Commit | Line | Data |
---|---|---|
2ba45a60 DM |
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/float_dsp.h" | |
92 | #include "libavutil/intfloat.h" | |
93 | #include "libavutil/intreadwrite.h" | |
94 | #include "avcodec.h" | |
95 | #include "internal.h" | |
96 | #include "get_bits.h" | |
97 | #include "put_bits.h" | |
98 | #include "wmaprodata.h" | |
99 | #include "sinewin.h" | |
100 | #include "wma.h" | |
101 | #include "wma_common.h" | |
102 | ||
103 | /** current decoder limitations */ | |
104 | #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels | |
105 | #define MAX_SUBFRAMES 32 ///< max number of subframes per channel | |
106 | #define MAX_BANDS 29 ///< max number of scale factor bands | |
107 | #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size | |
108 | ||
109 | #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size | |
110 | #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size | |
111 | #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size | |
112 | #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size | |
113 | #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes | |
114 | ||
115 | ||
116 | #define VLCBITS 9 | |
117 | #define SCALEVLCBITS 8 | |
118 | #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS) | |
119 | #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS) | |
120 | #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS) | |
121 | #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS) | |
122 | #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS) | |
123 | ||
124 | static VLC sf_vlc; ///< scale factor DPCM vlc | |
125 | static VLC sf_rl_vlc; ///< scale factor run length vlc | |
126 | static VLC vec4_vlc; ///< 4 coefficients per symbol | |
127 | static VLC vec2_vlc; ///< 2 coefficients per symbol | |
128 | static VLC vec1_vlc; ///< 1 coefficient per symbol | |
129 | static VLC coef_vlc[2]; ///< coefficient run length vlc codes | |
130 | static float sin64[33]; ///< sine table for decorrelation | |
131 | ||
132 | /** | |
133 | * @brief frame specific decoder context for a single channel | |
134 | */ | |
135 | typedef struct { | |
136 | int16_t prev_block_len; ///< length of the previous block | |
137 | uint8_t transmit_coefs; | |
138 | uint8_t num_subframes; | |
139 | uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples | |
140 | uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame | |
141 | uint8_t cur_subframe; ///< current subframe number | |
142 | uint16_t decoded_samples; ///< number of already processed samples | |
143 | uint8_t grouped; ///< channel is part of a group | |
144 | int quant_step; ///< quantization step for the current subframe | |
145 | int8_t reuse_sf; ///< share scale factors between subframes | |
146 | int8_t scale_factor_step; ///< scaling step for the current subframe | |
147 | int max_scale_factor; ///< maximum scale factor for the current subframe | |
148 | int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values | |
149 | int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling) | |
150 | int* scale_factors; ///< pointer to the scale factor values used for decoding | |
151 | uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block | |
152 | float* coeffs; ///< pointer to the subframe decode buffer | |
153 | uint16_t num_vec_coeffs; ///< number of vector coded coefficients | |
154 | DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer | |
155 | } WMAProChannelCtx; | |
156 | ||
157 | /** | |
158 | * @brief channel group for channel transformations | |
159 | */ | |
160 | typedef struct { | |
161 | uint8_t num_channels; ///< number of channels in the group | |
162 | int8_t transform; ///< transform on / off | |
163 | int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band | |
164 | float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS]; | |
165 | float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients | |
166 | } WMAProChannelGrp; | |
167 | ||
168 | /** | |
169 | * @brief main decoder context | |
170 | */ | |
171 | typedef struct WMAProDecodeCtx { | |
172 | /* generic decoder variables */ | |
173 | AVCodecContext* avctx; ///< codec context for av_log | |
174 | AVFloatDSPContext fdsp; | |
175 | uint8_t frame_data[MAX_FRAMESIZE + | |
176 | FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data | |
177 | PutBitContext pb; ///< context for filling the frame_data buffer | |
178 | FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size | |
179 | DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer | |
180 | const float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes | |
181 | ||
182 | /* frame size dependent frame information (set during initialization) */ | |
183 | uint32_t decode_flags; ///< used compression features | |
184 | uint8_t len_prefix; ///< frame is prefixed with its length | |
185 | uint8_t dynamic_range_compression; ///< frame contains DRC data | |
186 | uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0]) | |
187 | uint16_t samples_per_frame; ///< number of samples to output | |
188 | uint16_t log2_frame_size; | |
189 | int8_t lfe_channel; ///< lfe channel index | |
190 | uint8_t max_num_subframes; | |
191 | uint8_t subframe_len_bits; ///< number of bits used for the subframe length | |
192 | uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1 | |
193 | uint16_t min_samples_per_subframe; | |
194 | int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size | |
195 | int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4) | |
196 | int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix | |
197 | int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values | |
198 | ||
199 | /* packet decode state */ | |
200 | GetBitContext pgb; ///< bitstream reader context for the packet | |
201 | int next_packet_start; ///< start offset of the next wma packet in the demuxer packet | |
202 | uint8_t packet_offset; ///< frame offset in the packet | |
203 | uint8_t packet_sequence_number; ///< current packet number | |
204 | int num_saved_bits; ///< saved number of bits | |
205 | int frame_offset; ///< frame offset in the bit reservoir | |
206 | int subframe_offset; ///< subframe offset in the bit reservoir | |
207 | uint8_t packet_loss; ///< set in case of bitstream error | |
208 | uint8_t packet_done; ///< set when a packet is fully decoded | |
209 | ||
210 | /* frame decode state */ | |
211 | uint32_t frame_num; ///< current frame number (not used for decoding) | |
212 | GetBitContext gb; ///< bitstream reader context | |
213 | int buf_bit_size; ///< buffer size in bits | |
214 | uint8_t drc_gain; ///< gain for the DRC tool | |
215 | int8_t skip_frame; ///< skip output step | |
216 | int8_t parsed_all_subframes; ///< all subframes decoded? | |
217 | ||
218 | /* subframe/block decode state */ | |
219 | int16_t subframe_len; ///< current subframe length | |
220 | int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe | |
221 | int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS]; | |
222 | int8_t num_bands; ///< number of scale factor bands | |
223 | int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream | |
224 | int16_t* cur_sfb_offsets; ///< sfb offsets for the current block | |
225 | uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables | |
226 | int8_t esc_len; ///< length of escaped coefficients | |
227 | ||
228 | uint8_t num_chgroups; ///< number of channel groups | |
229 | WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information | |
230 | ||
231 | WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data | |
232 | } WMAProDecodeCtx; | |
233 | ||
234 | ||
235 | /** | |
236 | *@brief helper function to print the most important members of the context | |
