2 * Wmapro compatible decoder
3 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson
6 * This file is part of Libav.
8 * Libav 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.
13 * Libav 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.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
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
32 * - windowing and overlapp-add
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
40 * The number of samples and a few other decode flags are stored
41 * as extradata that has to be passed to the decoder.
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.
47 * Example wmapro bitstream (in samples):
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 * ---------------------------------------------------
58 * The frame layouts for the individual channels of a wma frame does not need
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.
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
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
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.
89 #include "libavutil/float_dsp.h"
90 #include "libavutil/intfloat.h"
91 #include "libavutil/intreadwrite.h"
96 #include "wmaprodata.h"
99 #include "wma_common.h"
101 /** current decoder limitations */
102 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
103 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
104 #define MAX_BANDS 29 ///< max number of scale factor bands
105 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
107 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
108 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
109 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
110 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
114 #define SCALEVLCBITS 8
115 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
116 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
117 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
118 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
119 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
121 static VLC sf_vlc; ///< scale factor DPCM vlc
122 static VLC sf_rl_vlc; ///< scale factor run length vlc
123 static VLC vec4_vlc; ///< 4 coefficients per symbol
124 static VLC vec2_vlc; ///< 2 coefficients per symbol
125 static VLC vec1_vlc; ///< 1 coefficient per symbol
126 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
127 static float sin64[33]; ///< sinus table for decorrelation
130 * @brief frame specific decoder context for a single channel
133 int16_t prev_block_len; ///< length of the previous block
134 uint8_t transmit_coefs;
135 uint8_t num_subframes;
136 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
137 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
138 uint8_t cur_subframe; ///< current subframe number
139 uint16_t decoded_samples; ///< number of already processed samples
140 uint8_t grouped; ///< channel is part of a group
141 int quant_step; ///< quantization step for the current subframe
142 int8_t reuse_sf; ///< share scale factors between subframes
143 int8_t scale_factor_step; ///< scaling step for the current subframe
144 int max_scale_factor; ///< maximum scale factor for the current subframe
145 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
146 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
147 int* scale_factors; ///< pointer to the scale factor values used for decoding
148 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
149 float* coeffs; ///< pointer to the subframe decode buffer
150 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
151 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
155 * @brief channel group for channel transformations
158 uint8_t num_channels; ///< number of channels in the group
159 int8_t transform; ///< transform on / off
160 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
161 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
162 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
166 * @brief main decoder context
168 typedef struct WMAProDecodeCtx {
169 /* generic decoder variables */
170 AVCodecContext* avctx; ///< codec context for av_log
171 AVFloatDSPContext fdsp;
172 uint8_t frame_data[MAX_FRAMESIZE +
173 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
174 PutBitContext pb; ///< context for filling the frame_data buffer
175 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
176 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
177 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
179 /* frame size dependent frame information (set during initialization) */
180 uint32_t decode_flags; ///< used compression features
181 uint8_t len_prefix; ///< frame is prefixed with its length
182 uint8_t dynamic_range_compression; ///< frame contains DRC data
183 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
184 uint16_t samples_per_frame; ///< number of samples to output
185 uint16_t log2_frame_size;
186 int8_t lfe_channel; ///< lfe channel index
187 uint8_t max_num_subframes;
188 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
189 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
190 uint16_t min_samples_per_subframe;
191 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
192 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
193 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
194 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
196 /* packet decode state */
197 GetBitContext pgb; ///< bitstream reader context for the packet
198 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
199 uint8_t packet_offset; ///< frame offset in the packet
200 uint8_t packet_sequence_number; ///< current packet number
201 int num_saved_bits; ///< saved number of bits
202 int frame_offset; ///< frame offset in the bit reservoir
203 int subframe_offset; ///< subframe offset in the bit reservoir
204 uint8_t packet_loss; ///< set in case of bitstream error
205 uint8_t packet_done; ///< set when a packet is fully decoded
207 /* frame decode state */
208 uint32_t frame_num; ///< current frame number (not used for decoding)
209 GetBitContext gb; ///< bitstream reader context
210 int buf_bit_size; ///< buffer size in bits
211 uint8_t drc_gain; ///< gain for the DRC tool
212 int8_t skip_frame; ///< skip output step
213 int8_t parsed_all_subframes; ///< all subframes decoded?
215 /* subframe/block decode state */
216 int16_t subframe_len; ///< current subframe length
217 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
218 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
219 int8_t num_bands; ///< number of scale factor bands
220 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
221 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
222 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
223 int8_t esc_len; ///< length of escaped coefficients
225 uint8_t num_chgroups; ///< number of channel groups
226 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
228 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
233 *@brief helper function to print the most important members of the context
236 static av_cold void dump_context(WMAProDecodeCtx *s)
238 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
239 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
241 PRINT("ed sample bit depth", s->bits_per_sample);
242 PRINT_HEX("ed decode flags", s->decode_flags);
243 PRINT("samples per frame", s->samples_per_frame);
244 PRINT("log2 frame size", s->log2_frame_size);
245 PRINT("max num subframes", s->max_num_subframes);
246 PRINT("len prefix", s->len_prefix);
247 PRINT("num channels", s->avctx->channels);
251 *@brief Uninitialize the decoder and free all resources.
