Statistics
| Branch: | Revision:

ffmpeg / libavcodec / wmaprodec.c @ 20169324

History | View | Annotate | Download (58.9 KB)

1
/*
2
 * Wmapro compatible decoder
3
 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4
 * Copyright (c) 2008 - 2009 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  libavcodec/wmaprodec.c
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 "avcodec.h"
90
#include "internal.h"
91
#include "get_bits.h"
92
#include "put_bits.h"
93
#include "wmaprodata.h"
94
#include "dsputil.h"
95
#include "wma.h"
96

    
97
/** current decoder limitations */
98
#define WMAPRO_MAX_CHANNELS    8                             ///< max number of handled channels
99
#define MAX_SUBFRAMES  32                                    ///< max number of subframes per channel
100
#define MAX_BANDS      29                                    ///< max number of scale factor bands
101
#define MAX_FRAMESIZE  16384                                 ///< maximum compressed frame size
102

    
103
#define WMAPRO_BLOCK_MAX_BITS 12                                           ///< log2 of max block size
104
#define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS)                 ///< maximum block size
105
#define WMAPRO_BLOCK_SIZES    (WMAPRO_BLOCK_MAX_BITS - BLOCK_MIN_BITS + 1) ///< possible block sizes
106

    
107

    
108
#define VLCBITS            9
109
#define SCALEVLCBITS       8
110
#define VEC4MAXDEPTH    ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
111
#define VEC2MAXDEPTH    ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
112
#define VEC1MAXDEPTH    ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
113
#define SCALEMAXDEPTH   ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
114
#define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
115

    
116
static VLC              sf_vlc;           ///< scale factor DPCM vlc
117
static VLC              sf_rl_vlc;        ///< scale factor run length vlc
118
static VLC              vec4_vlc;         ///< 4 coefficients per symbol
119
static VLC              vec2_vlc;         ///< 2 coefficients per symbol
120
static VLC              vec1_vlc;         ///< 1 coefficient per symbol
121
static VLC              coef_vlc[2];      ///< coefficient run length vlc codes
122
static float            sin64[33];        ///< sinus table for decorrelation
123

    
124
/**
125
 * @brief frame specific decoder context for a single channel
126
 */
127
typedef struct {
128
    int16_t  prev_block_len;                          ///< length of the previous block
129
    uint8_t  transmit_coefs;
130
    uint8_t  num_subframes;
131
    uint16_t subframe_len[MAX_SUBFRAMES];             ///< subframe length in samples
132
    uint16_t subframe_offset[MAX_SUBFRAMES];          ///< subframe positions in the current frame
133
    uint8_t  cur_subframe;                            ///< current subframe number
134
    uint16_t decoded_samples;                         ///< number of already processed samples
135
    uint8_t  grouped;                                 ///< channel is part of a group
136
    int      quant_step;                              ///< quantization step for the current subframe
137
    int8_t   reuse_sf;                                ///< share scale factors between subframes
138
    int8_t   scale_factor_step;                       ///< scaling step for the current subframe
139
    int      max_scale_factor;                        ///< maximum scale factor for the current subframe
140
    int      scale_factors[MAX_BANDS];                ///< scale factor values for the current subframe
141
    int      saved_scale_factors[MAX_BANDS];          ///< scale factors from a previous subframe
142
    uint8_t  table_idx;                               ///< index in sf_offsets for the scale factor reference block
143
    float*   coeffs;                                  ///< pointer to the subframe decode buffer
144
    DECLARE_ALIGNED_16(float, out[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]); ///< output buffer
145
} WMAProChannelCtx;
146

    
147
/**
148
 * @brief channel group for channel transformations
149
 */
150
typedef struct {
151
    uint8_t num_channels;                                     ///< number of channels in the group
152
    int8_t  transform;                                        ///< transform on / off
153
    int8_t  transform_band[MAX_BANDS];                        ///< controls if the transform is enabled for a certain band
154
    float   decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
155
    float*  channel_data[WMAPRO_MAX_CHANNELS];                ///< transformation coefficients
156
} WMAProChannelGrp;
157

    
158
/**
159
 * @brief main decoder context
160
 */
161
typedef struct WMAProDecodeCtx {
162
    /* generic decoder variables */
163
    AVCodecContext*  avctx;                         ///< codec context for av_log
164
    DSPContext       dsp;                           ///< accelerated DSP functions
165
    uint8_t          frame_data[MAX_FRAMESIZE +
166
                      FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
167
    PutBitContext    pb;                            ///< context for filling the frame_data buffer
168
    MDCTContext      mdct_ctx[WMAPRO_BLOCK_SIZES];  ///< MDCT context per block size
169
    DECLARE_ALIGNED_16(float, tmp[WMAPRO_BLOCK_MAX_SIZE]); ///< IMDCT output buffer
170
    float*           windows[WMAPRO_BLOCK_SIZES];   ///< windows for the different block sizes
171

    
172
    /* frame size dependent frame information (set during initialization) */
173
    uint32_t         decode_flags;                  ///< used compression features
174
    uint8_t          len_prefix;                    ///< frame is prefixed with its length
175
    uint8_t          dynamic_range_compression;     ///< frame contains DRC data
176
    uint8_t          bits_per_sample;               ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
177
    uint16_t         samples_per_frame;             ///< number of samples to output
178
    uint16_t         log2_frame_size;
179
    int8_t           num_channels;                  ///< number of channels in the stream (same as AVCodecContext.num_channels)
180
    int8_t           lfe_channel;                   ///< lfe channel index
181
    uint8_t          max_num_subframes;
182
    uint8_t          subframe_len_bits;             ///< number of bits used for the subframe length
183
    uint8_t          max_subframe_len_bit;          ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
184
    uint16_t         min_samples_per_subframe;
185
    int8_t           num_sfb[WMAPRO_BLOCK_SIZES];   ///< scale factor bands per block size
186
    int16_t          sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS];                    ///< scale factor band offsets (multiples of 4)
187
    int8_t           sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
188
    int16_t          subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
189

    
190
    /* packet decode state */
191
    GetBitContext    pgb;                           ///< bitstream reader context for the packet
192
    uint8_t          packet_sequence_number;        ///< current packet number
193
    int              num_saved_bits;                ///< saved number of bits
194
    int              frame_offset;                  ///< frame offset in the bit reservoir
195
    int              subframe_offset;               ///< subframe offset in the bit reservoir
196
    uint8_t          packet_loss;                   ///< set in case of bitstream error
197

    
198
    /* frame decode state */
199
    uint32_t         frame_num;                     ///< current frame number (not used for decoding)
200
    GetBitContext    gb;                            ///< bitstream reader context
201
    int              buf_bit_size;                  ///< buffer size in bits
202
    float*           samples;                       ///< current samplebuffer pointer
203
    float*           samples_end;                   ///< maximum samplebuffer pointer
204
    uint8_t          drc_gain;                      ///< gain for the DRC tool
205
    int8_t           skip_frame;                    ///< skip output step
206
    int8_t           parsed_all_subframes;          ///< all subframes decoded?
207

