blob: 5c99628c52d28e868106ac849b3926bbab050ee8
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 | |
110 | #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size |
111 | #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size |
112 | #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size |
113 | #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size |
114 | #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes |
115 | |
116 | |
117 | #define VLCBITS 9 |
118 | #define SCALEVLCBITS 8 |
119 | #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS) |
120 | #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS) |
121 | #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS) |
122 | #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS) |
123 | #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS) |
124 | |
125 | static VLC sf_vlc; ///< scale factor DPCM vlc |
126 | static VLC sf_rl_vlc; ///< scale factor run length vlc |
127 | static VLC vec4_vlc; ///< 4 coefficients per symbol |
128 | static VLC vec2_vlc; ///< 2 coefficients per symbol |
129 | static VLC vec1_vlc; ///< 1 coefficient per symbol |
130 | static VLC coef_vlc[2]; ///< coefficient run length vlc codes |
131 | static float sin64[33]; ///< sine table for decorrelation |
132 | |
133 | /** |
134 | * @brief frame specific decoder context for a single channel |
135 | */ |
136 | typedef struct WMAProChannelCtx { |
137 | int16_t prev_block_len; ///< length of the previous block |
138 | uint8_t transmit_coefs; |
139 | uint8_t num_subframes; |
140 | uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples |
141 | uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame |
142 | uint8_t cur_subframe; ///< current subframe number |
143 | uint16_t decoded_samples; ///< number of already processed samples |
144 | uint8_t grouped; ///< channel is part of a group |
145 | int quant_step; ///< quantization step for the current subframe |
146 | int8_t reuse_sf; ///< share scale factors between subframes |
147 | int8_t scale_factor_step; ///< scaling step for the current subframe |
148 | int max_scale_factor; ///< maximum scale factor for the current subframe |
149 | int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values |
150 | int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling) |
151 | int* scale_factors; ///< pointer to the scale factor values used for decoding |
152 | uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block |
153 | float* coeffs; ///< pointer to the subframe decode buffer |
154 | uint16_t num_vec_coeffs; ///< number of vector coded coefficients |
155 | DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer |
156 | } WMAProChannelCtx; |
157 | |
158 | /** |
159 | * @brief channel group for channel transformations |
160 | */ |
161 | typedef struct WMAProChannelGrp { |
162 | uint8_t num_channels; ///< number of channels in the group |
163 | int8_t transform; ///< transform on / off |
164 | int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band |
165 | float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS]; |
166 | float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients |
167 | } WMAProChannelGrp; |
168 | |
169 | /** |
170 | * @brief main decoder context |
171 | */ |
172 | typedef struct WMAProDecodeCtx { |
173 | /* generic decoder variables */ |
174 | AVCodecContext* avctx; ///< codec context for av_log |
175 | AVFloatDSPContext *fdsp; |
176 | uint8_t frame_data[MAX_FRAMESIZE + |
177 | AV_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data |
178 | PutBitContext pb; ///< context for filling the frame_data buffer |
179 | FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size |
180 | DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer |
181 | const float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes |
182 | |
183 | /* frame size dependent frame information (set during initialization) */ |
184 | uint32_t decode_flags; ///< used compression features |
185 | uint8_t len_prefix; ///< frame is prefixed with its length |
186 | uint8_t dynamic_range_compression; ///< frame contains DRC data |
187 | uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0]) |
188 | uint16_t samples_per_frame; ///< number of samples to output |
189 | uint16_t log2_frame_size; |
190 | int8_t lfe_channel; ///< lfe channel index |
191 | uint8_t max_num_subframes; |
192 | uint8_t subframe_len_bits; ///< number of bits used for the subframe length |
193 | uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1 |
194 | uint16_t min_samples_per_subframe; |
195 | int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size |
196 | int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4) |
197 | int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix |
198 | int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values |
199 | |
200 | /* packet decode state */ |
201 | GetBitContext pgb; ///< bitstream reader context for the packet |
202 | int next_packet_start; ///< start offset of the next wma packet in the demuxer packet |
203 | uint8_t packet_offset; ///< frame offset in the packet |
204 | uint8_t packet_sequence_number; ///< current packet number |
205 | int num_saved_bits; ///< saved number of bits |
206 | int frame_offset; ///< frame offset in the bit reservoir |
207 | int subframe_offset; ///< subframe offset in the bit reservoir |
208 | uint8_t packet_loss; ///< set in case of bitstream error |
209 | uint8_t packet_done; ///< set when a packet is fully decoded |
210 | |
211 | /* frame decode state */ |
212 | uint32_t frame_num; ///< current frame number (not used for decoding) |
213 | GetBitContext gb; ///< bitstream reader context |
214 | int buf_bit_size; ///< buffer size in bits |
215 | uint8_t drc_gain; ///< gain for the DRC tool |
216 | int8_t skip_frame; ///< skip output step |
217 | int8_t parsed_all_subframes; ///< all subframes decoded? |
218 | uint8_t skip_packets; |
219 | |
220 | /* subframe/block decode state */ |
221 | int16_t subframe_len; ///< current subframe length |
222 | int8_t nb_channels; ///< number of channels in stream (XMA1/2) |
223 | int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe |
224 | int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS]; |
225 | int8_t num_bands; ///< number of scale factor bands |
226 | int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream |
227 | int16_t* cur_sfb_offsets; ///< sfb offsets for the current block |
228 | uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables |
229 | int8_t esc_len; ///< length of escaped coefficients |
230 | |
231 | uint8_t num_chgroups; ///< number of channel groups |
232 | WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information |
233 | |
234 | WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data |
235 | } WMAProDecodeCtx; |
236 | |
237 | typedef struct XMADecodeCtx { |
238 | WMAProDecodeCtx xma[4]; |
239 | AVFrame *frames[4]; |
240 | int current_stream; |
241 | float samples[8][512 * 64]; |
242 | int offset[4]; |
243 | } XMADecodeCtx; |
244 | |
245 | /** |
246 | *@brief helper function to print the most important members of the context |
247 | *@param s context |
248 | */ |
249 | static av_cold void dump_context(WMAProDecodeCtx *s) |
250 | { |
251 | #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b); |
252 | #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b); |
253 | |
254 | PRINT("ed sample bit depth", s->bits_per_sample); |
255 | PRINT_HEX("ed decode flags", s->decode_flags); |
256 | PRINT("samples per frame", s->samples_per_frame); |
257 | PRINT("log2 frame size", s->log2_frame_size); |
258 | PRINT("max num subframes", s->max_num_subframes); |
259 | PRINT("len prefix", s->len_prefix); |
260 | PRINT("num channels", s->nb_channels); |
261 | } |
262 | |
263 | /** |
264 | *@brief Uninitialize the decoder and free all resources. |
265 | *@param avctx codec context |
266 | *@return 0 on success, < 0 otherwise |
267 | */ |
268 | static av_cold int decode_end(WMAProDecodeCtx *s) |
269 | { |
270 | int i; |
271 | |
272 | av_freep(&s->fdsp); |
273 | |
274 | for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) |
275 | ff_mdct_end(&s->mdct_ctx[i]); |
276 | |
277 | return 0; |
278 | } |
279 | |
280 | static av_cold int wmapro_decode_end(AVCodecContext *avctx) |
281 | { |
282 | WMAProDecodeCtx *s = avctx->priv_data; |
283 | |
284 | decode_end(s); |
285 | |
286 | return 0; |
287 | } |
288 | |
289 | static av_cold int get_rate(AVCodecContext *avctx) |
290 | { |
