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 FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
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/intreadwrite.h"
94 #include "wmaprodata.h"
96 #include "fmtconvert.h"
100 /** current decoder limitations */
101 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
102 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
103 #define MAX_BANDS 29 ///< max number of scale factor bands
104 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
106 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
107 #define WMAPRO_BLOCK_MAX_BITS 12 ///< log2 of max block size
108 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
109 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
113 #define SCALEVLCBITS 8
114 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
115 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
116 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
117 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
118 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
120 static VLC sf_vlc; ///< scale factor DPCM vlc
121 static VLC sf_rl_vlc; ///< scale factor run length vlc
122 static VLC vec4_vlc; ///< 4 coefficients per symbol
123 static VLC vec2_vlc; ///< 2 coefficients per symbol
124 static VLC vec1_vlc; ///< 1 coefficient per symbol
125 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
126 static float sin64[33]; ///< sinus table for decorrelation
129 * @brief frame specific decoder context for a single channel
132 int16_t prev_block_len; ///< length of the previous block
133 uint8_t transmit_coefs;
134 uint8_t num_subframes;
135 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
136 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
137 uint8_t cur_subframe; ///< current subframe number
138 uint16_t decoded_samples; ///< number of already processed samples
139 uint8_t grouped; ///< channel is part of a group
140 int quant_step; ///< quantization step for the current subframe
141 int8_t reuse_sf; ///< share scale factors between subframes
142 int8_t scale_factor_step; ///< scaling step for the current subframe
143 int max_scale_factor; ///< maximum scale factor for the current subframe
144 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
145 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
146 int* scale_factors; ///< pointer to the scale factor values used for decoding
147 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
148 float* coeffs; ///< pointer to the subframe decode buffer
149 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
150 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
154 * @brief channel group for channel transformations
157 uint8_t num_channels; ///< number of channels in the group
158 int8_t transform; ///< transform on / off
159 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
160 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
161 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
165 * @brief main decoder context
167 typedef struct WMAProDecodeCtx {
168 /* generic decoder variables */
169 AVCodecContext* avctx; ///< codec context for av_log
170 DSPContext dsp; ///< accelerated DSP functions
171 FmtConvertContext fmt_conv;
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 num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
187 int8_t lfe_channel; ///< lfe channel index
188 uint8_t max_num_subframes;
189 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
190 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
191 uint16_t min_samples_per_subframe;
192 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
193 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
194 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
195 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
197 /* packet decode state */
198 GetBitContext pgb; ///< bitstream reader context for the packet
199 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
200 uint8_t packet_offset; ///< frame offset in the packet
201 uint8_t packet_sequence_number; ///< current packet number
202 int num_saved_bits; ///< saved number of bits
203 int frame_offset; ///< frame offset in the bit reservoir
204 int subframe_offset; ///< subframe offset in the bit reservoir
205 uint8_t packet_loss; ///< set in case of bitstream error
206 uint8_t packet_done; ///< set when a packet is fully decoded
208 /* frame decode state */
209 uint32_t frame_num; ///< current frame number (not used for decoding)
210 GetBitContext gb; ///< bitstream reader context
211 int buf_bit_size; ///< buffer size in bits
212 float* samples; ///< current samplebuffer pointer
213 float* samples_end; ///< maximum samplebuffer pointer
214 uint8_t drc_gain; ///< gain for the DRC tool
215 int8_t skip_frame; ///< skip output step
216 int8_t parsed_all_subframes; ///< all subframes decoded?
218 /* subframe/block decode state */
219 int16_t subframe_len; ///< current subframe length
220 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
221 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
222 int8_t num_bands; ///< number of scale factor bands
223 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
224 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
225 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
226 int8_t esc_len; ///< length of escaped coefficients
228 uint8_t num_chgroups; ///< number of channel groups
229 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
231 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
236 *@brief helper function to print the most important members of the context
239 static void av_cold dump_context(WMAProDecodeCtx *s)
241 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
242 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
244 PRINT("ed sample bit depth", s->bits_per_sample);
245 PRINT_HEX("ed decode flags", s->decode_flags);
246 PRINT("samples per frame", s->samples_per_frame);
247 PRINT("log2 frame size", s->log2_frame_size);
248 PRINT("max num subframes", s->max_num_subframes);
249 PRINT("len prefix", s->len_prefix);
250 PRINT("num channels", s->num_channels);
254 *@brief Uninitialize the decoder and free all resources.
255 *@param avctx codec context
256 *@return 0 on success, < 0 otherwise
258 static av_cold int decode_end(AVCodecContext *avctx)
260 WMAProDecodeCtx *s = avctx->priv_data;
263 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
264 ff_mdct_end(&s->mdct_ctx[i]);
270 *@brief Initialize the decoder.