252 *@param avctx codec context
253 *@return 0 on success, < 0 otherwise
255 static av_cold int decode_end(AVCodecContext *avctx)
257 WMAProDecodeCtx *s = avctx->priv_data;
260 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
261 ff_mdct_end(&s->mdct_ctx[i]);
267 *@brief Initialize the decoder.
268 *@param avctx codec context
269 *@return 0 on success, -1 otherwise
271 static av_cold int decode_init(AVCodecContext *avctx)
273 WMAProDecodeCtx *s = avctx->priv_data;
274 uint8_t *edata_ptr = avctx->extradata;
275 unsigned int channel_mask;
277 int log2_max_num_subframes;
278 int num_possible_block_sizes;
280 if (!avctx->block_align) {
281 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
282 return AVERROR(EINVAL);
286 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
288 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
290 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
292 if (avctx->extradata_size >= 18) {
293 s->decode_flags = AV_RL16(edata_ptr+14);
294 channel_mask = AV_RL32(edata_ptr+2);
295 s->bits_per_sample = AV_RL16(edata_ptr);
296 /** dump the extradata */
297 for (i = 0; i < avctx->extradata_size; i++)
298 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
299 av_dlog(avctx, "\n");
302 avpriv_request_sample(avctx, "Unknown extradata size");
303 return AVERROR_PATCHWELCOME;
307 s->log2_frame_size = av_log2(avctx->block_align) + 4;
310 s->skip_frame = 1; /* skip first frame */
312 s->len_prefix = (s->decode_flags & 0x40);
315 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
316 if (bits > WMAPRO_BLOCK_MAX_BITS) {
317 avpriv_request_sample(avctx, "14-bit block sizes");
318 return AVERROR_PATCHWELCOME;
320 s->samples_per_frame = 1 << bits;
323 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
324 s->max_num_subframes = 1 << log2_max_num_subframes;
325 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
326 s->max_subframe_len_bit = 1;
327 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
329 num_possible_block_sizes = log2_max_num_subframes + 1;
330 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
331 s->dynamic_range_compression = (s->decode_flags & 0x80);
333 if (s->max_num_subframes > MAX_SUBFRAMES) {
334 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
335 s->max_num_subframes);
336 return AVERROR_INVALIDDATA;
339 if (s->avctx->sample_rate <= 0) {
340 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
341 return AVERROR_INVALIDDATA;
344 if (avctx->channels < 0) {
345 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
347 return AVERROR_INVALIDDATA;
348 } else if (avctx->channels > WMAPRO_MAX_CHANNELS) {
349 avpriv_request_sample(avctx,
350 "More than %d channels", WMAPRO_MAX_CHANNELS);
351 return AVERROR_PATCHWELCOME;
354 /** init previous block len */
355 for (i = 0; i < avctx->channels; i++)
356 s->channel[i].prev_block_len = s->samples_per_frame;
358 /** extract lfe channel position */
361 if (channel_mask & 8) {
363 for (mask = 1; mask < 16; mask <<= 1) {
364 if (channel_mask & mask)
369 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
370 scale_huffbits, 1, 1,
371 scale_huffcodes, 2, 2, 616);
373 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
374 scale_rl_huffbits, 1, 1,
375 scale_rl_huffcodes, 4, 4, 1406);
377 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
378 coef0_huffbits, 1, 1,
379 coef0_huffcodes, 4, 4, 2108);
381 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
382 coef1_huffbits, 1, 1,
383 coef1_huffcodes, 4, 4, 3912);
385 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
387 vec4_huffcodes, 2, 2, 604);
389 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
391 vec2_huffcodes, 2, 2, 562);
393 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
395 vec1_huffcodes, 2, 2, 562);
397 /** calculate number of scale factor bands and their offsets
398 for every possible block size */
399 for (i = 0; i < num_possible_block_sizes; i++) {
400 int subframe_len = s->samples_per_frame >> i;
404 s->sfb_offsets[i][0] = 0;
406 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
407 int offset = (subframe_len * 2 * critical_freq[x])
408 / s->avctx->sample_rate + 2;
410 if (offset > s->sfb_offsets[i][band - 1])
411 s->sfb_offsets[i][band++] = offset;
413 s->sfb_offsets[i][band - 1] = subframe_len;
414 s->num_sfb[i] = band - 1;
418 /** Scale factors can be shared between blocks of different size
419 as every block has a different scale factor band layout.
420 The matrix sf_offsets is needed to find the correct scale factor.
423 for (i = 0; i < num_possible_block_sizes; i++) {
425 for (b = 0; b < s->num_sfb[i]; b++) {
427 int offset = ((s->sfb_offsets[i][b]
428 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
429 for (x = 0; x < num_possible_block_sizes; x++) {
431 while (s->sfb_offsets[x][v + 1] << x < offset)
433 s->sf_offsets[i][x][b] = v;
438 /** init MDCT, FIXME: only init needed sizes */
439 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
440 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
441 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
442 / (1 << (s->bits_per_sample - 1)));
444 /** init MDCT windows: simple sinus window */
445 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
446 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
447 ff_init_ff_sine_windows(win_idx);
448 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
451 /** calculate subwoofer cutoff values */
452 for (i = 0; i < num_possible_block_sizes; i++) {
453 int block_size = s->samples_per_frame >> i;
454 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
455 / s->avctx->sample_rate;
456 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
459 /** calculate sine values for the decorrelation matrix */
460 for (i = 0; i < 33; i++)
461 sin64[i] = sin(i*M_PI / 64.0);
463 if (avctx->debug & FF_DEBUG_BITSTREAM)
466 avctx->channel_layout = channel_mask;
472 *@brief Decode the subframe length.
