Line data Source code
1 : /*
2 : * Wmapro compatible decoder
3 : * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 : * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson
5 : *
6 : * This file is part of FFmpeg.
7 : *
8 : * FFmpeg is free software; you can redistribute it and/or
9 : * modify it under the terms of the GNU Lesser General Public
10 : * License as published by the Free Software Foundation; either
11 : * version 2.1 of the License, or (at your option) any later version.
12 : *
13 : * FFmpeg is distributed in the hope that it will be useful,
14 : * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 : * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 : * Lesser General Public License for more details.
17 : *
18 : * You should have received a copy of the GNU Lesser General Public
19 : * License along with FFmpeg; if not, write to the Free Software
20 : * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 : */
22 :
23 : /**
24 : * @file
25 : * @brief wmapro decoder implementation
26 : * Wmapro is an MDCT based codec comparable to wma standard or AAC.
27 : * The decoding therefore consists of the following steps:
28 : * - bitstream decoding
29 : * - reconstruction of per-channel data
30 : * - rescaling and inverse quantization
31 : * - IMDCT
32 : * - windowing and overlapp-add
33 : *
34 : * The compressed wmapro bitstream is split into individual packets.
35 : * Every such packet contains one or more wma frames.
36 : * The compressed frames may have a variable length and frames may
37 : * cross packet boundaries.
38 : * Common to all wmapro frames is the number of samples that are stored in
39 : * a frame.
40 : * The number of samples and a few other decode flags are stored
41 : * as extradata that has to be passed to the decoder.
42 : *
43 : * The wmapro frames themselves are again split into a variable number of
44 : * subframes. Every subframe contains the data for 2^N time domain samples
45 : * where N varies between 7 and 12.
46 : *
47 : * Example wmapro bitstream (in samples):
48 : *
49 : * || packet 0 || packet 1 || packet 2 packets
50 : * ---------------------------------------------------
51 : * || frame 0 || frame 1 || frame 2 || frames
52 : * ---------------------------------------------------
53 : * || | | || | | | || || subframes of channel 0
54 : * ---------------------------------------------------
55 : * || | | || | | | || || subframes of channel 1
56 : * ---------------------------------------------------
57 : *
58 : * The frame layouts for the individual channels of a wma frame does not need
59 : * to be the same.
60 : *
61 : * However, if the offsets and lengths of several subframes of a frame are the
62 : * same, the subframes of the channels can be grouped.
63 : * Every group may then use special coding techniques like M/S stereo coding
64 : * to improve the compression ratio. These channel transformations do not
65 : * need to be applied to a whole subframe. Instead, they can also work on
66 : * individual scale factor bands (see below).
67 : * The coefficients that carry the audio signal in the frequency domain
68 : * are transmitted as huffman-coded vectors with 4, 2 and 1 elements.
69 : * In addition to that, the encoder can switch to a runlevel coding scheme
70 : * by transmitting subframe_length / 128 zero coefficients.
71 : *
72 : * Before the audio signal can be converted to the time domain, the
73 : * coefficients have to be rescaled and inverse quantized.
74 : * A subframe is therefore split into several scale factor bands that get
75 : * scaled individually.
76 : * Scale factors are submitted for every frame but they might be shared
77 : * between the subframes of a channel. Scale factors are initially DPCM-coded.
78 : * Once scale factors are shared, the differences are transmitted as runlevel
79 : * codes.
80 : * Every subframe length and offset combination in the frame layout shares a
81 : * common quantization factor that can be adjusted for every channel by a
82 : * modifier.
83 : * After the inverse quantization, the coefficients get processed by an IMDCT.
84 : * The resulting values are then windowed with a sine window and the first half
85 : * of the values are added to the second half of the output from the previous
86 : * subframe in order to reconstruct the output samples.
87 : */
88 :
89 : #include <inttypes.h>
90 :
91 : #include "libavutil/ffmath.h"
92 : #include "libavutil/float_dsp.h"
93 : #include "libavutil/intfloat.h"
94 : #include "libavutil/intreadwrite.h"
95 : #include "avcodec.h"
96 : #include "internal.h"
97 : #include "get_bits.h"
98 : #include "put_bits.h"
99 : #include "wmaprodata.h"
100 : #include "sinewin.h"
101 : #include "wma.h"
102 : #include "wma_common.h"
103 :
104 : /** current decoder limitations */
105 : #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
106 : #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
107 : #define MAX_BANDS 29 ///< max number of scale factor bands
108 : #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
109 : #define XMA_MAX_STREAMS 8
110 : #define XMA_MAX_CHANNELS_STREAM 2
111 : #define XMA_MAX_CHANNELS (XMA_MAX_STREAMS * XMA_MAX_CHANNELS_STREAM)
112 :
113 : #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
114 : #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
115 : #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size
116 : #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
117 : #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
118 :
119 :
120 : #define VLCBITS 9
121 : #define SCALEVLCBITS 8
122 : #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
123 : #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
124 : #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
125 : #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
126 : #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
127 :
128 : static VLC sf_vlc; ///< scale factor DPCM vlc
129 : static VLC sf_rl_vlc; ///< scale factor run length vlc
130 : static VLC vec4_vlc; ///< 4 coefficients per symbol
131 : static VLC vec2_vlc; ///< 2 coefficients per symbol
132 : static VLC vec1_vlc; ///< 1 coefficient per symbol
133 : static VLC coef_vlc[2]; ///< coefficient run length vlc codes
134 : static float sin64[33]; ///< sine table for decorrelation
135 :
136 : /**
137 : * @brief frame specific decoder context for a single channel
138 : */
139 : typedef struct WMAProChannelCtx {
140 : int16_t prev_block_len; ///< length of the previous block
141 : uint8_t transmit_coefs;
142 : uint8_t num_subframes;
143 : uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
144 : uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
145 : uint8_t cur_subframe; ///< current subframe number
146 : uint16_t decoded_samples; ///< number of already processed samples
147 : uint8_t grouped; ///< channel is part of a group
148 : int quant_step; ///< quantization step for the current subframe
149 : int8_t reuse_sf; ///< share scale factors between subframes
150 : int8_t scale_factor_step; ///< scaling step for the current subframe
151 : int max_scale_factor; ///< maximum scale factor for the current subframe
152 : int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
153 : int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
154 : int* scale_factors; ///< pointer to the scale factor values used for decoding
155 : uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
156 : float* coeffs; ///< pointer to the subframe decode buffer
157 : uint16_t num_vec_coeffs; ///< number of vector coded coefficients
158 : DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
159 : } WMAProChannelCtx;
160 :
161 : /**
162 : * @brief channel group for channel transformations
163 : */
164 : typedef struct WMAProChannelGrp {
165 : uint8_t num_channels; ///< number of channels in the group
166 : int8_t transform; ///< transform on / off
167 : int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
168 : float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
169 : float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
170 : } WMAProChannelGrp;
171 :
172 : /**
173 : * @brief main decoder context
174 : */
175 : typedef struct WMAProDecodeCtx {
176 : /* generic decoder variables */
177 : AVCodecContext* avctx; ///< codec context for av_log
178 : AVFloatDSPContext *fdsp;
179 : uint8_t frame_data[MAX_FRAMESIZE +
180 : AV_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
181 : PutBitContext pb; ///< context for filling the frame_data buffer
182 : FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
183 : DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
184 : const float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
185 :
186 : /* frame size dependent frame information (set during initialization) */
187 : uint32_t decode_flags; ///< used compression features
188 : uint8_t len_prefix; ///< frame is prefixed with its length
189 : uint8_t dynamic_range_compression; ///< frame contains DRC data
190 : uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
191 : uint16_t samples_per_frame; ///< number of samples to output
192 : uint16_t log2_frame_size;
193 : int8_t lfe_channel; ///< lfe channel index
194 : uint8_t max_num_subframes;
195 : uint8_t subframe_len_bits; ///< number of bits used for the subframe length
196 : uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
197 : uint16_t min_samples_per_subframe;
198 : int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
199 : int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
200 : int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
201 : int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
202 :
203 : /* packet decode state */
204 : GetBitContext pgb; ///< bitstream reader context for the packet
205 : int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
206 : uint8_t packet_offset; ///< frame offset in the packet
207 : uint8_t packet_sequence_number; ///< current packet number
208 : int num_saved_bits; ///< saved number of bits
209 : int frame_offset; ///< frame offset in the bit reservoir
210 : int subframe_offset; ///< subframe offset in the bit reservoir
211 : uint8_t packet_loss; ///< set in case of bitstream error
212 : uint8_t packet_done; ///< set when a packet is fully decoded
213 :
214 : /* frame decode state */
215 : uint32_t frame_num; ///< current frame number (not used for decoding)
216 : GetBitContext gb; ///< bitstream reader context
217 : int buf_bit_size; ///< buffer size in bits
218 : uint8_t drc_gain; ///< gain for the DRC tool
219 : int8_t skip_frame; ///< skip output step
220 : int8_t parsed_all_subframes; ///< all subframes decoded?
221 : uint8_t skip_packets; ///< packets to skip to find next packet in a stream (XMA1/2)
222 :
223 : /* subframe/block decode state */
224 : int16_t subframe_len; ///< current subframe length
225 : int8_t nb_channels; ///< number of channels in stream (XMA1/2)
226 : int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
227 : int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
228 : int8_t num_bands; ///< number of scale factor bands
229 : int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
230 : int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
231 : uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
232 : int8_t esc_len; ///< length of escaped coefficients
233 :
234 : uint8_t num_chgroups; ///< number of channel groups
235 : WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
236 :
237 : WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
238 : } WMAProDecodeCtx;
239 :
240 : typedef struct XMADecodeCtx {
241 : WMAProDecodeCtx xma[XMA_MAX_STREAMS];
242 : AVFrame *frames[XMA_MAX_STREAMS];
243 : int current_stream;
244 : int num_streams;
245 : float samples[XMA_MAX_CHANNELS][512 * 64];
246 : int offset[XMA_MAX_STREAMS];
247 : int start_channel[XMA_MAX_STREAMS];
248 : } XMADecodeCtx;
249 :
250 : /**
251 : *@brief helper function to print the most important members of the context
252 : *@param s context
253 : */
254 0 : static av_cold void dump_context(WMAProDecodeCtx *s)
255 : {
256 : #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
257 : #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b);
258 :
259 0 : PRINT("ed sample bit depth", s->bits_per_sample);
260 0 : PRINT_HEX("ed decode flags", s->decode_flags);
261 0 : PRINT("samples per frame", s->samples_per_frame);
262 0 : PRINT("log2 frame size", s->log2_frame_size);
263 0 : PRINT("max num subframes", s->max_num_subframes);
264 0 : PRINT("len prefix", s->len_prefix);
265 0 : PRINT("num channels", s->nb_channels);
266 0 : }
267 :
268 : /**
269 : *@brief Uninitialize the decoder and free all resources.