    
208
    /* subframe/block decode state */
209
    int16_t          subframe_len;                  ///< current subframe length
210
    int8_t           channels_for_cur_subframe;     ///< number of channels that contain the subframe
211
    int8_t           channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
212
    int8_t           num_bands;                     ///< number of scale factor bands
213
    int16_t*         cur_sfb_offsets;               ///< sfb offsets for the current block
214
    uint8_t          table_idx;                     ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
215
    int8_t           esc_len;                       ///< length of escaped coefficients
216

    
217
    uint8_t          num_chgroups;                  ///< number of channel groups
218
    WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS];  ///< channel group information
219

    
220
    WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS];  ///< per channel data
221
} WMAProDecodeCtx;
222

    
223

    
224
/**
225
 *@brief helper function to print the most important members of the context
226
 *@param s context
227
 */
228
static void av_cold dump_context(WMAProDecodeCtx *s)
229
{
230
#define PRINT(a, b)     av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
231
#define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
232

    
233
    PRINT("ed sample bit depth", s->bits_per_sample);
234
    PRINT_HEX("ed decode flags", s->decode_flags);
235
    PRINT("samples per frame",   s->samples_per_frame);
236
    PRINT("log2 frame size",     s->log2_frame_size);
237
    PRINT("max num subframes",   s->max_num_subframes);
238
    PRINT("len prefix",          s->len_prefix);
239
    PRINT("num channels",        s->num_channels);
240
}
241

    
242
/**
243
 *@brief Uninitialize the decoder and free all resources.
244
 *@param avctx codec context
245
 *@return 0 on success, < 0 otherwise
246
 */
247
static av_cold int decode_end(AVCodecContext *avctx)
248
{
249
    WMAProDecodeCtx *s = avctx->priv_data;
250
    int i;
251

    
252
    for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
253
        ff_mdct_end(&s->mdct_ctx[i]);
254

    
255
    return 0;
256
}
257

    
258
/**
259
 *@brief Initialize the decoder.
260
 *@param avctx codec context
261
 *@return 0 on success, -1 otherwise
262
 */
263
static av_cold int decode_init(AVCodecContext *avctx)
264
{
265
    WMAProDecodeCtx *s = avctx->priv_data;
266
    uint8_t *edata_ptr = avctx->extradata;
267
    unsigned int channel_mask;
268
    int i;
269
    int log2_max_num_subframes;
270
    int num_possible_block_sizes;
271

    
272
    s->avctx = avctx;
273
    dsputil_init(&s->dsp, avctx);
274
    init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
275

    
276
    avctx->sample_fmt = SAMPLE_FMT_FLT;
277

    
278
    if (avctx->extradata_size >= 18) {
279
        s->decode_flags    = AV_RL16(edata_ptr+14);
280
        channel_mask       = AV_RL32(edata_ptr+2);
281
        s->bits_per_sample = AV_RL16(edata_ptr);
282
        /** dump the extradata */
283
        for (i = 0; i < avctx->extradata_size; i++)
284
            dprintf(avctx, "[%x] ", avctx->extradata[i]);
285
        dprintf(avctx, "\n");
286

    
287
    } else {
288
        av_log_ask_for_sample(avctx, "Unknown extradata size\n");
289
        return AVERROR_INVALIDDATA;
290
    }
291

    
292
    /** generic init */
293
    s->log2_frame_size = av_log2(avctx->block_align) + 4;
294

    
295
    /** frame info */
296
    s->skip_frame  = 1; /** skip first frame */
297
    s->packet_loss = 1;
298
    s->len_prefix  = (s->decode_flags & 0x40);
299

    
300
    if (!s->len_prefix) {
301
        av_log_ask_for_sample(avctx, "no length prefix\n");
302
        return AVERROR_INVALIDDATA;
303
    }
304

    
305
    /** get frame len */
306
    s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
307
                                                          3, s->decode_flags);
308

    
309
    /** init previous block len */
310
    for (i = 0; i < avctx->channels; i++)
311
        s->channel[i].prev_block_len = s->samples_per_frame;
312

    
313
    /** subframe info */
314
    log2_max_num_subframes       = ((s->decode_flags & 0x38) >> 3);
315
    s->max_num_subframes         = 1 << log2_max_num_subframes;
316
    if (s->max_num_subframes == 16)
317
        s->max_subframe_len_bit = 1;
318
    s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
319

    
320
    num_possible_block_sizes     = log2_max_num_subframes + 1;
321
    s->min_samples_per_subframe  = s->samples_per_frame / s->max_num_subframes;
322
    s->dynamic_range_compression = (s->decode_flags & 0x80);
323

    
324
    if (s->max_num_subframes > MAX_SUBFRAMES) {
325
        av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
326
               s->max_num_subframes);
327
        return AVERROR_INVALIDDATA;
328
    }
329

    
330
    s->num_channels = avctx->channels;
331

    
332
    /** extract lfe channel position */
333
    s->lfe_channel = -1;
334

    
335
    if (channel_mask & 8) {
336
        unsigned int mask;
337
        for (mask = 1; mask < 16; mask <<= 1) {
338
            if (channel_mask & mask)
339
                ++s->lfe_channel;
340
        }
341
    }
342

    
343
    if (s->num_channels < 0 || s->num_channels > WMAPRO_MAX_CHANNELS) {
344
        av_log_ask_for_sample(avctx, "invalid number of channels\n");
345
        return AVERROR_NOTSUPP;
346
    }
347

    
348
    INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
349
                    scale_huffbits, 1, 1,
350
                    scale_huffcodes, 2, 2, 616);
351

    
352
    INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
353
                    scale_rl_huffbits, 1, 1,
354
                    scale_rl_huffcodes, 4, 4, 1406);
355

    
356
    INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
357
                    coef0_huffbits, 1, 1,
358
                    coef0_huffcodes, 4, 4, 2108);
359

    
360
    INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
361
                    coef1_huffbits, 1, 1,
362
                    coef1_huffcodes, 4, 4, 3912);
363

    
364
    INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
365
                    vec4_huffbits, 1, 1,
366
                    vec4_huffcodes, 2, 2, 604);
367

    
368
    INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
369
                    vec2_huffbits, 1, 1,
370
                    vec2_huffcodes, 2, 2, 562);
371

    
372
    INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
373
                    vec1_huffbits, 1, 1,
374
                    vec1_huffcodes, 2, 2, 562);
375

    
376
    /** calculate number of scale factor bands and their offsets
377
        for every possible block size */
378
    for (i = 0; i < num_possible_block_sizes; i++) {
379
        int subframe_len = s->samples_per_frame >> i;
380
        int x;
381
        int band = 1;
382

    
383
        s->sfb_offsets[i][0] = 0;
384

    
385
        for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
386
            int offset = (subframe_len * 2 * critical_freq[x])
387
                          / s->avctx->sample_rate + 2;
388
            offset &= ~3;
389
            if (offset > s->sfb_offsets[i][band - 1])
390
                s->sfb_offsets[i][band++] = offset;
391
        }
392
        s->sfb_offsets[i][band - 1] = subframe_len;
393
        s->num_sfb[i]               = band - 1;
394
    }
395

    
396

    
397
    /** Scale factors can be shared between blocks of different size
398
        as every block has a different scale factor band layout.
399
        The matrix sf_offsets is needed to find the correct scale factor.
400
     */
401