474 *@param offset sample offset in the frame
475 *@return decoded subframe length on success, < 0 in case of an error
477 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
479 int frame_len_shift = 0;
482 /** no need to read from the bitstream when only one length is possible */
483 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
484 return s->min_samples_per_subframe;
486 /** 1 bit indicates if the subframe is of maximum length */
487 if (s->max_subframe_len_bit) {
488 if (get_bits1(&s->gb))
489 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
491 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
493 subframe_len = s->samples_per_frame >> frame_len_shift;
495 /** sanity check the length */
496 if (subframe_len < s->min_samples_per_subframe ||
497 subframe_len > s->samples_per_frame) {
498 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
500 return AVERROR_INVALIDDATA;
506 *@brief Decode how the data in the frame is split into subframes.
507 * Every WMA frame contains the encoded data for a fixed number of
508 * samples per channel. The data for every channel might be split
509 * into several subframes. This function will reconstruct the list of
510 * subframes for every channel.
512 * If the subframes are not evenly split, the algorithm estimates the
513 * channels with the lowest number of total samples.
514 * Afterwards, for each of these channels a bit is read from the
515 * bitstream that indicates if the channel contains a subframe with the
516 * next subframe size that is going to be read from the bitstream or not.
517 * If a channel contains such a subframe, the subframe size gets added to
518 * the channel's subframe list.
519 * The algorithm repeats these steps until the frame is properly divided
520 * between the individual channels.
523 *@return 0 on success, < 0 in case of an error
525 static int decode_tilehdr(WMAProDecodeCtx *s)
527 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
528 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
529 int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */
530 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
531 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
534 /* Should never consume more than 3073 bits (256 iterations for the
535 * while loop when always the minimum amount of 128 samples is subtracted
536 * from missing samples in the 8 channel case).
537 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
540 /** reset tiling information */
541 for (c = 0; c < s->avctx->channels; c++)
542 s->channel[c].num_subframes = 0;
544 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
545 fixed_channel_layout = 1;
547 /** loop until the frame data is split between the subframes */
551 /** check which channels contain the subframe */
552 for (c = 0; c < s->avctx->channels; c++) {
553 if (num_samples[c] == min_channel_len) {
554 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
555 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
556 contains_subframe[c] = 1;
558 contains_subframe[c] = get_bits1(&s->gb);
560 contains_subframe[c] = 0;
563 /** get subframe length, subframe_len == 0 is not allowed */
564 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
565 return AVERROR_INVALIDDATA;
567 /** add subframes to the individual channels and find new min_channel_len */
568 min_channel_len += subframe_len;
569 for (c = 0; c < s->avctx->channels; c++) {
570 WMAProChannelCtx* chan = &s->channel[c];
572 if (contains_subframe[c]) {
573 if (chan->num_subframes >= MAX_SUBFRAMES) {
574 av_log(s->avctx, AV_LOG_ERROR,
575 "broken frame: num subframes > 31\n");
576 return AVERROR_INVALIDDATA;
578 chan->subframe_len[chan->num_subframes] = subframe_len;
579 num_samples[c] += subframe_len;
580 ++chan->num_subframes;
581 if (num_samples[c] > s->samples_per_frame) {
582 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
583 "channel len > samples_per_frame\n");
584 return AVERROR_INVALIDDATA;
586 } else if (num_samples[c] <= min_channel_len) {
587 if (num_samples[c] < min_channel_len) {
588 channels_for_cur_subframe = 0;
589 min_channel_len = num_samples[c];
591 ++channels_for_cur_subframe;
594 } while (min_channel_len < s->samples_per_frame);
596 for (c = 0; c < s->avctx->channels; c++) {
599 for (i = 0; i < s->channel[c].num_subframes; i++) {
600 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
601 " len %i\n", s->frame_num, c, i,
602 s->channel[c].subframe_len[i]);
603 s->channel[c].subframe_offset[i] = offset;
604 offset += s->channel[c].subframe_len[i];
612 *@brief Calculate a decorrelation matrix from the bitstream parameters.
