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/intfloat.h"
90 #include "libavutil/intreadwrite.h"
95 #include "wmaprodata.h"
97 #include "fmtconvert.h"
100 #include "wma_common.h"
102 /** current decoder limitations */
103 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
104 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
105 #define MAX_BANDS 29 ///< max number of scale factor bands
106 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
108 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
109 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
110 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
111 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
115 #define SCALEVLCBITS 8
116 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
117 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
118 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
119 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
120 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
122 static VLC sf_vlc; ///< scale factor DPCM vlc
123 static VLC sf_rl_vlc; ///< scale factor run length vlc
124 static VLC vec4_vlc; ///< 4 coefficients per symbol
125 static VLC vec2_vlc; ///< 2 coefficients per symbol
126 static VLC vec1_vlc; ///< 1 coefficient per symbol
127 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
128 static float sin64[33]; ///< sinus table for decorrelation
131 * @brief frame specific decoder context for a single channel
134 int16_t prev_block_len; ///< length of the previous block
135 uint8_t transmit_coefs;
136 uint8_t num_subframes;
137 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
138 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
139 uint8_t cur_subframe; ///< current subframe number
140 uint16_t decoded_samples; ///< number of already processed samples
141 uint8_t grouped; ///< channel is part of a group
142 int quant_step; ///< quantization step for the current subframe
143 int8_t reuse_sf; ///< share scale factors between subframes
144 int8_t scale_factor_step; ///< scaling step for the current subframe
145 int max_scale_factor; ///< maximum scale factor for the current subframe
146 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
147 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
148 int* scale_factors; ///< pointer to the scale factor values used for decoding
149 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
150 float* coeffs; ///< pointer to the subframe decode buffer
151 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
152 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
156 * @brief channel group for channel transformations
159 uint8_t num_channels; ///< number of channels in the group
160 int8_t transform; ///< transform on / off
161 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
162 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
163 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
167 * @brief main decoder context
169 typedef struct WMAProDecodeCtx {
170 /* generic decoder variables */
171 AVCodecContext* avctx; ///< codec context for av_log
172 AVFrame frame; ///< AVFrame for decoded output
173 DSPContext dsp; ///< accelerated DSP functions
174 FmtConvertContext fmt_conv;
175 uint8_t frame_data[MAX_FRAMESIZE +
176 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
177 PutBitContext pb; ///< context for filling the frame_data buffer
178 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
179 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
180 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
182 /* frame size dependent frame information (set during initialization) */
183 uint32_t decode_flags; ///< used compression features
184 uint8_t len_prefix; ///< frame is prefixed with its length
185 uint8_t dynamic_range_compression; ///< frame contains DRC data
186 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
187 uint16_t samples_per_frame; ///< number of samples to output
188 uint16_t log2_frame_size;
189 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
190 int8_t lfe_channel; ///< lfe channel index
191 uint8_t max_num_subframes;
192 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
193 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
194 uint16_t min_samples_per_subframe;
195 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
196 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
197 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
198 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
200 /* packet decode state */
201 GetBitContext pgb; ///< bitstream reader context for the packet
202 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
203 uint8_t packet_offset; ///< frame offset in the packet
204 uint8_t packet_sequence_number; ///< current packet number
205 int num_saved_bits; ///< saved number of bits
206 int frame_offset; ///< frame offset in the bit reservoir
207 int subframe_offset; ///< subframe offset in the bit reservoir
208 uint8_t packet_loss; ///< set in case of bitstream error
209 uint8_t packet_done; ///< set when a packet is fully decoded
211 /* frame decode state */
212 uint32_t frame_num; ///< current frame number (not used for decoding)
213 GetBitContext gb; ///< bitstream reader context
214 int buf_bit_size; ///< buffer size in bits
215 uint8_t drc_gain; ///< gain for the DRC tool
216 int8_t skip_frame; ///< skip output step
217 int8_t parsed_all_subframes; ///< all subframes decoded?
219 /* subframe/block decode state */
220 int16_t subframe_len; ///< current subframe length
221 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
222 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
223 int8_t num_bands; ///< number of scale factor bands
224 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
225 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
226 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
227 int8_t esc_len; ///< length of escaped coefficients
229 uint8_t num_chgroups; ///< number of channel groups
230 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
232 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
237 *@brief helper function to print the most important members of the context
240 static av_cold void dump_context(WMAProDecodeCtx *s)
242 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
243 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
245 PRINT("ed sample bit depth", s->bits_per_sample);
246 PRINT_HEX("ed decode flags", s->decode_flags);
247 PRINT("samples per frame", s->samples_per_frame);
248 PRINT("log2 frame size", s->log2_frame_size);
249 PRINT("max num subframes", s->max_num_subframes);
250 PRINT("len prefix", s->len_prefix);
251 PRINT("num channels", s->num_channels);
255 *@brief Uninitialize the decoder and free all resources.
256 *@param avctx codec context
257 *@return 0 on success, < 0 otherwise
259 static av_cold int decode_end(AVCodecContext *avctx)
261 WMAProDecodeCtx *s = avctx->priv_data;
264 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
265 ff_mdct_end(&s->mdct_ctx[i]);
271 *@brief Initialize the decoder.