270 : *@param avctx codec context
271 : *@return 0 on success, < 0 otherwise
272 : */
273 8 : static av_cold int decode_end(WMAProDecodeCtx *s)
274 : {
275 : int i;
276 :
277 8 : av_freep(&s->fdsp);
278 :
279 72 : for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
280 64 : ff_mdct_end(&s->mdct_ctx[i]);
281 :
282 8 : return 0;
283 : }
284 :
285 8 : static av_cold int wmapro_decode_end(AVCodecContext *avctx)
286 : {
287 8 : WMAProDecodeCtx *s = avctx->priv_data;
288 :
289 8 : decode_end(s);
290 :
291 8 : return 0;
292 : }
293 :
294 40 : static av_cold int get_rate(AVCodecContext *avctx)
295 : {
296 40 : if (avctx->codec_id != AV_CODEC_ID_WMAPRO) { // XXX: is this really only for XMA?
297 0 : if (avctx->sample_rate > 44100)
298 0 : return 48000;
299 0 : else if (avctx->sample_rate > 32000)
300 0 : return 44100;
301 0 : else if (avctx->sample_rate > 24000)
302 0 : return 32000;
303 0 : return 24000;
304 : }
305 :
306 40 : return avctx->sample_rate;
307 : }
308 :
309 : /**
310 : *@brief Initialize the decoder.
311 : *@param avctx codec context
312 : *@return 0 on success, -1 otherwise
313 : */
314 8 : static av_cold int decode_init(WMAProDecodeCtx *s, AVCodecContext *avctx, int num_stream)
315 : {
316 8 : uint8_t *edata_ptr = avctx->extradata;
317 : unsigned int channel_mask;
318 : int i, bits;
319 : int log2_max_num_subframes;
320 : int num_possible_block_sizes;
321 :
322 8 : if (avctx->codec_id == AV_CODEC_ID_XMA1 || avctx->codec_id == AV_CODEC_ID_XMA2)
323 0 : avctx->block_align = 2048;
324 :
325 8 : if (!avctx->block_align) {
326 0 : av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
327 0 : return AVERROR(EINVAL);
328 : }
329 :
330 8 : s->avctx = avctx;
331 :
332 8 : init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
333 :
334 8 : avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
335 :
336 : /** dump the extradata */
337 8 : av_log(avctx, AV_LOG_DEBUG, "extradata:\n");
338 152 : for (i = 0; i < avctx->extradata_size; i++)
339 144 : av_log(avctx, AV_LOG_DEBUG, "[%x] ", avctx->extradata[i]);
340 8 : av_log(avctx, AV_LOG_DEBUG, "\n");
341 :
342 8 : if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size == 34) { /* XMA2WAVEFORMATEX */
343 0 : s->decode_flags = 0x10d6;
344 0 : s->bits_per_sample = 16;
345 0 : channel_mask = 0; //AV_RL32(edata_ptr+2); /* not always in expected order */
346 0 : if ((num_stream+1) * XMA_MAX_CHANNELS_STREAM > avctx->channels) /* stream config is 2ch + 2ch + ... + 1/2ch */
347 0 : s->nb_channels = 1;
348 : else
349 0 : s->nb_channels = 2;
350 8 : } else if (avctx->codec_id == AV_CODEC_ID_XMA2) { /* XMA2WAVEFORMAT */
351 0 : s->decode_flags = 0x10d6;
352 0 : s->bits_per_sample = 16;
353 0 : channel_mask = 0; /* would need to aggregate from all streams */
354 0 : s->nb_channels = edata_ptr[32 + ((edata_ptr[0]==3)?0:8) + 4*num_stream + 0]; /* nth stream config */
355 8 : } else if (avctx->codec_id == AV_CODEC_ID_XMA1) { /* XMAWAVEFORMAT */
356 0 : s->decode_flags = 0x10d6;
357 0 : s->bits_per_sample = 16;
358 0 : channel_mask = 0; /* would need to aggregate from all streams */
359 0 : s->nb_channels = edata_ptr[8 + 20*num_stream + 17]; /* nth stream config */
360 8 : } else if (avctx->codec_id == AV_CODEC_ID_WMAPRO && avctx->extradata_size >= 18) {
361 8 : s->decode_flags = AV_RL16(edata_ptr+14);
362 8 : channel_mask = AV_RL32(edata_ptr+2);
363 8 : s->bits_per_sample = AV_RL16(edata_ptr);
364 8 : s->nb_channels = avctx->channels;
365 :
366 16 : if (s->bits_per_sample > 32 || s->bits_per_sample < 1) {
367 0 : avpriv_request_sample(avctx, "bits per sample is %d", s->bits_per_sample);
368 0 : return AVERROR_PATCHWELCOME;
369 : }
370 : } else {
371 0 : avpriv_request_sample(avctx, "Unknown extradata size");
372 0 : return AVERROR_PATCHWELCOME;
373 : }
374 :
375 : /** generic init */
376 8 : s->log2_frame_size = av_log2(avctx->block_align) + 4;
377 8 : if (s->log2_frame_size > 25) {
378 0 : avpriv_request_sample(avctx, "Large block align");
379 0 : return AVERROR_PATCHWELCOME;
380 : }
381 :
382 : /** frame info */
383 8 : if (avctx->codec_id != AV_CODEC_ID_WMAPRO)
384 0 : s->skip_frame = 0;
385 : else
386 8 : s->skip_frame = 1; /* skip first frame */
387 :
388 8 : s->packet_loss = 1;
389 8 : s->len_prefix = (s->decode_flags & 0x40);
390 :
391 : /** get frame len */
392 8 : if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
393 8 : bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
394 8 : if (bits > WMAPRO_BLOCK_MAX_BITS) {
395 0 : avpriv_request_sample(avctx, "14-bit block sizes");
396 0 : return AVERROR_PATCHWELCOME;
397 : }
398 8 : s->samples_per_frame = 1 << bits;
399 : } else {
400 0 : s->samples_per_frame = 512;
401 : }
402 :
403 : /** subframe info */
404 8 : log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
405 8 : s->max_num_subframes = 1 << log2_max_num_subframes;
406 8 : if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
407 8 : s->max_subframe_len_bit = 1;
408 8 : s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
409 :
410 8 : num_possible_block_sizes = log2_max_num_subframes + 1;
411 8 : s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
412 8 : s->dynamic_range_compression = (s->decode_flags & 0x80);
413 :
414 8 : if (s->max_num_subframes > MAX_SUBFRAMES) {
415 0 : av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n",
416 0 : s->max_num_subframes);
417 0 : return AVERROR_INVALIDDATA;
418 : }
419 :
420 8 : if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) {
421 0 : av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
422 0 : s->min_samples_per_subframe);
423 0 : return AVERROR_INVALIDDATA;
424 : }
425 :
426 8 : if (s->avctx->sample_rate <= 0) {
427 0 : av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
428 0 : return AVERROR_INVALIDDATA;
429 : }
430 :
431 8 : if (s->nb_channels <= 0) {
432 0 : av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
433 0 : s->nb_channels);
434 0 : return AVERROR_INVALIDDATA;
435 8 : } else if (avctx->codec_id != AV_CODEC_ID_WMAPRO && s->nb_channels > XMA_MAX_CHANNELS_STREAM) {
436 0 : av_log(avctx, AV_LOG_ERROR, "invalid number of channels per XMA stream %d\n",
437 0 : s->nb_channels);
438 0 : return AVERROR_INVALIDDATA;
439 8 : } else if (s->nb_channels > WMAPRO_MAX_CHANNELS) {
440 0 : avpriv_request_sample(avctx,
441 : "More than %d channels", WMAPRO_MAX_CHANNELS);
442 0 : return AVERROR_PATCHWELCOME;
443 : }
444 :
445 : /** init previous block len */
446 32 : for (i = 0; i < s->nb_channels; i++)
447 24 : s->channel[i].prev_block_len = s->samples_per_frame;
448 :
449 : /** extract lfe channel position */
450 8 : s->lfe_channel = -1;
451 :
452 8 : if (channel_mask & 8) {
453 : unsigned int mask;
454 10 : for (mask = 1; mask < 16; mask <<= 1) {
455 8 : if (channel_mask & mask)
456 8 : ++s->lfe_channel;
457 : }
458 : }
459 :
460 8 : INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
461 : scale_huffbits, 1, 1,
462 : scale_huffcodes, 2, 2, 616);
463 :
464 8 : INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
465 : scale_rl_huffbits, 1, 1,
466 : scale_rl_huffcodes, 4, 4, 1406);
467 :
468 8 : INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
469 : coef0_huffbits, 1, 1,
470 : coef0_huffcodes, 4, 4, 2108);
471 :
472 8 : INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
473 : coef1_huffbits, 1, 1,
474 : coef1_huffcodes, 4, 4, 3912);
475 :
476 8 : INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
477 : vec4_huffbits, 1, 1,
478 : vec4_huffcodes, 2, 2, 604);
479 :
480 8 : INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
481 : vec2_huffbits, 1, 1,
482 : vec2_huffcodes, 2, 2, 562);
483 :
484 8 : INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
485 : vec1_huffbits, 1, 1,
486 : vec1_huffcodes, 2, 2, 562);
487 :
488 : /** calculate number of scale factor bands and their offsets
489 : for every possible block size */
490 48 : for (i = 0; i < num_possible_block_sizes; i++) {
491 40 : int subframe_len = s->samples_per_frame >> i;
492 : int x;
493 40 : int band = 1;
494 40 : int rate = get_rate(avctx);
495 :
496 40 : s->sfb_offsets[i][0] = 0;
497 :
498 1040 : for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
499 1040 : int offset = (subframe_len * 2 * critical_freq[x]) / rate + 2;
500 1040 : offset &= ~3;
501 1040 : if (offset > s->sfb_offsets[i][band - 1])
502 860 : s->sfb_offsets[i][band++] = offset;
503 :
504 1040 : if (offset >= subframe_len)
505 40 : break;
506 : }
507 40 : s->sfb_offsets[i][band - 1] = subframe_len;
508 40 : s->num_sfb[i] = band - 1;
509 40 : if (s->num_sfb[i] <= 0) {
510 0 : av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n");
511 0 : return AVERROR_INVALIDDATA;
512 : }
513 : }
514 :
515 :
516 : /** Scale factors can be shared between blocks of different size
517 : as every block has a different scale factor band layout.