613 *@param s codec context
614 *@param chgroup channel group for which the matrix needs to be calculated
616 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
617 WMAProChannelGrp *chgroup)
621 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
622 memset(chgroup->decorrelation_matrix, 0, s->avctx->channels *
623 s->avctx->channels * sizeof(*chgroup->decorrelation_matrix));
625 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
626 rotation_offset[i] = get_bits(&s->gb, 6);
628 for (i = 0; i < chgroup->num_channels; i++)
629 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
630 get_bits1(&s->gb) ? 1.0 : -1.0;
632 for (i = 1; i < chgroup->num_channels; i++) {
634 for (x = 0; x < i; x++) {
636 for (y = 0; y < i + 1; y++) {
637 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
638 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
639 int n = rotation_offset[offset + x];
645 cosv = sin64[32 - n];
647 sinv = sin64[64 - n];
648 cosv = -sin64[n - 32];
651 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
652 (v1 * sinv) - (v2 * cosv);
653 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
654 (v1 * cosv) + (v2 * sinv);
662 *@brief Decode channel transformation parameters
663 *@param s codec context
664 *@return 0 in case of success, < 0 in case of bitstream errors
666 static int decode_channel_transform(WMAProDecodeCtx* s)
669 /* should never consume more than 1921 bits for the 8 channel case
670 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
671 * + MAX_CHANNELS + MAX_BANDS + 1)
674 /** in the one channel case channel transforms are pointless */
676 if (s->avctx->channels > 1) {
677 int remaining_channels = s->channels_for_cur_subframe;
679 if (get_bits1(&s->gb)) {
680 avpriv_request_sample(s->avctx,
681 "Channel transform bit");
682 return AVERROR_PATCHWELCOME;
685 for (s->num_chgroups = 0; remaining_channels &&
686 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
687 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
688 float** channel_data = chgroup->channel_data;
689 chgroup->num_channels = 0;
690 chgroup->transform = 0;
692 /** decode channel mask */
693 if (remaining_channels > 2) {
694 for (i = 0; i < s->channels_for_cur_subframe; i++) {
695 int channel_idx = s->channel_indexes_for_cur_subframe[i];
696 if (!s->channel[channel_idx].grouped
697 && get_bits1(&s->gb)) {
698 ++chgroup->num_channels;
699 s->channel[channel_idx].grouped = 1;
700 *channel_data++ = s->channel[channel_idx].coeffs;
704 chgroup->num_channels = remaining_channels;
705 for (i = 0; i < s->channels_for_cur_subframe; i++) {
706 int channel_idx = s->channel_indexes_for_cur_subframe[i];
707 if (!s->channel[channel_idx].grouped)
708 *channel_data++ = s->channel[channel_idx].coeffs;
709 s->channel[channel_idx].grouped = 1;
713 /** decode transform type */
714 if (chgroup->num_channels == 2) {
715 if (get_bits1(&s->gb)) {
716 if (get_bits1(&s->gb)) {
717 avpriv_request_sample(s->avctx,
718 "Unknown channel transform type");
719 return AVERROR_PATCHWELCOME;
722 chgroup->transform = 1;
723 if (s->avctx->channels == 2) {
724 chgroup->decorrelation_matrix[0] = 1.0;
725 chgroup->decorrelation_matrix[1] = -1.0;
726 chgroup->decorrelation_matrix[2] = 1.0;
727 chgroup->decorrelation_matrix[3] = 1.0;
730 chgroup->decorrelation_matrix[0] = 0.70703125;
731 chgroup->decorrelation_matrix[1] = -0.70703125;
732 chgroup->decorrelation_matrix[2] = 0.70703125;
733 chgroup->decorrelation_matrix[3] = 0.70703125;
736 } else if (chgroup->num_channels > 2) {
737 if (get_bits1(&s->gb)) {
738 chgroup->transform = 1;
739 if (get_bits1(&s->gb)) {
740 decode_decorrelation_matrix(s, chgroup);
742 /** FIXME: more than 6 coupled channels not supported */
743 if (chgroup->num_channels > 6) {
744 avpriv_request_sample(s->avctx,
745 "Coupled channels > 6");
747 memcpy(chgroup->decorrelation_matrix,
748 default_decorrelation[chgroup->num_channels],
749 chgroup->num_channels * chgroup->num_channels *
750 sizeof(*chgroup->decorrelation_matrix));
756 /** decode transform on / off */
757 if (chgroup->transform) {
758 if (!get_bits1(&s->gb)) {
760 /** transform can be enabled for individual bands */
761 for (i = 0; i < s->num_bands; i++) {
762 chgroup->transform_band[i] = get_bits1(&s->gb);
765 memset(chgroup->transform_band, 1, s->num_bands);
768 remaining_channels -= chgroup->num_channels;
775 *@brief Extract the coefficients from the bitstream.