272 *@param avctx codec context
273 *@return 0 on success, -1 otherwise
275 static av_cold int decode_init(AVCodecContext *avctx)
277 WMAProDecodeCtx *s = avctx->priv_data;
278 uint8_t *edata_ptr = avctx->extradata;
279 unsigned int channel_mask;
281 int log2_max_num_subframes;
282 int num_possible_block_sizes;
285 ff_dsputil_init(&s->dsp, avctx);
286 ff_fmt_convert_init(&s->fmt_conv, avctx);
287 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
289 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
291 if (avctx->extradata_size >= 18) {
292 s->decode_flags = AV_RL16(edata_ptr+14);
293 channel_mask = AV_RL32(edata_ptr+2);
294 s->bits_per_sample = AV_RL16(edata_ptr);
295 /** dump the extradata */
296 for (i = 0; i < avctx->extradata_size; i++)
297 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
298 av_dlog(avctx, "\n");
301 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
302 return AVERROR_INVALIDDATA;
306 s->log2_frame_size = av_log2(avctx->block_align) + 4;
309 s->skip_frame = 1; /* skip first frame */
311 s->len_prefix = (s->decode_flags & 0x40);
314 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
315 if (bits > WMAPRO_BLOCK_MAX_BITS) {
316 av_log_missing_feature(avctx, "14-bits block sizes", 1);
317 return AVERROR_INVALIDDATA;
319 s->samples_per_frame = 1 << bits;
322 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
323 s->max_num_subframes = 1 << log2_max_num_subframes;
324 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
325 s->max_subframe_len_bit = 1;
326 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
328 num_possible_block_sizes = log2_max_num_subframes + 1;
329 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
330 s->dynamic_range_compression = (s->decode_flags & 0x80);
332 if (s->max_num_subframes > MAX_SUBFRAMES) {
333 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
334 s->max_num_subframes);
335 return AVERROR_INVALIDDATA;
338 s->num_channels = avctx->channels;
340 if (s->num_channels < 0) {
341 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
342 return AVERROR_INVALIDDATA;
343 } else if (s->num_channels > WMAPRO_MAX_CHANNELS) {
344 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
345 return AVERROR_PATCHWELCOME;
348 /** init previous block len */
349 for (i = 0; i < s->num_channels; i++)
350 s->channel[i].prev_block_len = s->samples_per_frame;
352 /** extract lfe channel position */
355 if (channel_mask & 8) {
357 for (mask = 1; mask < 16; mask <<= 1) {
358 if (channel_mask & mask)
363 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
364 scale_huffbits, 1, 1,
365 scale_huffcodes, 2, 2, 616);
367 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
368 scale_rl_huffbits, 1, 1,
369 scale_rl_huffcodes, 4, 4, 1406);
371 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
372 coef0_huffbits, 1, 1,
373 coef0_huffcodes, 4, 4, 2108);
375 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
376 coef1_huffbits, 1, 1,
377 coef1_huffcodes, 4, 4, 3912);
379 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
381 vec4_huffcodes, 2, 2, 604);
383 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
385 vec2_huffcodes, 2, 2, 562);
387 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
389 vec1_huffcodes, 2, 2, 562);
391 /** calculate number of scale factor bands and their offsets
392 for every possible block size */
393 for (i = 0; i < num_possible_block_sizes; i++) {
394 int subframe_len = s->samples_per_frame >> i;
398 s->sfb_offsets[i][0] = 0;
400 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
401 int offset = (subframe_len * 2 * critical_freq[x])
402 / s->avctx->sample_rate + 2;
404 if (offset > s->sfb_offsets[i][band - 1])
405 s->sfb_offsets[i][band++] = offset;
407 s->sfb_offsets[i][band - 1] = subframe_len;
408 s->num_sfb[i] = band - 1;
412 /** Scale factors can be shared between blocks of different size
413 as every block has a different scale factor band layout.
414 The matrix sf_offsets is needed to find the correct scale factor.
417 for (i = 0; i < num_possible_block_sizes; i++) {
419 for (b = 0; b < s->num_sfb[i]; b++) {
421 int offset = ((s->sfb_offsets[i][b]
422 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
423 for (x = 0; x < num_possible_block_sizes; x++) {
425 while (s->sfb_offsets[x][v + 1] << x < offset)
427 s->sf_offsets[i][x][b] = v;
432 /** init MDCT, FIXME: only init needed sizes */
433 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
434 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
435 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
436 / (1 << (s->bits_per_sample - 1)));
438 /** init MDCT windows: simple sinus window */
439 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
440 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
441 ff_init_ff_sine_windows(win_idx);
442 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
445 /** calculate subwoofer cutoff values */
446 for (i = 0; i < num_possible_block_sizes; i++) {
447 int block_size = s->samples_per_frame >> i;
448 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
449 / s->avctx->sample_rate;
450 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
453 /** calculate sine values for the decorrelation matrix */
454 for (i = 0; i < 33; i++)
455 sin64[i] = sin(i*M_PI / 64.0);
457 if (avctx->debug & FF_DEBUG_BITSTREAM)
460 avctx->channel_layout = channel_mask;
462 avcodec_get_frame_defaults(&s->frame);
463 avctx->coded_frame = &s->frame;
469 *@brief Decode the subframe length.