237 | *@param s context | |
238 | */ | |
239 | static av_cold void dump_context(WMAProDecodeCtx *s) | |
240 | { | |
241 | #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b); | |
242 | #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b); | |
243 | ||
244 | PRINT("ed sample bit depth", s->bits_per_sample); | |
245 | PRINT_HEX("ed decode flags", s->decode_flags); | |
246 | PRINT("samples per frame", s->samples_per_frame); | |
247 | PRINT("log2 frame size", s->log2_frame_size); | |
248 | PRINT("max num subframes", s->max_num_subframes); | |
249 | PRINT("len prefix", s->len_prefix); | |
250 | PRINT("num channels", s->avctx->channels); | |
251 | } | |
252 | ||
253 | /** | |
254 | *@brief Uninitialize the decoder and free all resources. | |
255 | *@param avctx codec context | |
256 | *@return 0 on success, < 0 otherwise | |
257 | */ | |
258 | static av_cold int decode_end(AVCodecContext *avctx) | |
259 | { | |
260 | WMAProDecodeCtx *s = avctx->priv_data; | |
261 | int i; | |
262 | ||
263 | for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) | |
264 | ff_mdct_end(&s->mdct_ctx[i]); | |
265 | ||
266 | return 0; | |
267 | } | |
268 | ||
269 | /** | |
270 | *@brief Initialize the decoder. | |
271 | *@param avctx codec context | |
272 | *@return 0 on success, -1 otherwise | |
273 | */ | |
274 | static av_cold int decode_init(AVCodecContext *avctx) | |
275 | { | |
276 | WMAProDecodeCtx *s = avctx->priv_data; | |
277 | uint8_t *edata_ptr = avctx->extradata; | |
278 | unsigned int channel_mask; | |
279 | int i, bits; | |
280 | int log2_max_num_subframes; | |
281 | int num_possible_block_sizes; | |
282 | ||
283 | if (!avctx->block_align) { | |
284 | av_log(avctx, AV_LOG_ERROR, "block_align is not set\n"); | |
285 | return AVERROR(EINVAL); | |
286 | } | |
287 | ||
288 | s->avctx = avctx; | |
289 | avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT); | |
290 | ||
291 | init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE); | |
292 | ||
293 | avctx->sample_fmt = AV_SAMPLE_FMT_FLTP; | |
294 | ||
295 | if (avctx->extradata_size >= 18) { | |
296 | s->decode_flags = AV_RL16(edata_ptr+14); | |
297 | channel_mask = AV_RL32(edata_ptr+2); | |
298 | s->bits_per_sample = AV_RL16(edata_ptr); | |
299 | /** dump the extradata */ | |
300 | for (i = 0; i < avctx->extradata_size; i++) | |
301 | av_dlog(avctx, "[%x] ", avctx->extradata[i]); | |
302 | av_dlog(avctx, "\n"); | |
303 | ||
304 | } else { | |
305 | avpriv_request_sample(avctx, "Unknown extradata size"); | |
306 | return AVERROR_PATCHWELCOME; | |
307 | } | |
308 | ||
309 | /** generic init */ | |
310 | s->log2_frame_size = av_log2(avctx->block_align) + 4; | |
311 | if (s->log2_frame_size > 25) { | |
312 | avpriv_request_sample(avctx, "Large block align"); | |
313 | return AVERROR_PATCHWELCOME; | |
314 | } | |
315 | ||
316 | /** frame info */ | |
317 | s->skip_frame = 1; /* skip first frame */ | |
318 | s->packet_loss = 1; | |
319 | s->len_prefix = (s->decode_flags & 0x40); | |
320 | ||
321 | /** get frame len */ | |
322 | bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags); | |
323 | if (bits > WMAPRO_BLOCK_MAX_BITS) { | |
324 | avpriv_request_sample(avctx, "14-bit block sizes"); | |
325 | return AVERROR_PATCHWELCOME; | |
326 | } | |
327 | s->samples_per_frame = 1 << bits; | |
328 | ||
329 | /** subframe info */ | |
330 | log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3); | |
331 | s->max_num_subframes = 1 << log2_max_num_subframes; | |
332 | if (s->max_num_subframes == 16 || s->max_num_subframes == 4) | |
333 | s->max_subframe_len_bit = 1; | |
334 | s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1; | |
335 | ||
336 | num_possible_block_sizes = log2_max_num_subframes + 1; | |
337 | s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes; | |
338 | s->dynamic_range_compression = (s->decode_flags & 0x80); | |
339 | ||
340 | if (s->max_num_subframes > MAX_SUBFRAMES) { | |
341 | av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n", | |
342 | s->max_num_subframes); | |
343 | return AVERROR_INVALIDDATA; | |
344 | } | |
345 | ||
346 | if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) { | |
347 | av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n", | |
348 | s->min_samples_per_subframe); | |
349 | return AVERROR_INVALIDDATA; | |
350 | } | |
351 | ||
352 | if (s->avctx->sample_rate <= 0) { | |
353 | av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n"); | |
354 | return AVERROR_INVALIDDATA; | |
355 | } | |
356 | ||
357 | if (avctx->channels < 0) { | |
358 | av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", | |
359 | avctx->channels); | |
360 | return AVERROR_INVALIDDATA; | |
361 | } else if (avctx->channels > WMAPRO_MAX_CHANNELS) { | |
362 | avpriv_request_sample(avctx, | |
363 | "More than %d channels", WMAPRO_MAX_CHANNELS); | |
364 | return AVERROR_PATCHWELCOME; | |
365 | } | |
366 | ||
367 | /** init previous block len */ | |
368 | for (i = 0; i < avctx->channels; i++) | |
369 | s->channel[i].prev_block_len = s->samples_per_frame; | |
370 | ||
371 | /** extract lfe channel position */ | |
372 | s->lfe_channel = -1; | |
373 | ||
374 | if (channel_mask & 8) { | |
375 | unsigned int mask; | |
376 | for (mask = 1; mask < 16; mask <<= 1) { | |
377 | if (channel_mask & mask) | |
378 | ++s->lfe_channel; | |
379 | } | |
380 | } | |
381 | ||
382 | INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE, | |
383 | scale_huffbits, 1, 1, | |
384 | scale_huffcodes, 2, 2, 616); | |
385 | ||
386 | INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE, | |
387 | scale_rl_huffbits, 1, 1, | |
388 | scale_rl_huffcodes, 4, 4, 1406); | |
389 | ||
390 | INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE, | |
391 | coef0_huffbits, 1, 1, | |
392 | coef0_huffcodes, 4, 4, 2108); | |
393 | ||
394 | INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE, | |
395 | coef1_huffbits, 1, 1, | |
396 | coef1_huffcodes, 4, 4, 3912); | |
397 | ||
398 | INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE, | |
399 | vec4_huffbits, 1, 1, | |
400 | vec4_huffcodes, 2, 2, 604); | |
401 | ||
402 | INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE, | |
403 | vec2_huffbits, 1, 1, | |
404 | vec2_huffcodes, 2, 2, 562); | |
405 | ||
406 | INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE, | |
407 | vec1_huffbits, 1, 1, | |
408 | vec1_huffcodes, 2, 2, 562); | |
409 | ||
410 | /** calculate number of scale factor bands and their offsets | |
411 | for every possible block size */ | |
412 | for (i = 0; i < num_possible_block_sizes; i++) { | |
413 | int subframe_len = s->samples_per_frame >> i; | |
414 | int x; | |
415 | int band = 1; | |
416 | ||
417 | s->sfb_offsets[i][0] = 0; | |
418 | ||
419 | for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) { | |
420 | int offset = (subframe_len * 2 * critical_freq[x]) | |
421 | / s->avctx->sample_rate + 2; | |
422 | offset &= ~3; | |
423 | if (offset > s->sfb_offsets[i][band - 1]) | |
424 | s->sfb_offsets[i][band++] = offset; | |
425 | } | |
426 | s->sfb_offsets[i][band - 1] = subframe_len; | |
427 | s->num_sfb[i] = band - 1; | |
428 | if (s->num_sfb[i] <= 0) { | |
429 | av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n"); | |
430 | return AVERROR_INVALIDDATA; | |
431 | } | |
432 | } | |
433 | ||
434 | ||
435 | /** Scale factors can be shared between blocks of different size | |
436 | as every block has a different scale factor band layout. | |
437 | The matrix sf_offsets is needed to find the correct scale factor. | |
438 | */ | |
439 | ||
440 | for (i = 0; i < num_possible_block_sizes; i++) { | |
441 | int b; | |
442 | for (b = 0; b < s->num_sfb[i]; b++) { | |
443 | int x; | |
444 | int offset = ((s->sfb_offsets[i][b] | |
