream_frame_ptr = 0;
1815 ret = AVERROR_INVALIDDATA;
1816 }
1817
1818 /* copy stream samples (1/2ch) to sample buffer (Nch) */
1819 if (got_stream_frame_ptr) {
1820 int start_ch = s->start_channel[s->current_stream];
1821 memcpy(&s->samples[start_ch + 0][s->offset[s->current_stream] * 512],
1822 s->frames[s->current_stream]->extended_data[0], 512 * 4);
1823 if (s->xma[s->current_stream].nb_channels > 1)
1824 memcpy(&s->samples[start_ch + 1][s->offset[s->current_stream] * 512],
1825 s->frames[s->current_stream]->extended_data[1], 512 * 4);
1826 s->offset[s->current_stream]++;
1827 } else if (ret < 0) {
1828 memset(s->offset, 0, sizeof(s->offset));
1829 s->current_stream = 0;
1830 return ret;
1831 }
1832
1833 /* find next XMA packet's owner stream, and update.
1834 * XMA streams find their packets following packet_skips
1835 * (at start there is one packet per stream, then interleave non-linearly). */
1836 if (s->xma[s->current_stream].packet_done ||
1837 s->xma[s->current_stream].packet_loss) {
1838
1839 /* select stream with 0 skip_packets (= uses next packet) */
1840 if (s->xma[s->current_stream].skip_packets != 0) {
1841 int min[2];
1842
1843 min[0] = s->xma[0].skip_packets;
1844 min[1] = i = 0;
1845
1846 for (i = 1; i < s->num_streams; i++) {
1847 if (s->xma[i].skip_packets < min[0]) {
1848 min[0] = s->xma[i].skip_packets;
1849 min[1] = i;
1850 }
1851 }
1852
1853 s->current_stream = min[1];
1854 }
1855
1856 /* all other streams skip next packet */
1857 for (i = 0; i < s->num_streams; i++) {
1858 s->xma[i].skip_packets = FFMAX(0, s->xma[i].skip_packets - 1);
1859 }
1860
1861 /* copy samples from buffer to output if possible */
1862 for (i = 0; i < s->num_streams; i++) {
1863 offset = FFMIN(offset, s->offset[i]);
1864 }
1865 if (offset > 0) {
1866 int bret;
1867
1868 frame->nb_samples = 512 * offset;
1869 if ((bret = ff_get_buffer(avctx, frame, 0)) < 0)
1870 return bret;
1871
1872 /* copy samples buffer (Nch) to frame samples (Nch), move unconsumed samples */
1873 for (i = 0; i < s->num_streams; i++) {
1874 int start_ch = s->start_channel[i];
1875 memcpy(frame->extended_data[start_ch + 0], s->samples[start_ch + 0], frame->nb_samples * 4);
1876 if (s->xma[i].nb_channels > 1)
1877 memcpy(frame->extended_data[start_ch + 1], s->samples[start_ch + 1], frame->nb_samples * 4);
1878
1879 s->offset[i] -= offset;
1880 if (s->offset[i]) {
1881 memmove(s->samples[start_ch + 0], s->samples[start_ch + 0] + frame->nb_samples, s->offset[i] * 4 * 512);
1882 if (s->xma[i].nb_channels > 1)
1883 memmove(s->samples[start_ch + 1], s->samples[start_ch + 1] + frame->nb_samples, s->offset[i] * 4 * 512);
1884 }
1885 }
1886
1887 *got_frame_ptr = 1;
1888 }
1889 }
1890
1891 return ret;
1892 }
1893
xma_decode_init(AVCodecContext * avctx)1894 static av_cold int xma_decode_init(AVCodecContext *avctx)
1895 {
1896 XMADecodeCtx *s = avctx->priv_data;
1897 int i, ret, start_channels = 0;
1898
1899 if (avctx->channels <= 0 || avctx->extradata_size == 0)
1900 return AVERROR_INVALIDDATA;
1901
1902 /* get stream config */
1903 if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size == 34) { /* XMA2WAVEFORMATEX */
1904 s->num_streams = (avctx->channels + 1) / 2;
1905 } else if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size >= 2) { /* XMA2WAVEFORMAT */
1906 s->num_streams = avctx->extradata[1];
1907 if (avctx->extradata_size != (32 + ((avctx->extradata[0]==3)?0:8) + 4*s->num_streams)) {
1908 av_log(avctx, AV_LOG_ERROR, "Incorrect XMA2 extradata size\n");
1909 s->num_streams = 0;
1910 return AVERROR(EINVAL);
1911 }
1912 } else if (avctx->codec_id == AV_CODEC_ID_XMA1 && avctx->extradata_size >= 4) { /* XMAWAVEFORMAT */
1913 s->num_streams = avctx->extradata[4];
1914 if (avctx->extradata_size != (8 + 20*s->num_streams)) {
1915 av_log(avctx, AV_LOG_ERROR, "Incorrect XMA1 extradata size\n");
1916 s->num_streams = 0;
1917 return AVERROR(EINVAL);
1918 }
1919 } else {
1920 av_log(avctx, AV_LOG_ERROR, "Incorrect XMA config\n");