471 *@param offset sample offset in the frame
472 *@return decoded subframe length on success, < 0 in case of an error
474 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
476 int frame_len_shift = 0;
479 /** no need to read from the bitstream when only one length is possible */
480 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
481 return s->min_samples_per_subframe;
483 /** 1 bit indicates if the subframe is of maximum length */
484 if (s->max_subframe_len_bit) {
485 if (get_bits1(&s->gb))
486 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
488 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
490 subframe_len = s->samples_per_frame >> frame_len_shift;
492 /** sanity check the length */
493 if (subframe_len < s->min_samples_per_subframe ||
494 subframe_len > s->samples_per_frame) {
495 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
497 return AVERROR_INVALIDDATA;
503 *@brief Decode how the data in the frame is split into subframes.
504 * Every WMA frame contains the encoded data for a fixed number of
505 * samples per channel. The data for every channel might be split
506 * into several subframes. This function will reconstruct the list of
507 * subframes for every channel.
509 * If the subframes are not evenly split, the algorithm estimates the
510 * channels with the lowest number of total samples.
511 * Afterwards, for each of these channels a bit is read from the
512 * bitstream that indicates if the channel contains a subframe with the
513 * next subframe size that is going to be read from the bitstream or not.
514 * If a channel contains such a subframe, the subframe size gets added to
515 * the channel's subframe list.
516 * The algorithm repeats these steps until the frame is properly divided
517 * between the individual channels.
520 *@return 0 on success, < 0 in case of an error
522 static int decode_tilehdr(WMAProDecodeCtx *s)
524 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
525 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
526 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
527 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
528 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
531 /* Should never consume more than 3073 bits (256 iterations for the
532 * while loop when always the minimum amount of 128 samples is substracted
533 * from missing samples in the 8 channel case).
534 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
537 /** reset tiling information */
538 for (c = 0; c < s->num_channels; c++)
539 s->channel[c].num_subframes = 0;
541 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
542 fixed_channel_layout = 1;
544 /** loop until the frame data is split between the subframes */
548 /** check which channels contain the subframe */
549 for (c = 0; c < s->num_channels; c++) {
550 if (num_samples[c] == min_channel_len) {
551 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
552 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
553 contains_subframe[c] = 1;
555 contains_subframe[c] = get_bits1(&s->gb);
557 contains_subframe[c] = 0;
560 /** get subframe length, subframe_len == 0 is not allowed */
561 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
562 return AVERROR_INVALIDDATA;
564 /** add subframes to the individual channels and find new min_channel_len */
565 min_channel_len += subframe_len;
566 for (c = 0; c < s->num_channels; c++) {
567 WMAProChannelCtx* chan = &s->channel[c];
569 if (contains_subframe[c]) {
570 if (chan->num_subframes >= MAX_SUBFRAMES) {
571 av_log(s->avctx, AV_LOG_ERROR,
572 "broken frame: num subframes > 31\n");
573 return AVERROR_INVALIDDATA;
575 chan->subframe_len[chan->num_subframes] = subframe_len;
576 num_samples[c] += subframe_len;
577 ++chan->num_subframes;
578 if (num_samples[c] > s->samples_per_frame) {
579 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
580 "channel len > samples_per_frame\n");
581 return AVERROR_INVALIDDATA;
583 } else if (num_samples[c] <= min_channel_len) {
584 if (num_samples[c] < min_channel_len) {
585 channels_for_cur_subframe = 0;
586 min_channel_len = num_samples[c];
588 ++channels_for_cur_subframe;
591 } while (min_channel_len < s->samples_per_frame);
593 for (c = 0; c < s->num_channels; c++) {
596 for (i = 0; i < s->channel[c].num_subframes; i++) {
597 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
598 " len %i\n", s->frame_num, c, i,
599 s->channel[c].subframe_len[i]);
600 s->channel[c].subframe_offset[i] = offset;
601 offset += s->channel[c].subframe_len[i];
609 *@brief Calculate a decorrelation matrix from the bitstream parameters.