291 | if (avctx->codec_id != AV_CODEC_ID_WMAPRO) { // XXX: is this really only for XMA? |
292 | if (avctx->sample_rate > 44100) |
293 | return 48000; |
294 | else if (avctx->sample_rate > 32000) |
295 | return 44100; |
296 | else if (avctx->sample_rate > 24000) |
297 | return 32000; |
298 | return 24000; |
299 | } |
300 | |
301 | return avctx->sample_rate; |
302 | } |
303 | |
304 | /** |
305 | *@brief Initialize the decoder. |
306 | *@param avctx codec context |
307 | *@return 0 on success, -1 otherwise |
308 | */ |
309 | static av_cold int decode_init(WMAProDecodeCtx *s, AVCodecContext *avctx) |
310 | { |
311 | uint8_t *edata_ptr = avctx->extradata; |
312 | unsigned int channel_mask; |
313 | int i, bits; |
314 | int log2_max_num_subframes; |
315 | int num_possible_block_sizes; |
316 | |
317 | if (avctx->codec_id == AV_CODEC_ID_XMA1 || avctx->codec_id == AV_CODEC_ID_XMA2) |
318 | avctx->block_align = 2048; |
319 | |
320 | if (!avctx->block_align) { |
321 | av_log(avctx, AV_LOG_ERROR, "block_align is not set\n"); |
322 | return AVERROR(EINVAL); |
323 | } |
324 | |
325 | s->avctx = avctx; |
326 | |
327 | init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE); |
328 | |
329 | avctx->sample_fmt = AV_SAMPLE_FMT_FLTP; |
330 | |
331 | /** dump the extradata */ |
332 | av_log(avctx, AV_LOG_DEBUG, "extradata:\n"); |
333 | for (i = 0; i < avctx->extradata_size; i++) |
334 | av_log(avctx, AV_LOG_DEBUG, "[%x] ", avctx->extradata[i]); |
335 | av_log(avctx, AV_LOG_DEBUG, "\n"); |
336 | if (avctx->codec_id == AV_CODEC_ID_XMA2 && (!avctx->extradata || avctx->extradata_size >= 6)) { |
337 | s->decode_flags = 0x10d6; |
338 | channel_mask = avctx->extradata ? AV_RL32(edata_ptr+2) : 0; |
339 | s->bits_per_sample = 16; |
340 | } else if (avctx->codec_id == AV_CODEC_ID_XMA1) { |
341 | s->decode_flags = 0x10d6; |
342 | s->bits_per_sample = 16; |
343 | channel_mask = 0; |
344 | } else if (avctx->codec_id == AV_CODEC_ID_WMAPRO && avctx->extradata_size >= 18) { |
345 | s->decode_flags = AV_RL16(edata_ptr+14); |
346 | channel_mask = AV_RL32(edata_ptr+2); |
347 | s->bits_per_sample = AV_RL16(edata_ptr); |
348 | |
349 | if (s->bits_per_sample > 32 || s->bits_per_sample < 1) { |
350 | avpriv_request_sample(avctx, "bits per sample is %d", s->bits_per_sample); |
351 | return AVERROR_PATCHWELCOME; |
352 | } |
353 | } else { |
354 | avpriv_request_sample(avctx, "Unknown extradata size"); |
355 | return AVERROR_PATCHWELCOME; |
356 | } |
357 | |
358 | if (avctx->codec_id != AV_CODEC_ID_WMAPRO && avctx->channels > 2) { |
359 | s->nb_channels = 2; |
360 | } else { |
361 | s->nb_channels = avctx->channels; |
362 | } |
363 | |
364 | /** generic init */ |
365 | s->log2_frame_size = av_log2(avctx->block_align) + 4; |
366 | if (s->log2_frame_size > 25) { |
367 | avpriv_request_sample(avctx, "Large block align"); |
368 | return AVERROR_PATCHWELCOME; |
369 | } |
370 | |
371 | /** frame info */ |
372 | if (avctx->codec_id != AV_CODEC_ID_WMAPRO) |
373 | s->skip_frame = 0; |
374 | else |
375 | s->skip_frame = 1; /* skip first frame */ |
376 | |
377 | s->packet_loss = 1; |
378 | s->len_prefix = (s->decode_flags & 0x40); |
379 | |
380 | /** get frame len */ |
381 | if (avctx->codec_id == AV_CODEC_ID_WMAPRO) { |
382 | bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags); |
383 | if (bits > WMAPRO_BLOCK_MAX_BITS) { |
384 | avpriv_request_sample(avctx, "14-bit block sizes"); |
385 | return AVERROR_PATCHWELCOME; |
386 | } |
387 | s->samples_per_frame = 1 << bits; |
388 | } else { |
389 | s->samples_per_frame = 512; |
390 | } |
391 | |
392 | /** subframe info */ |
393 | log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3); |
394 | s->max_num_subframes = 1 << log2_max_num_subframes; |
395 | if (s->max_num_subframes == 16 || s->max_num_subframes == 4) |
396 | s->max_subframe_len_bit = 1; |
397 | s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1; |
398 | |
399 | num_possible_block_sizes = log2_max_num_subframes + 1; |
400 | s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes; |
401 | s->dynamic_range_compression = (s->decode_flags & 0x80); |
402 | |
403 | if (s->max_num_subframes > MAX_SUBFRAMES) { |
404 | av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n", |
405 | s->max_num_subframes); |
406 | return AVERROR_INVALIDDATA; |
407 | } |
408 | |
409 | if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) { |
410 | av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n", |
411 | s->min_samples_per_subframe); |
412 | return AVERROR_INVALIDDATA; |
413 | } |
414 | |
415 | if (s->avctx->sample_rate <= 0) { |
416 | av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n"); |
417 | return AVERROR_INVALIDDATA; |
418 | } |
419 | |
420 | if (s->nb_channels <= 0) { |
421 | av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", |
422 | s->nb_channels); |
423 | return AVERROR_INVALIDDATA; |
424 | } else if (s->nb_channels > WMAPRO_MAX_CHANNELS) { |
425 | avpriv_request_sample(avctx, |
426 | "More than %d channels", WMAPRO_MAX_CHANNELS); |
427 | return AVERROR_PATCHWELCOME; |
428 | } |
429 | |
430 | /** init previous block len */ |
431 | for (i = 0; i < s->nb_channels; i++) |
432 | s->channel[i].prev_block_len = s->samples_per_frame; |
433 | |
434 | /** extract lfe channel position */ |
435 | s->lfe_channel = -1; |
436 | |
437 | if (channel_mask & 8) { |
438 | unsigned int mask; |
439 | for (mask = 1; mask < 16; mask <<= 1) { |
440 | if (channel_mask & mask) |
441 | ++s->lfe_channel; |
442 | } |
443 | } |
444 | |
445 | INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE, |
446 | scale_huffbits, 1, 1, |
447 | scale_huffcodes, 2, 2, 616); |
448 | |
449 | INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE, |
450 | scale_rl_huffbits, 1, 1, |
451 | scale_rl_huffcodes, 4, 4, 1406); |
452 | |
453 | INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE, |
454 | coef0_huffbits, 1, 1, |
455 | coef0_huffcodes, 4, 4, 2108); |
456 | |
457 | INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE, |
458 | coef1_huffbits, 1, 1, |
459 | coef1_huffcodes, 4, 4, 3912); |
460 | |
461 | INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE, |
462 | vec4_huffbits, 1, 1, |
463 | vec4_huffcodes, 2, 2, 604); |
464 | |
465 | INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE, |
466 | vec2_huffbits, 1, 1, |
467 | vec2_huffcodes, 2, 2, 562); |
468 | |
469 | INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE, |
470 | vec1_huffbits, 1, 1, |
471 | vec1_huffcodes, 2, 2, 562); |
472 | |
473 | /** calculate number of scale factor bands and their offsets |
474 | for every possible block size */ |
475 | for (i = 0; i < num_possible_block_sizes; i++) { |
476 | int subframe_len = s->samples_per_frame >> i; |
477 | int x; |
478 | int band = 1; |
479 | int rate = get_rate(avctx); |
480 | |
481 | s->sfb_offsets[i][0] = 0; |
482 | |
483 | for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) { |
484 | int offset = (subframe_len * 2 * critical_freq[x]) / rate + 2; |
485 | offset &= ~3; |
486 | if (offset > s->sfb_offsets[i][band - 1]) |
487 | s->sfb_offsets[i][band++] = offset; |
488 | |
489 | if (offset >= subframe_len) |
490 | break; |
491 | } |
492 | s->sfb_offsets[i][band - 1] = subframe_len; |
493 | s->num_sfb[i] = band - 1; |
494 | if (s->num_sfb[i] <= 0) { |
495 | av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n"); |
496 | return AVERROR_INVALIDDATA; |
497 | } |
498 | } |
499 | |
500 | |
501 | /** Scale factors can be shared between blocks of different size |
502 | as every block has a different scale factor band layout. |
503 | The matrix sf_offsets is needed to find the correct scale factor. |
504 | */ |
505 | |
506 | for (i = 0; i < num_possible_block_sizes; i++) { |
507 | int b; |
508 | for (b = 0; b < s->num_sfb[i]; b++) { |
509 | int x; |
510 | int offset = ((s->sfb_offsets[i][b] |
511 | + s->sfb_offsets[i][b + 1] - 1) << i) >> 1; |
512 | for (x = 0; x < num_possible_block_sizes; x++) { |
513 | int v = 0; |
514 | while (s->sfb_offsets[x][v + 1] << x < offset) { |
515 | v++; |
516 | av_assert0(v < MAX_BANDS); |
517 | } |
518 | s->sf_offsets[i][x][b] = v; |
519 | } |
520 | } |
521 | } |
522 | |
523 | s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT); |
524 | if (!s->fdsp) |
525 | return AVERROR(ENOMEM); |
526 | |
527 | /** init MDCT, FIXME: only init needed sizes */ |
528 | for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) |
529 | ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1, |
530 | 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1)) |