    
402
    for (i = 0; i < num_possible_block_sizes; i++) {
403
        int b;
404
        for (b = 0; b < s->num_sfb[i]; b++) {
405
            int x;
406
            int offset = ((s->sfb_offsets[i][b]
407
                           + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
408
            for (x = 0; x < num_possible_block_sizes; x++) {
409
                int v = 0;
410
                while (s->sfb_offsets[x][v + 1] << x < offset)
411
                    ++v;
412
                s->sf_offsets[i][x][b] = v;
413
            }
414
        }
415
    }
416

    
417
    /** init MDCT, FIXME: only init needed sizes */
418
    for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
419
        ff_mdct_init(&s->mdct_ctx[i], BLOCK_MIN_BITS+1+i, 1,
420
                     1.0 / (1 << (BLOCK_MIN_BITS + i - 1))
421
                     / (1 << (s->bits_per_sample - 1)));
422

    
423
    /** init MDCT windows: simple sinus window */
424
    for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
425
        const int n       = 1 << (WMAPRO_BLOCK_MAX_BITS - i);
426
        const int win_idx = WMAPRO_BLOCK_MAX_BITS - i - 7;
427
        ff_sine_window_init(ff_sine_windows[win_idx], n);
428
        s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
429
    }
430

    
431
    /** calculate subwoofer cutoff values */
432
    for (i = 0; i < num_possible_block_sizes; i++) {
433
        int block_size = s->samples_per_frame >> i;
434
        int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
435
                     / s->avctx->sample_rate;
436
        s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
437
    }
438

    
439
    /** calculate sine values for the decorrelation matrix */
440
    for (i = 0; i < 33; i++)
441
        sin64[i] = sin(i*M_PI / 64.0);
442

    
443
    if (avctx->debug & FF_DEBUG_BITSTREAM)
444
        dump_context(s);
445

    
446
    avctx->channel_layout = channel_mask;
447
    return 0;
448
}
449

    
450
/**
451
 *@brief Decode the subframe length.
452
 *@param s context
453
 *@param offset sample offset in the frame
454
 *@return decoded subframe length on success, < 0 in case of an error
455
 */
456
static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
457
{
458
    int frame_len_shift = 0;
459
    int subframe_len;
460

    
461
    /** no need to read from the bitstream when only one length is possible */
462
    if (offset == s->samples_per_frame - s->min_samples_per_subframe)
463
        return s->min_samples_per_subframe;
464

    
465
    /** 1 bit indicates if the subframe is of maximum length */
466
    if (s->max_subframe_len_bit) {
467
        if (get_bits1(&s->gb))
468
            frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
469
    } else
470
        frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
471

    
472
    subframe_len = s->samples_per_frame >> frame_len_shift;
473

    
474
    /** sanity check the length */
475
    if (subframe_len < s->min_samples_per_subframe ||
476
        subframe_len > s->samples_per_frame) {
477
        av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
478
               subframe_len);
479
        return AVERROR_INVALIDDATA;
480
    }
481
    return subframe_len;
482
}
483

    
484
/**
485
 *@brief Decode how the data in the frame is split into subframes.
486
 *       Every WMA frame contains the encoded data for a fixed number of
487
 *       samples per channel. The data for every channel might be split
488
 *       into several subframes. This function will reconstruct the list of
489
 *       subframes for every channel.
490
 *
491
 *       If the subframes are not evenly split, the algorithm estimates the
492
 *       channels with the lowest number of total samples.
493
 *       Afterwards, for each of these channels a bit is read from the
494
 *       bitstream that indicates if the channel contains a subframe with the
495
 *       next subframe size that is going to be read from the bitstream or not.
496
 *       If a channel contains such a subframe, the subframe size gets added to
497
 *       the channel's subframe list.
498
 *       The algorithm repeats these steps until the frame is properly divided
499
 *       between the individual channels.
500
 *
501
 *@param s context
502
 *@return 0 on success, < 0 in case of an error
503
 */
504
static int decode_tilehdr(WMAProDecodeCtx *s)
505
{
506
    uint16_t num_samples[WMAPRO_MAX_CHANNELS];        /** sum of samples for all currently known subframes of a channel */
507
    uint8_t  contains_subframe[WMAPRO_MAX_CHANNELS];  /** flag indicating if a channel contains the current subframe */
508
    int channels_for_cur_subframe = s->num_channels;  /** number of channels that contain the current subframe */
509
    int fixed_channel_layout = 0;                     /** flag indicating that all channels use the same subframe offsets and sizes */
510
    int min_channel_len = 0;                          /** smallest sum of samples (channels with this length will be processed first) */
511
    int c;
512

    
513
    /* Should never consume more than 3073 bits (256 iterations for the
514
     * while loop when always the minimum amount of 128 samples is substracted
515
     * from missing samples in the 8 channel case).
516
     * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS  + 4)
517
     */
518

    
519
    /** reset tiling information */
520
    for (c = 0; c < s->num_channels; c++)
521
        s->channel[c].num_subframes = 0;
522

    
523
    memset(num_samples, 0, sizeof(num_samples));
524

    
525
    if (s->max_num_subframes == 1 || get_bits1(&s->gb))
526
        fixed_channel_layout = 1;
527

    
528
    /** loop until the frame data is split between the subframes */
529
    do {
530
        int subframe_len;
531

    
532
        /** check which channels contain the subframe */
533
        for (c = 0; c < s->num_channels; c++) {
534
            if (num_samples[c] == min_channel_len) {
535
                if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
536
                   (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
537
                    contains_subframe[c] = 1;
538
                else
539
                    contains_subframe[c] = get_bits1(&s->gb);
540
            } else
541
                contains_subframe[c] = 0;
542
        }
543

    
544
        /** get subframe length, subframe_len == 0 is not allowed */
545
        if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
546
            return AVERROR_INVALIDDATA;
547

    
548
        /** add subframes to the individual channels and find new min_channel_len */
549
        min_channel_len += subframe_len;
550
        for (c = 0; c < s->num_channels; c++) {
551
            WMAProChannelCtx* chan = &s->channel[c];
552

    
553
            if (contains_subframe[c]) {
554
                if (chan->num_subframes >= MAX_SUBFRAMES) {
555
                    av_log(s->avctx, AV_LOG_ERROR,
556
                           "broken frame: num subframes > 31\n");
557
                    return AVERROR_INVALIDDATA;
558
                }
559
                chan->subframe_len[chan->num_subframes] = subframe_len;
560
                num_samples[c] += subframe_len;
561
                ++chan->num_subframes;
562
                if (num_samples[c] > s->samples_per_frame) {
563
                    av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
564
                           "channel len > samples_per_frame\n");
565
                    return AVERROR_INVALIDDATA;
566
                }
567
            } else if (num_samples[c] <= min_channel_len) {
568
                if (num_samples[c] < min_channel_len) {
569
                    channels_for_cur_subframe = 0;
570
                    min_channel_len = num_samples[c];
571
                }
572
                ++channels_for_cur_subframe;
573
            }
574
        }
575
    } while (min_channel_len < s->samples_per_frame);
576

    
577
    for (c = 0; c < s->num_channels; c++) {
578
        int i;
579
        int offset = 0;
580
        for (i = 0; i < s->channel[c].num_subframes; i++) {
581
            dprintf(s->avctx, "frame[%i] channel[%i] subframe[%i]"
582
                    " len %i\n", s->frame_num, c, i,
583
                    s->channel[c].subframe_len[i]);
584
            s->channel[c].subframe_offset[i] = offset;
585
            offset += s->channel[c].subframe_len[i];
586
        }
587
    }
588