610 *@param s codec context
611 *@param chgroup channel group for which the matrix needs to be calculated
613 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
614 WMAProChannelGrp *chgroup)
618 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
619 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
620 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
622 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
623 rotation_offset[i] = get_bits(&s->gb, 6);
625 for (i = 0; i < chgroup->num_channels; i++)
626 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
627 get_bits1(&s->gb) ? 1.0 : -1.0;
629 for (i = 1; i < chgroup->num_channels; i++) {
631 for (x = 0; x < i; x++) {
633 for (y = 0; y < i + 1; y++) {
634 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
635 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
636 int n = rotation_offset[offset + x];
642 cosv = sin64[32 - n];
644 sinv = sin64[64 - n];
645 cosv = -sin64[n - 32];
648 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
649 (v1 * sinv) - (v2 * cosv);
650 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
651 (v1 * cosv) + (v2 * sinv);
659 *@brief Decode channel transformation parameters
660 *@param s codec context
661 *@return 0 in case of success, < 0 in case of bitstream errors
663 static int decode_channel_transform(WMAProDecodeCtx* s)
666 /* should never consume more than 1921 bits for the 8 channel case
667 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
668 * + MAX_CHANNELS + MAX_BANDS + 1)
671 /** in the one channel case channel transforms are pointless */
673 if (s->num_channels > 1) {
674 int remaining_channels = s->channels_for_cur_subframe;
676 if (get_bits1(&s->gb)) {
677 av_log_ask_for_sample(s->avctx,
678 "unsupported channel transform bit\n");
679 return AVERROR_INVALIDDATA;
682 for (s->num_chgroups = 0; remaining_channels &&
683 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
684 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
685 float** channel_data = chgroup->channel_data;
686 chgroup->num_channels = 0;
687 chgroup->transform = 0;
689 /** decode channel mask */
690 if (remaining_channels > 2) {
691 for (i = 0; i < s->channels_for_cur_subframe; i++) {
692 int channel_idx = s->channel_indexes_for_cur_subframe[i];
693 if (!s->channel[channel_idx].grouped
694 && get_bits1(&s->gb)) {
695 ++chgroup->num_channels;
696 s->channel[channel_idx].grouped = 1;
697 *channel_data++ = s->channel[channel_idx].coeffs;
701 chgroup->num_channels = remaining_channels;
702 for (i = 0; i < s->channels_for_cur_subframe; i++) {
703 int channel_idx = s->channel_indexes_for_cur_subframe[i];
704 if (!s->channel[channel_idx].grouped)
705 *channel_data++ = s->channel[channel_idx].coeffs;
706 s->channel[channel_idx].grouped = 1;
710 /** decode transform type */
711 if (chgroup->num_channels == 2) {
712 if (get_bits1(&s->gb)) {
713 if (get_bits1(&s->gb)) {
714 av_log_ask_for_sample(s->avctx,
715 "unsupported channel transform type\n");
718 chgroup->transform = 1;
719 if (s->num_channels == 2) {
720 chgroup->decorrelation_matrix[0] = 1.0;
721 chgroup->decorrelation_matrix[1] = -1.0;
722 chgroup->decorrelation_matrix[2] = 1.0;
723 chgroup->decorrelation_matrix[3] = 1.0;
726 chgroup->decorrelation_matrix[0] = 0.70703125;
727 chgroup->decorrelation_matrix[1] = -0.70703125;
728 chgroup->decorrelation_matrix[2] = 0.70703125;
729 chgroup->decorrelation_matrix[3] = 0.70703125;
732 } else if (chgroup->num_channels > 2) {
733 if (get_bits1(&s->gb)) {
734 chgroup->transform = 1;
735 if (get_bits1(&s->gb)) {
736 decode_decorrelation_matrix(s, chgroup);
738 /** FIXME: more than 6 coupled channels not supported */
739 if (chgroup->num_channels > 6) {
740 av_log_ask_for_sample(s->avctx,
741 "coupled channels > 6\n");
743 memcpy(chgroup->decorrelation_matrix,
744 default_decorrelation[chgroup->num_channels],
745 chgroup->num_channels * chgroup->num_channels *
746 sizeof(*chgroup->decorrelation_matrix));
752 /** decode transform on / off */
753 if (chgroup->transform) {
754 if (!get_bits1(&s->gb)) {
756 /** transform can be enabled for individual bands */
757 for (i = 0; i < s->num_bands; i++) {
758 chgroup->transform_band[i] = get_bits1(&s->gb);
761 memset(chgroup->transform_band, 1, s->num_bands);
764 remaining_channels -= chgroup->num_channels;
771 *@brief Extract the coefficients from the bitstream.
772 *@param s codec context
773 *@param c current channel number
774 *@return 0 on success, < 0 in case of bitstream errors
776 static int decode_coeffs(WMAProDecodeCtx *s, int c)
778 /* Integers 0..15 as single-precision floats. The table saves a
779 costly int to float conversion, and storing the values as
780 integers allows fast sign-flipping. */
781 static const uint32_t fval_tab[16] = {
782 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
783 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
784 0x41000000, 0x41100000, 0x41200000, 0x41300000,
785 0x41400000, 0x41500000, 0x41600000, 0x41700000,
789 WMAProChannelCtx* ci = &s->channel[c];
796 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
798 vlctable = get_bits1(&s->gb);
799 vlc = &coef_vlc[vlctable];
809 /** decode vector coefficients (consumes up to 167 bits per iteration for
810 4 vector coded large values) */
811 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
812 (cur_coeff + 3 < ci->num_vec_coeffs)) {
817 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
819 if (idx == HUFF_VEC4_SIZE - 1) {
820 for (i = 0; i < 4; i += 2) {
821 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
822 if (idx == HUFF_VEC2_SIZE - 1) {
824 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
825 if (v0 == HUFF_VEC1_SIZE - 1)
826 v0 += ff_wma_get_large_val(&s->gb);
827 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
828 if (v1 == HUFF_VEC1_SIZE - 1)
829 v1 += ff_wma_get_large_val(&s->gb);
830 vals[i ] = av_float2int(v0);
831 vals[i+1] = av_float2int(v1);
833 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
834 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
838 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
839 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
840 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
841 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
845 for (i = 0; i < 4; i++) {
847 uint32_t sign = get_bits1(&s->gb) - 1;
848 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
851 ci->coeffs[cur_coeff] = 0;
852 /** switch to run level mode when subframe_len / 128 zeros
853 were found in a row */
854 rl_mode |= (++num_zeros > s->subframe_len >> 8);
860 /** decode run level coded coefficients */
861 if (cur_coeff < s->subframe_len) {
862 memset(&ci->coeffs[cur_coeff], 0,
863 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
864 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
865 level, run, 1, ci->coeffs,
866 cur_coeff, s->subframe_len,
867 s->subframe_len, s->esc_len, 0))
868 return AVERROR_INVALIDDATA;
875 *@brief Extract scale factors from the bitstream.