531 | / (1 << (s->bits_per_sample - 1))); |
532 | |
533 | /** init MDCT windows: simple sine window */ |
534 | for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) { |
535 | const int win_idx = WMAPRO_BLOCK_MAX_BITS - i; |
536 | ff_init_ff_sine_windows(win_idx); |
537 | s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx]; |
538 | } |
539 | |
540 | /** calculate subwoofer cutoff values */ |
541 | for (i = 0; i < num_possible_block_sizes; i++) { |
542 | int block_size = s->samples_per_frame >> i; |
543 | int cutoff = (440*block_size + 3LL * (s->avctx->sample_rate >> 1) - 1) |
544 | / s->avctx->sample_rate; |
545 | s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size); |
546 | } |
547 | |
548 | /** calculate sine values for the decorrelation matrix */ |
549 | for (i = 0; i < 33; i++) |
550 | sin64[i] = sin(i*M_PI / 64.0); |
551 | |
552 | if (avctx->debug & FF_DEBUG_BITSTREAM) |
553 | dump_context(s); |
554 | |
555 | avctx->channel_layout = channel_mask; |
556 | |
557 | return 0; |
558 | } |
559 | |
560 | /** |
561 | *@brief Initialize the decoder. |
562 | *@param avctx codec context |
563 | *@return 0 on success, -1 otherwise |
564 | */ |
565 | static av_cold int wmapro_decode_init(AVCodecContext *avctx) |
566 | { |
567 | WMAProDecodeCtx *s = avctx->priv_data; |
568 | |
569 | return decode_init(s, avctx); |
570 | } |
571 | |
572 | /** |
573 | *@brief Decode the subframe length. |
574 | *@param s context |
575 | *@param offset sample offset in the frame |
576 | *@return decoded subframe length on success, < 0 in case of an error |
577 | */ |
578 | static int decode_subframe_length(WMAProDecodeCtx *s, int offset) |
579 | { |
580 | int frame_len_shift = 0; |
581 | int subframe_len; |
582 | |
583 | /** no need to read from the bitstream when only one length is possible */ |
584 | if (offset == s->samples_per_frame - s->min_samples_per_subframe) |
585 | return s->min_samples_per_subframe; |
586 | |
587 | if (get_bits_left(&s->gb) < 1) |
588 | return AVERROR_INVALIDDATA; |
589 | |
590 | /** 1 bit indicates if the subframe is of maximum length */ |
591 | if (s->max_subframe_len_bit) { |
592 | if (get_bits1(&s->gb)) |
593 | frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1); |
594 | } else |
595 | frame_len_shift = get_bits(&s->gb, s->subframe_len_bits); |
596 | |
597 | subframe_len = s->samples_per_frame >> frame_len_shift; |
598 | |
599 | /** sanity check the length */ |
600 | if (subframe_len < s->min_samples_per_subframe || |
601 | subframe_len > s->samples_per_frame) { |
602 | av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n", |
603 | subframe_len); |
604 | return AVERROR_INVALIDDATA; |
605 | } |
606 | return subframe_len; |
607 | } |
608 | |
609 | /** |
610 | *@brief Decode how the data in the frame is split into subframes. |
611 | * Every WMA frame contains the encoded data for a fixed number of |
612 | * samples per channel. The data for every channel might be split |
613 | * into several subframes. This function will reconstruct the list of |
614 | * subframes for every channel. |
615 | * |
616 | * If the subframes are not evenly split, the algorithm estimates the |
617 | * channels with the lowest number of total samples. |
618 | * Afterwards, for each of these channels a bit is read from the |
619 | * bitstream that indicates if the channel contains a subframe with the |
620 | * next subframe size that is going to be read from the bitstream or not. |
621 | * If a channel contains such a subframe, the subframe size gets added to |
622 | * the channel's subframe list. |
623 | * The algorithm repeats these steps until the frame is properly divided |
624 | * between the individual channels. |
625 | * |
626 | *@param s context |
627 | *@return 0 on success, < 0 in case of an error |
628 | */ |
629 | static int decode_tilehdr(WMAProDecodeCtx *s) |
630 | { |
631 | uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */ |
632 | uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */ |
633 | int channels_for_cur_subframe = s->nb_channels; /**< number of channels that contain the current subframe */ |
634 | int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */ |
635 | int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */ |
636 | int c; |
637 | |
638 | /* Should never consume more than 3073 bits (256 iterations for the |
639 | * while loop when always the minimum amount of 128 samples is subtracted |
640 | * from missing samples in the 8 channel case). |
641 | * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4) |
642 | */ |
643 | |
644 | /** reset tiling information */ |
645 | for (c = 0; c < s->nb_channels; c++) |
646 | s->channel[c].num_subframes = 0; |
647 | |
648 | if (s->max_num_subframes == 1 || get_bits1(&s->gb)) |
649 | fixed_channel_layout = 1; |
650 | |
651 | /** loop until the frame data is split between the subframes */ |
652 | do { |
653 | int subframe_len; |
654 | |
655 | /** check which channels contain the subframe */ |
656 | for (c = 0; c < s->nb_channels; c++) { |
657 | if (num_samples[c] == min_channel_len) { |
658 | if (fixed_channel_layout || channels_for_cur_subframe == 1 || |
659 | (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) |
660 | contains_subframe[c] = 1; |
661 | else |
662 | contains_subframe[c] = get_bits1(&s->gb); |
663 | } else |
664 | contains_subframe[c] = 0; |
665 | } |
666 | |
667 | /** get subframe length, subframe_len == 0 is not allowed */ |
668 | if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0) |
669 | return AVERROR_INVALIDDATA; |
670 | |
671 | /** add subframes to the individual channels and find new min_channel_len */ |
672 | min_channel_len += subframe_len; |
673 | for (c = 0; c < s->nb_channels; c++) { |
674 | WMAProChannelCtx* chan = &s->channel[c]; |
675 | |
676 | if (contains_subframe[c]) { |
677 | if (chan->num_subframes >= MAX_SUBFRAMES) { |
678 | av_log(s->avctx, AV_LOG_ERROR, |
679 | "broken frame: num subframes > 31\n"); |
680 | return AVERROR_INVALIDDATA; |
681 | } |
682 | chan->subframe_len[chan->num_subframes] = subframe_len; |
683 | num_samples[c] += subframe_len; |
684 | ++chan->num_subframes; |
685 | if (num_samples[c] > s->samples_per_frame) { |
686 | av_log(s->avctx, AV_LOG_ERROR, "broken frame: " |
687 | "channel len > samples_per_frame\n"); |
688 | return AVERROR_INVALIDDATA; |
689 | } |
690 | } else if (num_samples[c] <= min_channel_len) { |
691 | if (num_samples[c] < min_channel_len) { |
692 | channels_for_cur_subframe = 0; |
693 | min_channel_len = num_samples[c]; |
694 | } |
695 | ++channels_for_cur_subframe; |
696 | } |
697 | } |
698 | } while (min_channel_len < s->samples_per_frame); |
699 | |
700 | for (c = 0; c < s->nb_channels; c++) { |
701 | int i; |
702 | int offset = 0; |
703 | for (i = 0; i < s->channel[c].num_subframes; i++) { |
704 | ff_dlog(s->avctx, "frame[%"PRIu32"] channel[%i] subframe[%i]" |
705 | " len %i\n", s->frame_num, c, i, |
706 | s->channel[c].subframe_len[i]); |
707 | s->channel[c].subframe_offset[i] = offset; |
708 | offset += s->channel[c].subframe_len[i]; |
709 | } |
710 | } |
711 | |
712 | return 0; |
713 | } |
714 | |
715 | /** |
716 | *@brief Calculate a decorrelation matrix from the bitstream parameters. |
717 | *@param s codec context |
718 | *@param chgroup channel group for which the matrix needs to be calculated |
719 | */ |
720 | static void decode_decorrelation_matrix(WMAProDecodeCtx *s, |
721 | WMAProChannelGrp *chgroup) |
722 | { |
723 | int i; |
724 | int offset = 0; |
725 | int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS]; |
726 | memset(chgroup->decorrelation_matrix, 0, s->nb_channels * |
727 | s->nb_channels * sizeof(*chgroup->decorrelation_matrix)); |
728 | |
729 | for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++) |
730 | rotation_offset[i] = get_bits(&s->gb, 6); |
731 | |
732 | for (i = 0; i < chgroup->num_channels; i++) |
733 | chgroup->decorrelation_matrix[chgroup->num_channels * i + i] = |
734 | get_bits1(&s->gb) ? 1.0 : -1.0; |
735 | |
736 | for (i = 1; i < chgroup->num_channels; i++) { |
737 | int x; |
738 | for (x = 0; x < i; x++) { |
739 | int y; |
740 | for (y = 0; y < i + 1; y++) { |
741 | float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y]; |
742 | float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y]; |
743 | int n = rotation_offset[offset + x]; |
744 | float sinv; |
745 | float cosv; |
746 | |
747 | if (n < 32) { |
748 | sinv = sin64[n]; |
749 | cosv = sin64[32 - n]; |
750 | } else { |
751 | sinv = sin64[64 - n]; |