518 : The matrix sf_offsets is needed to find the correct scale factor.
519 : */
520 :
521 48 : for (i = 0; i < num_possible_block_sizes; i++) {
522 : int b;
523 900 : for (b = 0; b < s->num_sfb[i]; b++) {
524 : int x;
525 1720 : int offset = ((s->sfb_offsets[i][b]
526 860 : + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
527 5160 : for (x = 0; x < num_possible_block_sizes; x++) {
528 4300 : int v = 0;
529 52476 : while (s->sfb_offsets[x][v + 1] << x < offset) {
530 43876 : v++;
531 43876 : av_assert0(v < MAX_BANDS);
532 : }
533 4300 : s->sf_offsets[i][x][b] = v;
534 : }
535 : }
536 : }
537 :
538 8 : s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
539 8 : if (!s->fdsp)
540 0 : return AVERROR(ENOMEM);
541 :
542 : /** init MDCT, FIXME: only init needed sizes */
543 72 : for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
544 64 : ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
545 64 : 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
546 64 : / (1 << (s->bits_per_sample - 1)));
547 :
548 : /** init MDCT windows: simple sine window */
549 72 : for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
550 64 : const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
551 64 : ff_init_ff_sine_windows(win_idx);
552 64 : s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
553 : }
554 :
555 : /** calculate subwoofer cutoff values */
556 48 : for (i = 0; i < num_possible_block_sizes; i++) {
557 40 : int block_size = s->samples_per_frame >> i;
558 80 : int cutoff = (440*block_size + 3LL * (s->avctx->sample_rate >> 1) - 1)
559 40 : / s->avctx->sample_rate;
560 40 : s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
561 : }
562 :
563 : /** calculate sine values for the decorrelation matrix */
564 272 : for (i = 0; i < 33; i++)
565 264 : sin64[i] = sin(i*M_PI / 64.0);
566 :
567 8 : if (avctx->debug & FF_DEBUG_BITSTREAM)
568 0 : dump_context(s);
569 :
570 8 : avctx->channel_layout = channel_mask;
571 :
572 8 : return 0;
573 : }
574 :
575 : /**
576 : *@brief Initialize the decoder.
577 : *@param avctx codec context
578 : *@return 0 on success, -1 otherwise
579 : */
580 8 : static av_cold int wmapro_decode_init(AVCodecContext *avctx)
581 : {
582 8 : WMAProDecodeCtx *s = avctx->priv_data;
583 :
584 8 : return decode_init(s, avctx, 0);
585 : }
586 :
587 : /**
588 : *@brief Decode the subframe length.
589 : *@param s context
590 : *@param offset sample offset in the frame
591 : *@return decoded subframe length on success, < 0 in case of an error
592 : */
593 527 : static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
594 : {
595 527 : int frame_len_shift = 0;
596 : int subframe_len;
597 :
598 : /** no need to read from the bitstream when only one length is possible */
599 527 : if (offset == s->samples_per_frame - s->min_samples_per_subframe)
600 9 : return s->min_samples_per_subframe;
601 :
602 518 : if (get_bits_left(&s->gb) < 1)
603 0 : return AVERROR_INVALIDDATA;
604 :
605 : /** 1 bit indicates if the subframe is of maximum length */
606 518 : if (s->max_subframe_len_bit) {
607 518 : if (get_bits1(&s->gb))
608 279 : frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
609 : } else
610 0 : frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
611 :
612 518 : subframe_len = s->samples_per_frame >> frame_len_shift;
613 :
614 : /** sanity check the length */
615 1036 : if (subframe_len < s->min_samples_per_subframe ||
616 518 : subframe_len > s->samples_per_frame) {
617 0 : av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
618 : subframe_len);
619 0 : return AVERROR_INVALIDDATA;
620 : }
621 518 : return subframe_len;
622 : }
623 :
624 : /**
625 : *@brief Decode how the data in the frame is split into subframes.
626 : * Every WMA frame contains the encoded data for a fixed number of
627 : * samples per channel. The data for every channel might be split
628 : * into several subframes. This function will reconstruct the list of
629 : * subframes for every channel.
630 : *
631 : * If the subframes are not evenly split, the algorithm estimates the
632 : * channels with the lowest number of total samples.
633 : * Afterwards, for each of these channels a bit is read from the
634 : * bitstream that indicates if the channel contains a subframe with the
635 : * next subframe size that is going to be read from the bitstream or not.
636 : * If a channel contains such a subframe, the subframe size gets added to
637 : * the channel's subframe list.
638 : * The algorithm repeats these steps until the frame is properly divided
639 : * between the individual channels.
640 : *
641 : *@param s context
642 : *@return 0 on success, < 0 in case of an error
643 : */
644 257 : static int decode_tilehdr(WMAProDecodeCtx *s)
645 : {
646 257 : uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
647 : uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
648 257 : int channels_for_cur_subframe = s->nb_channels; /**< number of channels that contain the current subframe */
649 257 : int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
650 257 : int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
651 : int c;
652 :
653 : /* Should never consume more than 3073 bits (256 iterations for the
654 : * while loop when always the minimum amount of 128 samples is subtracted
655 : * from missing samples in the 8 channel case).
656 : * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
657 : */
658 :
659 : /** reset tiling information */
660 1183 : for (c = 0; c < s->nb_channels; c++)
661 926 : s->channel[c].num_subframes = 0;
662 :
663 257 : if (s->max_num_subframes == 1 || get_bits1(&s->gb))
664 225 : fixed_channel_layout = 1;
665 :
666 : /** loop until the frame data is split between the subframes */
667 : do {
668 : int subframe_len;
669 :
670 : /** check which channels contain the subframe */
671 2813 : for (c = 0; c < s->nb_channels; c++) {
672 2286 : if (num_samples[c] == min_channel_len) {
673 1501 : if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
674 224 : (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
675 1053 : contains_subframe[c] = 1;
676 : else
677 224 : contains_subframe[c] = get_bits1(&s->gb);
678 : } else
679 1009 : contains_subframe[c] = 0;
680 : }
681 :
682 : /** get subframe length, subframe_len == 0 is not allowed */
683 527 : if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
684 0 : return AVERROR_INVALIDDATA;
685 :
686 : /** add subframes to the individual channels and find new min_channel_len */
687 527 : min_channel_len += subframe_len;
688 2813 : for (c = 0; c < s->nb_channels; c++) {
689 2286 : WMAProChannelCtx* chan = &s->channel[c];
690 :
691 2286 : if (contains_subframe[c]) {
692 1234 : if (chan->num_subframes >= MAX_SUBFRAMES) {
693 0 : av_log(s->avctx, AV_LOG_ERROR,
694 : "broken frame: num subframes > 31\n");
695 0 : return AVERROR_INVALIDDATA;
696 : }
697 1234 : chan->subframe_len[chan->num_subframes] = subframe_len;
698 1234 : num_samples[c] += subframe_len;
699 1234 : ++chan->num_subframes;
700 1234 : if (num_samples[c] > s->samples_per_frame) {
701 0 : av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
702 : "channel len > samples_per_frame\n");
703 0 : return AVERROR_INVALIDDATA;
704 : }
705 1052 : } else if (num_samples[c] <= min_channel_len) {
706 211 : if (num_samples[c] < min_channel_len) {
707 43 : channels_for_cur_subframe = 0;
708 43 : min_channel_len = num_samples[c];
709 : }
710 211 : ++channels_for_cur_subframe;
711 : }
712 : }
713 527 : } while (min_channel_len < s->samples_per_frame);
714 :
715 1183 : for (c = 0; c < s->nb_channels; c++) {
716 : int i;
717 926 : int offset = 0;
718 2160 : for (i = 0; i < s->channel[c].num_subframes; i++) {
719 : ff_dlog(s->avctx, "frame[%"PRIu32"] channel[%i] subframe[%i]"
720 : " len %i\n", s->frame_num, c, i,
721 : s->channel[c].subframe_len[i]);
722 1234 : s->channel[c].subframe_offset[i] = offset;
723 1234 : offset += s->channel[c].subframe_len[i];
724 : }
725 : }
726 :
727 257 : return 0;
728 : }
729 :
730 : /**
731 : *@brief Calculate a decorrelation matrix from the bitstream parameters.