876 *@param s codec context
877 *@return 0 on success, < 0 in case of bitstream errors
879 static int decode_scale_factors(WMAProDecodeCtx* s)
883 /** should never consume more than 5344 bits
884 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
887 for (i = 0; i < s->channels_for_cur_subframe; i++) {
888 int c = s->channel_indexes_for_cur_subframe[i];
891 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
892 sf_end = s->channel[c].scale_factors + s->num_bands;
894 /** resample scale factors for the new block size
895 * as the scale factors might need to be resampled several times
896 * before some new values are transmitted, a backup of the last
897 * transmitted scale factors is kept in saved_scale_factors
899 if (s->channel[c].reuse_sf) {
900 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
902 for (b = 0; b < s->num_bands; b++)
903 s->channel[c].scale_factors[b] =
904 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
907 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
909 if (!s->channel[c].reuse_sf) {
911 /** decode DPCM coded scale factors */
912 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
913 val = 45 / s->channel[c].scale_factor_step;
914 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
915 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
920 /** run level decode differences to the resampled factors */
921 for (i = 0; i < s->num_bands; i++) {
927 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
930 uint32_t code = get_bits(&s->gb, 14);
932 sign = (code & 1) - 1;
933 skip = (code & 0x3f) >> 1;
934 } else if (idx == 1) {
937 skip = scale_rl_run[idx];
938 val = scale_rl_level[idx];
939 sign = get_bits1(&s->gb)-1;
943 if (i >= s->num_bands) {
944 av_log(s->avctx, AV_LOG_ERROR,
945 "invalid scale factor coding\n");
946 return AVERROR_INVALIDDATA;
948 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
952 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
953 s->channel[c].table_idx = s->table_idx;
954 s->channel[c].reuse_sf = 1;
957 /** calculate new scale factor maximum */
958 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
959 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
960 s->channel[c].max_scale_factor =
961 FFMAX(s->channel[c].max_scale_factor, *sf);
969 *@brief Reconstruct the individual channel data.
970 *@param s codec context
972 static void inverse_channel_transform(WMAProDecodeCtx *s)
976 for (i = 0; i < s->num_chgroups; i++) {
977 if (s->chgroup[i].transform) {
978 float data[WMAPRO_MAX_CHANNELS];
979 const int num_channels = s->chgroup[i].num_channels;
980 float** ch_data = s->chgroup[i].channel_data;
981 float** ch_end = ch_data + num_channels;
982 const int8_t* tb = s->chgroup[i].transform_band;
985 /** multichannel decorrelation */
986 for (sfb = s->cur_sfb_offsets;
987 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
990 /** multiply values with the decorrelation_matrix */
991 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
992 const float* mat = s->chgroup[i].decorrelation_matrix;
993 const float* data_end = data + num_channels;
994 float* data_ptr = data;
997 for (ch = ch_data; ch < ch_end; ch++)
998 *data_ptr++ = (*ch)[y];
1000 for (ch = ch_data; ch < ch_end; ch++) {
1003 while (data_ptr < data_end)
1004 sum += *data_ptr++ * *mat++;
1009 } else if (s->num_channels == 2) {
1010 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1011 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1012 ch_data[0] + sfb[0],
1014 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1015 ch_data[1] + sfb[0],
1024 *@brief Apply sine window and reconstruct the output buffer.
1025 *@param s codec context
1027 static void wmapro_window(WMAProDecodeCtx *s)
1030 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1031 int c = s->channel_indexes_for_cur_subframe[i];
1033 int winlen = s->channel[c].prev_block_len;
1034 float* start = s->channel[c].coeffs - (winlen >> 1);
1036 if (s->subframe_len < winlen) {
1037 start += (winlen - s->subframe_len) >> 1;
1038 winlen = s->subframe_len;
1041 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1045 s->dsp.vector_fmul_window(start, start, start + winlen,
1048 s->channel[c].prev_block_len = s->subframe_len;
1053 *@brief Decode a single subframe (block).