    
589
    return 0;
590
}
591

    
592
/**
593
 *@brief Calculate a decorrelation matrix from the bitstream parameters.
594
 *@param s codec context
595
 *@param chgroup channel group for which the matrix needs to be calculated
596
 */
597
static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
598
                                        WMAProChannelGrp *chgroup)
599
{
600
    int i;
601
    int offset = 0;
602
    int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
603
    memset(chgroup->decorrelation_matrix, 0, s->num_channels *
604
           s->num_channels * sizeof(*chgroup->decorrelation_matrix));
605

    
606
    for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
607
        rotation_offset[i] = get_bits(&s->gb, 6);
608

    
609
    for (i = 0; i < chgroup->num_channels; i++)
610
        chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
611
            get_bits1(&s->gb) ? 1.0 : -1.0;
612

    
613
    for (i = 1; i < chgroup->num_channels; i++) {
614
        int x;
615
        for (x = 0; x < i; x++) {
616
            int y;
617
            for (y = 0; y < i + 1; y++) {
618
                float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
619
                float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
620
                int n = rotation_offset[offset + x];
621
                float sinv;
622
                float cosv;
623

    
624
                if (n < 32) {
625
                    sinv = sin64[n];
626
                    cosv = sin64[32 - n];
627
                } else {
628
                    sinv =  sin64[64 -  n];
629
                    cosv = -sin64[n  - 32];
630
                }
631

    
632
                chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
633
                                               (v1 * sinv) - (v2 * cosv);
634
                chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
635
                                               (v1 * cosv) + (v2 * sinv);
636
            }
637
        }
638
        offset += i;
639
    }
640
}
641

    
642
/**
643
 *@brief Decode channel transformation parameters
644
 *@param s codec context
645
 *@return 0 in case of success, < 0 in case of bitstream errors
646
 */
647
static int decode_channel_transform(WMAProDecodeCtx* s)
648
{
649
    int i;
650
    /* should never consume more than 1921 bits for the 8 channel case
651
     * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
652
     * + MAX_CHANNELS + MAX_BANDS + 1)
653
     */
654

    
655
    /** in the one channel case channel transforms are pointless */
656
    s->num_chgroups = 0;
657
    if (s->num_channels > 1) {
658
        int remaining_channels = s->channels_for_cur_subframe;
659

    
660
        if (get_bits1(&s->gb)) {
661
            av_log_ask_for_sample(s->avctx,
662
                                  "unsupported channel transform bit\n");
663
            return AVERROR_INVALIDDATA;
664
        }
665

    
666
        for (s->num_chgroups = 0; remaining_channels &&
667
             s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
668
            WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
669
            float** channel_data = chgroup->channel_data;
670
            chgroup->num_channels = 0;
671
            chgroup->transform = 0;
672

    
673
            /** decode channel mask */
674
            if (remaining_channels > 2) {
675
                for (i = 0; i < s->channels_for_cur_subframe; i++) {
676
                    int channel_idx = s->channel_indexes_for_cur_subframe[i];
677
                    if (!s->channel[channel_idx].grouped
678
                        && get_bits1(&s->gb)) {
679
                        ++chgroup->num_channels;
680
                        s->channel[channel_idx].grouped = 1;
681
                        *channel_data++ = s->channel[channel_idx].coeffs;
682
                    }
683
                }
684
            } else {
685
                chgroup->num_channels = remaining_channels;
686
                for (i = 0; i < s->channels_for_cur_subframe; i++) {
687
                    int channel_idx = s->channel_indexes_for_cur_subframe[i];
688
                    if (!s->channel[channel_idx].grouped)
689
                        *channel_data++ = s->channel[channel_idx].coeffs;
690
                    s->channel[channel_idx].grouped = 1;
691
                }
692
            }
693

    
694
            /** decode transform type */
695
            if (chgroup->num_channels == 2) {
696
                if (get_bits1(&s->gb)) {
697
                    if (get_bits1(&s->gb)) {
698
                        av_log_ask_for_sample(s->avctx,
699
                                              "unsupported channel transform type\n");
700
                    }
701
                } else {
702
                    chgroup->transform = 1;
703
                    if (s->num_channels == 2) {
704
                        chgroup->decorrelation_matrix[0] =  1.0;
705
                        chgroup->decorrelation_matrix[1] = -1.0;
706
                        chgroup->decorrelation_matrix[2] =  1.0;
707
                        chgroup->decorrelation_matrix[3] =  1.0;
708
                    } else {
709
                        /** cos(pi/4) */
710
                        chgroup->decorrelation_matrix[0] =  0.70703125;
711
                        chgroup->decorrelation_matrix[1] = -0.70703125;
712
                        chgroup->decorrelation_matrix[2] =  0.70703125;
713
                        chgroup->decorrelation_matrix[3] =  0.70703125;
714
                    }
715
                }
716
            } else if (chgroup->num_channels > 2) {
717
                if (get_bits1(&s->gb)) {
718
                    chgroup->transform = 1;
719
                    if (get_bits1(&s->gb)) {
720
                        decode_decorrelation_matrix(s, chgroup);
721
                    } else {
722
                        /** FIXME: more than 6 coupled channels not supported */
723
                        if (chgroup->num_channels > 6) {
724
                            av_log_ask_for_sample(s->avctx,
725
                                                  "coupled channels > 6\n");
726
                        } else {
727
                            memcpy(chgroup->decorrelation_matrix,
728
                                   default_decorrelation[chgroup->num_channels],
729
                                   chgroup->num_channels * chgroup->num_channels *
730
                                   sizeof(*chgroup->decorrelation_matrix));
731
                        }
732
                    }
733
                }
734
            }
735

    
736
            /** decode transform on / off */
737
            if (chgroup->transform) {
738
                if (!get_bits1(&s->gb)) {
739
                    int i;
740
                    /** transform can be enabled for individual bands */
741
                    for (i = 0; i < s->num_bands; i++) {
742
                        chgroup->transform_band[i] = get_bits1(&s->gb);
743
                    }
744
                } else {
745
                    memset(chgroup->transform_band, 1, s->num_bands);
746
                }
747
            }
748
            remaining_channels -= chgroup->num_channels;
749
        }
750
    }
751
    return 0;
752
}
753

    
754
/**
755
 *@brief Extract the coefficients from the bitstream.
756
 *@param s codec context
757
 *@param c current channel number
758
 *@return 0 on success, < 0 in case of bitstream errors
759
 */
760
static int decode_coeffs(WMAProDecodeCtx *s, int c)
761
{
762
    int vlctable;
763
    VLC* vlc;
764
    WMAProChannelCtx* ci = &s->channel[c];
765
    int rl_mode = 0;
766
    int cur_coeff = 0;
767
    int num_zeros = 0;
768
    const uint16_t* run;
769
    const uint16_t* level;
770

    
771
    dprintf(s->avctx, "decode coefficients for channel %i\n", c);
772

    
773
    vlctable = get_bits1(&s->gb);
774
    vlc = &coef_vlc[vlctable];
775

    
776
    if (vlctable) {
777
        run = coef1_run;
778
        level = coef1_level;
779
    } else {
780
        run = coef0_run;
781
        level = coef0_level;
782
    }
783