752 | cosv = -sin64[n - 32]; |
753 | } |
754 | |
755 | chgroup->decorrelation_matrix[y + x * chgroup->num_channels] = |
756 | (v1 * sinv) - (v2 * cosv); |
757 | chgroup->decorrelation_matrix[y + i * chgroup->num_channels] = |
758 | (v1 * cosv) + (v2 * sinv); |
759 | } |
760 | } |
761 | offset += i; |
762 | } |
763 | } |
764 | |
765 | /** |
766 | *@brief Decode channel transformation parameters |
767 | *@param s codec context |
768 | *@return >= 0 in case of success, < 0 in case of bitstream errors |
769 | */ |
770 | static int decode_channel_transform(WMAProDecodeCtx* s) |
771 | { |
772 | int i; |
773 | /* should never consume more than 1921 bits for the 8 channel case |
774 | * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS |
775 | * + MAX_CHANNELS + MAX_BANDS + 1) |
776 | */ |
777 | |
778 | /** in the one channel case channel transforms are pointless */ |
779 | s->num_chgroups = 0; |
780 | if (s->nb_channels > 1) { |
781 | int remaining_channels = s->channels_for_cur_subframe; |
782 | |
783 | if (get_bits1(&s->gb)) { |
784 | avpriv_request_sample(s->avctx, |
785 | "Channel transform bit"); |
786 | return AVERROR_PATCHWELCOME; |
787 | } |
788 | |
789 | for (s->num_chgroups = 0; remaining_channels && |
790 | s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) { |
791 | WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups]; |
792 | float** channel_data = chgroup->channel_data; |
793 | chgroup->num_channels = 0; |
794 | chgroup->transform = 0; |
795 | |
796 | /** decode channel mask */ |
797 | if (remaining_channels > 2) { |
798 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
799 | int channel_idx = s->channel_indexes_for_cur_subframe[i]; |
800 | if (!s->channel[channel_idx].grouped |
801 | && get_bits1(&s->gb)) { |
802 | ++chgroup->num_channels; |
803 | s->channel[channel_idx].grouped = 1; |
804 | *channel_data++ = s->channel[channel_idx].coeffs; |
805 | } |
806 | } |
807 | } else { |
808 | chgroup->num_channels = remaining_channels; |
809 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
810 | int channel_idx = s->channel_indexes_for_cur_subframe[i]; |
811 | if (!s->channel[channel_idx].grouped) |
812 | *channel_data++ = s->channel[channel_idx].coeffs; |
813 | s->channel[channel_idx].grouped = 1; |
814 | } |
815 | } |
816 | |
817 | /** decode transform type */ |
818 | if (chgroup->num_channels == 2) { |
819 | if (get_bits1(&s->gb)) { |
820 | if (get_bits1(&s->gb)) { |
821 | avpriv_request_sample(s->avctx, |
822 | "Unknown channel transform type"); |
823 | return AVERROR_PATCHWELCOME; |
824 | } |
825 | } else { |
826 | chgroup->transform = 1; |
827 | if (s->nb_channels == 2) { |
828 | chgroup->decorrelation_matrix[0] = 1.0; |
829 | chgroup->decorrelation_matrix[1] = -1.0; |
830 | chgroup->decorrelation_matrix[2] = 1.0; |
831 | chgroup->decorrelation_matrix[3] = 1.0; |
832 | } else { |
833 | /** cos(pi/4) */ |
834 | chgroup->decorrelation_matrix[0] = 0.70703125; |
835 | chgroup->decorrelation_matrix[1] = -0.70703125; |
836 | chgroup->decorrelation_matrix[2] = 0.70703125; |
837 | chgroup->decorrelation_matrix[3] = 0.70703125; |
838 | } |
839 | } |
840 | } else if (chgroup->num_channels > 2) { |
841 | if (get_bits1(&s->gb)) { |
842 | chgroup->transform = 1; |
843 | if (get_bits1(&s->gb)) { |
844 | decode_decorrelation_matrix(s, chgroup); |
845 | } else { |
846 | /** FIXME: more than 6 coupled channels not supported */ |
847 | if (chgroup->num_channels > 6) { |
848 | avpriv_request_sample(s->avctx, |
849 | "Coupled channels > 6"); |
850 | } else { |
851 | memcpy(chgroup->decorrelation_matrix, |
852 | default_decorrelation[chgroup->num_channels], |
853 | chgroup->num_channels * chgroup->num_channels * |
854 | sizeof(*chgroup->decorrelation_matrix)); |
855 | } |
856 | } |
857 | } |
858 | } |
859 | |
860 | /** decode transform on / off */ |
861 | if (chgroup->transform) { |
862 | if (!get_bits1(&s->gb)) { |
863 | int i; |
864 | /** transform can be enabled for individual bands */ |
865 | for (i = 0; i < s->num_bands; i++) { |
866 | chgroup->transform_band[i] = get_bits1(&s->gb); |
867 | } |
868 | } else { |
869 | memset(chgroup->transform_band, 1, s->num_bands); |
870 | } |
871 | } |
872 | remaining_channels -= chgroup->num_channels; |
873 | } |
874 | } |
875 | return 0; |
876 | } |
877 | |
878 | /** |
879 | *@brief Extract the coefficients from the bitstream. |
880 | *@param s codec context |
881 | *@param c current channel number |
882 | *@return 0 on success, < 0 in case of bitstream errors |
883 | */ |
884 | static int decode_coeffs(WMAProDecodeCtx *s, int c) |
885 | { |
886 | /* Integers 0..15 as single-precision floats. The table saves a |
887 | costly int to float conversion, and storing the values as |
888 | integers allows fast sign-flipping. */ |
889 | static const uint32_t fval_tab[16] = { |
890 | 0x00000000, 0x3f800000, 0x40000000, 0x40400000, |
891 | 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000, |
892 | 0x41000000, 0x41100000, 0x41200000, 0x41300000, |
893 | 0x41400000, 0x41500000, 0x41600000, 0x41700000, |
894 | }; |
895 | int vlctable; |
896 | VLC* vlc; |
897 | WMAProChannelCtx* ci = &s->channel[c]; |
898 | int rl_mode = 0; |
899 | int cur_coeff = 0; |
900 | int num_zeros = 0; |
901 | const uint16_t* run; |
902 | const float* level; |
903 | |
904 | ff_dlog(s->avctx, "decode coefficients for channel %i\n", c); |
905 | |
906 | vlctable = get_bits1(&s->gb); |
907 | vlc = &coef_vlc[vlctable]; |
908 | |
909 | if (vlctable) { |
910 | run = coef1_run; |
911 | level = coef1_level; |
912 | } else { |
913 | run = coef0_run; |
914 | level = coef0_level; |
915 | } |
916 | |
917 | /** decode vector coefficients (consumes up to 167 bits per iteration for |
918 | 4 vector coded large values) */ |
919 | while ((s->transmit_num_vec_coeffs || !rl_mode) && |
920 | (cur_coeff + 3 < ci->num_vec_coeffs)) { |
921 | uint32_t vals[4]; |
922 | int i; |
923 | unsigned int idx; |
924 | |
925 | idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH); |
926 | |
927 | if (idx == HUFF_VEC4_SIZE - 1) { |
928 | for (i = 0; i < 4; i += 2) { |
929 | idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH); |
930 | if (idx == HUFF_VEC2_SIZE - 1) { |
931 | uint32_t v0, v1; |
932 | v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH); |
933 | if (v0 == HUFF_VEC1_SIZE - 1) |
934 | v0 += ff_wma_get_large_val(&s->gb); |
935 | v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH); |
936 | if (v1 == HUFF_VEC1_SIZE - 1) |
937 | v1 += ff_wma_get_large_val(&s->gb); |
938 | vals[i ] = av_float2int(v0); |
939 | vals[i+1] = av_float2int(v1); |
940 | } else { |
941 | vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ]; |
942 | vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF]; |
943 | } |
944 | } |
945 | } else { |
946 | vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ]; |
947 | vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF]; |
948 | vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF]; |
949 | vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF]; |
950 | } |
951 | |
952 | /** decode sign */ |
953 | for (i = 0; i < 4; i++) { |
954 | if (vals[i]) { |
955 | uint32_t sign = get_bits1(&s->gb) - 1; |
956 | AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31); |
957 | num_zeros = 0; |
958 | } else { |
959 | ci->coeffs[cur_coeff] = 0; |
960 | /** switch to run level mode when subframe_len / 128 zeros |
961 | were found in a row */ |
962 | rl_mode |= (++num_zeros > s->subframe_len >> 8); |
963 | } |
964 | ++cur_coeff; |
965 | } |
966 | } |
967 | |
968 | /** decode run level coded coefficients */ |
969 | if (cur_coeff < s->subframe_len) { |
970 | memset(&ci->coeffs[cur_coeff], 0, |
971 | sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff)); |
972 | if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc, |
973 | level, run, 1, ci->coeffs, |
974 | cur_coeff, s->subframe_len, |
975 | s->subframe_len, s->esc_len, 0)) |
976 | return AVERROR_INVALIDDATA; |
977 | } |
978 | |
979 | return 0; |
980 | } |
981 | |
982 | /** |
983 | *@brief Extract scale factors from the bitstream. |
984 | *@param s codec context |
985 | *@return 0 on success, < 0 in case of bitstream errors |
986 | */ |
987 | static int decode_scale_factors(WMAProDecodeCtx* s) |
988 | { |
989 | int i; |
990 | |
991 | /** should never consume more than 5344 bits |
992 | * MAX_CHANNELS * (1 + MAX_BANDS * 23) |
993 | */ |
994 | |
995 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
996 | int c = s->channel_indexes_for_cur_subframe[i]; |