1054 *@param s codec context
1055 *@return 0 on success, < 0 when decoding failed
1057 static int decode_subframe(WMAProDecodeCtx *s)
1059 int offset = s->samples_per_frame;
1060 int subframe_len = s->samples_per_frame;
1062 int total_samples = s->samples_per_frame * s->num_channels;
1063 int transmit_coeffs = 0;
1064 int cur_subwoofer_cutoff;
1066 s->subframe_offset = get_bits_count(&s->gb);
1068 /** reset channel context and find the next block offset and size
1069 == the next block of the channel with the smallest number of
1072 for (i = 0; i < s->num_channels; i++) {
1073 s->channel[i].grouped = 0;
1074 if (offset > s->channel[i].decoded_samples) {
1075 offset = s->channel[i].decoded_samples;
1077 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1082 "processing subframe with offset %i len %i\n", offset, subframe_len);
1084 /** get a list of all channels that contain the estimated block */
1085 s->channels_for_cur_subframe = 0;
1086 for (i = 0; i < s->num_channels; i++) {
1087 const int cur_subframe = s->channel[i].cur_subframe;
1088 /** substract already processed samples */
1089 total_samples -= s->channel[i].decoded_samples;
1091 /** and count if there are multiple subframes that match our profile */
1092 if (offset == s->channel[i].decoded_samples &&
1093 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1094 total_samples -= s->channel[i].subframe_len[cur_subframe];
1095 s->channel[i].decoded_samples +=
1096 s->channel[i].subframe_len[cur_subframe];
1097 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1098 ++s->channels_for_cur_subframe;
1102 /** check if the frame will be complete after processing the
1105 s->parsed_all_subframes = 1;
1108 av_dlog(s->avctx, "subframe is part of %i channels\n",
1109 s->channels_for_cur_subframe);
1111 /** calculate number of scale factor bands and their offsets */
1112 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1113 s->num_bands = s->num_sfb[s->table_idx];
1114 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1115 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1117 /** configure the decoder for the current subframe */
1118 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1119 int c = s->channel_indexes_for_cur_subframe[i];
1121 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1125 s->subframe_len = subframe_len;
1126 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1128 /** skip extended header if any */
1129 if (get_bits1(&s->gb)) {
1131 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1132 int len = get_bits(&s->gb, 4);
1133 num_fill_bits = get_bits(&s->gb, len) + 1;
1136 if (num_fill_bits >= 0) {
1137 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1138 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1139 return AVERROR_INVALIDDATA;
1142 skip_bits_long(&s->gb, num_fill_bits);
1146 /** no idea for what the following bit is used */
1147 if (get_bits1(&s->gb)) {
1148 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1149 return AVERROR_INVALIDDATA;
1153 if (decode_channel_transform(s) < 0)
1154 return AVERROR_INVALIDDATA;
1157 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1158 int c = s->channel_indexes_for_cur_subframe[i];
1159 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1160 transmit_coeffs = 1;
1163 if (transmit_coeffs) {
1165 int quant_step = 90 * s->bits_per_sample >> 4;
1167 /** decode number of vector coded coefficients */
1168 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1169 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1170 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1171 int c = s->channel_indexes_for_cur_subframe[i];
1172 s->channel[c].num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1175 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1176 int c = s->channel_indexes_for_cur_subframe[i];
1177 s->channel[c].num_vec_coeffs = s->subframe_len;
1180 /** decode quantization step */
1181 step = get_sbits(&s->gb, 6);
1183 if (step == -32 || step == 31) {
1184 const int sign = (step == 31) - 1;
1186 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1187 (step = get_bits(&s->gb, 5)) == 31) {
1190 quant_step += ((quant + step) ^ sign) - sign;
1192 if (quant_step < 0) {
1193 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1196 /** decode quantization step modifiers for every channel */
1198 if (s->channels_for_cur_subframe == 1) {
1199 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1201 int modifier_len = get_bits(&s->gb, 3);
1202 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1203 int c = s->channel_indexes_for_cur_subframe[i];
1204 s->channel[c].quant_step = quant_step;
1205 if (get_bits1(&s->gb)) {
1207 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1209 ++s->channel[c].quant_step;
1214 /** decode scale factors */
1215 if (decode_scale_factors(s) < 0)
1216 return AVERROR_INVALIDDATA;
1219 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1220 get_bits_count(&s->gb) - s->subframe_offset);
1222 /** parse coefficients */
1223 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1224 int c = s->channel_indexes_for_cur_subframe[i];
1225 if (s->channel[c].transmit_coefs &&
1226 get_bits_count(&s->gb) < s->num_saved_bits) {
1227 decode_coeffs(s, c);
1229 memset(s->channel[c].coeffs, 0,
1230 sizeof(*s->channel[c].coeffs) * subframe_len);
1233 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1234 get_bits_count(&s->gb) - s->subframe_offset);
1236 if (transmit_coeffs) {
1237 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1238 /** reconstruct the per channel data */
1239 inverse_channel_transform(s);
1240 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1241 int c = s->channel_indexes_for_cur_subframe[i];
1242 const int* sf = s->channel[c].scale_factors;
1245 if (c == s->lfe_channel)
1246 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1247 (subframe_len - cur_subwoofer_cutoff));
1249 /** inverse quantization and rescaling */
1250 for (b = 0; b < s->num_bands; b++) {
1251 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1252 const int exp = s->channel[c].quant_step -
1253 (s->channel[c].max_scale_factor - *sf++) *
1254 s->channel[c].scale_factor_step;
1255 const float quant = pow(10.0, exp / 20.0);
1256 int start = s->cur_sfb_offsets[b];
1257 s->dsp.vector_fmul_scalar(s->tmp + start,
1258 s->channel[c].coeffs + start,
1259 quant, end - start);
1262 /** apply imdct (imdct_half == DCTIV with reverse) */
1263 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1267 /** window and overlapp-add */
1270 /** handled one subframe */
1271 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1272 int c = s->channel_indexes_for_cur_subframe[i];
1273 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1274 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1275 return AVERROR_INVALIDDATA;
1277 ++s->channel[c].cur_subframe;
1284 *@brief Decode one WMA frame.