445 | + s->sfb_offsets[i][b + 1] - 1) << i) >> 1; | |
446 | for (x = 0; x < num_possible_block_sizes; x++) { | |
447 | int v = 0; | |
448 | while (s->sfb_offsets[x][v + 1] << x < offset) { | |
449 | v++; | |
450 | av_assert0(v < MAX_BANDS); | |
451 | } | |
452 | s->sf_offsets[i][x][b] = v; | |
453 | } | |
454 | } | |
455 | } | |
456 | ||
457 | /** init MDCT, FIXME: only init needed sizes */ | |
458 | for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) | |
459 | ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1, | |
460 | 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1)) | |
461 | / (1 << (s->bits_per_sample - 1))); | |
462 | ||
463 | /** init MDCT windows: simple sine window */ | |
464 | for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) { | |
465 | const int win_idx = WMAPRO_BLOCK_MAX_BITS - i; | |
466 | ff_init_ff_sine_windows(win_idx); | |
467 | s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx]; | |
468 | } | |
469 | ||
470 | /** calculate subwoofer cutoff values */ | |
471 | for (i = 0; i < num_possible_block_sizes; i++) { | |
472 | int block_size = s->samples_per_frame >> i; | |
473 | int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1) | |
474 | / s->avctx->sample_rate; | |
475 | s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size); | |
476 | } | |
477 | ||
478 | /** calculate sine values for the decorrelation matrix */ | |
479 | for (i = 0; i < 33; i++) | |
480 | sin64[i] = sin(i*M_PI / 64.0); | |
481 | ||
482 | if (avctx->debug & FF_DEBUG_BITSTREAM) | |
483 | dump_context(s); | |
484 | ||
485 | avctx->channel_layout = channel_mask; | |
486 | ||
487 | return 0; | |
488 | } | |
489 | ||
490 | /** | |
491 | *@brief Decode the subframe length. | |
492 | *@param s context | |
493 | *@param offset sample offset in the frame | |
494 | *@return decoded subframe length on success, < 0 in case of an error | |
495 | */ | |
496 | static int decode_subframe_length(WMAProDecodeCtx *s, int offset) | |
497 | { | |
498 | int frame_len_shift = 0; | |
499 | int subframe_len; | |
500 | ||
501 | /** no need to read from the bitstream when only one length is possible */ | |
502 | if (offset == s->samples_per_frame - s->min_samples_per_subframe) | |
503 | return s->min_samples_per_subframe; | |
504 | ||
505 | if (get_bits_left(&s->gb) < 1) | |
506 | return AVERROR_INVALIDDATA; | |
507 | ||
508 | /** 1 bit indicates if the subframe is of maximum length */ | |
509 | if (s->max_subframe_len_bit) { | |
510 | if (get_bits1(&s->gb)) | |
511 | frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1); | |
512 | } else | |
513 | frame_len_shift = get_bits(&s->gb, s->subframe_len_bits); | |
514 | ||
515 | subframe_len = s->samples_per_frame >> frame_len_shift; | |
516 | ||
517 | /** sanity check the length */ | |
518 | if (subframe_len < s->min_samples_per_subframe || | |
519 | subframe_len > s->samples_per_frame) { | |
520 | av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n", | |
521 | subframe_len); | |
522 | return AVERROR_INVALIDDATA; | |
523 | } | |
524 | return subframe_len; | |
525 | } | |
526 | ||
527 | /** | |
528 | *@brief Decode how the data in the frame is split into subframes. | |
529 | * Every WMA frame contains the encoded data for a fixed number of | |
530 | * samples per channel. The data for every channel might be split | |
531 | * into several subframes. This function will reconstruct the list of | |
532 | * subframes for every channel. | |
533 | * | |
534 | * If the subframes are not evenly split, the algorithm estimates the | |
535 | * channels with the lowest number of total samples. | |
536 | * Afterwards, for each of these channels a bit is read from the | |
537 | * bitstream that indicates if the channel contains a subframe with the | |
538 | * next subframe size that is going to be read from the bitstream or not. | |
539 | * If a channel contains such a subframe, the subframe size gets added to | |
540 | * the channel's subframe list. | |
541 | * The algorithm repeats these steps until the frame is properly divided | |
542 | * between the individual channels. | |
543 | * | |
544 | *@param s context | |
545 | *@return 0 on success, < 0 in case of an error | |
546 | */ | |
547 | static int decode_tilehdr(WMAProDecodeCtx *s) | |
548 | { | |
549 | uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */ | |
550 | uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */ | |
551 | int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */ | |
552 | int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */ | |
553 | int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */ | |
554 | int c; | |
555 | ||
556 | /* Should never consume more than 3073 bits (256 iterations for the | |
557 | * while loop when always the minimum amount of 128 samples is subtracted | |
558 | * from missing samples in the 8 channel case). | |
559 | * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4) | |
560 | */ | |
561 | ||
562 | /** reset tiling information */ | |
563 | for (c = 0; c < s->avctx->channels; c++) | |
564 | s->channel[c].num_subframes = 0; | |
565 | ||
566 | if (s->max_num_subframes == 1 || get_bits1(&s->gb)) | |
567 | fixed_channel_layout = 1; | |
568 | ||
569 | /** loop until the frame data is split between the subframes */ | |
570 | do { | |
571 | int subframe_len; | |
572 | ||
573 | /** check which channels contain the subframe */ | |
574 | for (c = 0; c < s->avctx->channels; c++) { | |
575 | if (num_samples[c] == min_channel_len) { | |
576 | if (fixed_channel_layout || channels_for_cur_subframe == 1 || | |
577 | (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe)) | |
578 | contains_subframe[c] = 1; | |
579 | else | |
580 | contains_subframe[c] = get_bits1(&s->gb); | |
581 | } else | |
582 | contains_subframe[c] = 0; | |
583 | } | |
584 | ||
585 | /** get subframe length, subframe_len == 0 is not allowed */ | |
586 | if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0) | |
587 | return AVERROR_INVALIDDATA; | |
588 | ||
589 | /** add subframes to the individual channels and find new min_channel_len */ | |
590 | min_channel_len += subframe_len; | |
591 | for (c = 0; c < s->avctx->channels; c++) { | |
592 | WMAProChannelCtx* chan = &s->channel[c]; | |
593 | ||
594 | if (contains_subframe[c]) { | |
595 | if (chan->num_subframes >= MAX_SUBFRAMES) { | |
596 | av_log(s->avctx, AV_LOG_ERROR, | |
597 | "broken frame: num subframes > 31\n"); | |
598 | return AVERROR_INVALIDDATA; | |
599 | } | |
600 | chan->subframe_len[chan->num_subframes] = subframe_len; | |
601 | num_samples[c] += subframe_len; | |
602 | ++chan->num_subframes; | |
603 | if (num_samples[c] > s->samples_per_frame) { | |
604 | av_log(s->avctx, AV_LOG_ERROR, "broken frame: " | |
605 | "channel len > samples_per_frame\n"); | |
606 | return AVERROR_INVALIDDATA; | |
607 | } | |
608 | } else if (num_samples[c] <= min_channel_len) { | |
609 | if (num_samples[c] < min_channel_len) { | |
610 | channels_for_cur_subframe = 0; | |
611 | min_channel_len = num_samples[c]; | |
612 | } | |
613 | ++channels_for_cur_subframe; | |
614 | } | |
615 | } | |
616 | } while (min_channel_len < s->samples_per_frame); | |
617 | ||
618 | for (c = 0; c < s->avctx->channels; c++) { | |
619 | int i; | |
620 | int offset = 0; | |
621 | for (i = 0; i < s->channel[c].num_subframes; i++) { | |
622 | av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]" | |
623 | " len %i\n", s->frame_num, c, i, | |
624 | s->channel[c].subframe_len[i]); | |
625 | s->channel[c].subframe_offset[i] = offset; | |
626 | offset += s->channel[c].subframe_len[i]; | |
627 | } | |
628 | } | |
629 | ||
630 | return 0; | |
631 | } | |
632 | ||
633 | /** | |
634 | *@brief Calculate a decorrelation matrix from the bitstream parameters. | |