997 | int* sf; |
998 | int* sf_end; |
999 | s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx]; |
1000 | sf_end = s->channel[c].scale_factors + s->num_bands; |
1001 | |
1002 | /** resample scale factors for the new block size |
1003 | * as the scale factors might need to be resampled several times |
1004 | * before some new values are transmitted, a backup of the last |
1005 | * transmitted scale factors is kept in saved_scale_factors |
1006 | */ |
1007 | if (s->channel[c].reuse_sf) { |
1008 | const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx]; |
1009 | int b; |
1010 | for (b = 0; b < s->num_bands; b++) |
1011 | s->channel[c].scale_factors[b] = |
1012 | s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++]; |
1013 | } |
1014 | |
1015 | if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) { |
1016 | |
1017 | if (!s->channel[c].reuse_sf) { |
1018 | int val; |
1019 | /** decode DPCM coded scale factors */ |
1020 | s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1; |
1021 | val = 45 / s->channel[c].scale_factor_step; |
1022 | for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) { |
1023 | val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60; |
1024 | *sf = val; |
1025 | } |
1026 | } else { |
1027 | int i; |
1028 | /** run level decode differences to the resampled factors */ |
1029 | for (i = 0; i < s->num_bands; i++) { |
1030 | int idx; |
1031 | int skip; |
1032 | int val; |
1033 | int sign; |
1034 | |
1035 | idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH); |
1036 | |
1037 | if (!idx) { |
1038 | uint32_t code = get_bits(&s->gb, 14); |
1039 | val = code >> 6; |
1040 | sign = (code & 1) - 1; |
1041 | skip = (code & 0x3f) >> 1; |
1042 | } else if (idx == 1) { |
1043 | break; |
1044 | } else { |
1045 | skip = scale_rl_run[idx]; |
1046 | val = scale_rl_level[idx]; |
1047 | sign = get_bits1(&s->gb)-1; |
1048 | } |
1049 | |
1050 | i += skip; |
1051 | if (i >= s->num_bands) { |
1052 | av_log(s->avctx, AV_LOG_ERROR, |
1053 | "invalid scale factor coding\n"); |
1054 | return AVERROR_INVALIDDATA; |
1055 | } |
1056 | s->channel[c].scale_factors[i] += (val ^ sign) - sign; |
1057 | } |
1058 | } |
1059 | /** swap buffers */ |
1060 | s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx; |
1061 | s->channel[c].table_idx = s->table_idx; |
1062 | s->channel[c].reuse_sf = 1; |
1063 | } |
1064 | |
1065 | /** calculate new scale factor maximum */ |
1066 | s->channel[c].max_scale_factor = s->channel[c].scale_factors[0]; |
1067 | for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) { |
1068 | s->channel[c].max_scale_factor = |
1069 | FFMAX(s->channel[c].max_scale_factor, *sf); |
1070 | } |
1071 | |
1072 | } |
1073 | return 0; |
1074 | } |
1075 | |
1076 | /** |
1077 | *@brief Reconstruct the individual channel data. |
1078 | *@param s codec context |
1079 | */ |
1080 | static void inverse_channel_transform(WMAProDecodeCtx *s) |
1081 | { |
1082 | int i; |
1083 | |
1084 | for (i = 0; i < s->num_chgroups; i++) { |
1085 | if (s->chgroup[i].transform) { |
1086 | float data[WMAPRO_MAX_CHANNELS]; |
1087 | const int num_channels = s->chgroup[i].num_channels; |
1088 | float** ch_data = s->chgroup[i].channel_data; |
1089 | float** ch_end = ch_data + num_channels; |
1090 | const int8_t* tb = s->chgroup[i].transform_band; |
1091 | int16_t* sfb; |
1092 | |
1093 | /** multichannel decorrelation */ |
1094 | for (sfb = s->cur_sfb_offsets; |
1095 | sfb < s->cur_sfb_offsets + s->num_bands; sfb++) { |
1096 | int y; |
1097 | if (*tb++ == 1) { |
1098 | /** multiply values with the decorrelation_matrix */ |
1099 | for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) { |
1100 | const float* mat = s->chgroup[i].decorrelation_matrix; |
1101 | const float* data_end = data + num_channels; |
1102 | float* data_ptr = data; |
1103 | float** ch; |
1104 | |
1105 | for (ch = ch_data; ch < ch_end; ch++) |
1106 | *data_ptr++ = (*ch)[y]; |
1107 | |
1108 | for (ch = ch_data; ch < ch_end; ch++) { |
1109 | float sum = 0; |
1110 | data_ptr = data; |
1111 | while (data_ptr < data_end) |
1112 | sum += *data_ptr++ * *mat++; |
1113 | |
1114 | (*ch)[y] = sum; |
1115 | } |
1116 | } |
1117 | } else if (s->nb_channels == 2) { |
1118 | int len = FFMIN(sfb[1], s->subframe_len) - sfb[0]; |
1119 | s->fdsp->vector_fmul_scalar(ch_data[0] + sfb[0], |
1120 | ch_data[0] + sfb[0], |
1121 | 181.0 / 128, len); |
1122 | s->fdsp->vector_fmul_scalar(ch_data[1] + sfb[0], |
1123 | ch_data[1] + sfb[0], |
1124 | 181.0 / 128, len); |
1125 | } |
1126 | } |
1127 | } |
1128 | } |
1129 | } |
1130 | |
1131 | /** |
1132 | *@brief Apply sine window and reconstruct the output buffer. |
1133 | *@param s codec context |
1134 | */ |
1135 | static void wmapro_window(WMAProDecodeCtx *s) |
1136 | { |
1137 | int i; |
1138 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
1139 | int c = s->channel_indexes_for_cur_subframe[i]; |
1140 | const float* window; |
1141 | int winlen = s->channel[c].prev_block_len; |
1142 | float* start = s->channel[c].coeffs - (winlen >> 1); |
1143 | |
1144 | if (s->subframe_len < winlen) { |
1145 | start += (winlen - s->subframe_len) >> 1; |
1146 | winlen = s->subframe_len; |
1147 | } |
1148 | |
1149 | window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS]; |
1150 | |
1151 | winlen >>= 1; |
1152 | |
1153 | s->fdsp->vector_fmul_window(start, start, start + winlen, |
1154 | window, winlen); |
1155 | |
1156 | s->channel[c].prev_block_len = s->subframe_len; |
1157 | } |
1158 | } |
1159 | |
1160 | /** |
1161 | *@brief Decode a single subframe (block). |
1162 | *@param s codec context |
1163 | *@return 0 on success, < 0 when decoding failed |
1164 | */ |
1165 | static int decode_subframe(WMAProDecodeCtx *s) |
1166 | { |
1167 | int offset = s->samples_per_frame; |
1168 | int subframe_len = s->samples_per_frame; |
1169 | int i; |
1170 | int total_samples = s->samples_per_frame * s->nb_channels; |
1171 | int transmit_coeffs = 0; |
1172 | int cur_subwoofer_cutoff; |
1173 | |
1174 | s->subframe_offset = get_bits_count(&s->gb); |
1175 | |
1176 | /** reset channel context and find the next block offset and size |
1177 | == the next block of the channel with the smallest number of |
1178 | decoded samples |
1179 | */ |
1180 | for (i = 0; i < s->nb_channels; i++) { |
1181 | s->channel[i].grouped = 0; |
1182 | if (offset > s->channel[i].decoded_samples) { |
1183 | offset = s->channel[i].decoded_samples; |
1184 | subframe_len = |
1185 | s->channel[i].subframe_len[s->channel[i].cur_subframe]; |
1186 | } |
1187 | } |
1188 | |
1189 | ff_dlog(s->avctx, |
1190 | "processing subframe with offset %i len %i\n", offset, subframe_len); |
1191 | |
1192 | /** get a list of all channels that contain the estimated block */ |
1193 | s->channels_for_cur_subframe = 0; |
1194 | for (i = 0; i < s->nb_channels; i++) { |
1195 | const int cur_subframe = s->channel[i].cur_subframe; |
1196 | /** subtract already processed samples */ |
1197 | total_samples -= s->channel[i].decoded_samples; |
1198 | |
1199 | /** and count if there are multiple subframes that match our profile */ |
1200 | if (offset == s->channel[i].decoded_samples && |
1201 | subframe_len == s->channel[i].subframe_len[cur_subframe]) { |
1202 | total_samples -= s->channel[i].subframe_len[cur_subframe]; |
1203 | s->channel[i].decoded_samples += |
1204 | s->channel[i].subframe_len[cur_subframe]; |
1205 | s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i; |
1206 | ++s->channels_for_cur_subframe; |
1207 | } |
1208 | } |
1209 | |
1210 | /** check if the frame will be complete after processing the |
1211 | estimated block */ |
1212 | if (!total_samples) |
1213 | s->parsed_all_subframes = 1; |
1214 | |
1215 | |
1216 | ff_dlog(s->avctx, "subframe is part of %i channels\n", |
1217 | s->channels_for_cur_subframe); |
1218 | |
1219 | /** calculate number of scale factor bands and their offsets */ |
1220 | s->table_idx = av_log2(s->samples_per_frame/subframe_len); |
1221 | s->num_bands = s->num_sfb[s->table_idx]; |
1222 | s->cur_sfb_offsets = s->sfb_offsets[s->table_idx]; |
1223 | cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx]; |
1224 | |
1225 | /** configure the decoder for the current subframe */ |
1226 | offset += s->samples_per_frame >> 1; |
1227 | |
1228 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
1229 | int c = s->channel_indexes_for_cur_subframe[i]; |
1230 | |
1231 | s->channel[c].coeffs = &s->channel[c].out[offset]; |
1232 | } |
1233 | |
1234 | s->subframe_len = subframe_len; |