1285 *@param s codec context
1286 *@return 0 if the trailer bit indicates that this is the last frame,
1287 * 1 if there are additional frames
1289 static int decode_frame(WMAProDecodeCtx *s, int *got_frame_ptr)
1291 AVCodecContext *avctx = s->avctx;
1292 GetBitContext* gb = &s->gb;
1293 int more_frames = 0;
1296 const float *out_ptr[WMAPRO_MAX_CHANNELS];
1299 /** get frame length */
1301 len = get_bits(gb, s->log2_frame_size);
1303 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1305 /** decode tile information */
1306 if (decode_tilehdr(s)) {
1311 /** read postproc transform */
1312 if (s->num_channels > 1 && get_bits1(gb)) {
1313 if (get_bits1(gb)) {
1314 for (i = 0; i < s->num_channels * s->num_channels; i++)
1319 /** read drc info */
1320 if (s->dynamic_range_compression) {
1321 s->drc_gain = get_bits(gb, 8);
1322 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1325 /** no idea what these are for, might be the number of samples
1326 that need to be skipped at the beginning or end of a stream */
1327 if (get_bits1(gb)) {
1330 /** usually true for the first frame */
1331 if (get_bits1(gb)) {
1332 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1333 av_dlog(s->avctx, "start skip: %i\n", skip);
1336 /** sometimes true for the last frame */
1337 if (get_bits1(gb)) {
1338 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1339 av_dlog(s->avctx, "end skip: %i\n", skip);
1344 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1345 get_bits_count(gb) - s->frame_offset);
1347 /** reset subframe states */
1348 s->parsed_all_subframes = 0;
1349 for (i = 0; i < s->num_channels; i++) {
1350 s->channel[i].decoded_samples = 0;
1351 s->channel[i].cur_subframe = 0;
1352 s->channel[i].reuse_sf = 0;
1355 /** decode all subframes */
1356 while (!s->parsed_all_subframes) {
1357 if (decode_subframe(s) < 0) {
1363 /* get output buffer */
1364 s->frame.nb_samples = s->samples_per_frame;
1365 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
1366 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1370 samples = (float *)s->frame.data[0];
1372 /** interleave samples and write them to the output buffer */
1373 for (i = 0; i < s->num_channels; i++)
1374 out_ptr[i] = s->channel[i].out;
1375 s->fmt_conv.float_interleave(samples, out_ptr, s->samples_per_frame,
1378 for (i = 0; i < s->num_channels; i++) {
1379 /** reuse second half of the IMDCT output for the next frame */
1380 memcpy(&s->channel[i].out[0],
1381 &s->channel[i].out[s->samples_per_frame],
1382 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1385 if (s->skip_frame) {
1392 if (s->len_prefix) {
1393 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1394 /** FIXME: not sure if this is always an error */
1395 av_log(s->avctx, AV_LOG_ERROR,
1396 "frame[%i] would have to skip %i bits\n", s->frame_num,
1397 len - (get_bits_count(gb) - s->frame_offset) - 1);
1402 /** skip the rest of the frame data */
1403 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1405 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1409 /** decode trailer bit */
1410 more_frames = get_bits1(gb);
1417 *@brief Calculate remaining input buffer length.
1418 *@param s codec context
1419 *@param gb bitstream reader context
1420 *@return remaining size in bits
1422 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1424 return s->buf_bit_size - get_bits_count(gb);
1428 *@brief Fill the bit reservoir with a (partial) frame.