732 : *@param s codec context
733 : *@param chgroup channel group for which the matrix needs to be calculated
734 : */
735 79 : static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
736 : WMAProChannelGrp *chgroup)
737 : {
738 : int i;
739 79 : int offset = 0;
740 : int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
741 158 : memset(chgroup->decorrelation_matrix, 0, s->nb_channels *
742 79 : s->nb_channels * sizeof(*chgroup->decorrelation_matrix));
743 :
744 316 : for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
745 237 : rotation_offset[i] = get_bits(&s->gb, 6);
746 :
747 316 : for (i = 0; i < chgroup->num_channels; i++)
748 237 : chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
749 237 : get_bits1(&s->gb) ? 1.0 : -1.0;
750 :
751 237 : for (i = 1; i < chgroup->num_channels; i++) {
752 : int x;
753 395 : for (x = 0; x < i; x++) {
754 : int y;
755 869 : for (y = 0; y < i + 1; y++) {
756 632 : float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
757 632 : float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
758 632 : int n = rotation_offset[offset + x];
759 : float sinv;
760 : float cosv;
761 :
762 632 : if (n < 32) {
763 442 : sinv = sin64[n];
764 442 : cosv = sin64[32 - n];
765 : } else {
766 190 : sinv = sin64[64 - n];
767 190 : cosv = -sin64[n - 32];
768 : }
769 :
770 1264 : chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
771 1264 : (v1 * sinv) - (v2 * cosv);
772 1264 : chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
773 1264 : (v1 * cosv) + (v2 * sinv);
774 : }
775 : }
776 158 : offset += i;
777 : }
778 79 : }
779 :
780 : /**
781 : *@brief Decode channel transformation parameters
782 : *@param s codec context
783 : *@return >= 0 in case of success, < 0 in case of bitstream errors
784 : */
785 527 : static int decode_channel_transform(WMAProDecodeCtx* s)
786 : {
787 : int i;
788 : /* should never consume more than 1921 bits for the 8 channel case
789 : * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
790 : * + MAX_CHANNELS + MAX_BANDS + 1)
791 : */
792 :
793 : /** in the one channel case channel transforms are pointless */
794 527 : s->num_chgroups = 0;
795 527 : if (s->nb_channels > 1) {
796 527 : int remaining_channels = s->channels_for_cur_subframe;
797 :
798 527 : if (get_bits1(&s->gb)) {
799 0 : avpriv_request_sample(s->avctx,
800 : "Channel transform bit");
801 0 : return AVERROR_PATCHWELCOME;
802 : }
803 :
804 2312 : for (s->num_chgroups = 0; remaining_channels &&
805 1258 : s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
806 629 : WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
807 629 : float** channel_data = chgroup->channel_data;
808 629 : chgroup->num_channels = 0;
809 629 : chgroup->transform = 0;
810 :
811 : /** decode channel mask */
812 629 : if (remaining_channels > 2) {
813 1217 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
814 1035 : int channel_idx = s->channel_indexes_for_cur_subframe[i];
815 1035 : if (!s->channel[channel_idx].grouped
816 819 : && get_bits1(&s->gb)) {
817 536 : ++chgroup->num_channels;
818 536 : s->channel[channel_idx].grouped = 1;
819 536 : *channel_data++ = s->channel[channel_idx].coeffs;
820 : }
821 : }
822 : } else {
823 447 : chgroup->num_channels = remaining_channels;
824 1222 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
825 775 : int channel_idx = s->channel_indexes_for_cur_subframe[i];
826 775 : if (!s->channel[channel_idx].grouped)
827 698 : *channel_data++ = s->channel[channel_idx].coeffs;
828 775 : s->channel[channel_idx].grouped = 1;
829 : }
830 : }
831 :
832 : /** decode transform type */
833 629 : if (chgroup->num_channels == 2) {
834 276 : if (get_bits1(&s->gb)) {
835 38 : if (get_bits1(&s->gb)) {
836 0 : avpriv_request_sample(s->avctx,
837 : "Unknown channel transform type");
838 0 : return AVERROR_PATCHWELCOME;
839 : }
840 : } else {
841 238 : chgroup->transform = 1;
842 238 : if (s->nb_channels == 2) {
843 215 : chgroup->decorrelation_matrix[0] = 1.0;
844 215 : chgroup->decorrelation_matrix[1] = -1.0;
845 215 : chgroup->decorrelation_matrix[2] = 1.0;
846 215 : chgroup->decorrelation_matrix[3] = 1.0;
847 : } else {
848 : /** cos(pi/4) */
849 23 : chgroup->decorrelation_matrix[0] = 0.70703125;
850 23 : chgroup->decorrelation_matrix[1] = -0.70703125;
851 23 : chgroup->decorrelation_matrix[2] = 0.70703125;
852 23 : chgroup->decorrelation_matrix[3] = 0.70703125;
853 : }
854 : }
855 353 : } else if (chgroup->num_channels > 2) {
856 157 : if (get_bits1(&s->gb)) {
857 79 : chgroup->transform = 1;
858 79 : if (get_bits1(&s->gb)) {
859 79 : decode_decorrelation_matrix(s, chgroup);
860 : } else {
861 : /** FIXME: more than 6 coupled channels not supported */
862 0 : if (chgroup->num_channels > 6) {
863 0 : avpriv_request_sample(s->avctx,
864 : "Coupled channels > 6");
865 : } else {
866 0 : memcpy(chgroup->decorrelation_matrix,
867 0 : default_decorrelation[chgroup->num_channels],
868 0 : chgroup->num_channels * chgroup->num_channels *
869 : sizeof(*chgroup->decorrelation_matrix));
870 : }
871 : }
872 : }
873 : }
874 :
875 : /** decode transform on / off */
876 629 : if (chgroup->transform) {
877 317 : if (!get_bits1(&s->gb)) {
878 : int i;
879 : /** transform can be enabled for individual bands */
880 0 : for (i = 0; i < s->num_bands; i++) {
881 0 : chgroup->transform_band[i] = get_bits1(&s->gb);
882 : }
883 : } else {
884 317 : memset(chgroup->transform_band, 1, s->num_bands);
885 : }
886 : }
887 629 : remaining_channels -= chgroup->num_channels;
888 : }
889 : }
890 527 : return 0;
891 : }
892 :
893 : /**
894 : *@brief Extract the coefficients from the bitstream.
895 : *@param s codec context
896 : *@param c current channel number
897 : *@return 0 on success, < 0 in case of bitstream errors
898 : */
899 1038 : static int decode_coeffs(WMAProDecodeCtx *s, int c)
900 : {
901 : /* Integers 0..15 as single-precision floats. The table saves a
902 : costly int to float conversion, and storing the values as
903 : integers allows fast sign-flipping. */
904 : static const uint32_t fval_tab[16] = {
905 : 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
906 : 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
907 : 0x41000000, 0x41100000, 0x41200000, 0x41300000,
908 : 0x41400000, 0x41500000, 0x41600000, 0x41700000,
909 : };
910 : int vlctable;
911 : VLC* vlc;
912 1038 : WMAProChannelCtx* ci = &s->channel[c];
913 1038 : int rl_mode = 0;
914 1038 : int cur_coeff = 0;
915 1038 : int num_zeros = 0;
916 : const uint16_t* run;
917 : const float* level;
918 :
919 : ff_dlog(s->avctx, "decode coefficients for channel %i\n", c);
920 :
921 1038 : vlctable = get_bits1(&s->gb);
922 1038 : vlc = &coef_vlc[vlctable];
923 :
924 1038 : if (vlctable) {
925 321 : run = coef1_run;
926 321 : level = coef1_level;
927 : } else {
928 717 : run = coef0_run;
929 717 : level = coef0_level;
930 : }
931 :
932 : /** decode vector coefficients (consumes up to 167 bits per iteration for
933 : 4 vector coded large values) */
934 78502 : while ((s->transmit_num_vec_coeffs || !rl_mode) &&
935 38213 : (cur_coeff + 3 < ci->num_vec_coeffs)) {
936 : uint32_t vals[4];
937 : int i;
938 : unsigned int idx;
939 :
940 38213 : idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
941 :
942 38213 : if (idx == HUFF_VEC4_SIZE - 1) {
943 65430 : for (i = 0; i < 4; i += 2) {
944 43620 : idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
945 43620 : if (idx == HUFF_VEC2_SIZE - 1) {
946 : uint32_t v0, v1;
947 20544 : v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
948 20544 : if (v0 == HUFF_VEC1_SIZE - 1)
949 1608 : v0 += ff_wma_get_large_val(&s->gb);
950 20544 : v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
951 20544 : if (v1 == HUFF_VEC1_SIZE - 1)
952 1665 : v1 += ff_wma_get_large_val(&s->gb);
953 20544 : vals[i ] = av_float2int(v0);
954 20544 : vals[i+1] = av_float2int(v1);
955 : } else {
956 23076 : vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
957 23076 : vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
958 : }
959 : }
960 : } else {
961 16403 : vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
962 16403 : vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
963 16403 : vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
964 16403 : vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
965 : }
966 :
967 : /** decode sign */
968 191065 : for (i = 0; i < 4; i++) {
969 152852 : if (vals[i]) {
970 111800 : uint32_t sign = get_bits1(&s->gb) - 1;
971 111800 : AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
972 111800 : num_zeros = 0;
973 : } else {
974 41052 : ci->coeffs[cur_coeff] = 0;
975 : /** switch to run level mode when subframe_len / 128 zeros
976 : were found in a row */
977 41052 : rl_mode |= (++num_zeros > s->subframe_len >> 8);
978 : }
979 152852 : ++cur_coeff;
980 : }
981 : }
982 :
983 : /** decode run level coded coefficients */
984 1038 : if (cur_coeff < s->subframe_len) {
985 1038 : memset(&ci->coeffs[cur_coeff], 0,
986 1038 : sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
987 4152 : if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
988 1038 : level, run, 1, ci->coeffs,
989 1038 : cur_coeff, s->subframe_len,
990 2076 : s->subframe_len, s->esc_len, 0))
991 0 : return AVERROR_INVALIDDATA;
992 : }
993 :
994 1038 : return 0;
995 : }
996 :
997 : /**
998 : *@brief Extract scale factors from the bitstream.