776 *@param s codec context
777 *@param c current channel number
778 *@return 0 on success, < 0 in case of bitstream errors
780 static int decode_coeffs(WMAProDecodeCtx *s, int c)
782 /* Integers 0..15 as single-precision floats. The table saves a
783 costly int to float conversion, and storing the values as
784 integers allows fast sign-flipping. */
785 static const uint32_t fval_tab[16] = {
786 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
787 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
788 0x41000000, 0x41100000, 0x41200000, 0x41300000,
789 0x41400000, 0x41500000, 0x41600000, 0x41700000,
793 WMAProChannelCtx* ci = &s->channel[c];
800 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
802 vlctable = get_bits1(&s->gb);
803 vlc = &coef_vlc[vlctable];
813 /** decode vector coefficients (consumes up to 167 bits per iteration for
814 4 vector coded large values) */
815 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
816 (cur_coeff + 3 < ci->num_vec_coeffs)) {
821 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
823 if (idx == HUFF_VEC4_SIZE - 1) {
824 for (i = 0; i < 4; i += 2) {
825 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
826 if (idx == HUFF_VEC2_SIZE - 1) {
828 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
829 if (v0 == HUFF_VEC1_SIZE - 1)
830 v0 += ff_wma_get_large_val(&s->gb);
831 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
832 if (v1 == HUFF_VEC1_SIZE - 1)
833 v1 += ff_wma_get_large_val(&s->gb);
834 vals[i ] = av_float2int(v0);
835 vals[i+1] = av_float2int(v1);
837 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
838 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
842 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
843 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
844 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
845 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
849 for (i = 0; i < 4; i++) {
851 uint32_t sign = get_bits1(&s->gb) - 1;
852 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
855 ci->coeffs[cur_coeff] = 0;
856 /** switch to run level mode when subframe_len / 128 zeros
857 were found in a row */
858 rl_mode |= (++num_zeros > s->subframe_len >> 8);
864 /** decode run level coded coefficients */
865 if (cur_coeff < s->subframe_len) {
866 memset(&ci->coeffs[cur_coeff], 0,
867 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
868 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
869 level, run, 1, ci->coeffs,
870 cur_coeff, s->subframe_len,
871 s->subframe_len, s->esc_len, 0))
872 return AVERROR_INVALIDDATA;
879 *@brief Extract scale factors from the bitstream.
880 *@param s codec context
881 *@return 0 on success, < 0 in case of bitstream errors
883 static int decode_scale_factors(WMAProDecodeCtx* s)
887 /** should never consume more than 5344 bits
888 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
891 for (i = 0; i < s->channels_for_cur_subframe; i++) {
892 int c = s->channel_indexes_for_cur_subframe[i];
895 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
896 sf_end = s->channel[c].scale_factors + s->num_bands;
898 /** resample scale factors for the new block size
899 * as the scale factors might need to be resampled several times
900 * before some new values are transmitted, a backup of the last
901 * transmitted scale factors is kept in saved_scale_factors
903 if (s->channel[c].reuse_sf) {
904 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
906 for (b = 0; b < s->num_bands; b++)
907 s->channel[c].scale_factors[b] =
908 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
911 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
913 if (!s->channel[c].reuse_sf) {
915 /** decode DPCM coded scale factors */
916 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
917 val = 45 / s->channel[c].scale_factor_step;
918 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
919 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
924 /** run level decode differences to the resampled factors */
925 for (i = 0; i < s->num_bands; i++) {
931 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
934 uint32_t code = get_bits(&s->gb, 14);
936 sign = (code & 1) - 1;
937 skip = (code & 0x3f) >> 1;
938 } else if (idx == 1) {
941 skip = scale_rl_run[idx];
942 val = scale_rl_level[idx];
943 sign = get_bits1(&s->gb)-1;
947 if (i >= s->num_bands) {
948 av_log(s->avctx, AV_LOG_ERROR,
949 "invalid scale factor coding\n");
950 return AVERROR_INVALIDDATA;
952 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
956 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
957 s->channel[c].table_idx = s->table_idx;
958 s->channel[c].reuse_sf = 1;
961 /** calculate new scale factor maximum */
962 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
963 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
964 s->channel[c].max_scale_factor =
965 FFMAX(s->channel[c].max_scale_factor, *sf);
973 *@brief Reconstruct the individual channel data.
974 *@param s codec context
976 static void inverse_channel_transform(WMAProDecodeCtx *s)
980 for (i = 0; i < s->num_chgroups; i++) {
981 if (s->chgroup[i].transform) {
982 float data[WMAPRO_MAX_CHANNELS];
983 const int num_channels = s->chgroup[i].num_channels;
984 float** ch_data = s->chgroup[i].channel_data;
985 float** ch_end = ch_data + num_channels;
986 const int8_t* tb = s->chgroup[i].transform_band;
989 /** multichannel decorrelation */
990 for (sfb = s->cur_sfb_offsets;
991 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
994 /** multiply values with the decorrelation_matrix */
995 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
996 const float* mat = s->chgroup[i].decorrelation_matrix;
997 const float* data_end = data + num_channels;
998 float* data_ptr = data;
1001 for (ch = ch_data; ch < ch_end; ch++)
1002 *data_ptr++ = (*ch)[y];
1004 for (ch = ch_data; ch < ch_end; ch++) {
1007 while (data_ptr < data_end)
1008 sum += *data_ptr++ * *mat++;
1013 } else if (s->avctx->channels == 2) {
1014 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1015 s->fdsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1016 ch_data[0] + sfb[0],
1018 s->fdsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1019 ch_data[1] + sfb[0],
1028 *@brief Apply sine window and reconstruct the output buffer.
1029 *@param s codec context
1031 static void wmapro_window(WMAProDecodeCtx *s)
1034 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1035 int c = s->channel_indexes_for_cur_subframe[i];
1037 int winlen = s->channel[c].prev_block_len;
1038 float* start = s->channel[c].coeffs - (winlen >> 1);
1040 if (s->subframe_len < winlen) {
1041 start += (winlen - s->subframe_len) >> 1;
1042 winlen = s->subframe_len;
1045 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1049 s->fdsp.vector_fmul_window(start, start, start + winlen,
1052 s->channel[c].prev_block_len = s->subframe_len;
1057 *@brief Decode a single subframe (block).