635 | *@param s codec context | |
636 | *@param chgroup channel group for which the matrix needs to be calculated | |
637 | */ | |
638 | static void decode_decorrelation_matrix(WMAProDecodeCtx *s, | |
639 | WMAProChannelGrp *chgroup) | |
640 | { | |
641 | int i; | |
642 | int offset = 0; | |
643 | int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS]; | |
644 | memset(chgroup->decorrelation_matrix, 0, s->avctx->channels * | |
645 | s->avctx->channels * sizeof(*chgroup->decorrelation_matrix)); | |
646 | ||
647 | for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++) | |
648 | rotation_offset[i] = get_bits(&s->gb, 6); | |
649 | ||
650 | for (i = 0; i < chgroup->num_channels; i++) | |
651 | chgroup->decorrelation_matrix[chgroup->num_channels * i + i] = | |
652 | get_bits1(&s->gb) ? 1.0 : -1.0; | |
653 | ||
654 | for (i = 1; i < chgroup->num_channels; i++) { | |
655 | int x; | |
656 | for (x = 0; x < i; x++) { | |
657 | int y; | |
658 | for (y = 0; y < i + 1; y++) { | |
659 | float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y]; | |
660 | float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y]; | |
661 | int n = rotation_offset[offset + x]; | |
662 | float sinv; | |
663 | float cosv; | |
664 | ||
665 | if (n < 32) { | |
666 | sinv = sin64[n]; | |
667 | cosv = sin64[32 - n]; | |
668 | } else { | |
669 | sinv = sin64[64 - n]; | |
670 | cosv = -sin64[n - 32]; | |
671 | } | |
672 | ||
673 | chgroup->decorrelation_matrix[y + x * chgroup->num_channels] = | |
674 | (v1 * sinv) - (v2 * cosv); | |
675 | chgroup->decorrelation_matrix[y + i * chgroup->num_channels] = | |
676 | (v1 * cosv) + (v2 * sinv); | |
677 | } | |
678 | } | |
679 | offset += i; | |
680 | } | |
681 | } | |
682 | ||
683 | /** | |
684 | *@brief Decode channel transformation parameters | |
685 | *@param s codec context | |
686 | *@return >= 0 in case of success, < 0 in case of bitstream errors | |
687 | */ | |
688 | static int decode_channel_transform(WMAProDecodeCtx* s) | |
689 | { | |
690 | int i; | |
691 | /* should never consume more than 1921 bits for the 8 channel case | |
692 | * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS | |
693 | * + MAX_CHANNELS + MAX_BANDS + 1) | |
694 | */ | |
695 | ||
696 | /** in the one channel case channel transforms are pointless */ | |
697 | s->num_chgroups = 0; | |
698 | if (s->avctx->channels > 1) { | |
699 | int remaining_channels = s->channels_for_cur_subframe; | |
700 | ||
701 | if (get_bits1(&s->gb)) { | |
702 | avpriv_request_sample(s->avctx, | |
703 | "Channel transform bit"); | |
704 | return AVERROR_PATCHWELCOME; | |
705 | } | |
706 | ||
707 | for (s->num_chgroups = 0; remaining_channels && | |
708 | s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) { | |
709 | WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups]; | |
710 | float** channel_data = chgroup->channel_data; | |
711 | chgroup->num_channels = 0; | |
712 | chgroup->transform = 0; | |
713 | ||
714 | /** decode channel mask */ | |
715 | if (remaining_channels > 2) { | |
716 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
717 | int channel_idx = s->channel_indexes_for_cur_subframe[i]; | |
718 | if (!s->channel[channel_idx].grouped | |
719 | && get_bits1(&s->gb)) { | |
720 | ++chgroup->num_channels; | |
721 | s->channel[channel_idx].grouped = 1; | |
722 | *channel_data++ = s->channel[channel_idx].coeffs; | |
723 | } | |
724 | } | |
725 | } else { | |
726 | chgroup->num_channels = remaining_channels; | |
727 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
728 | int channel_idx = s->channel_indexes_for_cur_subframe[i]; | |
729 | if (!s->channel[channel_idx].grouped) | |
730 | *channel_data++ = s->channel[channel_idx].coeffs; | |
731 | s->channel[channel_idx].grouped = 1; | |
732 | } | |
733 | } | |
734 | ||
735 | /** decode transform type */ | |
736 | if (chgroup->num_channels == 2) { | |
737 | if (get_bits1(&s->gb)) { | |
738 | if (get_bits1(&s->gb)) { | |
739 | avpriv_request_sample(s->avctx, | |
740 | "Unknown channel transform type"); | |
741 | return AVERROR_PATCHWELCOME; | |
742 | } | |
743 | } else { | |
744 | chgroup->transform = 1; | |
745 | if (s->avctx->channels == 2) { | |
746 | chgroup->decorrelation_matrix[0] = 1.0; | |
747 | chgroup->decorrelation_matrix[1] = -1.0; | |
748 | chgroup->decorrelation_matrix[2] = 1.0; | |
749 | chgroup->decorrelation_matrix[3] = 1.0; | |
750 | } else { | |
751 | /** cos(pi/4) */ | |
752 | chgroup->decorrelation_matrix[0] = 0.70703125; | |
753 | chgroup->decorrelation_matrix[1] = -0.70703125; | |
754 | chgroup->decorrelation_matrix[2] = 0.70703125; | |
755 | chgroup->decorrelation_matrix[3] = 0.70703125; | |
756 | } | |
757 | } | |
758 | } else if (chgroup->num_channels > 2) { | |
759 | if (get_bits1(&s->gb)) { | |
760 | chgroup->transform = 1; | |
761 | if (get_bits1(&s->gb)) { | |
762 | decode_decorrelation_matrix(s, chgroup); | |
763 | } else { | |
764 | /** FIXME: more than 6 coupled channels not supported */ | |
765 | if (chgroup->num_channels > 6) { | |
766 | avpriv_request_sample(s->avctx, | |
767 | "Coupled channels > 6"); | |
768 | } else { | |
769 | memcpy(chgroup->decorrelation_matrix, | |
770 | default_decorrelation[chgroup->num_channels], | |
771 | chgroup->num_channels * chgroup->num_channels * | |
772 | sizeof(*chgroup->decorrelation_matrix)); | |
773 | } | |
774 | } | |
775 | } | |
776 | } | |
777 | ||
778 | /** decode transform on / off */ | |
779 | if (chgroup->transform) { | |
780 | if (!get_bits1(&s->gb)) { | |
781 | int i; | |
782 | /** transform can be enabled for individual bands */ | |
783 | for (i = 0; i < s->num_bands; i++) { | |
784 | chgroup->transform_band[i] = get_bits1(&s->gb); | |
785 | } | |
786 | } else { | |
787 | memset(chgroup->transform_band, 1, s->num_bands); | |
788 | } | |
789 | } | |
790 | remaining_channels -= chgroup->num_channels; | |
791 | } | |
792 | } | |
793 | return 0; | |
794 | } | |
795 | ||
796 | /** | |
797 | *@brief Extract the coefficients from the bitstream. | |
798 | *@param s codec context | |
799 | *@param c current channel number | |
800 | *@return 0 on success, < 0 in case of bitstream errors | |
801 | */ | |
802 | static int decode_coeffs(WMAProDecodeCtx *s, int c) | |
803 | { | |
804 | /* Integers 0..15 as single-precision floats. The table saves a | |
805 | costly int to float conversion, and storing the values as | |
806 | integers allows fast sign-flipping. */ | |
807 | static const uint32_t fval_tab[16] = { | |
808 | 0x00000000, 0x3f800000, 0x40000000, 0x40400000, | |
809 | 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000, | |
810 | 0x41000000, 0x41100000, 0x41200000, 0x41300000, | |
811 | 0x41400000, 0x41500000, 0x41600000, 0x41700000, | |
812 | }; | |
813 | int vlctable; | |
814 | VLC* vlc; | |
815 | WMAProChannelCtx* ci = &s->channel[c]; | |
816 | int rl_mode = 0; | |
817 | int cur_coeff = 0; | |
818 | int num_zeros = 0; | |
819 | const uint16_t* run; | |
820 | const float* level; | |
821 | ||
822 | av_dlog(s->avctx, "decode coefficients for channel %i\n", c); | |
823 | ||
824 | vlctable = get_bits1(&s->gb); | |
825 | vlc = &coef_vlc[vlctable]; | |
826 | ||
827 | if (vlctable) { | |
828 | run = coef1_run; | |
829 | level = coef1_level; | |
830 | } else { | |
831 | run = coef0_run; | |
832 | level = coef0_level; | |
833 | } | |
834 | ||
835 | /** decode vector coefficients (consumes up to 167 bits per iteration for | |
836 | 4 vector coded large values) */ | |
837 | while ((s->transmit_num_vec_coeffs || !rl_mode) && | |
838 | (cur_coeff + 3 < ci->num_vec_coeffs)) { | |
839 | uint32_t vals[4]; | |
840 | int i; | |
841 | unsigned int idx; | |
842 | ||
843 | idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH); | |
844 | ||