999 : *@param s codec context
1000 : *@return 0 on success, < 0 in case of bitstream errors
1001 : */
1002 521 : static int decode_scale_factors(WMAProDecodeCtx* s)
1003 : {
1004 : int i;
1005 :
1006 : /** should never consume more than 5344 bits
1007 : * MAX_CHANNELS * (1 + MAX_BANDS * 23)
1008 : */
1009 :
1010 1745 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
1011 1224 : int c = s->channel_indexes_for_cur_subframe[i];
1012 : int* sf;
1013 : int* sf_end;
1014 1224 : s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
1015 1224 : sf_end = s->channel[c].scale_factors + s->num_bands;
1016 :
1017 : /** resample scale factors for the new block size
1018 : * as the scale factors might need to be resampled several times
1019 : * before some new values are transmitted, a backup of the last
1020 : * transmitted scale factors is kept in saved_scale_factors
1021 : */
1022 1224 : if (s->channel[c].reuse_sf) {
1023 298 : const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
1024 : int b;
1025 5471 : for (b = 0; b < s->num_bands; b++)
1026 10346 : s->channel[c].scale_factors[b] =
1027 10346 : s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
1028 : }
1029 :
1030 1224 : if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
1031 :
1032 1206 : if (!s->channel[c].reuse_sf) {
1033 : int val;
1034 : /** decode DPCM coded scale factors */
1035 926 : s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
1036 926 : val = 45 / s->channel[c].scale_factor_step;
1037 24757 : for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
1038 23831 : val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
1039 23831 : *sf = val;
1040 : }
1041 : } else {
1042 : int i;
1043 : /** run level decode differences to the resampled factors */
1044 1904 : for (i = 0; i < s->num_bands; i++) {
1045 : int idx;
1046 : int skip;
1047 : int val;
1048 : int sign;
1049 :
1050 1904 : idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
1051 :
1052 1904 : if (!idx) {
1053 41 : uint32_t code = get_bits(&s->gb, 14);
1054 41 : val = code >> 6;
1055 41 : sign = (code & 1) - 1;
1056 41 : skip = (code & 0x3f) >> 1;
1057 1863 : } else if (idx == 1) {
1058 280 : break;
1059 : } else {
1060 1583 : skip = scale_rl_run[idx];
1061 1583 : val = scale_rl_level[idx];
1062 1583 : sign = get_bits1(&s->gb)-1;
1063 : }
1064 :
1065 1624 : i += skip;
1066 1624 : if (i >= s->num_bands) {
1067 0 : av_log(s->avctx, AV_LOG_ERROR,
1068 : "invalid scale factor coding\n");
1069 0 : return AVERROR_INVALIDDATA;
1070 : }
1071 1624 : s->channel[c].scale_factors[i] += (val ^ sign) - sign;
1072 : }
1073 : }
1074 : /** swap buffers */
1075 1206 : s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
1076 1206 : s->channel[c].table_idx = s->table_idx;
1077 1206 : s->channel[c].reuse_sf = 1;
1078 : }
1079 :
1080 : /** calculate new scale factor maximum */
1081 1224 : s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
1082 29004 : for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
1083 27780 : s->channel[c].max_scale_factor =
1084 27780 : FFMAX(s->channel[c].max_scale_factor, *sf);
1085 : }
1086 :
1087 : }
1088 521 : return 0;
1089 : }
1090 :
1091 : /**
1092 : *@brief Reconstruct the individual channel data.
1093 : *@param s codec context
1094 : */
1095 521 : static void inverse_channel_transform(WMAProDecodeCtx *s)
1096 : {
1097 : int i;
1098 :
1099 1144 : for (i = 0; i < s->num_chgroups; i++) {
1100 623 : if (s->chgroup[i].transform) {
1101 : float data[WMAPRO_MAX_CHANNELS];
1102 317 : const int num_channels = s->chgroup[i].num_channels;
1103 317 : float** ch_data = s->chgroup[i].channel_data;
1104 317 : float** ch_end = ch_data + num_channels;
1105 317 : const int8_t* tb = s->chgroup[i].transform_band;
1106 : int16_t* sfb;
1107 :
1108 : /** multichannel decorrelation */
1109 8330 : for (sfb = s->cur_sfb_offsets;
1110 7696 : sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1111 : int y;
1112 7696 : if (*tb++ == 1) {
1113 : /** multiply values with the decorrelation_matrix */
1114 530064 : for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1115 522368 : const float* mat = s->chgroup[i].decorrelation_matrix;
1116 522368 : const float* data_end = data + num_channels;
1117 522368 : float* data_ptr = data;
1118 : float** ch;
1119 :
1120 1728896 : for (ch = ch_data; ch < ch_end; ch++)
1121 1206528 : *data_ptr++ = (*ch)[y];
1122 :
1123 1728896 : for (ch = ch_data; ch < ch_end; ch++) {
1124 1206528 : float sum = 0;
1125 1206528 : data_ptr = data;
1126 5311488 : while (data_ptr < data_end)
1127 2898432 : sum += *data_ptr++ * *mat++;
1128 :
1129 1206528 : (*ch)[y] = sum;
1130 : }
1131 : }
1132 0 : } else if (s->nb_channels == 2) {
1133 0 : int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1134 0 : s->fdsp->vector_fmul_scalar(ch_data[0] + sfb[0],
1135 0 : ch_data[0] + sfb[0],
1136 : 181.0 / 128, len);
1137 0 : s->fdsp->vector_fmul_scalar(ch_data[1] + sfb[0],
1138 0 : ch_data[1] + sfb[0],
1139 : 181.0 / 128, len);
1140 : }
1141 : }
1142 : }
1143 : }
1144 521 : }
1145 :
1146 : /**
1147 : *@brief Apply sine window and reconstruct the output buffer.
1148 : *@param s codec context
1149 : */
1150 527 : static void wmapro_window(WMAProDecodeCtx *s)
1151 : {
1152 : int i;
1153 1761 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
1154 1234 : int c = s->channel_indexes_for_cur_subframe[i];
1155 : const float* window;
1156 1234 : int winlen = s->channel[c].prev_block_len;
1157 1234 : float* start = s->channel[c].coeffs - (winlen >> 1);
1158 :
1159 1234 : if (s->subframe_len < winlen) {
1160 142 : start += (winlen - s->subframe_len) >> 1;
1161 142 : winlen = s->subframe_len;
1162 : }
1163 :
1164 1234 : window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1165 :
1166 1234 : winlen >>= 1;
1167 :
1168 1234 : s->fdsp->vector_fmul_window(start, start, start + winlen,
1169 : window, winlen);
1170 :
1171 1234 : s->channel[c].prev_block_len = s->subframe_len;
1172 : }
1173 527 : }
1174 :
1175 : /**
1176 : *@brief Decode a single subframe (block).
1177 : *@param s codec context
1178 : *@return 0 on success, < 0 when decoding failed
1179 : */
1180 527 : static int decode_subframe(WMAProDecodeCtx *s)
1181 : {
1182 527 : int offset = s->samples_per_frame;
1183 527 : int subframe_len = s->samples_per_frame;
1184 : int i;
1185 527 : int total_samples = s->samples_per_frame * s->nb_channels;
1186 527 : int transmit_coeffs = 0;
1187 : int cur_subwoofer_cutoff;
1188 :
1189 527 : s->subframe_offset = get_bits_count(&s->gb);
1190 :
1191 : /** reset channel context and find the next block offset and size
1192 : == the next block of the channel with the smallest number of
1193 : decoded samples
1194 : */
1195 2813 : for (i = 0; i < s->nb_channels; i++) {
1196 2286 : s->channel[i].grouped = 0;
1197 2286 : if (offset > s->channel[i].decoded_samples) {
1198 539 : offset = s->channel[i].decoded_samples;
1199 539 : subframe_len =
1200 539 : s->channel[i].subframe_len[s->channel[i].cur_subframe];
1201 : }
1202 : }
1203 :
1204 : ff_dlog(s->avctx,
1205 : "processing subframe with offset %i len %i\n", offset, subframe_len);
1206 :
1207 : /** get a list of all channels that contain the estimated block */
1208 527 : s->channels_for_cur_subframe = 0;
1209 2813 : for (i = 0; i < s->nb_channels; i++) {
1210 2286 : const int cur_subframe = s->channel[i].cur_subframe;
1211 : /** subtract already processed samples */
1212 2286 : total_samples -= s->channel[i].decoded_samples;
1213 :
1214 : /** and count if there are multiple subframes that match our profile */
1215 3563 : if (offset == s->channel[i].decoded_samples &&
1216 1277 : subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1217 1234 : total_samples -= s->channel[i].subframe_len[cur_subframe];
1218 2468 : s->channel[i].decoded_samples +=
1219 1234 : s->channel[i].subframe_len[cur_subframe];
1220 1234 : s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1221 1234 : ++s->channels_for_cur_subframe;
1222 : }
1223 : }
1224 :
1225 : /** check if the frame will be complete after processing the
1226 : estimated block */
1227 527 : if (!total_samples)
1228 257 : s->parsed_all_subframes = 1;
1229 :
1230 :
1231 : ff_dlog(s->avctx, "subframe is part of %i channels\n",
1232 : s->channels_for_cur_subframe);
1233 :
1234 : /** calculate number of scale factor bands and their offsets */
1235 527 : s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1236 527 : s->num_bands = s->num_sfb[s->table_idx];
1237 527 : s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1238 527 : cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1239 :
1240 : /** configure the decoder for the current subframe */
1241 527 : offset += s->samples_per_frame >> 1;
1242 :
1243 1761 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
1244 1234 : int c = s->channel_indexes_for_cur_subframe[i];
1245 :
1246 1234 : s->channel[c].coeffs = &s->channel[c].out[offset];
1247 : }
1248 :
1249 527 : s->subframe_len = subframe_len;
1250 527 : s->esc_len = av_log2(s->subframe_len - 1) + 1;
1251 :
1252 : /** skip extended header if any */
1253 527 : if (get_bits1(&s->gb)) {
1254 : int num_fill_bits;
1255 158 : if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1256 158 : int len = get_bits(&s->gb, 4);
1257 158 : num_fill_bits = get_bitsz(&s->gb, len) + 1;
1258 : }
1259 :
1260 158 : if (num_fill_bits >= 0) {
1261 158 : if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1262 0 : av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1263 0 : return AVERROR_INVALIDDATA;
1264 : }
1265 :
1266 158 : skip_bits_long(&s->gb, num_fill_bits);
1267 : }
1268 : }
1269 :
1270 : /** no idea for what the following bit is used */