1235 | s->esc_len = av_log2(s->subframe_len - 1) + 1; |
1236 | |
1237 | /** skip extended header if any */ |
1238 | if (get_bits1(&s->gb)) { |
1239 | int num_fill_bits; |
1240 | if (!(num_fill_bits = get_bits(&s->gb, 2))) { |
1241 | int len = get_bits(&s->gb, 4); |
1242 | num_fill_bits = get_bitsz(&s->gb, len) + 1; |
1243 | } |
1244 | |
1245 | if (num_fill_bits >= 0) { |
1246 | if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) { |
1247 | av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n"); |
1248 | return AVERROR_INVALIDDATA; |
1249 | } |
1250 | |
1251 | skip_bits_long(&s->gb, num_fill_bits); |
1252 | } |
1253 | } |
1254 | |
1255 | /** no idea for what the following bit is used */ |
1256 | if (get_bits1(&s->gb)) { |
1257 | avpriv_request_sample(s->avctx, "Reserved bit"); |
1258 | return AVERROR_PATCHWELCOME; |
1259 | } |
1260 | |
1261 | |
1262 | if (decode_channel_transform(s) < 0) |
1263 | return AVERROR_INVALIDDATA; |
1264 | |
1265 | |
1266 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
1267 | int c = s->channel_indexes_for_cur_subframe[i]; |
1268 | if ((s->channel[c].transmit_coefs = get_bits1(&s->gb))) |
1269 | transmit_coeffs = 1; |
1270 | } |
1271 | |
1272 | av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE); |
1273 | if (transmit_coeffs) { |
1274 | int step; |
1275 | int quant_step = 90 * s->bits_per_sample >> 4; |
1276 | |
1277 | /** decode number of vector coded coefficients */ |
1278 | if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) { |
1279 | int num_bits = av_log2((s->subframe_len + 3)/4) + 1; |
1280 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
1281 | int c = s->channel_indexes_for_cur_subframe[i]; |
1282 | int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2; |
1283 | if (num_vec_coeffs > s->subframe_len) { |
1284 | av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs); |
1285 | return AVERROR_INVALIDDATA; |
1286 | } |
1287 | av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out)); |
1288 | s->channel[c].num_vec_coeffs = num_vec_coeffs; |
1289 | } |
1290 | } else { |
1291 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
1292 | int c = s->channel_indexes_for_cur_subframe[i]; |
1293 | s->channel[c].num_vec_coeffs = s->subframe_len; |
1294 | } |
1295 | } |
1296 | /** decode quantization step */ |
1297 | step = get_sbits(&s->gb, 6); |
1298 | quant_step += step; |
1299 | if (step == -32 || step == 31) { |
1300 | const int sign = (step == 31) - 1; |
1301 | int quant = 0; |
1302 | while (get_bits_count(&s->gb) + 5 < s->num_saved_bits && |
1303 | (step = get_bits(&s->gb, 5)) == 31) { |
1304 | quant += 31; |
1305 | } |
1306 | quant_step += ((quant + step) ^ sign) - sign; |
1307 | } |
1308 | if (quant_step < 0) { |
1309 | av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n"); |
1310 | } |
1311 | |
1312 | /** decode quantization step modifiers for every channel */ |
1313 | |
1314 | if (s->channels_for_cur_subframe == 1) { |
1315 | s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step; |
1316 | } else { |
1317 | int modifier_len = get_bits(&s->gb, 3); |
1318 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
1319 | int c = s->channel_indexes_for_cur_subframe[i]; |
1320 | s->channel[c].quant_step = quant_step; |
1321 | if (get_bits1(&s->gb)) { |
1322 | if (modifier_len) { |
1323 | s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1; |
1324 | } else |
1325 | ++s->channel[c].quant_step; |
1326 | } |
1327 | } |
1328 | } |
1329 | |
1330 | /** decode scale factors */ |
1331 | if (decode_scale_factors(s) < 0) |
1332 | return AVERROR_INVALIDDATA; |
1333 | } |
1334 | |
1335 | ff_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n", |
1336 | get_bits_count(&s->gb) - s->subframe_offset); |
1337 | |
1338 | /** parse coefficients */ |
1339 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
1340 | int c = s->channel_indexes_for_cur_subframe[i]; |
1341 | if (s->channel[c].transmit_coefs && |
1342 | get_bits_count(&s->gb) < s->num_saved_bits) { |
1343 | decode_coeffs(s, c); |
1344 | } else |
1345 | memset(s->channel[c].coeffs, 0, |
1346 | sizeof(*s->channel[c].coeffs) * subframe_len); |
1347 | } |
1348 | |
1349 | ff_dlog(s->avctx, "BITSTREAM: subframe length was %i\n", |
1350 | get_bits_count(&s->gb) - s->subframe_offset); |
1351 | |
1352 | if (transmit_coeffs) { |
1353 | FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS]; |
1354 | /** reconstruct the per channel data */ |
1355 | inverse_channel_transform(s); |
1356 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
1357 | int c = s->channel_indexes_for_cur_subframe[i]; |
1358 | const int* sf = s->channel[c].scale_factors; |
1359 | int b; |
1360 | |
1361 | if (c == s->lfe_channel) |
1362 | memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) * |
1363 | (subframe_len - cur_subwoofer_cutoff)); |
1364 | |
1365 | /** inverse quantization and rescaling */ |
1366 | for (b = 0; b < s->num_bands; b++) { |
1367 | const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len); |
1368 | const int exp = s->channel[c].quant_step - |
1369 | (s->channel[c].max_scale_factor - *sf++) * |
1370 | s->channel[c].scale_factor_step; |
1371 | const float quant = ff_exp10(exp / 20.0); |
1372 | int start = s->cur_sfb_offsets[b]; |
1373 | s->fdsp->vector_fmul_scalar(s->tmp + start, |
1374 | s->channel[c].coeffs + start, |
1375 | quant, end - start); |
1376 | } |
1377 | |
1378 | /** apply imdct (imdct_half == DCTIV with reverse) */ |
1379 | mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp); |
1380 | } |
1381 | } |
1382 | |
1383 | /** window and overlapp-add */ |
1384 | wmapro_window(s); |
1385 | |
1386 | /** handled one subframe */ |
1387 | for (i = 0; i < s->channels_for_cur_subframe; i++) { |
1388 | int c = s->channel_indexes_for_cur_subframe[i]; |
1389 | if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) { |
1390 | av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n"); |
1391 | return AVERROR_INVALIDDATA; |
1392 | } |
1393 | ++s->channel[c].cur_subframe; |
1394 | } |
1395 | |
1396 | return 0; |
1397 | } |
1398 | |
1399 | /** |
1400 | *@brief Decode one WMA frame. |
1401 | *@param s codec context |
1402 | *@return 0 if the trailer bit indicates that this is the last frame, |
1403 | * 1 if there are additional frames |
1404 | */ |
1405 | static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr) |
1406 | { |
1407 | GetBitContext* gb = &s->gb; |
1408 | int more_frames = 0; |
1409 | int len = 0; |
1410 | int i; |
1411 | |
1412 | /** get frame length */ |
1413 | if (s->len_prefix) |
1414 | len = get_bits(gb, s->log2_frame_size); |
1415 | |
1416 | ff_dlog(s->avctx, "decoding frame with length %x\n", len); |
1417 | |
1418 | /** decode tile information */ |
1419 | if (decode_tilehdr(s)) { |
1420 | s->packet_loss = 1; |
1421 | return 0; |
1422 | } |
1423 | |
1424 | /** read postproc transform */ |
1425 | if (s->nb_channels > 1 && get_bits1(gb)) { |
1426 | if (get_bits1(gb)) { |
1427 | for (i = 0; i < s->nb_channels * s->nb_channels; i++) |
1428 | skip_bits(gb, 4); |
1429 | } |
1430 | } |
1431 | |
1432 | /** read drc info */ |
1433 | if (s->dynamic_range_compression) { |
1434 | s->drc_gain = get_bits(gb, 8); |
1435 | ff_dlog(s->avctx, "drc_gain %i\n", s->drc_gain); |
1436 | } |
1437 | |
1438 | /** no idea what these are for, might be the number of samples |
1439 | that need to be skipped at the beginning or end of a stream */ |
1440 | if (get_bits1(gb)) { |
1441 | int av_unused skip; |
1442 | |
1443 | /** usually true for the first frame */ |
1444 | if (get_bits1(gb)) { |
1445 | skip = get_bits(gb, av_log2(s->samples_per_frame * 2)); |
1446 | ff_dlog(s->avctx, "start skip: %i\n", skip); |
1447 | } |
1448 | |
1449 | /** sometimes true for the last frame */ |
1450 | if (get_bits1(gb)) { |
1451 | skip = get_bits(gb, av_log2(s->samples_per_frame * 2)); |
1452 | ff_dlog(s->avctx, "end skip: %i\n", skip); |
1453 | } |
1454 | |
1455 | } |
1456 | |
1457 | ff_dlog(s->avctx, "BITSTREAM: frame header length was %i\n", |
1458 | get_bits_count(gb) - s->frame_offset); |
1459 | |
1460 | /** reset subframe states */ |
1461 | s->parsed_all_subframes = 0; |
1462 | for (i = 0; i < s->nb_channels; i++) { |
1463 | s->channel[i].decoded_samples = 0; |
1464 | s->channel[i].cur_subframe = 0; |
1465 | s->channel[i].reuse_sf = 0; |
1466 | } |
1467 | |
1468 | /** decode all subframes */ |
1469 | while (!s->parsed_all_subframes) { |
1470 | if (decode_subframe(s) < 0) { |
1471 | s->packet_loss = 1; |
1472 | return 0; |
1473 | } |
1474 | } |
1475 | |