845 | if (idx == HUFF_VEC4_SIZE - 1) { | |
846 | for (i = 0; i < 4; i += 2) { | |
847 | idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH); | |
848 | if (idx == HUFF_VEC2_SIZE - 1) { | |
849 | uint32_t v0, v1; | |
850 | v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH); | |
851 | if (v0 == HUFF_VEC1_SIZE - 1) | |
852 | v0 += ff_wma_get_large_val(&s->gb); | |
853 | v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH); | |
854 | if (v1 == HUFF_VEC1_SIZE - 1) | |
855 | v1 += ff_wma_get_large_val(&s->gb); | |
856 | vals[i ] = av_float2int(v0); | |
857 | vals[i+1] = av_float2int(v1); | |
858 | } else { | |
859 | vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ]; | |
860 | vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF]; | |
861 | } | |
862 | } | |
863 | } else { | |
864 | vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ]; | |
865 | vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF]; | |
866 | vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF]; | |
867 | vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF]; | |
868 | } | |
869 | ||
870 | /** decode sign */ | |
871 | for (i = 0; i < 4; i++) { | |
872 | if (vals[i]) { | |
873 | uint32_t sign = get_bits1(&s->gb) - 1; | |
874 | AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31); | |
875 | num_zeros = 0; | |
876 | } else { | |
877 | ci->coeffs[cur_coeff] = 0; | |
878 | /** switch to run level mode when subframe_len / 128 zeros | |
879 | were found in a row */ | |
880 | rl_mode |= (++num_zeros > s->subframe_len >> 8); | |
881 | } | |
882 | ++cur_coeff; | |
883 | } | |
884 | } | |
885 | ||
886 | /** decode run level coded coefficients */ | |
887 | if (cur_coeff < s->subframe_len) { | |
888 | memset(&ci->coeffs[cur_coeff], 0, | |
889 | sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff)); | |
890 | if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc, | |
891 | level, run, 1, ci->coeffs, | |
892 | cur_coeff, s->subframe_len, | |
893 | s->subframe_len, s->esc_len, 0)) | |
894 | return AVERROR_INVALIDDATA; | |
895 | } | |
896 | ||
897 | return 0; | |
898 | } | |
899 | ||
900 | /** | |
901 | *@brief Extract scale factors from the bitstream. | |
902 | *@param s codec context | |
903 | *@return 0 on success, < 0 in case of bitstream errors | |
904 | */ | |
905 | static int decode_scale_factors(WMAProDecodeCtx* s) | |
906 | { | |
907 | int i; | |
908 | ||
909 | /** should never consume more than 5344 bits | |
910 | * MAX_CHANNELS * (1 + MAX_BANDS * 23) | |
911 | */ | |
912 | ||
913 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
914 | int c = s->channel_indexes_for_cur_subframe[i]; | |
915 | int* sf; | |
916 | int* sf_end; | |
917 | s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx]; | |
918 | sf_end = s->channel[c].scale_factors + s->num_bands; | |
919 | ||
920 | /** resample scale factors for the new block size | |
921 | * as the scale factors might need to be resampled several times | |
922 | * before some new values are transmitted, a backup of the last | |
923 | * transmitted scale factors is kept in saved_scale_factors | |
924 | */ | |
925 | if (s->channel[c].reuse_sf) { | |
926 | const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx]; | |
927 | int b; | |
928 | for (b = 0; b < s->num_bands; b++) | |
929 | s->channel[c].scale_factors[b] = | |
930 | s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++]; | |
931 | } | |
932 | ||
933 | if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) { | |
934 | ||
935 | if (!s->channel[c].reuse_sf) { | |
936 | int val; | |
937 | /** decode DPCM coded scale factors */ | |
938 | s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1; | |
939 | val = 45 / s->channel[c].scale_factor_step; | |
940 | for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) { | |
941 | val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60; | |
942 | *sf = val; | |
943 | } | |
944 | } else { | |
945 | int i; | |
946 | /** run level decode differences to the resampled factors */ | |
947 | for (i = 0; i < s->num_bands; i++) { | |
948 | int idx; | |
949 | int skip; | |
950 | int val; | |
951 | int sign; | |
952 | ||
953 | idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH); | |
954 | ||
955 | if (!idx) { | |
956 | uint32_t code = get_bits(&s->gb, 14); | |
957 | val = code >> 6; | |
958 | sign = (code & 1) - 1; | |
959 | skip = (code & 0x3f) >> 1; | |
960 | } else if (idx == 1) { | |
961 | break; | |
962 | } else { | |
963 | skip = scale_rl_run[idx]; | |
964 | val = scale_rl_level[idx]; | |
965 | sign = get_bits1(&s->gb)-1; | |
966 | } | |
967 | ||
968 | i += skip; | |
969 | if (i >= s->num_bands) { | |
970 | av_log(s->avctx, AV_LOG_ERROR, | |
971 | "invalid scale factor coding\n"); | |
972 | return AVERROR_INVALIDDATA; | |
973 | } | |
974 | s->channel[c].scale_factors[i] += (val ^ sign) - sign; | |
975 | } | |
976 | } | |
977 | /** swap buffers */ | |
978 | s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx; | |
979 | s->channel[c].table_idx = s->table_idx; | |
980 | s->channel[c].reuse_sf = 1; | |
981 | } | |
982 | ||
983 | /** calculate new scale factor maximum */ | |
984 | s->channel[c].max_scale_factor = s->channel[c].scale_factors[0]; | |
985 | for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) { | |
986 | s->channel[c].max_scale_factor = | |
987 | FFMAX(s->channel[c].max_scale_factor, *sf); | |
988 | } | |
989 | ||
990 | } | |
991 | return 0; | |
992 | } | |
993 | ||
994 | /** | |
995 | *@brief Reconstruct the individual channel data. | |
996 | *@param s codec context | |
997 | */ | |
998 | static void inverse_channel_transform(WMAProDecodeCtx *s) | |
999 | { | |
1000 | int i; | |
1001 | ||
1002 | for (i = 0; i < s->num_chgroups; i++) { | |
1003 | if (s->chgroup[i].transform) { | |
1004 | float data[WMAPRO_MAX_CHANNELS]; | |
1005 | const int num_channels = s->chgroup[i].num_channels; | |
1006 | float** ch_data = s->chgroup[i].channel_data; | |
1007 | float** ch_end = ch_data + num_channels; | |
1008 | const int8_t* tb = s->chgroup[i].transform_band; | |
1009 | int16_t* sfb; | |
1010 | ||
1011 | /** multichannel decorrelation */ | |
1012 | for (sfb = s->cur_sfb_offsets; | |
1013 | sfb < s->cur_sfb_offsets + s->num_bands; sfb++) { | |
1014 | int y; | |
1015 | if (*tb++ == 1) { | |
1016 | /** multiply values with the decorrelation_matrix */ | |
1017 | for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) { | |
1018 | const float* mat = s->chgroup[i].decorrelation_matrix; | |
1019 | const float* data_end = data + num_channels; | |
1020 | float* data_ptr = data; | |
1021 | float** ch; | |
1022 | ||
1023 | for (ch = ch_data; ch < ch_end; ch++) | |
1024 | *data_ptr++ = (*ch)[y]; | |
1025 | ||
1026 | for (ch = ch_data; ch < ch_end; ch++) { | |
1027 | float sum = 0; | |
1028 | data_ptr = data; | |
1029 | while (data_ptr < data_end) | |
1030 | sum += *data_ptr++ * *mat++; | |
1031 | ||
1032 | (*ch)[y] = sum; | |
1033 | } | |
1034 | } | |
1035 | } else if (s->avctx->channels == 2) { | |
1036 | int len = FFMIN(sfb[1], s->subframe_len) - sfb[0]; | |
1037 | s->fdsp.vector_fmul_scalar(ch_data[0] + sfb[0], | |
1038 | ch_data[0] + sfb[0], | |
1039 | 181.0 / 128, len); | |
1040 | s->fdsp.vector_fmul_scalar(ch_data[1] + sfb[0], | |
1041 | ch_data[1] + sfb[0], | |
1042 | 181.0 / 128, len); | |
1043 | } | |
1044 | } | |
1045 | } | |
1046 | } | |
1047 | } | |
1048 | ||
1049 | /** | |
1050 | *@brief Apply sine window and reconstruct the output buffer. | |
1051 | *@param s codec context | |
1052 | */ | |
1053 | static void wmapro_window(WMAProDecodeCtx *s) | |
1054 | { | |
1055 | int i; | |
1056 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
1057 | int c = s->channel_indexes_for_cur_subframe[i]; | |
1058 | const float* window; | |