    
784
    /** decode vector coefficients (consumes up to 167 bits per iteration for
785
      4 vector coded large values) */
786
    while (!rl_mode && cur_coeff + 3 < s->subframe_len) {
787
        int vals[4];
788
        int i;
789
        unsigned int idx;
790

    
791
        idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
792

    
793
        if (idx == HUFF_VEC4_SIZE - 1) {
794
            for (i = 0; i < 4; i += 2) {
795
                idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
796
                if (idx == HUFF_VEC2_SIZE - 1) {
797
                    vals[i] = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
798
                    if (vals[i] == HUFF_VEC1_SIZE - 1)
799
                        vals[i] += ff_wma_get_large_val(&s->gb);
800
                    vals[i+1] = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
801
                    if (vals[i+1] == HUFF_VEC1_SIZE - 1)
802
                        vals[i+1] += ff_wma_get_large_val(&s->gb);
803
                } else {
804
                    vals[i]   = symbol_to_vec2[idx] >> 4;
805
                    vals[i+1] = symbol_to_vec2[idx] & 0xF;
806
                }
807
            }
808
        } else {
809
            vals[0] =  symbol_to_vec4[idx] >> 12;
810
            vals[1] = (symbol_to_vec4[idx] >> 8) & 0xF;
811
            vals[2] = (symbol_to_vec4[idx] >> 4) & 0xF;
812
            vals[3] =  symbol_to_vec4[idx]       & 0xF;
813
        }
814

    
815
        /** decode sign */
816
        for (i = 0; i < 4; i++) {
817
            if (vals[i]) {
818
                int sign = get_bits1(&s->gb) - 1;
819
                ci->coeffs[cur_coeff] = (vals[i] ^ sign) - sign;
820
                num_zeros = 0;
821
            } else {
822
                ci->coeffs[cur_coeff] = 0;
823
                /** switch to run level mode when subframe_len / 128 zeros
824
                    were found in a row */
825
                rl_mode |= (++num_zeros > s->subframe_len >> 8);
826
            }
827
            ++cur_coeff;
828
        }
829
    }
830

    
831
    /** decode run level coded coefficients */
832
    if (rl_mode) {
833
        memset(&ci->coeffs[cur_coeff], 0,
834
               sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
835
        if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
836
                                    level, run, 1, ci->coeffs,
837
                                    cur_coeff, s->subframe_len,
838
                                    s->subframe_len, s->esc_len, 0))
839
            return AVERROR_INVALIDDATA;
840
    }
841

    
842
    return 0;
843
}
844

    
845
/**
846
 *@brief Extract scale factors from the bitstream.
847
 *@param s codec context
848
 *@return 0 on success, < 0 in case of bitstream errors
849
 */
850
static int decode_scale_factors(WMAProDecodeCtx* s)
851
{
852
    int i;
853

    
854
    /** should never consume more than 5344 bits
855
     *  MAX_CHANNELS * (1 +  MAX_BANDS * 23)
856
     */
857

    
858
    for (i = 0; i < s->channels_for_cur_subframe; i++) {
859
        int c = s->channel_indexes_for_cur_subframe[i];
860
        int* sf;
861
        int* sf_end = s->channel[c].scale_factors + s->num_bands;
862

    
863
        /** resample scale factors for the new block size
864
         *  as the scale factors might need to be resampled several times
865
         *  before some  new values are transmitted, a backup of the last
866
         *  transmitted scale factors is kept in saved_scale_factors
867
         */
868
        if (s->channel[c].reuse_sf) {
869
            const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
870
            int b;
871
            for (b = 0; b < s->num_bands; b++)
872
                s->channel[c].scale_factors[b] =
873
                                   s->channel[c].saved_scale_factors[*sf_offsets++];
874
        }
875

    
876
        if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
877

    
878
            if (!s->channel[c].reuse_sf) {
879
                int val;
880
                /** decode DPCM coded scale factors */
881
                s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
882
                val = 45 / s->channel[c].scale_factor_step;
883
                for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
884
                    val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
885
                    *sf = val;
886
                }
887
            } else {
888
                int i;
889
                /** run level decode differences to the resampled factors */
890
                for (i = 0; i < s->num_bands; i++) {
891
                    int idx;
892
                    int skip;
893
                    int val;
894
                    int sign;
895

    
896
                    idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
897

    
898
                    if (!idx) {
899
                        uint32_t code = get_bits(&s->gb, 14);
900
                        val  =  code >> 6;
901
                        sign = (code & 1) - 1;
902
                        skip = (code & 0x3f) >> 1;
903
                    } else if (idx == 1) {
904
                        break;
905
                    } else {
906
                        skip = scale_rl_run[idx];
907
                        val  = scale_rl_level[idx];
908
                        sign = get_bits1(&s->gb)-1;
909
                    }
910

    
911
                    i += skip;
912
                    if (i >= s->num_bands) {
913
                        av_log(s->avctx, AV_LOG_ERROR,
914
                               "invalid scale factor coding\n");
915
                        return AVERROR_INVALIDDATA;
916
                    }
917
                    s->channel[c].scale_factors[i] += (val ^ sign) - sign;
918
                }
919
            }
920

    
921
            /** save transmitted scale factors so that they can be reused for
922
                the next subframe */
923
            memcpy(s->channel[c].saved_scale_factors,
924
                   s->channel[c].scale_factors, s->num_bands *
925
                   sizeof(*s->channel[c].saved_scale_factors));
926
            s->channel[c].table_idx = s->table_idx;
927
            s->channel[c].reuse_sf  = 1;
928
        }
929

    
930
        /** calculate new scale factor maximum */
931
        s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
932
        for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
933
            s->channel[c].max_scale_factor =
934
                FFMAX(s->channel[c].max_scale_factor, *sf);
935
        }
936

    
937
    }
938
    return 0;
939
}
940

    
941
/**
942
 *@brief Reconstruct the individual channel data.
943
 *@param s codec context
944
 */
945
static void inverse_channel_transform(WMAProDecodeCtx *s)
946
{
947
    int i;
948

    
949
    for (i = 0; i < s->num_chgroups; i++) {
950
        if (s->chgroup[i].transform) {
951
            float data[WMAPRO_MAX_CHANNELS];
952
            const int num_channels = s->chgroup[i].num_channels;
953
            float** ch_data = s->chgroup[i].channel_data;
954
            float** ch_end = ch_data + num_channels;
955
            const int8_t* tb = s->chgroup[i].transform_band;
956
            int16_t* sfb;
957

    
958
            /** multichannel decorrelation */
959
            for (sfb = s->cur_sfb_offsets;
960
                 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
961
                int y;
962
                if (*tb++ == 1) {
963
                    /** multiply values with the decorrelation_matrix */
964
                    for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
965
                        const float* mat = s->chgroup[i].decorrelation_matrix;
966
                        const float* data_end = data + num_channels;
967
                        float* data_ptr = data;
968
                        float** ch;
969

    
970
                        for (ch = ch_data; ch < ch_end; ch++)
971
                            *data_ptr++ = (*ch)[y];
972