1058 *@param s codec context
1059 *@return 0 on success, < 0 when decoding failed
1061 static int decode_subframe(WMAProDecodeCtx *s)
1063 int offset = s->samples_per_frame;
1064 int subframe_len = s->samples_per_frame;
1066 int total_samples = s->samples_per_frame * s->avctx->channels;
1067 int transmit_coeffs = 0;
1068 int cur_subwoofer_cutoff;
1070 s->subframe_offset = get_bits_count(&s->gb);
1072 /** reset channel context and find the next block offset and size
1073 == the next block of the channel with the smallest number of
1076 for (i = 0; i < s->avctx->channels; i++) {
1077 s->channel[i].grouped = 0;
1078 if (offset > s->channel[i].decoded_samples) {
1079 offset = s->channel[i].decoded_samples;
1081 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1086 "processing subframe with offset %i len %i\n", offset, subframe_len);
1088 /** get a list of all channels that contain the estimated block */
1089 s->channels_for_cur_subframe = 0;
1090 for (i = 0; i < s->avctx->channels; i++) {
1091 const int cur_subframe = s->channel[i].cur_subframe;
1092 /** subtract already processed samples */
1093 total_samples -= s->channel[i].decoded_samples;
1095 /** and count if there are multiple subframes that match our profile */
1096 if (offset == s->channel[i].decoded_samples &&
1097 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1098 total_samples -= s->channel[i].subframe_len[cur_subframe];
1099 s->channel[i].decoded_samples +=
1100 s->channel[i].subframe_len[cur_subframe];
1101 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1102 ++s->channels_for_cur_subframe;
1106 /** check if the frame will be complete after processing the
1109 s->parsed_all_subframes = 1;
1112 av_dlog(s->avctx, "subframe is part of %i channels\n",
1113 s->channels_for_cur_subframe);
1115 /** calculate number of scale factor bands and their offsets */
1116 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1117 s->num_bands = s->num_sfb[s->table_idx];
1118 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1119 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1121 /** configure the decoder for the current subframe */
1122 offset += s->samples_per_frame >> 1;
1124 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1125 int c = s->channel_indexes_for_cur_subframe[i];
1127 s->channel[c].coeffs = &s->channel[c].out[offset];
1130 s->subframe_len = subframe_len;
1131 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1133 /** skip extended header if any */
1134 if (get_bits1(&s->gb)) {
1136 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1137 int len = get_bits(&s->gb, 4);
1138 num_fill_bits = get_bits(&s->gb, len) + 1;
1141 if (num_fill_bits >= 0) {
1142 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1143 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1144 return AVERROR_INVALIDDATA;
1147 skip_bits_long(&s->gb, num_fill_bits);
1151 /** no idea for what the following bit is used */
1152 if (get_bits1(&s->gb)) {
1153 avpriv_request_sample(s->avctx, "Reserved bit");
1154 return AVERROR_PATCHWELCOME;
1158 if (decode_channel_transform(s) < 0)
1159 return AVERROR_INVALIDDATA;
1162 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1163 int c = s->channel_indexes_for_cur_subframe[i];
1164 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1165 transmit_coeffs = 1;
1168 if (transmit_coeffs) {
1170 int quant_step = 90 * s->bits_per_sample >> 4;
1172 /** decode number of vector coded coefficients */
1173 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1174 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1175 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1176 int c = s->channel_indexes_for_cur_subframe[i];
1177 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1178 if (num_vec_coeffs + offset > FF_ARRAY_ELEMS(s->channel[c].out)) {
1179 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1180 return AVERROR_INVALIDDATA;
1182 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1185 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1186 int c = s->channel_indexes_for_cur_subframe[i];
1187 s->channel[c].num_vec_coeffs = s->subframe_len;
1190 /** decode quantization step */
1191 step = get_sbits(&s->gb, 6);
1193 if (step == -32 || step == 31) {
1194 const int sign = (step == 31) - 1;
1196 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1197 (step = get_bits(&s->gb, 5)) == 31) {
1200 quant_step += ((quant + step) ^ sign) - sign;
1202 if (quant_step < 0) {
1203 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1206 /** decode quantization step modifiers for every channel */
1208 if (s->channels_for_cur_subframe == 1) {
1209 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1211 int modifier_len = get_bits(&s->gb, 3);
1212 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1213 int c = s->channel_indexes_for_cur_subframe[i];
1214 s->channel[c].quant_step = quant_step;
1215 if (get_bits1(&s->gb)) {
1217 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1219 ++s->channel[c].quant_step;
1224 /** decode scale factors */
1225 if (decode_scale_factors(s) < 0)
1226 return AVERROR_INVALIDDATA;
1229 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1230 get_bits_count(&s->gb) - s->subframe_offset);
1232 /** parse coefficients */
1233 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1234 int c = s->channel_indexes_for_cur_subframe[i];
1235 if (s->channel[c].transmit_coefs &&
1236 get_bits_count(&s->gb) < s->num_saved_bits) {
1237 decode_coeffs(s, c);
1239 memset(s->channel[c].coeffs, 0,
1240 sizeof(*s->channel[c].coeffs) * subframe_len);
1243 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1244 get_bits_count(&s->gb) - s->subframe_offset);
1246 if (transmit_coeffs) {
1247 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1248 /** reconstruct the per channel data */
1249 inverse_channel_transform(s);
1250 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1251 int c = s->channel_indexes_for_cur_subframe[i];
1252 const int* sf = s->channel[c].scale_factors;
1255 if (c == s->lfe_channel)
1256 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1257 (subframe_len - cur_subwoofer_cutoff));
1259 /** inverse quantization and rescaling */
1260 for (b = 0; b < s->num_bands; b++) {
1261 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1262 const int exp = s->channel[c].quant_step -
1263 (s->channel[c].max_scale_factor - *sf++) *
1264 s->channel[c].scale_factor_step;
1265 const float quant = pow(10.0, exp / 20.0);
1266 int start = s->cur_sfb_offsets[b];
1267 s->fdsp.vector_fmul_scalar(s->tmp + start,
1268 s->channel[c].coeffs + start,
1269 quant, end - start);
1272 /** apply imdct (imdct_half == DCTIV with reverse) */
1273 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1277 /** window and overlapp-add */
1280 /** handled one subframe */
1281 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1282 int c = s->channel_indexes_for_cur_subframe[i];
1283 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1284 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1285 return AVERROR_INVALIDDATA;
1287 ++s->channel[c].cur_subframe;
1294 *@brief Decode one WMA frame.