1921 return AVERROR(EINVAL);
1922 }
1923
1924 /* encoder supports up to 64 streams / 64*2 channels (would have to alloc arrays) */
1925 if (avctx->channels > XMA_MAX_CHANNELS || s->num_streams > XMA_MAX_STREAMS ||
1926 s->num_streams <= 0
1927 ) {
1928 avpriv_request_sample(avctx, "More than %d channels in %d streams", XMA_MAX_CHANNELS, s->num_streams);
1929 s->num_streams = 0;
1930 return AVERROR_PATCHWELCOME;
1931 }
1932
1933 /* init all streams (several streams of 1/2ch make Nch files) */
1934 for (i = 0; i < s->num_streams; i++) {
1935 ret = decode_init(&s->xma[i], avctx, i);
1936 if (ret < 0)
1937 return ret;
1938 s->frames[i] = av_frame_alloc();
1939 if (!s->frames[i])
1940 return AVERROR(ENOMEM);
1941
1942 s->start_channel[i] = start_channels;
1943 start_channels += s->xma[i].nb_channels;
1944 }
1945 if (start_channels != avctx->channels)
1946 return AVERROR_INVALIDDATA;
1947
1948 return ret;
1949 }
1950
xma_decode_end(AVCodecContext * avctx)1951 static av_cold int xma_decode_end(AVCodecContext *avctx)
1952 {
1953 XMADecodeCtx *s = avctx->priv_data;
1954 int i;
1955
1956 for (i = 0; i < s->num_streams; i++) {
1957 decode_end(&s->xma[i]);
1958 av_frame_free(&s->frames[i]);
1959 }
1960 s->num_streams = 0;
1961
1962 return 0;
1963 }
1964
flush(WMAProDecodeCtx * s)1965 static void flush(WMAProDecodeCtx *s)
1966 {
1967 int i;
1968 /** reset output buffer as a part of it is used during the windowing of a
1969 new frame */
1970 for (i = 0; i < s->nb_channels; i++)
1971 memset(s->channel[i].out, 0, s->samples_per_frame *
1972 sizeof(*s->channel[i].out));
1973 s->packet_loss = 1;
1974 s->skip_packets = 0;
1975 s->eof_done = 0;
1976 }
1977
1978
1979 /**
1980 *@brief Clear decoder buffers (for seeking).
1981 *@param avctx codec context
1982 */
wmapro_flush(AVCodecContext * avctx)1983 static void wmapro_flush(AVCodecContext *avctx)
1984 {
1985 WMAProDecodeCtx *s = avctx->priv_data;
1986
1987 flush(s);
1988 }
1989
xma_flush(AVCodecContext * avctx)1990 static void xma_flush(AVCodecContext *avctx)
1991 {
1992 XMADecodeCtx *s = avctx->priv_data;
1993 int i;
1994
1995 for (i = 0; i < s->num_streams; i++)
1996 flush(&s->xma[i]);
1997
1998 memset(s->offset, 0, sizeof(s->offset));
1999 s->current_stream = 0;
2000 }
2001
2002
2003 /**
2004 *@brief wmapro decoder
2005 */
2006 AVCodec ff_wmapro_decoder = {
2007 .name = "wmapro",
2008 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
2009 .type = AVMEDIA_TYPE_AUDIO,
2010 .id = AV_CODEC_ID_WMAPRO,
2011 .priv_data_size = sizeof(WMAProDecodeCtx),
2012 .init = wmapro_decode_init,
2013 .close = wmapro_decode_end,
2014 .decode = wmapro_decode_packet,
2015 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
2016 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
2017 .flush = wmapro_flush,
2018 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
2019 AV_SAMPLE_FMT_NONE },
2020 };
2021
2022 AVCodec ff_xma1_decoder = {
2023 .name = "xma1",
2024 .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 1"),
2025 .type = AVMEDIA_TYPE_AUDIO,
2026 .id = AV_CODEC_ID_XMA1,
2027 .priv_data_size = sizeof(XMADecodeCtx),
2028 .init = xma_decode_init,
2029 .close = xma_decode_end,
2030 .decode = xma_decode_packet,
2031 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
2032 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
2033 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
2034 AV_SAMPLE_FMT_NONE },
2035 };
2036
2037 AVCodec ff_xma2_decoder = {
2038 .name = "xma2",
2039 .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 2"),
2040 .type = AVMEDIA_TYPE_AUDIO,
2041 .id = AV_CODEC_ID_XMA2,
2042 .priv_data_size = sizeof(XMADecodeCtx),
2043 .init = xma_decode_init,
2044 .close = xma_decode_end,
2045 .decode = xma_decode_packet,
2046 .flush = xma_flush,
2047 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
2048 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
2049 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
2050 AV_SAMPLE_FMT_NONE },
2051 };
2052