1059 | int winlen = s->channel[c].prev_block_len; | |
1060 | float* start = s->channel[c].coeffs - (winlen >> 1); | |
1061 | ||
1062 | if (s->subframe_len < winlen) { | |
1063 | start += (winlen - s->subframe_len) >> 1; | |
1064 | winlen = s->subframe_len; | |
1065 | } | |
1066 | ||
1067 | window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS]; | |
1068 | ||
1069 | winlen >>= 1; | |
1070 | ||
1071 | s->fdsp.vector_fmul_window(start, start, start + winlen, | |
1072 | window, winlen); | |
1073 | ||
1074 | s->channel[c].prev_block_len = s->subframe_len; | |
1075 | } | |
1076 | } | |
1077 | ||
1078 | /** | |
1079 | *@brief Decode a single subframe (block). | |
1080 | *@param s codec context | |
1081 | *@return 0 on success, < 0 when decoding failed | |
1082 | */ | |
1083 | static int decode_subframe(WMAProDecodeCtx *s) | |
1084 | { | |
1085 | int offset = s->samples_per_frame; | |
1086 | int subframe_len = s->samples_per_frame; | |
1087 | int i; | |
1088 | int total_samples = s->samples_per_frame * s->avctx->channels; | |
1089 | int transmit_coeffs = 0; | |
1090 | int cur_subwoofer_cutoff; | |
1091 | ||
1092 | s->subframe_offset = get_bits_count(&s->gb); | |
1093 | ||
1094 | /** reset channel context and find the next block offset and size | |
1095 | == the next block of the channel with the smallest number of | |
1096 | decoded samples | |
1097 | */ | |
1098 | for (i = 0; i < s->avctx->channels; i++) { | |
1099 | s->channel[i].grouped = 0; | |
1100 | if (offset > s->channel[i].decoded_samples) { | |
1101 | offset = s->channel[i].decoded_samples; | |
1102 | subframe_len = | |
1103 | s->channel[i].subframe_len[s->channel[i].cur_subframe]; | |
1104 | } | |
1105 | } | |
1106 | ||
1107 | av_dlog(s->avctx, | |
1108 | "processing subframe with offset %i len %i\n", offset, subframe_len); | |
1109 | ||
1110 | /** get a list of all channels that contain the estimated block */ | |
1111 | s->channels_for_cur_subframe = 0; | |
1112 | for (i = 0; i < s->avctx->channels; i++) { | |
1113 | const int cur_subframe = s->channel[i].cur_subframe; | |
1114 | /** subtract already processed samples */ | |
1115 | total_samples -= s->channel[i].decoded_samples; | |
1116 | ||
1117 | /** and count if there are multiple subframes that match our profile */ | |
1118 | if (offset == s->channel[i].decoded_samples && | |
1119 | subframe_len == s->channel[i].subframe_len[cur_subframe]) { | |
1120 | total_samples -= s->channel[i].subframe_len[cur_subframe]; | |
1121 | s->channel[i].decoded_samples += | |
1122 | s->channel[i].subframe_len[cur_subframe]; | |
1123 | s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i; | |
1124 | ++s->channels_for_cur_subframe; | |
1125 | } | |
1126 | } | |
1127 | ||
1128 | /** check if the frame will be complete after processing the | |
1129 | estimated block */ | |
1130 | if (!total_samples) | |
1131 | s->parsed_all_subframes = 1; | |
1132 | ||
1133 | ||
1134 | av_dlog(s->avctx, "subframe is part of %i channels\n", | |
1135 | s->channels_for_cur_subframe); | |
1136 | ||
1137 | /** calculate number of scale factor bands and their offsets */ | |
1138 | s->table_idx = av_log2(s->samples_per_frame/subframe_len); | |
1139 | s->num_bands = s->num_sfb[s->table_idx]; | |
1140 | s->cur_sfb_offsets = s->sfb_offsets[s->table_idx]; | |
1141 | cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx]; | |
1142 | ||
1143 | /** configure the decoder for the current subframe */ | |
1144 | offset += s->samples_per_frame >> 1; | |
1145 | ||
1146 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
1147 | int c = s->channel_indexes_for_cur_subframe[i]; | |
1148 | ||
1149 | s->channel[c].coeffs = &s->channel[c].out[offset]; | |
1150 | } | |
1151 | ||
1152 | s->subframe_len = subframe_len; | |
1153 | s->esc_len = av_log2(s->subframe_len - 1) + 1; | |
1154 | ||
1155 | /** skip extended header if any */ | |
1156 | if (get_bits1(&s->gb)) { | |
1157 | int num_fill_bits; | |
1158 | if (!(num_fill_bits = get_bits(&s->gb, 2))) { | |
1159 | int len = get_bits(&s->gb, 4); | |
1160 | num_fill_bits = (len ? get_bits(&s->gb, len) : 0) + 1; | |
1161 | } | |
1162 | ||
1163 | if (num_fill_bits >= 0) { | |
1164 | if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) { | |
1165 | av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n"); | |
1166 | return AVERROR_INVALIDDATA; | |
1167 | } | |
1168 | ||
1169 | skip_bits_long(&s->gb, num_fill_bits); | |
1170 | } | |
1171 | } | |
1172 | ||
1173 | /** no idea for what the following bit is used */ | |
1174 | if (get_bits1(&s->gb)) { | |
1175 | avpriv_request_sample(s->avctx, "Reserved bit"); | |
1176 | return AVERROR_PATCHWELCOME; | |
1177 | } | |
1178 | ||
1179 | ||
1180 | if (decode_channel_transform(s) < 0) | |
1181 | return AVERROR_INVALIDDATA; | |
1182 | ||
1183 | ||
1184 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
1185 | int c = s->channel_indexes_for_cur_subframe[i]; | |
1186 | if ((s->channel[c].transmit_coefs = get_bits1(&s->gb))) | |
1187 | transmit_coeffs = 1; | |
1188 | } | |
1189 | ||
1190 | av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE); | |
1191 | if (transmit_coeffs) { | |
1192 | int step; | |
1193 | int quant_step = 90 * s->bits_per_sample >> 4; | |
1194 | ||
1195 | /** decode number of vector coded coefficients */ | |
1196 | if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) { | |
1197 | int num_bits = av_log2((s->subframe_len + 3)/4) + 1; | |
1198 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
1199 | int c = s->channel_indexes_for_cur_subframe[i]; | |
1200 | int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2; | |
1201 | if (num_vec_coeffs > s->subframe_len) { | |
1202 | av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs); | |
1203 | return AVERROR_INVALIDDATA; | |
1204 | } | |
1205 | av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out)); | |
1206 | s->channel[c].num_vec_coeffs = num_vec_coeffs; | |
1207 | } | |
1208 | } else { | |
1209 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
1210 | int c = s->channel_indexes_for_cur_subframe[i]; | |
1211 | s->channel[c].num_vec_coeffs = s->subframe_len; | |
1212 | } | |
1213 | } | |
1214 | /** decode quantization step */ | |
1215 | step = get_sbits(&s->gb, 6); | |
1216 | quant_step += step; | |
1217 | if (step == -32 || step == 31) { | |
1218 | const int sign = (step == 31) - 1; | |
1219 | int quant = 0; | |
1220 | while (get_bits_count(&s->gb) + 5 < s->num_saved_bits && | |
1221 | (step = get_bits(&s->gb, 5)) == 31) { | |
1222 | quant += 31; | |
1223 | } | |
1224 | quant_step += ((quant + step) ^ sign) - sign; | |
1225 | } | |
1226 | if (quant_step < 0) { | |
1227 | av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n"); | |
1228 | } | |
1229 | ||
1230 | /** decode quantization step modifiers for every channel */ | |
1231 | ||
1232 | if (s->channels_for_cur_subframe == 1) { | |
1233 | s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step; | |
1234 | } else { | |
1235 | int modifier_len = get_bits(&s->gb, 3); | |
1236 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
1237 | int c = s->channel_indexes_for_cur_subframe[i]; | |
1238 | s->channel[c].quant_step = quant_step; | |
1239 | if (get_bits1(&s->gb)) { | |
1240 | if (modifier_len) { | |
1241 | s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1; | |
1242 | } else | |
1243 | ++s->channel[c].quant_step; | |
1244 | } | |
1245 | } | |
1246 | } | |
1247 | ||
1248 | /** decode scale factors */ | |
1249 | if (decode_scale_factors(s) < 0) | |
1250 | return AVERROR_INVALIDDATA; | |
1251 | } | |
1252 | ||
1253 | av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n", | |
1254 | get_bits_count(&s->gb) - s->subframe_offset); | |
1255 | ||
1256 | /** parse coefficients */ | |