1429 *@param s codec context
1430 *@param gb bitstream reader context
1431 *@param len length of the partial frame
1432 *@param append decides whether to reset the buffer or not
1434 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1439 /** when the frame data does not need to be concatenated, the input buffer
1440 is resetted and additional bits from the previous frame are copyed
1441 and skipped later so that a fast byte copy is possible */
1444 s->frame_offset = get_bits_count(gb) & 7;
1445 s->num_saved_bits = s->frame_offset;
1446 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1449 buflen = (s->num_saved_bits + len + 8) >> 3;
1451 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1452 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1457 s->num_saved_bits += len;
1459 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1462 int align = 8 - (get_bits_count(gb) & 7);
1463 align = FFMIN(align, len);
1464 put_bits(&s->pb, align, get_bits(gb, align));
1466 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1468 skip_bits_long(gb, len);
1471 PutBitContext tmp = s->pb;
1472 flush_put_bits(&tmp);
1475 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1476 skip_bits(&s->gb, s->frame_offset);
1480 *@brief Decode a single WMA packet.
1481 *@param avctx codec context
1482 *@param data the output buffer
1483 *@param avpkt input packet
1484 *@return number of bytes that were read from the input buffer
1486 static int decode_packet(AVCodecContext *avctx, void *data,
1487 int *got_frame_ptr, AVPacket* avpkt)
1489 WMAProDecodeCtx *s = avctx->priv_data;
1490 GetBitContext* gb = &s->pgb;
1491 const uint8_t* buf = avpkt->data;
1492 int buf_size = avpkt->size;
1493 int num_bits_prev_frame;
1494 int packet_sequence_number;
1498 if (s->packet_done || s->packet_loss) {
1501 /** sanity check for the buffer length */
1502 if (buf_size < avctx->block_align)
1505 s->next_packet_start = buf_size - avctx->block_align;
1506 buf_size = avctx->block_align;
1507 s->buf_bit_size = buf_size << 3;
1509 /** parse packet header */
1510 init_get_bits(gb, buf, s->buf_bit_size);
1511 packet_sequence_number = get_bits(gb, 4);
1514 /** get number of bits that need to be added to the previous frame */
1515 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1516 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1517 num_bits_prev_frame);
1519 /** check for packet loss */
1520 if (!s->packet_loss &&
1521 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1523 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1524 s->packet_sequence_number, packet_sequence_number);
1526 s->packet_sequence_number = packet_sequence_number;
1528 if (num_bits_prev_frame > 0) {
1529 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1530 if (num_bits_prev_frame >= remaining_packet_bits) {
1531 num_bits_prev_frame = remaining_packet_bits;
1535 /** append the previous frame data to the remaining data from the
1536 previous packet to create a full frame */
1537 save_bits(s, gb, num_bits_prev_frame, 1);
1538 av_dlog(avctx, "accumulated %x bits of frame data\n",
1539 s->num_saved_bits - s->frame_offset);
1541 /** decode the cross packet frame if it is valid */
1542 if (!s->packet_loss)
1543 decode_frame(s, got_frame_ptr);
1544 } else if (s->num_saved_bits - s->frame_offset) {
1545 av_dlog(avctx, "ignoring %x previously saved bits\n",
1546 s->num_saved_bits - s->frame_offset);
1549 if (s->packet_loss) {
1550 /** reset number of saved bits so that the decoder
1551 does not start to decode incomplete frames in the
1552 s->len_prefix == 0 case */
1553 s->num_saved_bits = 0;
1559 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1560 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1561 skip_bits(gb, s->packet_offset);
1562 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1563 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1564 frame_size <= remaining_bits(s, gb)) {
1565 save_bits(s, gb, frame_size, 0);
1566 s->packet_done = !decode_frame(s, got_frame_ptr);
1567 } else if (!s->len_prefix
1568 && s->num_saved_bits > get_bits_count(&s->gb)) {
1569 /** when the frames do not have a length prefix, we don't know
1570 the compressed length of the individual frames
1571 however, we know what part of a new packet belongs to the
1573 therefore we save the incoming packet first, then we append
1574 the "previous frame" data from the next packet so that
1575 we get a buffer that only contains full frames */
1576 s->packet_done = !decode_frame(s, got_frame_ptr);
1581 if (s->packet_done && !s->packet_loss &&
1582 remaining_bits(s, gb) > 0) {
1583 /** save the rest of the data so that it can be decoded
1584 with the next packet */
1585 save_bits(s, gb, remaining_bits(s, gb), 0);
1588 s->packet_offset = get_bits_count(gb) & 7;
1590 return AVERROR_INVALIDDATA;
1593 *(AVFrame *)data = s->frame;
1595 return get_bits_count(gb) >> 3;
1599 *@brief Clear decoder buffers (for seeking).
1600 *@param avctx codec context
1602 static void flush(AVCodecContext *avctx)
1604 WMAProDecodeCtx *s = avctx->priv_data;
1606 /** reset output buffer as a part of it is used during the windowing of a
1608 for (i = 0; i < s->num_channels; i++)
1609 memset(s->channel[i].out, 0, s->samples_per_frame *
1610 sizeof(*s->channel[i].out));
1616 *@brief wmapro decoder
1618 AVCodec ff_wmapro_decoder = {
1620 .type = AVMEDIA_TYPE_AUDIO,
1621 .id = CODEC_ID_WMAPRO,
1622 .priv_data_size = sizeof(WMAProDecodeCtx),
1623 .init = decode_init,
1624 .close = decode_end,
1625 .decode = decode_packet,
1626 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1628 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),