1476 | /** copy samples to the output buffer */ |
1477 | for (i = 0; i < s->nb_channels; i++) |
1478 | memcpy(frame->extended_data[i], s->channel[i].out, |
1479 | s->samples_per_frame * sizeof(*s->channel[i].out)); |
1480 | |
1481 | for (i = 0; i < s->nb_channels; i++) { |
1482 | /** reuse second half of the IMDCT output for the next frame */ |
1483 | memcpy(&s->channel[i].out[0], |
1484 | &s->channel[i].out[s->samples_per_frame], |
1485 | s->samples_per_frame * sizeof(*s->channel[i].out) >> 1); |
1486 | } |
1487 | |
1488 | if (s->skip_frame) { |
1489 | s->skip_frame = 0; |
1490 | *got_frame_ptr = 0; |
1491 | av_frame_unref(frame); |
1492 | } else { |
1493 | *got_frame_ptr = 1; |
1494 | } |
1495 | |
1496 | if (s->len_prefix) { |
1497 | if (len != (get_bits_count(gb) - s->frame_offset) + 2) { |
1498 | /** FIXME: not sure if this is always an error */ |
1499 | av_log(s->avctx, AV_LOG_ERROR, |
1500 | "frame[%"PRIu32"] would have to skip %i bits\n", |
1501 | s->frame_num, |
1502 | len - (get_bits_count(gb) - s->frame_offset) - 1); |
1503 | s->packet_loss = 1; |
1504 | return 0; |
1505 | } |
1506 | |
1507 | /** skip the rest of the frame data */ |
1508 | skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1); |
1509 | } else { |
1510 | while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) { |
1511 | } |
1512 | } |
1513 | |
1514 | /** decode trailer bit */ |
1515 | more_frames = get_bits1(gb); |
1516 | |
1517 | ++s->frame_num; |
1518 | return more_frames; |
1519 | } |
1520 | |
1521 | /** |
1522 | *@brief Calculate remaining input buffer length. |
1523 | *@param s codec context |
1524 | *@param gb bitstream reader context |
1525 | *@return remaining size in bits |
1526 | */ |
1527 | static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb) |
1528 | { |
1529 | return s->buf_bit_size - get_bits_count(gb); |
1530 | } |
1531 | |
1532 | /** |
1533 | *@brief Fill the bit reservoir with a (partial) frame. |
1534 | *@param s codec context |
1535 | *@param gb bitstream reader context |
1536 | *@param len length of the partial frame |
1537 | *@param append decides whether to reset the buffer or not |
1538 | */ |
1539 | static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len, |
1540 | int append) |
1541 | { |
1542 | int buflen; |
1543 | |
1544 | /** when the frame data does not need to be concatenated, the input buffer |
1545 | is reset and additional bits from the previous frame are copied |
1546 | and skipped later so that a fast byte copy is possible */ |
1547 | |
1548 | if (!append) { |
1549 | s->frame_offset = get_bits_count(gb) & 7; |
1550 | s->num_saved_bits = s->frame_offset; |
1551 | init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE); |
1552 | } |
1553 | |
1554 | buflen = (put_bits_count(&s->pb) + len + 8) >> 3; |
1555 | |
1556 | if (len <= 0 || buflen > MAX_FRAMESIZE) { |
1557 | avpriv_request_sample(s->avctx, "Too small input buffer"); |
1558 | s->packet_loss = 1; |
1559 | return; |
1560 | } |
1561 | |
1562 | av_assert0(len <= put_bits_left(&s->pb)); |
1563 | |
1564 | s->num_saved_bits += len; |
1565 | if (!append) { |
1566 | avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), |
1567 | s->num_saved_bits); |
1568 | } else { |
1569 | int align = 8 - (get_bits_count(gb) & 7); |
1570 | align = FFMIN(align, len); |
1571 | put_bits(&s->pb, align, get_bits(gb, align)); |
1572 | len -= align; |
1573 | avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len); |
1574 | } |
1575 | skip_bits_long(gb, len); |
1576 | |
1577 | { |
1578 | PutBitContext tmp = s->pb; |
1579 | flush_put_bits(&tmp); |
1580 | } |
1581 | |
1582 | init_get_bits(&s->gb, s->frame_data, s->num_saved_bits); |
1583 | skip_bits(&s->gb, s->frame_offset); |
1584 | } |
1585 | |
1586 | static int decode_packet(AVCodecContext *avctx, WMAProDecodeCtx *s, |
1587 | void *data, int *got_frame_ptr, AVPacket *avpkt) |
1588 | { |
1589 | GetBitContext* gb = &s->pgb; |
1590 | const uint8_t* buf = avpkt->data; |
1591 | int buf_size = avpkt->size; |
1592 | int num_bits_prev_frame; |
1593 | int packet_sequence_number; |
1594 | |
1595 | *got_frame_ptr = 0; |
1596 | |
1597 | if (s->packet_done || s->packet_loss) { |
1598 | s->packet_done = 0; |
1599 | |
1600 | /** sanity check for the buffer length */ |
1601 | if (avctx->codec_id == AV_CODEC_ID_WMAPRO && buf_size < avctx->block_align) { |
1602 | av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n", |
1603 | buf_size, avctx->block_align); |
1604 | return AVERROR_INVALIDDATA; |
1605 | } |
1606 | |
1607 | if (avctx->codec_id == AV_CODEC_ID_WMAPRO) { |
1608 | s->next_packet_start = buf_size - avctx->block_align; |
1609 | buf_size = avctx->block_align; |
1610 | } else { |
1611 | s->next_packet_start = buf_size - FFMIN(buf_size, avctx->block_align); |
1612 | buf_size = FFMIN(buf_size, avctx->block_align); |
1613 | } |
1614 | s->buf_bit_size = buf_size << 3; |
1615 | |
1616 | /** parse packet header */ |
1617 | init_get_bits(gb, buf, s->buf_bit_size); |
1618 | if (avctx->codec_id != AV_CODEC_ID_XMA2) { |
1619 | packet_sequence_number = get_bits(gb, 4); |
1620 | skip_bits(gb, 2); |
1621 | } else { |
1622 | int num_frames = get_bits(gb, 6); |
1623 | ff_dlog(avctx, "packet[%d]: number of frames %d\n", avctx->frame_number, num_frames); |
1624 | packet_sequence_number = 0; |
1625 | } |
1626 | |
1627 | /** get number of bits that need to be added to the previous frame */ |
1628 | num_bits_prev_frame = get_bits(gb, s->log2_frame_size); |
1629 | if (avctx->codec_id != AV_CODEC_ID_WMAPRO) { |
1630 | skip_bits(gb, 3); |
1631 | s->skip_packets = get_bits(gb, 8); |
1632 | ff_dlog(avctx, "packet[%d]: skip packets %d\n", avctx->frame_number, s->skip_packets); |
1633 | } |
1634 | |
1635 | ff_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number, |
1636 | num_bits_prev_frame); |
1637 | |
1638 | /** check for packet loss */ |
1639 | if (avctx->codec_id == AV_CODEC_ID_WMAPRO && !s->packet_loss && |
1640 | ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) { |
1641 | s->packet_loss = 1; |
1642 | av_log(avctx, AV_LOG_ERROR, |
1643 | "Packet loss detected! seq %"PRIx8" vs %x\n", |
1644 | s->packet_sequence_number, packet_sequence_number); |
1645 | } |
1646 | s->packet_sequence_number = packet_sequence_number; |
1647 | |
1648 | if (num_bits_prev_frame > 0) { |
1649 | int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb); |
1650 | if (num_bits_prev_frame >= remaining_packet_bits) { |
1651 | num_bits_prev_frame = remaining_packet_bits; |
1652 | s->packet_done = 1; |
1653 | } |
1654 | |
1655 | /** append the previous frame data to the remaining data from the |
1656 | previous packet to create a full frame */ |
1657 | save_bits(s, gb, num_bits_prev_frame, 1); |
1658 | ff_dlog(avctx, "accumulated %x bits of frame data\n", |
1659 | s->num_saved_bits - s->frame_offset); |
1660 | |
1661 | /** decode the cross packet frame if it is valid */ |
1662 | if (!s->packet_loss) |
1663 | decode_frame(s, data, got_frame_ptr); |
1664 | } else if (s->num_saved_bits - s->frame_offset) { |
1665 | ff_dlog(avctx, "ignoring %x previously saved bits\n", |
1666 | s->num_saved_bits - s->frame_offset); |
1667 | } |
1668 | |
1669 | if (s->packet_loss) { |
1670 | /** reset number of saved bits so that the decoder |
1671 | does not start to decode incomplete frames in the |
1672 | s->len_prefix == 0 case */ |
1673 | s->num_saved_bits = 0; |
1674 | s->packet_loss = 0; |
1675 | } |
1676 | } else { |
1677 | int frame_size; |
1678 | s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3; |
1679 | init_get_bits(gb, avpkt->data, s->buf_bit_size); |
1680 | skip_bits(gb, s->packet_offset); |
1681 | if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size && |
1682 | (frame_size = show_bits(gb, s->log2_frame_size)) && |
1683 | frame_size <= remaining_bits(s, gb)) { |
1684 | save_bits(s, gb, frame_size, 0); |
1685 | if (!s->packet_loss) |
1686 | s->packet_done = !decode_frame(s, data, got_frame_ptr); |
1687 | } else if (!s->len_prefix |
1688 | && s->num_saved_bits > get_bits_count(&s->gb)) { |
1689 | /** when the frames do not have a length prefix, we don't know |
1690 | the compressed length of the individual frames |
1691 | however, we know what part of a new packet belongs to the |
1692 | previous frame |
1693 | therefore we save the incoming packet first, then we append |
1694 | the "previous frame" data from the next packet so that |
1695 | we get a buffer that only contains full frames */ |
1696 | s->packet_done = !decode_frame(s, data, got_frame_ptr); |
1697 | } else { |