    
973
                        for (ch = ch_data; ch < ch_end; ch++) {
974
                            float sum = 0;
975
                            data_ptr = data;
976
                            while (data_ptr < data_end)
977
                                sum += *data_ptr++ * *mat++;
978

    
979
                            (*ch)[y] = sum;
980
                        }
981
                    }
982
                } else if (s->num_channels == 2) {
983
                    for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
984
                        ch_data[0][y] *= 181.0 / 128;
985
                        ch_data[1][y] *= 181.0 / 128;
986
                    }
987
                }
988
            }
989
        }
990
    }
991
}
992

    
993
/**
994
 *@brief Apply sine window and reconstruct the output buffer.
995
 *@param s codec context
996
 */
997
static void wmapro_window(WMAProDecodeCtx *s)
998
{
999
    int i;
1000
    for (i = 0; i < s->channels_for_cur_subframe; i++) {
1001
        int c = s->channel_indexes_for_cur_subframe[i];
1002
        float* window;
1003
        int winlen = s->channel[c].prev_block_len;
1004
        float* start = s->channel[c].coeffs - (winlen >> 1);
1005

    
1006
        if (s->subframe_len < winlen) {
1007
            start += (winlen - s->subframe_len) >> 1;
1008
            winlen = s->subframe_len;
1009
        }
1010

    
1011
        window = s->windows[av_log2(winlen) - BLOCK_MIN_BITS];
1012

    
1013
        winlen >>= 1;
1014

    
1015
        s->dsp.vector_fmul_window(start, start, start + winlen,
1016
                                  window, 0, winlen);
1017

    
1018
        s->channel[c].prev_block_len = s->subframe_len;
1019
    }
1020
}
1021

    
1022
/**
1023
 *@brief Decode a single subframe (block).
1024
 *@param s codec context
1025
 *@return 0 on success, < 0 when decoding failed
1026
 */
1027
static int decode_subframe(WMAProDecodeCtx *s)
1028
{
1029
    int offset = s->samples_per_frame;
1030
    int subframe_len = s->samples_per_frame;
1031
    int i;
1032
    int total_samples   = s->samples_per_frame * s->num_channels;
1033
    int transmit_coeffs = 0;
1034
    int cur_subwoofer_cutoff;
1035

    
1036
    s->subframe_offset = get_bits_count(&s->gb);
1037

    
1038
    /** reset channel context and find the next block offset and size
1039
        == the next block of the channel with the smallest number of
1040
        decoded samples
1041
    */
1042
    for (i = 0; i < s->num_channels; i++) {
1043
        s->channel[i].grouped = 0;
1044
        if (offset > s->channel[i].decoded_samples) {
1045
            offset = s->channel[i].decoded_samples;
1046
            subframe_len =
1047
                s->channel[i].subframe_len[s->channel[i].cur_subframe];
1048
        }
1049
    }
1050

    
1051
    dprintf(s->avctx,
1052
            "processing subframe with offset %i len %i\n", offset, subframe_len);
1053

    
1054
    /** get a list of all channels that contain the estimated block */
1055
    s->channels_for_cur_subframe = 0;
1056
    for (i = 0; i < s->num_channels; i++) {
1057
        const int cur_subframe = s->channel[i].cur_subframe;
1058
        /** substract already processed samples */
1059
        total_samples -= s->channel[i].decoded_samples;
1060

    
1061
        /** and count if there are multiple subframes that match our profile */
1062
        if (offset == s->channel[i].decoded_samples &&
1063
            subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1064
            total_samples -= s->channel[i].subframe_len[cur_subframe];
1065
            s->channel[i].decoded_samples +=
1066
                s->channel[i].subframe_len[cur_subframe];
1067
            s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1068
            ++s->channels_for_cur_subframe;
1069
        }
1070
    }
1071

    
1072
    /** check if the frame will be complete after processing the
1073
        estimated block */
1074
    if (!total_samples)
1075
        s->parsed_all_subframes = 1;
1076

    
1077

    
1078
    dprintf(s->avctx, "subframe is part of %i channels\n",
1079
            s->channels_for_cur_subframe);
1080

    
1081
    /** calculate number of scale factor bands and their offsets */
1082
    s->table_idx         = av_log2(s->samples_per_frame/subframe_len);
1083
    s->num_bands         = s->num_sfb[s->table_idx];
1084
    s->cur_sfb_offsets   = s->sfb_offsets[s->table_idx];
1085
    cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1086

    
1087
    /** configure the decoder for the current subframe */
1088
    for (i = 0; i < s->channels_for_cur_subframe; i++) {
1089
        int c = s->channel_indexes_for_cur_subframe[i];
1090

    
1091
        s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1092
                                                  + offset];
1093
    }
1094

    
1095
    s->subframe_len = subframe_len;
1096
    s->esc_len = av_log2(s->subframe_len - 1) + 1;
1097

    
1098
    /** skip extended header if any */
1099
    if (get_bits1(&s->gb)) {
1100
        int num_fill_bits;
1101
        if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1102
            int len = get_bits(&s->gb, 4);
1103
            num_fill_bits = get_bits(&s->gb, len) + 1;
1104
        }
1105

    
1106
        if (num_fill_bits >= 0) {
1107
            if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1108
                av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1109
                return AVERROR_INVALIDDATA;
1110
            }
1111

    
1112
            skip_bits_long(&s->gb, num_fill_bits);
1113
        }
1114
    }
1115

    
1116
    /** no idea for what the following bit is used */
1117
    if (get_bits1(&s->gb)) {
1118
        av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1119
        return AVERROR_INVALIDDATA;
1120
    }
1121

    
1122

    
1123
    if (decode_channel_transform(s) < 0)
1124
        return AVERROR_INVALIDDATA;
1125

    
1126

    
1127
    for (i = 0; i < s->channels_for_cur_subframe; i++) {
1128
        int c = s->channel_indexes_for_cur_subframe[i];
1129
        if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1130
            transmit_coeffs = 1;
1131
    }
1132

    
1133
    if (transmit_coeffs) {
1134
        int step;
1135
        int quant_step = 90 * s->bits_per_sample >> 4;
1136
        if ((get_bits1(&s->gb))) {
1137
            /** FIXME: might change run level mode decision */
1138
            av_log_ask_for_sample(s->avctx, "unsupported quant step coding\n");
1139
            return AVERROR_INVALIDDATA;
1140
        }
1141
        /** decode quantization step */
1142
        step = get_sbits(&s->gb, 6);
1143
        quant_step += step;
1144
        if (step == -32 || step == 31) {
1145
            const int sign = (step == 31) - 1;
1146
            int quant = 0;
1147
            while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1148
                   (step = get_bits(&s->gb, 5)) == 31) {
1149
                quant += 31;
1150
            }
1151
            quant_step += ((quant + step) ^ sign) - sign;
1152
        }
1153
        if (quant_step < 0) {
1154
            av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1155
        }
1156

    
1157
        /** decode quantization step modifiers for every channel */
1158

    
1159
        if (s->channels_for_cur_subframe == 1) {
1160
            s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1161
        } else {
1162
            int modifier_len = get_bits(&s->gb, 3);
1163
            for (i = 0; i < s->channels_for_cur_subframe; i++) {
1164
                int c = s->channel_indexes_for_cur_subframe[i];
1165
                s->channel[c].quant_step = quant_step;
1166
                if (get_bits1(&s->gb)) {
1167
                    if (modifier_len) {
1168
                        s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1169
                    } else
1170
                        ++s->channel[c].quant_step;
1171
                }
1172
            }
1173
        }
1174