1271 527 : if (get_bits1(&s->gb)) {
1272 0 : avpriv_request_sample(s->avctx, "Reserved bit");
1273 0 : return AVERROR_PATCHWELCOME;
1274 : }
1275 :
1276 :
1277 527 : if (decode_channel_transform(s) < 0)
1278 0 : return AVERROR_INVALIDDATA;
1279 :
1280 :
1281 1761 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
1282 1234 : int c = s->channel_indexes_for_cur_subframe[i];
1283 1234 : if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1284 1038 : transmit_coeffs = 1;
1285 : }
1286 :
1287 527 : av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1288 527 : if (transmit_coeffs) {
1289 : int step;
1290 521 : int quant_step = 90 * s->bits_per_sample >> 4;
1291 :
1292 : /** decode number of vector coded coefficients */
1293 521 : if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1294 0 : int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1295 0 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
1296 0 : int c = s->channel_indexes_for_cur_subframe[i];
1297 0 : int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1298 0 : if (num_vec_coeffs > s->subframe_len) {
1299 0 : av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1300 0 : return AVERROR_INVALIDDATA;
1301 : }
1302 0 : av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out));
1303 0 : s->channel[c].num_vec_coeffs = num_vec_coeffs;
1304 : }
1305 : } else {
1306 1745 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
1307 1224 : int c = s->channel_indexes_for_cur_subframe[i];
1308 1224 : s->channel[c].num_vec_coeffs = s->subframe_len;
1309 : }
1310 : }
1311 : /** decode quantization step */
1312 521 : step = get_sbits(&s->gb, 6);
1313 521 : quant_step += step;
1314 521 : if (step == -32 || step == 31) {
1315 23 : const int sign = (step == 31) - 1;
1316 23 : int quant = 0;
1317 69 : while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1318 23 : (step = get_bits(&s->gb, 5)) == 31) {
1319 0 : quant += 31;
1320 : }
1321 23 : quant_step += ((quant + step) ^ sign) - sign;
1322 : }
1323 521 : if (quant_step < 0) {
1324 0 : av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1325 : }
1326 :
1327 : /** decode quantization step modifiers for every channel */
1328 :
1329 521 : if (s->channels_for_cur_subframe == 1) {
1330 194 : s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1331 : } else {
1332 327 : int modifier_len = get_bits(&s->gb, 3);
1333 1357 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
1334 1030 : int c = s->channel_indexes_for_cur_subframe[i];
1335 1030 : s->channel[c].quant_step = quant_step;
1336 1030 : if (get_bits1(&s->gb)) {
1337 511 : if (modifier_len) {
1338 458 : s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1339 : } else
1340 53 : ++s->channel[c].quant_step;
1341 : }
1342 : }
1343 : }
1344 :
1345 : /** decode scale factors */
1346 521 : if (decode_scale_factors(s) < 0)
1347 0 : return AVERROR_INVALIDDATA;
1348 : }
1349 :
1350 : ff_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1351 : get_bits_count(&s->gb) - s->subframe_offset);
1352 :
1353 : /** parse coefficients */
1354 1761 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
1355 1234 : int c = s->channel_indexes_for_cur_subframe[i];
1356 2272 : if (s->channel[c].transmit_coefs &&
1357 1038 : get_bits_count(&s->gb) < s->num_saved_bits) {
1358 1038 : decode_coeffs(s, c);
1359 : } else
1360 196 : memset(s->channel[c].coeffs, 0,
1361 : sizeof(*s->channel[c].coeffs) * subframe_len);
1362 : }
1363 :
1364 : ff_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1365 : get_bits_count(&s->gb) - s->subframe_offset);
1366 :
1367 527 : if (transmit_coeffs) {
1368 521 : FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1369 : /** reconstruct the per channel data */
1370 521 : inverse_channel_transform(s);
1371 1745 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
1372 1224 : int c = s->channel_indexes_for_cur_subframe[i];
1373 1224 : const int* sf = s->channel[c].scale_factors;
1374 : int b;
1375 :
1376 1224 : if (c == s->lfe_channel)
1377 103 : memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1378 103 : (subframe_len - cur_subwoofer_cutoff));
1379 :
1380 : /** inverse quantization and rescaling */
1381 30228 : for (b = 0; b < s->num_bands; b++) {
1382 29004 : const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1383 58008 : const int exp = s->channel[c].quant_step -
1384 58008 : (s->channel[c].max_scale_factor - *sf++) *
1385 29004 : s->channel[c].scale_factor_step;
1386 29004 : const float quant = ff_exp10(exp / 20.0);
1387 29004 : int start = s->cur_sfb_offsets[b];
1388 58008 : s->fdsp->vector_fmul_scalar(s->tmp + start,
1389 29004 : s->channel[c].coeffs + start,
1390 : quant, end - start);
1391 : }
1392 :
1393 : /** apply imdct (imdct_half == DCTIV with reverse) */
1394 1224 : mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1395 : }
1396 : }
1397 :
1398 : /** window and overlapp-add */
1399 527 : wmapro_window(s);
1400 :
1401 : /** handled one subframe */
1402 1761 : for (i = 0; i < s->channels_for_cur_subframe; i++) {
1403 1234 : int c = s->channel_indexes_for_cur_subframe[i];
1404 1234 : if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1405 0 : av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1406 0 : return AVERROR_INVALIDDATA;
1407 : }
1408 1234 : ++s->channel[c].cur_subframe;
1409 : }
1410 :
1411 527 : return 0;
1412 : }
1413 :
1414 : /**
1415 : *@brief Decode one WMA frame.
1416 : *@param s codec context
1417 : *@return 0 if the trailer bit indicates that this is the last frame,
1418 : * 1 if there are additional frames
1419 : */
1420 257 : static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1421 : {
1422 257 : GetBitContext* gb = &s->gb;
1423 257 : int more_frames = 0;
1424 257 : int len = 0;
1425 : int i;
1426 :
1427 : /** get frame length */
1428 257 : if (s->len_prefix)
1429 257 : len = get_bits(gb, s->log2_frame_size);
1430 :
1431 : ff_dlog(s->avctx, "decoding frame with length %x\n", len);
1432 :
1433 : /** decode tile information */
1434 257 : if (decode_tilehdr(s)) {
1435 0 : s->packet_loss = 1;
1436 0 : return 0;
1437 : }
1438 :
1439 : /** read postproc transform */
1440 257 : if (s->nb_channels > 1 && get_bits1(gb)) {
1441 0 : if (get_bits1(gb)) {
1442 0 : for (i = 0; i < s->nb_channels * s->nb_channels; i++)
1443 0 : skip_bits(gb, 4);
1444 : }
1445 : }
1446 :
1447 : /** read drc info */
1448 257 : if (s->dynamic_range_compression) {
1449 257 : s->drc_gain = get_bits(gb, 8);
1450 : ff_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1451 : }
1452 :
1453 : /** no idea what these are for, might be the number of samples
1454 : that need to be skipped at the beginning or end of a stream */
1455 257 : if (get_bits1(gb)) {
1456 : int av_unused skip;
1457 :
1458 : /** usually true for the first frame */
1459 4 : if (get_bits1(gb)) {
1460 3 : skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1461 : ff_dlog(s->avctx, "start skip: %i\n", skip);
1462 : }
1463 :
1464 : /** sometimes true for the last frame */
1465 4 : if (get_bits1(gb)) {
1466 1 : skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1467 : ff_dlog(s->avctx, "end skip: %i\n", skip);
1468 : }
1469 :
1470 : }
1471 :
1472 : ff_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1473 : get_bits_count(gb) - s->frame_offset);
1474 :
1475 : /** reset subframe states */
1476 257 : s->parsed_all_subframes = 0;
1477 1183 : for (i = 0; i < s->nb_channels; i++) {
1478 926 : s->channel[i].decoded_samples = 0;
1479 926 : s->channel[i].cur_subframe = 0;
1480 926 : s->channel[i].reuse_sf = 0;
1481 : }
1482 :
1483 : /** decode all subframes */
1484 1041 : while (!s->parsed_all_subframes) {
1485 527 : if (decode_subframe(s) < 0) {
1486 0 : s->packet_loss = 1;
1487 0 : return 0;
1488 : }
1489 : }
1490 :
1491 : /** copy samples to the output buffer */
1492 1183 : for (i = 0; i < s->nb_channels; i++)
1493 926 : memcpy(frame->extended_data[i], s->channel[i].out,
1494 926 : s->samples_per_frame * sizeof(*s->channel[i].out));
1495 :
1496 1183 : for (i = 0; i < s->nb_channels; i++) {
1497 : /** reuse second half of the IMDCT output for the next frame */
1498 1852 : memcpy(&s->channel[i].out[0],
1499 926 : &s->channel[i].out[s->samples_per_frame],
1500 926 : s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1501 : }
1502 :
1503 257 : if (s->skip_frame) {
1504 3 : s->skip_frame = 0;
1505 3 : *got_frame_ptr = 0;
1506 3 : av_frame_unref(frame);
1507 : } else {
1508 254 : *got_frame_ptr = 1;
1509 : }
1510 :
1511 257 : if (s->len_prefix) {
1512 257 : if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1513 : /** FIXME: not sure if this is always an error */
1514 0 : av_log(s->avctx, AV_LOG_ERROR,
1515 : "frame[%"PRIu32"] would have to skip %i bits\n",
1516 : s->frame_num,
1517 0 : len - (get_bits_count(gb) - s->frame_offset) - 1);
1518 0 : s->packet_loss = 1;
1519 0 : return 0;
1520 : }
1521 :
1522 : /** skip the rest of the frame data */
1523 257 : skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1524 : } else {
1525 0 : while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1526 : }
1527 : }
1528 :
1529 : /** decode trailer bit */
1530 257 : more_frames = get_bits1(gb);
1531 :
1532 257 : ++s->frame_num;
1533 257 : return more_frames;
1534 : }
1535 :
1536 : /**
1537 : *@brief Calculate remaining input buffer length.
1538 : *@param s codec context
1539 : *@param gb bitstream reader context
1540 : *@return remaining size in bits
1541 : */
1542 849 : static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1543 : {
1544 849 : return s->buf_bit_size - get_bits_count(gb);
1545 : }
1546 :
1547 : /**
1548 : *@brief Fill the bit reservoir with a (partial) frame.