1295 *@param s codec context
1296 *@return 0 if the trailer bit indicates that this is the last frame,
1297 * 1 if there are additional frames
1299 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1301 AVCodecContext *avctx = s->avctx;
1302 GetBitContext* gb = &s->gb;
1303 int more_frames = 0;
1307 /** get frame length */
1309 len = get_bits(gb, s->log2_frame_size);
1311 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1313 /** decode tile information */
1314 if (decode_tilehdr(s)) {
1319 /** read postproc transform */
1320 if (s->avctx->channels > 1 && get_bits1(gb)) {
1321 if (get_bits1(gb)) {
1322 for (i = 0; i < avctx->channels * avctx->channels; i++)
1327 /** read drc info */
1328 if (s->dynamic_range_compression) {
1329 s->drc_gain = get_bits(gb, 8);
1330 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1333 /** no idea what these are for, might be the number of samples
1334 that need to be skipped at the beginning or end of a stream */
1335 if (get_bits1(gb)) {
1338 /** usually true for the first frame */
1339 if (get_bits1(gb)) {
1340 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1341 av_dlog(s->avctx, "start skip: %i\n", skip);
1344 /** sometimes true for the last frame */
1345 if (get_bits1(gb)) {
1346 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1347 av_dlog(s->avctx, "end skip: %i\n", skip);
1352 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1353 get_bits_count(gb) - s->frame_offset);
1355 /** reset subframe states */
1356 s->parsed_all_subframes = 0;
1357 for (i = 0; i < avctx->channels; i++) {
1358 s->channel[i].decoded_samples = 0;
1359 s->channel[i].cur_subframe = 0;
1360 s->channel[i].reuse_sf = 0;
1363 /** decode all subframes */
1364 while (!s->parsed_all_subframes) {
1365 if (decode_subframe(s) < 0) {
1371 /* get output buffer */
1372 frame->nb_samples = s->samples_per_frame;
1373 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1374 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1379 /** copy samples to the output buffer */
1380 for (i = 0; i < avctx->channels; i++)
1381 memcpy(frame->extended_data[i], s->channel[i].out,
1382 s->samples_per_frame * sizeof(*s->channel[i].out));
1384 for (i = 0; i < avctx->channels; i++) {
1385 /** reuse second half of the IMDCT output for the next frame */
1386 memcpy(&s->channel[i].out[0],
1387 &s->channel[i].out[s->samples_per_frame],
1388 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1391 if (s->skip_frame) {
1394 av_frame_unref(frame);
1399 if (s->len_prefix) {
1400 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1401 /** FIXME: not sure if this is always an error */
1402 av_log(s->avctx, AV_LOG_ERROR,
1403 "frame[%i] would have to skip %i bits\n", s->frame_num,
1404 len - (get_bits_count(gb) - s->frame_offset) - 1);
1409 /** skip the rest of the frame data */
1410 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1412 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1416 /** decode trailer bit */
1417 more_frames = get_bits1(gb);
1424 *@brief Calculate remaining input buffer length.
1425 *@param s codec context
1426 *@param gb bitstream reader context
1427 *@return remaining size in bits
1429 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1431 return s->buf_bit_size - get_bits_count(gb);
1435 *@brief Fill the bit reservoir with a (partial) frame.