1698 | s->packet_done = 1; |
1699 | } |
1700 | } |
1701 | |
1702 | if (remaining_bits(s, gb) < 0) { |
1703 | av_log(avctx, AV_LOG_ERROR, "Overread %d\n", -remaining_bits(s, gb)); |
1704 | s->packet_loss = 1; |
1705 | } |
1706 | |
1707 | if (s->packet_done && !s->packet_loss && |
1708 | remaining_bits(s, gb) > 0) { |
1709 | /** save the rest of the data so that it can be decoded |
1710 | with the next packet */ |
1711 | save_bits(s, gb, remaining_bits(s, gb), 0); |
1712 | } |
1713 | |
1714 | s->packet_offset = get_bits_count(gb) & 7; |
1715 | if (s->packet_loss) |
1716 | return AVERROR_INVALIDDATA; |
1717 | |
1718 | return get_bits_count(gb) >> 3; |
1719 | } |
1720 | |
1721 | /** |
1722 | *@brief Decode a single WMA packet. |
1723 | *@param avctx codec context |
1724 | *@param data the output buffer |
1725 | *@param avpkt input packet |
1726 | *@return number of bytes that were read from the input buffer |
1727 | */ |
1728 | static int wmapro_decode_packet(AVCodecContext *avctx, void *data, |
1729 | int *got_frame_ptr, AVPacket *avpkt) |
1730 | { |
1731 | WMAProDecodeCtx *s = avctx->priv_data; |
1732 | AVFrame *frame = data; |
1733 | int ret; |
1734 | |
1735 | /* get output buffer */ |
1736 | frame->nb_samples = s->samples_per_frame; |
1737 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) { |
1738 | s->packet_loss = 1; |
1739 | return 0; |
1740 | } |
1741 | |
1742 | return decode_packet(avctx, s, data, got_frame_ptr, avpkt); |
1743 | } |
1744 | |
1745 | static int xma_decode_packet(AVCodecContext *avctx, void *data, |
1746 | int *got_frame_ptr, AVPacket *avpkt) |
1747 | { |
1748 | XMADecodeCtx *s = avctx->priv_data; |
1749 | int got_stream_frame_ptr = 0; |
1750 | AVFrame *frame = data; |
1751 | int i, ret, offset = INT_MAX; |
1752 | |
1753 | ret = decode_packet(avctx, &s->xma[s->current_stream], s->frames[s->current_stream], |
1754 | &got_stream_frame_ptr, avpkt); |
1755 | |
1756 | if (got_stream_frame_ptr) { |
1757 | memcpy(&s->samples[s->current_stream * 2 + 0][s->offset[s->current_stream] * 512], |
1758 | s->frames[s->current_stream]->extended_data[0], 512 * 4); |
1759 | if (avctx->channels > 1) |
1760 | memcpy(&s->samples[s->current_stream * 2 + 1][s->offset[s->current_stream] * 512], |
1761 | s->frames[s->current_stream]->extended_data[1], 512 * 4); |
1762 | s->offset[s->current_stream]++; |
1763 | } else if (ret < 0) { |
1764 | memset(s->offset, 0, sizeof(s->offset)); |
1765 | s->current_stream = 0; |
1766 | return ret; |
1767 | } |
1768 | |
1769 | if (s->xma[s->current_stream].packet_done || |
1770 | s->xma[s->current_stream].packet_loss) { |
1771 | int bret; |
1772 | |
1773 | if (s->xma[s->current_stream].skip_packets == 0) { |
1774 | ; |
1775 | } else if (s->xma[0].skip_packets == 0 && avctx->channels >= 2) { |
1776 | s->current_stream = 0; |
1777 | } else if (s->xma[1].skip_packets == 0 && avctx->channels >= 4) { |
1778 | s->current_stream = 1; |
1779 | } else if (s->xma[2].skip_packets == 0 && avctx->channels >= 6) { |
1780 | s->current_stream = 2; |
1781 | } else if (s->xma[3].skip_packets == 0 && avctx->channels == 8) { |
1782 | s->current_stream = 3; |
1783 | } else { |
1784 | int min[2]; |
1785 | |
1786 | min[0] = s->xma[0].skip_packets; |
1787 | min[1] = i = 0; |
1788 | |
1789 | for (i = 1; i < avctx->channels / 2; i++) { |
1790 | if (s->xma[i].skip_packets < min[0]) { |
1791 | min[1] = i; |
1792 | min[0] = s->xma[i].skip_packets; |
1793 | } |
1794 | } |
1795 | |
1796 | s->current_stream = min[1]; |
1797 | } |
1798 | |
1799 | for (i = 0; i < avctx->channels / 2; i++) { |
1800 | s->xma[i].skip_packets = FFMAX(0, s->xma[i].skip_packets - 1); |
1801 | } |
1802 | |
1803 | for (i = 0; i < (avctx->channels + 1) / 2; i++) { |
1804 | offset = FFMIN(offset, s->offset[i]); |
1805 | } |
1806 | |
1807 | if (offset > 0) { |
1808 | frame->nb_samples = 512 * offset; |
1809 | if ((bret = ff_get_buffer(avctx, frame, 0)) < 0) |
1810 | return bret; |
1811 | |
1812 | for (i = 0; i < (avctx->channels + 1) / 2; i++) { |
1813 | memcpy(frame->extended_data[i * 2 + 0], s->samples[i * 2 + 0], frame->nb_samples * 4); |
1814 | if (avctx->channels > 1) |
1815 | memcpy(frame->extended_data[i * 2 + 1], s->samples[i * 2 + 1], frame->nb_samples * 4); |
1816 | s->offset[i] -= offset; |
1817 | if (s->offset[i]) { |
1818 | memmove(s->samples[i * 2 + 0], s->samples[i * 2 + 0] + frame->nb_samples, s->offset[i] * 4 * 512); |
1819 | if (avctx->channels > 1) |
1820 | memmove(s->samples[i * 2 + 1], s->samples[i * 2 + 1] + frame->nb_samples, s->offset[i] * 4 * 512); |
1821 | } |
1822 | } |
1823 | |
1824 | *got_frame_ptr = 1; |
1825 | } |
1826 | } |
1827 | |
1828 | return ret; |
1829 | } |
1830 | |
1831 | static av_cold int xma_decode_init(AVCodecContext *avctx) |
1832 | { |
1833 | XMADecodeCtx *s = avctx->priv_data; |
1834 | int i, ret; |
1835 | |
1836 | if (avctx->channels <= 0 || avctx->channels > 8) |
1837 | return AVERROR_INVALIDDATA; |
1838 | |
1839 | for (i = 0; i < (avctx->channels + 1) / 2; i++) { |
1840 | ret = decode_init(&s->xma[i], avctx); |
1841 | if (ret < 0) |
1842 | return ret; |
1843 | s->frames[i] = av_frame_alloc(); |
1844 | if (!s->frames[i]) |
1845 | return AVERROR(ENOMEM); |
1846 | s->frames[i]->nb_samples = 512; |
1847 | if ((ret = ff_get_buffer(avctx, s->frames[i], 0)) < 0) { |
1848 | return AVERROR(ENOMEM); |
1849 | } |
1850 | |
1851 | } |
1852 | |
1853 | return ret; |
1854 | } |
1855 | |
1856 | static av_cold int xma_decode_end(AVCodecContext *avctx) |
1857 | { |
1858 | XMADecodeCtx *s = avctx->priv_data; |
1859 | int i; |
1860 | |
1861 | for (i = 0; i < avctx->channels / 2; i++) { |
1862 | decode_end(&s->xma[i]); |
1863 | av_frame_free(&s->frames[i]); |
1864 | } |
1865 | |
1866 | return 0; |
1867 | } |
1868 | |
1869 | static void flush(WMAProDecodeCtx *s) |
1870 | { |
1871 | int i; |
1872 | /** reset output buffer as a part of it is used during the windowing of a |
1873 | new frame */ |
1874 | for (i = 0; i < s->nb_channels; i++) |
1875 | memset(s->channel[i].out, 0, s->samples_per_frame * |
1876 | sizeof(*s->channel[i].out)); |
1877 | s->packet_loss = 1; |
1878 | s->skip_packets = 0; |
1879 | } |
1880 | |
1881 | |
1882 | /** |
1883 | *@brief Clear decoder buffers (for seeking). |
1884 | *@param avctx codec context |
1885 | */ |
1886 | static void wmapro_flush(AVCodecContext *avctx) |
1887 | { |
1888 | WMAProDecodeCtx *s = avctx->priv_data; |
1889 | |
1890 | flush(s); |
1891 | } |
1892 | |
1893 | static void xma_flush(AVCodecContext *avctx) |
1894 | { |
1895 | XMADecodeCtx *s = avctx->priv_data; |
1896 | int i; |
1897 | for (i = 0; i < (avctx->channels + 1) / 2; i++) |
1898 | flush(&s->xma[i]); |
1899 | |
1900 | memset(s->offset, 0, sizeof(s->offset)); |
1901 | s->current_stream = 0; |
1902 | } |
1903 | |
1904 | |
1905 | /** |
1906 | *@brief wmapro decoder |
1907 | */ |
1908 | AVCodec ff_wmapro_decoder = { |
1909 | .name = "wmapro", |
1910 | .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"), |
1911 | .type = AVMEDIA_TYPE_AUDIO, |
1912 | .id = AV_CODEC_ID_WMAPRO, |
1913 | .priv_data_size = sizeof(WMAProDecodeCtx), |
1914 | .init = wmapro_decode_init, |
1915 | .close = wmapro_decode_end, |
1916 | .decode = wmapro_decode_packet, |
1917 | .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1, |
1918 | .flush = wmapro_flush, |
1919 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, |
1920 | AV_SAMPLE_FMT_NONE }, |
1921 | }; |
1922 | |
1923 | AVCodec ff_xma1_decoder = { |
1924 | .name = "xma1", |
1925 | .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 1"), |
1926 | .type = AVMEDIA_TYPE_AUDIO, |
1927 | .id = AV_CODEC_ID_XMA1, |
1928 | .priv_data_size = sizeof(XMADecodeCtx), |
1929 | .init = xma_decode_init, |
1930 | .close = xma_decode_end, |
1931 | .decode = xma_decode_packet, |
1932 | .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1, |
1933 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, |
1934 | AV_SAMPLE_FMT_NONE }, |
1935 | }; |
1936 | |
1937 | AVCodec ff_xma2_decoder = { |
1938 | .name = "xma2", |
1939 | .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 2"), |
1940 | .type = AVMEDIA_TYPE_AUDIO, |
1941 | .id = AV_CODEC_ID_XMA2, |
1942 | .priv_data_size = sizeof(XMADecodeCtx), |
1943 | .init = xma_decode_init, |
1944 | .close = xma_decode_end, |
1945 | .decode = xma_decode_packet, |
1946 | .flush = xma_flush, |
1947 | .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1, |
1948 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, |
1949 | AV_SAMPLE_FMT_NONE }, |
1950 | }; |
1951 |