1549 : *@param s codec context
1550 : *@param gb bitstream reader context
1551 : *@param len length of the partial frame
1552 : *@param append decides whether to reset the buffer or not
1553 : */
1554 281 : static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1555 : int append)
1556 : {
1557 : int buflen;
1558 :
1559 : /** when the frame data does not need to be concatenated, the input buffer
1560 : is reset and additional bits from the previous frame are copied
1561 : and skipped later so that a fast byte copy is possible */
1562 :
1563 281 : if (!append) {
1564 260 : s->frame_offset = get_bits_count(gb) & 7;
1565 260 : s->num_saved_bits = s->frame_offset;
1566 260 : init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1567 : }
1568 :
1569 281 : buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1570 :
1571 281 : if (len <= 0 || buflen > MAX_FRAMESIZE) {
1572 0 : avpriv_request_sample(s->avctx, "Too small input buffer");
1573 0 : s->packet_loss = 1;
1574 0 : return;
1575 : }
1576 :
1577 281 : av_assert0(len <= put_bits_left(&s->pb));
1578 :
1579 281 : s->num_saved_bits += len;
1580 281 : if (!append) {
1581 260 : avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1582 : s->num_saved_bits);
1583 : } else {
1584 21 : int align = 8 - (get_bits_count(gb) & 7);
1585 21 : align = FFMIN(align, len);
1586 21 : put_bits(&s->pb, align, get_bits(gb, align));
1587 21 : len -= align;
1588 21 : avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1589 : }
1590 281 : skip_bits_long(gb, len);
1591 :
1592 : {
1593 281 : PutBitContext tmp = s->pb;
1594 281 : flush_put_bits(&tmp);
1595 : }
1596 :
1597 281 : init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1598 281 : skip_bits(&s->gb, s->frame_offset);
1599 : }
1600 :
1601 283 : static int decode_packet(AVCodecContext *avctx, WMAProDecodeCtx *s,
1602 : void *data, int *got_frame_ptr, AVPacket *avpkt)
1603 : {
1604 283 : GetBitContext* gb = &s->pgb;
1605 283 : const uint8_t* buf = avpkt->data;
1606 283 : int buf_size = avpkt->size;
1607 : int num_bits_prev_frame;
1608 : int packet_sequence_number;
1609 :
1610 283 : *got_frame_ptr = 0;
1611 :
1612 283 : if (s->packet_done || s->packet_loss) {
1613 24 : s->packet_done = 0;
1614 :
1615 : /** sanity check for the buffer length */
1616 24 : if (avctx->codec_id == AV_CODEC_ID_WMAPRO && buf_size < avctx->block_align) {
1617 0 : av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1618 : buf_size, avctx->block_align);
1619 0 : return AVERROR_INVALIDDATA;
1620 : }
1621 :
1622 24 : if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
1623 24 : s->next_packet_start = buf_size - avctx->block_align;
1624 24 : buf_size = avctx->block_align;
1625 : } else {
1626 0 : s->next_packet_start = buf_size - FFMIN(buf_size, avctx->block_align);
1627 0 : buf_size = FFMIN(buf_size, avctx->block_align);
1628 : }
1629 24 : s->buf_bit_size = buf_size << 3;
1630 :
1631 : /** parse packet header */
1632 24 : init_get_bits(gb, buf, s->buf_bit_size);
1633 24 : if (avctx->codec_id != AV_CODEC_ID_XMA2) {
1634 24 : packet_sequence_number = get_bits(gb, 4);
1635 24 : skip_bits(gb, 2);
1636 : } else {
1637 0 : int num_frames = get_bits(gb, 6);
1638 : ff_dlog(avctx, "packet[%d]: number of frames %d\n", avctx->frame_number, num_frames);
1639 0 : packet_sequence_number = 0;
1640 : }
1641 :
1642 : /** get number of bits that need to be added to the previous frame */
1643 24 : num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1644 24 : if (avctx->codec_id != AV_CODEC_ID_WMAPRO) {
1645 0 : skip_bits(gb, 3);
1646 0 : s->skip_packets = get_bits(gb, 8);
1647 : ff_dlog(avctx, "packet[%d]: skip packets %d\n", avctx->frame_number, s->skip_packets);
1648 : }
1649 :
1650 : ff_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1651 : num_bits_prev_frame);
1652 :
1653 : /** check for packet loss */
1654 45 : if (avctx->codec_id == AV_CODEC_ID_WMAPRO && !s->packet_loss &&
1655 21 : ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1656 0 : s->packet_loss = 1;
1657 0 : av_log(avctx, AV_LOG_ERROR,
1658 : "Packet loss detected! seq %"PRIx8" vs %x\n",
1659 0 : s->packet_sequence_number, packet_sequence_number);
1660 : }
1661 24 : s->packet_sequence_number = packet_sequence_number;
1662 :
1663 24 : if (num_bits_prev_frame > 0) {
1664 21 : int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1665 21 : if (num_bits_prev_frame >= remaining_packet_bits) {
1666 0 : num_bits_prev_frame = remaining_packet_bits;
1667 0 : s->packet_done = 1;
1668 : }
1669 :
1670 : /** append the previous frame data to the remaining data from the
1671 : previous packet to create a full frame */
1672 21 : save_bits(s, gb, num_bits_prev_frame, 1);
1673 : ff_dlog(avctx, "accumulated %x bits of frame data\n",
1674 : s->num_saved_bits - s->frame_offset);
1675 :
1676 : /** decode the cross packet frame if it is valid */
1677 21 : if (!s->packet_loss)
1678 21 : decode_frame(s, data, got_frame_ptr);
1679 3 : } else if (s->num_saved_bits - s->frame_offset) {
1680 : ff_dlog(avctx, "ignoring %x previously saved bits\n",
1681 : s->num_saved_bits - s->frame_offset);
1682 : }
1683 :
1684 48 : if (s->packet_loss) {
1685 : /** reset number of saved bits so that the decoder
1686 : does not start to decode incomplete frames in the
1687 : s->len_prefix == 0 case */
1688 3 : s->num_saved_bits = 0;
1689 3 : s->packet_loss = 0;
1690 : }
1691 : } else {
1692 : int frame_size;
1693 259 : s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1694 259 : init_get_bits(gb, avpkt->data, s->buf_bit_size);
1695 259 : skip_bits(gb, s->packet_offset);
1696 518 : if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1697 518 : (frame_size = show_bits(gb, s->log2_frame_size)) &&
1698 259 : frame_size <= remaining_bits(s, gb)) {
1699 236 : save_bits(s, gb, frame_size, 0);
1700 472 : if (!s->packet_loss)
1701 236 : s->packet_done = !decode_frame(s, data, got_frame_ptr);
1702 23 : } else if (!s->len_prefix
1703 0 : && s->num_saved_bits > get_bits_count(&s->gb)) {
1704 : /** when the frames do not have a length prefix, we don't know
1705 : the compressed length of the individual frames
1706 : however, we know what part of a new packet belongs to the
1707 : previous frame
1708 : therefore we save the incoming packet first, then we append
1709 : the "previous frame" data from the next packet so that
1710 : we get a buffer that only contains full frames */
1711 0 : s->packet_done = !decode_frame(s, data, got_frame_ptr);
1712 : } else {
1713 23 : s->packet_done = 1;
1714 : }
1715 : }
1716 :
1717 283 : if (remaining_bits(s, gb) < 0) {
1718 0 : av_log(avctx, AV_LOG_ERROR, "Overread %d\n", -remaining_bits(s, gb));
1719 0 : s->packet_loss = 1;
1720 : }
1721 :
1722 307 : if (s->packet_done && !s->packet_loss &&
1723 24 : remaining_bits(s, gb) > 0) {
1724 : /** save the rest of the data so that it can be decoded
1725 : with the next packet */
1726 24 : save_bits(s, gb, remaining_bits(s, gb), 0);
1727 : }
1728 :
1729 283 : s->packet_offset = get_bits_count(gb) & 7;
1730 283 : if (s->packet_loss)
1731 0 : return AVERROR_INVALIDDATA;
1732 :
1733 283 : return get_bits_count(gb) >> 3;
1734 : }
1735 :
1736 : /**
1737 : *@brief Decode a single WMA packet.
1738 : *@param avctx codec context
1739 : *@param data the output buffer
1740 : *@param avpkt input packet
1741 : *@return number of bytes that were read from the input buffer
1742 : */
1743 283 : static int wmapro_decode_packet(AVCodecContext *avctx, void *data,
1744 : int *got_frame_ptr, AVPacket *avpkt)
1745 : {
1746 283 : WMAProDecodeCtx *s = avctx->priv_data;
1747 283 : AVFrame *frame = data;
1748 : int ret;
1749 :
1750 : /* get output buffer */
1751 283 : frame->nb_samples = s->samples_per_frame;
1752 283 : if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1753 0 : s->packet_loss = 1;
1754 0 : return 0;
1755 : }
1756 :
1757 283 : return decode_packet(avctx, s, data, got_frame_ptr, avpkt);
1758 : }
1759 :
1760 0 : static int xma_decode_packet(AVCodecContext *avctx, void *data,
1761 : int *got_frame_ptr, AVPacket *avpkt)
1762 : {
1763 0 : XMADecodeCtx *s = avctx->priv_data;
1764 0 : int got_stream_frame_ptr = 0;
1765 0 : AVFrame *frame = data;
1766 0 : int i, ret, offset = INT_MAX;
1767 :
1768 : /* decode current stream packet */
1769 0 : ret = decode_packet(avctx, &s->xma[s->current_stream], s->frames[s->current_stream],
1770 : &got_stream_frame_ptr, avpkt);
1771 :
1772 : /* copy stream samples (1/2ch) to sample buffer (Nch) */
1773 0 : if (got_stream_frame_ptr) {
1774 0 : int start_ch = s->start_channel[s->current_stream];
1775 0 : memcpy(&s->samples[start_ch + 0][s->offset[s->current_stream] * 512],
1776 0 : s->frames[s->current_stream]->extended_data[0], 512 * 4);
1777 0 : if (s->xma[s->current_stream].nb_channels > 1)
1778 0 : memcpy(&s->samples[start_ch + 1][s->offset[s->current_stream] * 512],
1779 0 : s->frames[s->current_stream]->extended_data[1], 512 * 4);
1780 0 : s->offset[s->current_stream]++;
1781 0 : } else if (ret < 0) {
1782 0 : memset(s->offset, 0, sizeof(s->offset));
1783 0 : s->current_stream = 0;
1784 0 : return ret;
1785 : }
1786 :
1787 : /* find next XMA packet's owner stream, and update.