    
1175
        /** decode scale factors */
1176
        if (decode_scale_factors(s) < 0)
1177
            return AVERROR_INVALIDDATA;
1178
    }
1179

    
1180
    dprintf(s->avctx, "BITSTREAM: subframe header length was %i\n",
1181
            get_bits_count(&s->gb) - s->subframe_offset);
1182

    
1183
    /** parse coefficients */
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 &&
1187
            get_bits_count(&s->gb) < s->num_saved_bits) {
1188
            decode_coeffs(s, c);
1189
        } else
1190
            memset(s->channel[c].coeffs, 0,
1191
                   sizeof(*s->channel[c].coeffs) * subframe_len);
1192
    }
1193

    
1194
    dprintf(s->avctx, "BITSTREAM: subframe length was %i\n",
1195
            get_bits_count(&s->gb) - s->subframe_offset);
1196

    
1197
    if (transmit_coeffs) {
1198
        /** reconstruct the per channel data */
1199
        inverse_channel_transform(s);
1200
        for (i = 0; i < s->channels_for_cur_subframe; i++) {
1201
            int c = s->channel_indexes_for_cur_subframe[i];
1202
            const int* sf = s->channel[c].scale_factors;
1203
            int b;
1204

    
1205
            if (c == s->lfe_channel)
1206
                memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1207
                       (subframe_len - cur_subwoofer_cutoff));
1208

    
1209
            /** inverse quantization and rescaling */
1210
            for (b = 0; b < s->num_bands; b++) {
1211
                const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1212
                const int exp = s->channel[c].quant_step -
1213
                            (s->channel[c].max_scale_factor - *sf++) *
1214
                            s->channel[c].scale_factor_step;
1215
                const float quant = pow(10.0, exp / 20.0);
1216
                int start;
1217

    
1218
                for (start = s->cur_sfb_offsets[b]; start < end; start++)
1219
                    s->tmp[start] = s->channel[c].coeffs[start] * quant;
1220
            }
1221

    
1222
            /** apply imdct (ff_imdct_half == DCTIV with reverse) */
1223
            ff_imdct_half(&s->mdct_ctx[av_log2(subframe_len) - BLOCK_MIN_BITS],
1224
                          s->channel[c].coeffs, s->tmp);
1225
        }
1226
    }
1227

    
1228
    /** window and overlapp-add */
1229
    wmapro_window(s);
1230

    
1231
    /** handled one subframe */
1232
    for (i = 0; i < s->channels_for_cur_subframe; i++) {
1233
        int c = s->channel_indexes_for_cur_subframe[i];
1234
        if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1235
            av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1236
            return AVERROR_INVALIDDATA;
1237
        }
1238
        ++s->channel[c].cur_subframe;
1239
    }
1240

    
1241
    return 0;
1242
}
1243

    
1244
/**
1245
 *@brief Decode one WMA frame.
1246
 *@param s codec context
1247
 *@return 0 if the trailer bit indicates that this is the last frame,
1248
 *        1 if there are additional frames
1249
 */
1250
static int decode_frame(WMAProDecodeCtx *s)
1251
{
1252
    GetBitContext* gb = &s->gb;
1253
    int more_frames = 0;
1254
    int len = 0;
1255
    int i;
1256

    
1257
    /** check for potential output buffer overflow */
1258
    if (s->num_channels * s->samples_per_frame > s->samples_end - s->samples) {
1259
        av_log(s->avctx, AV_LOG_ERROR,
1260
               "not enough space for the output samples\n");
1261
        s->packet_loss = 1;
1262
        return 0;
1263
    }
1264

    
1265
    /** get frame length */
1266
    if (s->len_prefix)
1267
        len = get_bits(gb, s->log2_frame_size);
1268

    
1269
    dprintf(s->avctx, "decoding frame with length %x\n", len);
1270

    
1271
    /** decode tile information */
1272
    if (decode_tilehdr(s)) {
1273
        s->packet_loss = 1;
1274
        return 0;
1275
    }
1276

    
1277
    /** read postproc transform */
1278
    if (s->num_channels > 1 && get_bits1(gb)) {
1279
        av_log_ask_for_sample(s->avctx, "Unsupported postproc transform found\n");
1280
        s->packet_loss = 1;
1281
        return 0;
1282
    }
1283

    
1284
    /** read drc info */
1285
    if (s->dynamic_range_compression) {
1286
        s->drc_gain = get_bits(gb, 8);
1287
        dprintf(s->avctx, "drc_gain %i\n", s->drc_gain);
1288
    }
1289

    
1290
    /** no idea what these are for, might be the number of samples
1291
        that need to be skipped at the beginning or end of a stream */
1292
    if (get_bits1(gb)) {
1293
        int skip;
1294

    
1295
        /** usually true for the first frame */
1296
        if (get_bits1(gb)) {
1297
            skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1298
            dprintf(s->avctx, "start skip: %i\n", skip);
1299
        }
1300

    
1301
        /** sometimes true for the last frame */
1302
        if (get_bits1(gb)) {
1303
            skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1304
            dprintf(s->avctx, "end skip: %i\n", skip);
1305
        }
1306

    
1307
    }
1308

    
1309
    dprintf(s->avctx, "BITSTREAM: frame header length was %i\n",
1310
            get_bits_count(gb) - s->frame_offset);
1311

    
1312
    /** reset subframe states */
1313
    s->parsed_all_subframes = 0;
1314
    for (i = 0; i < s->num_channels; i++) {
1315
        s->channel[i].decoded_samples = 0;
1316
        s->channel[i].cur_subframe    = 0;
1317
        s->channel[i].reuse_sf        = 0;
1318
    }
1319

    
1320
    /** decode all subframes */
1321
    while (!s->parsed_all_subframes) {
1322
        if (decode_subframe(s) < 0) {
1323
            s->packet_loss = 1;
1324
            return 0;
1325
        }
1326
    }
1327

    
1328
    /** interleave samples and write them to the output buffer */
1329
    for (i = 0; i < s->num_channels; i++) {
1330
        float* ptr;
1331
        int incr = s->num_channels;
1332
        float* iptr = s->channel[i].out;
1333
        int x;
1334

    
1335
        ptr = s->samples + i;
1336

    
1337
        for (x = 0; x < s->samples_per_frame; x++) {
1338
            *ptr = av_clipf(*iptr++, -1.0, 32767.0 / 32768.0);
1339
            ptr += incr;
1340
        }
1341

    
1342
        /** reuse second half of the IMDCT output for the next frame */
1343
        memcpy(&s->channel[i].out[0],
1344
               &s->channel[i].out[s->samples_per_frame],
1345
               s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1346
    }
1347

    
1348
    if (s->skip_frame) {
1349
        s->skip_frame = 0;
1350
    } else
1351
        s->samples += s->num_channels * s->samples_per_frame;
1352

    
1353
    if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1354
        /** FIXME: not sure if this is always an error */
1355
        av_log(s->avctx, AV_LOG_ERROR, "frame[%i] would have to skip %i bits\n",
1356
               s->frame_num, len - (get_bits_count(gb) - s->frame_offset) - 1);
1357
        s->packet_loss = 1;
1358
        return 0;
1359
    }
1360