1436 *@param s codec context
1437 *@param gb bitstream reader context
1438 *@param len length of the partial frame
1439 *@param append decides whether to reset the buffer or not
1441 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1446 /** when the frame data does not need to be concatenated, the input buffer
1447 is resetted and additional bits from the previous frame are copyed
1448 and skipped later so that a fast byte copy is possible */
1451 s->frame_offset = get_bits_count(gb) & 7;
1452 s->num_saved_bits = s->frame_offset;
1453 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1456 buflen = (s->num_saved_bits + len + 8) >> 3;
1458 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1459 avpriv_request_sample(s->avctx, "Too small input buffer");
1464 if (len > put_bits_left(&s->pb)) {
1465 av_log(s->avctx, AV_LOG_ERROR,
1466 "Cannot append %d bits, only %d bits available.\n",
1467 len, put_bits_left(&s->pb));
1472 s->num_saved_bits += len;
1474 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1477 int align = 8 - (get_bits_count(gb) & 7);
1478 align = FFMIN(align, len);
1479 put_bits(&s->pb, align, get_bits(gb, align));
1481 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1483 skip_bits_long(gb, len);
1486 PutBitContext tmp = s->pb;
1487 flush_put_bits(&tmp);
1490 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1491 skip_bits(&s->gb, s->frame_offset);
1495 *@brief Decode a single WMA packet.
1496 *@param avctx codec context
1497 *@param data the output buffer
1498 *@param avpkt input packet
1499 *@return number of bytes that were read from the input buffer
1501 static int decode_packet(AVCodecContext *avctx, void *data,
1502 int *got_frame_ptr, AVPacket* avpkt)
1504 WMAProDecodeCtx *s = avctx->priv_data;
1505 GetBitContext* gb = &s->pgb;
1506 const uint8_t* buf = avpkt->data;
1507 int buf_size = avpkt->size;
1508 int num_bits_prev_frame;
1509 int packet_sequence_number;
1513 if (s->packet_done || s->packet_loss) {
1516 /** sanity check for the buffer length */
1517 if (buf_size < avctx->block_align) {
1518 av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1519 buf_size, avctx->block_align);
1520 return AVERROR_INVALIDDATA;
1523 s->next_packet_start = buf_size - avctx->block_align;
1524 buf_size = avctx->block_align;
1525 s->buf_bit_size = buf_size << 3;
1527 /** parse packet header */
1528 init_get_bits(gb, buf, s->buf_bit_size);
1529 packet_sequence_number = get_bits(gb, 4);
1532 /** get number of bits that need to be added to the previous frame */
1533 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1534 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1535 num_bits_prev_frame);
1537 /** check for packet loss */
1538 if (!s->packet_loss &&
1539 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1541 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1542 s->packet_sequence_number, packet_sequence_number);
1544 s->packet_sequence_number = packet_sequence_number;
1546 if (num_bits_prev_frame > 0) {
1547 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1548 if (num_bits_prev_frame >= remaining_packet_bits) {
1549 num_bits_prev_frame = remaining_packet_bits;
1553 /** append the previous frame data to the remaining data from the
1554 previous packet to create a full frame */
1555 save_bits(s, gb, num_bits_prev_frame, 1);
1556 av_dlog(avctx, "accumulated %x bits of frame data\n",
1557 s->num_saved_bits - s->frame_offset);
1559 /** decode the cross packet frame if it is valid */
1560 if (!s->packet_loss)
1561 decode_frame(s, data, got_frame_ptr);
1562 } else if (s->num_saved_bits - s->frame_offset) {
1563 av_dlog(avctx, "ignoring %x previously saved bits\n",
1564 s->num_saved_bits - s->frame_offset);
1567 if (s->packet_loss) {
1568 /** reset number of saved bits so that the decoder
1569 does not start to decode incomplete frames in the
1570 s->len_prefix == 0 case */
1571 s->num_saved_bits = 0;
1577 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1578 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1579 skip_bits(gb, s->packet_offset);
1580 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1581 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1582 frame_size <= remaining_bits(s, gb)) {
1583 save_bits(s, gb, frame_size, 0);
1584 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1585 } else if (!s->len_prefix
1586 && s->num_saved_bits > get_bits_count(&s->gb)) {
1587 /** when the frames do not have a length prefix, we don't know
1588 the compressed length of the individual frames
1589 however, we know what part of a new packet belongs to the
1591 therefore we save the incoming packet first, then we append
1592 the "previous frame" data from the next packet so that
1593 we get a buffer that only contains full frames */
1594 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1599 if (s->packet_done && !s->packet_loss &&
1600 remaining_bits(s, gb) > 0) {
1601 /** save the rest of the data so that it can be decoded
1602 with the next packet */
1603 save_bits(s, gb, remaining_bits(s, gb), 0);
1606 s->packet_offset = get_bits_count(gb) & 7;
1608 return AVERROR_INVALIDDATA;
1610 return get_bits_count(gb) >> 3;
1614 *@brief Clear decoder buffers (for seeking).
1615 *@param avctx codec context
1617 static void flush(AVCodecContext *avctx)
1619 WMAProDecodeCtx *s = avctx->priv_data;
1621 /** reset output buffer as a part of it is used during the windowing of a
1623 for (i = 0; i < avctx->channels; i++)
1624 memset(s->channel[i].out, 0, s->samples_per_frame *
1625 sizeof(*s->channel[i].out));
1631 *@brief wmapro decoder
1633 AVCodec ff_wmapro_decoder = {
1635 .type = AVMEDIA_TYPE_AUDIO,
1636 .id = AV_CODEC_ID_WMAPRO,
1637 .priv_data_size = sizeof(WMAProDecodeCtx),
1638 .init = decode_init,
1639 .close = decode_end,
1640 .decode = decode_packet,
1641 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1643 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1644 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1645 AV_SAMPLE_FMT_NONE },