1788 : * XMA streams find their packets following packet_skips
1789 : * (at start there is one packet per stream, then interleave non-linearly). */
1790 0 : if (s->xma[s->current_stream].packet_done ||
1791 0 : s->xma[s->current_stream].packet_loss) {
1792 :
1793 : /* select stream with 0 skip_packets (= uses next packet) */
1794 0 : if (s->xma[s->current_stream].skip_packets != 0) {
1795 : int min[2];
1796 :
1797 0 : min[0] = s->xma[0].skip_packets;
1798 0 : min[1] = i = 0;
1799 :
1800 0 : for (i = 1; i < s->num_streams; i++) {
1801 0 : if (s->xma[i].skip_packets < min[0]) {
1802 0 : min[0] = s->xma[i].skip_packets;
1803 0 : min[1] = i;
1804 : }
1805 : }
1806 :
1807 0 : s->current_stream = min[1];
1808 : }
1809 :
1810 : /* all other streams skip next packet */
1811 0 : for (i = 0; i < s->num_streams; i++) {
1812 0 : s->xma[i].skip_packets = FFMAX(0, s->xma[i].skip_packets - 1);
1813 : }
1814 :
1815 : /* copy samples from buffer to output if possible */
1816 0 : for (i = 0; i < s->num_streams; i++) {
1817 0 : offset = FFMIN(offset, s->offset[i]);
1818 : }
1819 0 : if (offset > 0) {
1820 : int bret;
1821 :
1822 0 : frame->nb_samples = 512 * offset;
1823 0 : if ((bret = ff_get_buffer(avctx, frame, 0)) < 0)
1824 0 : return bret;
1825 :
1826 : /* copy samples buffer (Nch) to frame samples (Nch), move unconsumed samples */
1827 0 : for (i = 0; i < s->num_streams; i++) {
1828 0 : int start_ch = s->start_channel[i];
1829 0 : memcpy(frame->extended_data[start_ch + 0], s->samples[start_ch + 0], frame->nb_samples * 4);
1830 0 : if (s->xma[i].nb_channels > 1)
1831 0 : memcpy(frame->extended_data[start_ch + 1], s->samples[start_ch + 1], frame->nb_samples * 4);
1832 :
1833 0 : s->offset[i] -= offset;
1834 0 : if (s->offset[i]) {
1835 0 : memmove(s->samples[start_ch + 0], s->samples[start_ch + 0] + frame->nb_samples, s->offset[i] * 4 * 512);
1836 0 : if (s->xma[i].nb_channels > 1)
1837 0 : memmove(s->samples[start_ch + 1], s->samples[start_ch + 1] + frame->nb_samples, s->offset[i] * 4 * 512);
1838 : }
1839 : }
1840 :
1841 0 : *got_frame_ptr = 1;
1842 : }
1843 : }
1844 :
1845 0 : return ret;
1846 : }
1847 :
1848 0 : static av_cold int xma_decode_init(AVCodecContext *avctx)
1849 : {
1850 0 : XMADecodeCtx *s = avctx->priv_data;
1851 0 : int i, ret, start_channels = 0;
1852 :
1853 0 : if (avctx->channels <= 0 || avctx->extradata_size == 0)
1854 0 : return AVERROR_INVALIDDATA;
1855 :
1856 : /* get stream config */
1857 0 : if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size == 34) { /* XMA2WAVEFORMATEX */
1858 0 : s->num_streams = (avctx->channels + 1) / 2;
1859 0 : } else if (avctx->codec_id == AV_CODEC_ID_XMA2 && avctx->extradata_size >= 2) { /* XMA2WAVEFORMAT */
1860 0 : s->num_streams = avctx->extradata[1];
1861 0 : if (avctx->extradata_size != (32 + ((avctx->extradata[0]==3)?0:8) + 4*s->num_streams)) {
1862 0 : av_log(avctx, AV_LOG_ERROR, "Incorrect XMA2 extradata size\n");
1863 0 : return AVERROR(EINVAL);
1864 : }
1865 0 : } else if (avctx->codec_id == AV_CODEC_ID_XMA1 && avctx->extradata_size >= 4) { /* XMAWAVEFORMAT */
1866 0 : s->num_streams = avctx->extradata[4];
1867 0 : if (avctx->extradata_size != (8 + 20*s->num_streams)) {
1868 0 : av_log(avctx, AV_LOG_ERROR, "Incorrect XMA1 extradata size\n");
1869 0 : return AVERROR(EINVAL);
1870 : }
1871 : } else {
1872 0 : av_log(avctx, AV_LOG_ERROR, "Incorrect XMA config\n");
1873 0 : return AVERROR(EINVAL);
1874 : }
1875 :
1876 : /* encoder supports up to 64 streams / 64*2 channels (would have to alloc arrays) */
1877 0 : if (avctx->channels > XMA_MAX_CHANNELS || s->num_streams > XMA_MAX_STREAMS) {
1878 0 : avpriv_request_sample(avctx, "More than %d channels in %d streams", XMA_MAX_CHANNELS, s->num_streams);
1879 0 : return AVERROR_PATCHWELCOME;
1880 : }
1881 :
1882 : /* init all streams (several streams of 1/2ch make Nch files) */
1883 0 : for (i = 0; i < s->num_streams; i++) {
1884 0 : ret = decode_init(&s->xma[i], avctx, i);
1885 0 : if (ret < 0)
1886 0 : return ret;
1887 0 : s->frames[i] = av_frame_alloc();
1888 0 : if (!s->frames[i])
1889 0 : return AVERROR(ENOMEM);
1890 0 : s->frames[i]->nb_samples = 512;
1891 0 : if ((ret = ff_get_buffer(avctx, s->frames[i], 0)) < 0) {
1892 0 : return AVERROR(ENOMEM);
1893 : }
1894 :
1895 0 : s->start_channel[i] = start_channels;
1896 0 : start_channels += s->xma[i].nb_channels;
1897 : }
1898 :
1899 0 : return ret;
1900 : }
1901 :
1902 0 : static av_cold int xma_decode_end(AVCodecContext *avctx)
1903 : {
1904 0 : XMADecodeCtx *s = avctx->priv_data;
1905 : int i;
1906 :
1907 0 : for (i = 0; i < s->num_streams; i++) {
1908 0 : decode_end(&s->xma[i]);
1909 0 : av_frame_free(&s->frames[i]);
1910 : }
1911 :
1912 0 : return 0;
1913 : }
1914 :
1915 0 : static void flush(WMAProDecodeCtx *s)
1916 : {
1917 : int i;
1918 : /** reset output buffer as a part of it is used during the windowing of a
1919 : new frame */
1920 0 : for (i = 0; i < s->nb_channels; i++)
1921 0 : memset(s->channel[i].out, 0, s->samples_per_frame *
1922 : sizeof(*s->channel[i].out));
1923 0 : s->packet_loss = 1;
1924 0 : s->skip_packets = 0;
1925 0 : }
1926 :
1927 :
1928 : /**
1929 : *@brief Clear decoder buffers (for seeking).
1930 : *@param avctx codec context
1931 : */
1932 0 : static void wmapro_flush(AVCodecContext *avctx)
1933 : {
1934 0 : WMAProDecodeCtx *s = avctx->priv_data;
1935 :
1936 0 : flush(s);
1937 0 : }
1938 :
1939 0 : static void xma_flush(AVCodecContext *avctx)
1940 : {
1941 0 : XMADecodeCtx *s = avctx->priv_data;
1942 : int i;
1943 :
1944 0 : for (i = 0; i < s->num_streams; i++)
1945 0 : flush(&s->xma[i]);
1946 :
1947 0 : memset(s->offset, 0, sizeof(s->offset));
1948 0 : s->current_stream = 0;
1949 0 : }
1950 :
1951 :
1952 : /**
1953 : *@brief wmapro decoder
1954 : */
1955 : AVCodec ff_wmapro_decoder = {
1956 : .name = "wmapro",
1957 : .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1958 : .type = AVMEDIA_TYPE_AUDIO,
1959 : .id = AV_CODEC_ID_WMAPRO,
1960 : .priv_data_size = sizeof(WMAProDecodeCtx),
1961 : .init = wmapro_decode_init,
1962 : .close = wmapro_decode_end,
1963 : .decode = wmapro_decode_packet,
1964 : .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1965 : .flush = wmapro_flush,
1966 : .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1967 : AV_SAMPLE_FMT_NONE },
1968 : };
1969 :
1970 : AVCodec ff_xma1_decoder = {
1971 : .name = "xma1",
1972 : .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 1"),
1973 : .type = AVMEDIA_TYPE_AUDIO,
1974 : .id = AV_CODEC_ID_XMA1,
1975 : .priv_data_size = sizeof(XMADecodeCtx),
1976 : .init = xma_decode_init,
1977 : .close = xma_decode_end,
1978 : .decode = xma_decode_packet,
1979 : .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1980 : .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1981 : AV_SAMPLE_FMT_NONE },
1982 : };
1983 :
1984 : AVCodec ff_xma2_decoder = {
1985 : .name = "xma2",
1986 : .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 2"),
1987 : .type = AVMEDIA_TYPE_AUDIO,
1988 : .id = AV_CODEC_ID_XMA2,
1989 : .priv_data_size = sizeof(XMADecodeCtx),
1990 : .init = xma_decode_init,
1991 : .close = xma_decode_end,
1992 : .decode = xma_decode_packet,
1993 : .flush = xma_flush,
1994 : .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1995 : .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1996 : AV_SAMPLE_FMT_NONE },
1997 : };
|