    
1361
    /** skip the rest of the frame data */
1362
    skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1363

    
1364
    /** decode trailer bit */
1365
    more_frames = get_bits1(gb);
1366

    
1367
    ++s->frame_num;
1368
    return more_frames;
1369
}
1370

    
1371
/**
1372
 *@brief Calculate remaining input buffer length.
1373
 *@param s codec context
1374
 *@param gb bitstream reader context
1375
 *@return remaining size in bits
1376
 */
1377
static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1378
{
1379
    return s->buf_bit_size - get_bits_count(gb);
1380
}
1381

    
1382
/**
1383
 *@brief Fill the bit reservoir with a (partial) frame.
1384
 *@param s codec context
1385
 *@param gb bitstream reader context
1386
 *@param len length of the partial frame
1387
 *@param append decides wether to reset the buffer or not
1388
 */
1389
static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1390
                      int append)
1391
{
1392
    int buflen;
1393

    
1394
    /** when the frame data does not need to be concatenated, the input buffer
1395
        is resetted and additional bits from the previous frame are copyed
1396
        and skipped later so that a fast byte copy is possible */
1397

    
1398
    if (!append) {
1399
        s->frame_offset = get_bits_count(gb) & 7;
1400
        s->num_saved_bits = s->frame_offset;
1401
        init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1402
    }
1403

    
1404
    buflen = (s->num_saved_bits + len + 8) >> 3;
1405

    
1406
    if (len <= 0 || buflen > MAX_FRAMESIZE) {
1407
        av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1408
        s->packet_loss = 1;
1409
        return;
1410
    }
1411

    
1412
    s->num_saved_bits += len;
1413
    if (!append) {
1414
        ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1415
                     s->num_saved_bits);
1416
    } else {
1417
        int align = 8 - (get_bits_count(gb) & 7);
1418
        align = FFMIN(align, len);
1419
        put_bits(&s->pb, align, get_bits(gb, align));
1420
        len -= align;
1421
        ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1422
    }
1423
    skip_bits_long(gb, len);
1424

    
1425
    {
1426
        PutBitContext tmp = s->pb;
1427
        flush_put_bits(&tmp);
1428
    }
1429

    
1430
    init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1431
    skip_bits(&s->gb, s->frame_offset);
1432
}
1433

    
1434
/**
1435
 *@brief Decode a single WMA packet.
1436
 *@param avctx codec context
1437
 *@param data the output buffer
1438
 *@param data_size number of bytes that were written to the output buffer
1439
 *@param avpkt input packet
1440
 *@return number of bytes that were read from the input buffer
1441
 */
1442
static int decode_packet(AVCodecContext *avctx,
1443
                         void *data, int *data_size, AVPacket* avpkt)
1444
{
1445
    WMAProDecodeCtx *s = avctx->priv_data;
1446
    GetBitContext* gb    = &s->pgb;
1447
    const uint8_t* buf   = avpkt->data;
1448
    int buf_size         = avpkt->size;
1449
    int more_frames      = 1;
1450
    int num_bits_prev_frame;
1451
    int packet_sequence_number;
1452

    
1453
    s->samples      = data;
1454
    s->samples_end  = (float*)((int8_t*)data + *data_size);
1455
    s->buf_bit_size = buf_size << 3;
1456

    
1457

    
1458
    *data_size = 0;
1459

    
1460
    /** sanity check for the buffer length */
1461
    if (buf_size < avctx->block_align)
1462
        return 0;
1463

    
1464
    buf_size = avctx->block_align;
1465

    
1466
    /** parse packet header */
1467
    init_get_bits(gb, buf, s->buf_bit_size);
1468
    packet_sequence_number = get_bits(gb, 4);
1469
    skip_bits(gb, 2);
1470

    
1471
    /** get number of bits that need to be added to the previous frame */
1472
    num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1473
    dprintf(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1474
            num_bits_prev_frame);
1475

    
1476
    /** check for packet loss */
1477
    if (!s->packet_loss &&
1478
        ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1479
        s->packet_loss = 1;
1480
        av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1481
               s->packet_sequence_number, packet_sequence_number);
1482
    }
1483
    s->packet_sequence_number = packet_sequence_number;
1484

    
1485
    if (num_bits_prev_frame > 0) {
1486
        /** append the previous frame data to the remaining data from the
1487
            previous packet to create a full frame */
1488
        save_bits(s, gb, num_bits_prev_frame, 1);
1489
        dprintf(avctx, "accumulated %x bits of frame data\n",
1490
                s->num_saved_bits - s->frame_offset);
1491

    
1492
        /** decode the cross packet frame if it is valid */
1493
        if (!s->packet_loss)
1494
            decode_frame(s);
1495
    } else if (s->num_saved_bits - s->frame_offset) {
1496
        dprintf(avctx, "ignoring %x previously saved bits\n",
1497
                s->num_saved_bits - s->frame_offset);
1498
    }
1499

    
1500
    s->packet_loss = 0;
1501
    /** decode the rest of the packet */
1502
    while (!s->packet_loss && more_frames &&
1503
           remaining_bits(s, gb) > s->log2_frame_size) {
1504
        int frame_size = show_bits(gb, s->log2_frame_size);
1505

    
1506
        /** there is enough data for a full frame */
1507
        if (remaining_bits(s, gb) >= frame_size && frame_size > 0) {
1508
            save_bits(s, gb, frame_size, 0);
1509

    
1510
            /** decode the frame */
1511
            more_frames = decode_frame(s);
1512

    
1513
            if (!more_frames) {
1514
                dprintf(avctx, "no more frames\n");
1515
            }
1516
        } else
1517
            more_frames = 0;
1518
    }
1519

    
1520
    if (!s->packet_loss && remaining_bits(s, gb) > 0) {
1521
        /** save the rest of the data so that it can be decoded
1522
            with the next packet */
1523
        save_bits(s, gb, remaining_bits(s, gb), 0);
1524
    }
1525

    
1526
    *data_size = (int8_t *)s->samples - (int8_t *)data;
1527

    
1528
    return avctx->block_align;
1529
}
1530

    
1531
/**
1532
 *@brief Clear decoder buffers (for seeking).
1533
 *@param avctx codec context
1534
 */
1535
static void flush(AVCodecContext *avctx)
1536
{
1537
    WMAProDecodeCtx *s = avctx->priv_data;
1538
    int i;
1539
    /** reset output buffer as a part of it is used during the windowing of a
1540
        new frame */
1541
    for (i = 0; i < s->num_channels; i++)
1542
        memset(s->channel[i].out, 0, s->samples_per_frame *
1543
               sizeof(*s->channel[i].out));
1544
    s->packet_loss = 1;
1545
}
1546

    
1547

    
1548
/**
1549
 *@brief wmapro decoder
1550
 */
1551
AVCodec wmapro_decoder = {
1552
    "wmapro",
1553
    CODEC_TYPE_AUDIO,
1554
    CODEC_ID_WMAPRO,
1555
    sizeof(WMAProDecodeCtx),
1556
    decode_init,
1557
    NULL,
1558
    decode_end,
1559
    decode_packet,
1560
    .flush= flush,
1561
    .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1562
};