1257 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
1258 | int c = s->channel_indexes_for_cur_subframe[i]; | |
1259 | if (s->channel[c].transmit_coefs && | |
1260 | get_bits_count(&s->gb) < s->num_saved_bits) { | |
1261 | decode_coeffs(s, c); | |
1262 | } else | |
1263 | memset(s->channel[c].coeffs, 0, | |
1264 | sizeof(*s->channel[c].coeffs) * subframe_len); | |
1265 | } | |
1266 | ||
1267 | av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n", | |
1268 | get_bits_count(&s->gb) - s->subframe_offset); | |
1269 | ||
1270 | if (transmit_coeffs) { | |
1271 | FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS]; | |
1272 | /** reconstruct the per channel data */ | |
1273 | inverse_channel_transform(s); | |
1274 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
1275 | int c = s->channel_indexes_for_cur_subframe[i]; | |
1276 | const int* sf = s->channel[c].scale_factors; | |
1277 | int b; | |
1278 | ||
1279 | if (c == s->lfe_channel) | |
1280 | memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) * | |
1281 | (subframe_len - cur_subwoofer_cutoff)); | |
1282 | ||
1283 | /** inverse quantization and rescaling */ | |
1284 | for (b = 0; b < s->num_bands; b++) { | |
1285 | const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len); | |
1286 | const int exp = s->channel[c].quant_step - | |
1287 | (s->channel[c].max_scale_factor - *sf++) * | |
1288 | s->channel[c].scale_factor_step; | |
1289 | const float quant = pow(10.0, exp / 20.0); | |
1290 | int start = s->cur_sfb_offsets[b]; | |
1291 | s->fdsp.vector_fmul_scalar(s->tmp + start, | |
1292 | s->channel[c].coeffs + start, | |
1293 | quant, end - start); | |
1294 | } | |
1295 | ||
1296 | /** apply imdct (imdct_half == DCTIV with reverse) */ | |
1297 | mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp); | |
1298 | } | |
1299 | } | |
1300 | ||
1301 | /** window and overlapp-add */ | |
1302 | wmapro_window(s); | |
1303 | ||
1304 | /** handled one subframe */ | |
1305 | for (i = 0; i < s->channels_for_cur_subframe; i++) { | |
1306 | int c = s->channel_indexes_for_cur_subframe[i]; | |
1307 | if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) { | |
1308 | av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n"); | |
1309 | return AVERROR_INVALIDDATA; | |
1310 | } | |
1311 | ++s->channel[c].cur_subframe; | |
1312 | } | |
1313 | ||
1314 | return 0; | |
1315 | } | |
1316 | ||
1317 | /** | |
1318 | *@brief Decode one WMA frame. | |
1319 | *@param s codec context | |
1320 | *@return 0 if the trailer bit indicates that this is the last frame, | |
1321 | * 1 if there are additional frames | |
1322 | */ | |
1323 | static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr) | |
1324 | { | |
1325 | AVCodecContext *avctx = s->avctx; | |
1326 | GetBitContext* gb = &s->gb; | |
1327 | int more_frames = 0; | |
1328 | int len = 0; | |
1329 | int i, ret; | |
1330 | ||
1331 | /** get frame length */ | |
1332 | if (s->len_prefix) | |
1333 | len = get_bits(gb, s->log2_frame_size); | |
1334 | ||
1335 | av_dlog(s->avctx, "decoding frame with length %x\n", len); | |
1336 | ||
1337 | /** decode tile information */ | |
1338 | if (decode_tilehdr(s)) { | |
1339 | s->packet_loss = 1; | |
1340 | return 0; | |
1341 | } | |
1342 | ||
1343 | /** read postproc transform */ | |
1344 | if (s->avctx->channels > 1 && get_bits1(gb)) { | |
1345 | if (get_bits1(gb)) { | |
1346 | for (i = 0; i < avctx->channels * avctx->channels; i++) | |
1347 | skip_bits(gb, 4); | |
1348 | } | |
1349 | } | |
1350 | ||
1351 | /** read drc info */ | |
1352 | if (s->dynamic_range_compression) { | |
1353 | s->drc_gain = get_bits(gb, 8); | |
1354 | av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain); | |
1355 | } | |
1356 | ||
1357 | /** no idea what these are for, might be the number of samples | |
1358 | that need to be skipped at the beginning or end of a stream */ | |
1359 | if (get_bits1(gb)) { | |
1360 | int av_unused skip; | |
1361 | ||
1362 | /** usually true for the first frame */ | |
1363 | if (get_bits1(gb)) { | |
1364 | skip = get_bits(gb, av_log2(s->samples_per_frame * 2)); | |
1365 | av_dlog(s->avctx, "start skip: %i\n", skip); | |
1366 | } | |
1367 | ||
1368 | /** sometimes true for the last frame */ | |
1369 | if (get_bits1(gb)) { | |
1370 | skip = get_bits(gb, av_log2(s->samples_per_frame * 2)); | |
1371 | av_dlog(s->avctx, "end skip: %i\n", skip); | |
1372 | } | |
1373 | ||
1374 | } | |
1375 | ||
1376 | av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n", | |
1377 | get_bits_count(gb) - s->frame_offset); | |
1378 | ||
1379 | /** reset subframe states */ | |
1380 | s->parsed_all_subframes = 0; | |
1381 | for (i = 0; i < avctx->channels; i++) { | |
1382 | s->channel[i].decoded_samples = 0; | |
1383 | s->channel[i].cur_subframe = 0; | |
1384 | s->channel[i].reuse_sf = 0; | |
1385 | } | |
1386 | ||
1387 | /** decode all subframes */ | |
1388 | while (!s->parsed_all_subframes) { | |
1389 | if (decode_subframe(s) < 0) { | |
1390 | s->packet_loss = 1; | |
1391 | return 0; | |
1392 | } | |
1393 | } | |
1394 | ||
1395 | /* get output buffer */ | |
1396 | frame->nb_samples = s->samples_per_frame; | |
1397 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) { | |
1398 | s->packet_loss = 1; | |
1399 | return 0; | |
1400 | } | |
1401 | ||
1402 | /** copy samples to the output buffer */ | |
1403 | for (i = 0; i < avctx->channels; i++) | |
1404 | memcpy(frame->extended_data[i], s->channel[i].out, | |
1405 | s->samples_per_frame * sizeof(*s->channel[i].out)); | |
1406 | ||
1407 | for (i = 0; i < avctx->channels; i++) { | |
1408 | /** reuse second half of the IMDCT output for the next frame */ | |
1409 | memcpy(&s->channel[i].out[0], | |
1410 | &s->channel[i].out[s->samples_per_frame], | |
1411 | s->samples_per_frame * sizeof(*s->channel[i].out) >> 1); | |
1412 | } | |
1413 | ||
1414 | if (s->skip_frame) { | |
1415 | s->skip_frame = 0; | |
1416 | *got_frame_ptr = 0; | |
1417 | av_frame_unref(frame); | |
1418 | } else { | |
1419 | *got_frame_ptr = 1; | |
1420 | } | |
1421 | ||
1422 | if (s->len_prefix) { | |
1423 | if (len != (get_bits_count(gb) - s->frame_offset) + 2) { | |
1424 | /** FIXME: not sure if this is always an error */ | |
1425 | av_log(s->avctx, AV_LOG_ERROR, | |
1426 | "frame[%"PRIu32"] would have to skip %i bits\n", | |
1427 | s->frame_num, | |
1428 | len - (get_bits_count(gb) - s->frame_offset) - 1); | |
1429 | s->packet_loss = 1; | |
1430 | return 0; | |
1431 | } | |
1432 | ||
1433 | /** skip the rest of the frame data */ | |
1434 | skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1); | |
1435 | } else { | |
1436 | while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) { | |
1437 | } | |
1438 | } | |
1439 | ||
1440 | /** decode trailer bit */ | |
1441 | more_frames = get_bits1(gb); | |
1442 | ||
1443 | ++s->frame_num; | |
1444 | return more_frames; | |
1445 | } | |
1446 | ||
1447 | /** | |
1448 | *@brief Calculate remaining input buffer length. | |
1449 | *@param s codec context | |
1450 | *@param gb bitstream reader context | |
1451 | *@return remaining size in bits | |
1452 | */ | |
1453 | static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb) | |
1454 | { | |
1455 | return s->buf_bit_size - get_bits_count(gb); | |
1456 | } | |
1457 | ||
1458 | /** | |
1459 | *@brief Fill the bit reservoir with a (partial) frame. | |
1460 | *@param s codec context | |
1461 | *@param gb bitstream reader context | |
1462 | *@param len length of the partial frame | |
1463 | *@param append decides whether to reset the buffer or not | |
1464 | */ | |
1465 | static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len, | |
1466 | int append) | |
1467 | { | |
1468 | int buflen; | |
1469 | ||
1470 | /** when the frame data does not need to be concatenated, the input buffer | |
1471 | is reset and additional bits from the previous frame are copied | |
1472 | and skipped later so that a fast byte copy is possible */ | |
1473 | ||
1474 | if (!append) { | |
1475 | s->frame_offset = get_bits_count(gb) & 7; | |
1476 | s->num_saved_bits = s->frame_offset; | |
1477 | init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE); | |
1478 | } | |
1479 | ||
1480 | buflen = (put_bits_count(&s->pb) + len + 8) >> 3; | |
1481 | ||
1482 | if (len <= 0 || buflen > MAX_FRAMESIZE) { | |
1483 | avpriv_request_sample(s->avctx, "Too small input buffer"); | |
1484 | s->packet_loss = 1; | |
1485 | return; | |
1486 | } | |
1487 | ||
1488 | av_assert0(len <= put_bits_left(&s->pb)); | |
1489 | ||
1490 | s->num_saved_bits += len; | |
1491 | if (!append) { | |
1492 | avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), | |
1493 | s->num_saved_bits); | |
1494 | } else { | |
1495 | int align = 8 - (get_bits_count(gb) & 7); | |
1496 | align = FFMIN(align, len); | |
1497 | put_bits(&s->pb, align, get_bits(gb, align)); | |
1498 | len -= align; | |
1499 | avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len); | |
1500 | } | |
1501 | skip_bits_long(gb, len); | |
1502 | ||
1503 | { | |
1504 | PutBitContext tmp = s->pb; | |
1505 | flush_put_bits(&tmp); | |
1506 | } | |
1507 | ||
1508 | init_get_bits(&s->gb, s->frame_data, s->num_saved_bits); | |
1509 | skip_bits(&s->gb, s->frame_offset); | |
1510 | } | |
1511 | ||
1512 | /** | |
1513 | *@brief Decode a single WMA packet. | |
1514 | *@param avctx codec context | |
1515 | *@param data the output buffer | |
1516 | *@param avpkt input packet | |
1517 | *@return number of bytes that were read from the input buffer | |
1518 | */ | |
1519 | static int decode_packet(AVCodecContext *avctx, void *data, | |
1520 | int *got_frame_ptr, AVPacket* avpkt) | |
1521 | { | |
1522 | WMAProDecodeCtx *s = avctx->priv_data; | |
1523 | GetBitContext* gb = &s->pgb; | |
1524 | const uint8_t* buf = avpkt->data; | |
1525 | int buf_size = avpkt->size; | |
1526 | int num_bits_prev_frame; | |
1527 | int packet_sequence_number; | |
1528 | ||
1529 | *got_frame_ptr = 0; | |
1530 | ||
1531 | if (s->packet_done || s->packet_loss) { | |
1532 | s->packet_done = 0; | |
1533 | ||
1534 | /** sanity check for the buffer length */ | |
1535 | if (buf_size < avctx->block_align) { | |
1536 | av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n", | |
1537 | buf_size, avctx->block_align); | |
1538 | return AVERROR_INVALIDDATA; | |
1539 | } | |
1540 | ||
1541 | s->next_packet_start = buf_size - avctx->block_align; | |
1542 | buf_size = avctx->block_align; | |
1543 | s->buf_bit_size = buf_size << 3; | |
1544 | ||
1545 | /** parse packet header */ | |
1546 | init_get_bits(gb, buf, s->buf_bit_size); | |
1547 | packet_sequence_number = get_bits(gb, 4); | |
1548 | skip_bits(gb, 2); | |
1549 | ||
1550 | /** get number of bits that need to be added to the previous frame */ | |
1551 | num_bits_prev_frame = get_bits(gb, s->log2_frame_size); | |
1552 | av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number, | |
1553 | num_bits_prev_frame); | |
1554 | ||
1555 | /** check for packet loss */ | |
1556 | if (!s->packet_loss && | |
1557 | ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) { | |
1558 | s->packet_loss = 1; | |
1559 | av_log(avctx, AV_LOG_ERROR, | |
1560 | "Packet loss detected! seq %"PRIx8" vs %x\n", | |
1561 | s->packet_sequence_number, packet_sequence_number); | |
1562 | } | |
1563 | s->packet_sequence_number = packet_sequence_number; | |
1564 | ||
1565 | if (num_bits_prev_frame > 0) { | |
1566 | int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb); | |
1567 | if (num_bits_prev_frame >= remaining_packet_bits) { | |
1568 | num_bits_prev_frame = remaining_packet_bits; | |
1569 | s->packet_done = 1; | |
1570 | } | |
1571 | ||
1572 | /** append the previous frame data to the remaining data from the | |
1573 | previous packet to create a full frame */ | |
1574 | save_bits(s, gb, num_bits_prev_frame, 1); | |
1575 | av_dlog(avctx, "accumulated %x bits of frame data\n", | |
1576 | s->num_saved_bits - s->frame_offset); | |
1577 | ||
1578 | /** decode the cross packet frame if it is valid */ | |
1579 | if (!s->packet_loss) | |
1580 | decode_frame(s, data, got_frame_ptr); | |
1581 | } else if (s->num_saved_bits - s->frame_offset) { | |
1582 | av_dlog(avctx, "ignoring %x previously saved bits\n", | |
1583 | s->num_saved_bits - s->frame_offset); | |
1584 | } | |
1585 | ||
1586 | if (s->packet_loss) { | |
1587 | /** reset number of saved bits so that the decoder | |
1588 | does not start to decode incomplete frames in the | |
1589 | s->len_prefix == 0 case */ | |
1590 | s->num_saved_bits = 0; | |
1591 | s->packet_loss = 0; | |
1592 | } | |
1593 | ||
1594 | } else { | |
1595 | int frame_size; | |
1596 | s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3; | |
1597 | init_get_bits(gb, avpkt->data, s->buf_bit_size); | |
1598 | skip_bits(gb, s->packet_offset); | |
1599 | if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size && | |
1600 | (frame_size = show_bits(gb, s->log2_frame_size)) && | |
1601 | frame_size <= remaining_bits(s, gb)) { | |
1602 | save_bits(s, gb, frame_size, 0); | |
1603 | if (!s->packet_loss) | |
1604 | s->packet_done = !decode_frame(s, data, got_frame_ptr); | |
1605 | } else if (!s->len_prefix | |
1606 | && s->num_saved_bits > get_bits_count(&s->gb)) { | |
1607 | /** when the frames do not have a length prefix, we don't know | |
1608 | the compressed length of the individual frames | |
1609 | however, we know what part of a new packet belongs to the | |
1610 | previous frame | |
1611 | therefore we save the incoming packet first, then we append | |
1612 | the "previous frame" data from the next packet so that | |
1613 | we get a buffer that only contains full frames */ | |
1614 | s->packet_done = !decode_frame(s, data, got_frame_ptr); | |
1615 | } else | |
1616 | s->packet_done = 1; | |
1617 | } | |
1618 | ||
1619 | if (s->packet_done && !s->packet_loss && | |
1620 | remaining_bits(s, gb) > 0) { | |
1621 | /** save the rest of the data so that it can be decoded | |
1622 | with the next packet */ | |
1623 | save_bits(s, gb, remaining_bits(s, gb), 0); | |
1624 | } | |
1625 | ||
1626 | s->packet_offset = get_bits_count(gb) & 7; | |
1627 | if (s->packet_loss) | |
1628 | return AVERROR_INVALIDDATA; | |
1629 | ||
1630 | return get_bits_count(gb) >> 3; | |
1631 | } | |
1632 | ||
1633 | /** | |
1634 | *@brief Clear decoder buffers (for seeking). | |
1635 | *@param avctx codec context | |
1636 | */ | |
1637 | static void flush(AVCodecContext *avctx) | |
1638 | { | |
1639 | WMAProDecodeCtx *s = avctx->priv_data; | |
1640 | int i; | |
1641 | /** reset output buffer as a part of it is used during the windowing of a | |
1642 | new frame */ | |
1643 | for (i = 0; i < avctx->channels; i++) | |
1644 | memset(s->channel[i].out, 0, s->samples_per_frame * | |
1645 | sizeof(*s->channel[i].out)); | |
1646 | s->packet_loss = 1; | |
1647 | } | |
1648 | ||
1649 | ||
1650 | /** | |
1651 | *@brief wmapro decoder | |
1652 | */ | |
1653 | AVCodec ff_wmapro_decoder = { | |
1654 | .name = "wmapro", | |
1655 | .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"), | |
1656 | .type = AVMEDIA_TYPE_AUDIO, | |
1657 | .id = AV_CODEC_ID_WMAPRO, | |
1658 | .priv_data_size = sizeof(WMAProDecodeCtx), | |
1659 | .init = decode_init, | |
1660 | .close = decode_end, | |
1661 | .decode = decode_packet, | |
1662 | .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1, | |
1663 | .flush = flush, | |
1664 | .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP, | |
1665 | AV_SAMPLE_FMT_NONE }, | |
1666 | }; |