271 *@param avctx codec context
272 *@return 0 on success, -1 otherwise
274 static av_cold int decode_init(AVCodecContext *avctx)
276 WMAProDecodeCtx *s = avctx->priv_data;
277 uint8_t *edata_ptr = avctx->extradata;
278 unsigned int channel_mask;
280 int log2_max_num_subframes;
281 int num_possible_block_sizes;
284 dsputil_init(&s->dsp, avctx);
285 ff_fmt_convert_init(&s->fmt_conv, avctx);
286 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
288 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
290 if (avctx->extradata_size >= 18) {
291 s->decode_flags = AV_RL16(edata_ptr+14);
292 channel_mask = AV_RL32(edata_ptr+2);
293 s->bits_per_sample = AV_RL16(edata_ptr);
294 /** dump the extradata */
295 for (i = 0; i < avctx->extradata_size; i++)
296 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
297 av_dlog(avctx, "\n");
300 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
301 return AVERROR_INVALIDDATA;
305 s->log2_frame_size = av_log2(avctx->block_align) + 4;
308 s->skip_frame = 1; /* skip first frame */
310 s->len_prefix = (s->decode_flags & 0x40);
313 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
317 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
318 s->max_num_subframes = 1 << log2_max_num_subframes;
319 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
320 s->max_subframe_len_bit = 1;
321 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
323 num_possible_block_sizes = log2_max_num_subframes + 1;
324 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
325 s->dynamic_range_compression = (s->decode_flags & 0x80);
327 if (s->max_num_subframes > MAX_SUBFRAMES) {
328 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
329 s->max_num_subframes);
330 return AVERROR_INVALIDDATA;
333 s->num_channels = avctx->channels;
335 if (s->num_channels < 0) {
336 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
337 return AVERROR_INVALIDDATA;
338 } else if (s->num_channels > WMAPRO_MAX_CHANNELS) {
339 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
340 return AVERROR_PATCHWELCOME;
343 /** init previous block len */
344 for (i = 0; i < s->num_channels; i++)
345 s->channel[i].prev_block_len = s->samples_per_frame;
347 /** extract lfe channel position */
350 if (channel_mask & 8) {
352 for (mask = 1; mask < 16; mask <<= 1) {
353 if (channel_mask & mask)
358 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
359 scale_huffbits, 1, 1,
360 scale_huffcodes, 2, 2, 616);
362 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
363 scale_rl_huffbits, 1, 1,
364 scale_rl_huffcodes, 4, 4, 1406);
366 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
367 coef0_huffbits, 1, 1,
368 coef0_huffcodes, 4, 4, 2108);
370 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
371 coef1_huffbits, 1, 1,
372 coef1_huffcodes, 4, 4, 3912);
374 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
376 vec4_huffcodes, 2, 2, 604);
378 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
380 vec2_huffcodes, 2, 2, 562);
382 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
384 vec1_huffcodes, 2, 2, 562);
386 /** calculate number of scale factor bands and their offsets
387 for every possible block size */
388 for (i = 0; i < num_possible_block_sizes; i++) {
389 int subframe_len = s->samples_per_frame >> i;
393 s->sfb_offsets[i][0] = 0;
395 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
396 int offset = (subframe_len * 2 * critical_freq[x])
397 / s->avctx->sample_rate + 2;
399 if (offset > s->sfb_offsets[i][band - 1])
400 s->sfb_offsets[i][band++] = offset;
402 s->sfb_offsets[i][band - 1] = subframe_len;
403 s->num_sfb[i] = band - 1;
407 /** Scale factors can be shared between blocks of different size
408 as every block has a different scale factor band layout.
409 The matrix sf_offsets is needed to find the correct scale factor.
412 for (i = 0; i < num_possible_block_sizes; i++) {
414 for (b = 0; b < s->num_sfb[i]; b++) {
416 int offset = ((s->sfb_offsets[i][b]
417 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
418 for (x = 0; x < num_possible_block_sizes; x++) {
420 while (s->sfb_offsets[x][v + 1] << x < offset)
422 s->sf_offsets[i][x][b] = v;
427 /** init MDCT, FIXME: only init needed sizes */
428 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
429 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
430 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
431 / (1 << (s->bits_per_sample - 1)));
433 /** init MDCT windows: simple sinus window */
434 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
435 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
436 ff_init_ff_sine_windows(win_idx);
437 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
440 /** calculate subwoofer cutoff values */
441 for (i = 0; i < num_possible_block_sizes; i++) {
442 int block_size = s->samples_per_frame >> i;
443 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
444 / s->avctx->sample_rate;
445 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
448 /** calculate sine values for the decorrelation matrix */
449 for (i = 0; i < 33; i++)
450 sin64[i] = sin(i*M_PI / 64.0);
452 if (avctx->debug & FF_DEBUG_BITSTREAM)
455 avctx->channel_layout = channel_mask;
460 *@brief Decode the subframe length.
462 *@param offset sample offset in the frame
463 *@return decoded subframe length on success, < 0 in case of an error
465 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
467 int frame_len_shift = 0;
470 /** no need to read from the bitstream when only one length is possible */
471 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
472 return s->min_samples_per_subframe;
474 /** 1 bit indicates if the subframe is of maximum length */
475 if (s->max_subframe_len_bit) {
476 if (get_bits1(&s->gb))
477 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
479 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
481 subframe_len = s->samples_per_frame >> frame_len_shift;
483 /** sanity check the length */
484 if (subframe_len < s->min_samples_per_subframe ||
485 subframe_len > s->samples_per_frame) {
486 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
488 return AVERROR_INVALIDDATA;
494 *@brief Decode how the data in the frame is split into subframes.
495 * Every WMA frame contains the encoded data for a fixed number of
496 * samples per channel. The data for every channel might be split
497 * into several subframes. This function will reconstruct the list of
498 * subframes for every channel.
500 * If the subframes are not evenly split, the algorithm estimates the
501 * channels with the lowest number of total samples.
502 * Afterwards, for each of these channels a bit is read from the
503 * bitstream that indicates if the channel contains a subframe with the
504 * next subframe size that is going to be read from the bitstream or not.
505 * If a channel contains such a subframe, the subframe size gets added to
506 * the channel's subframe list.
507 * The algorithm repeats these steps until the frame is properly divided
508 * between the individual channels.
511 *@return 0 on success, < 0 in case of an error
513 static int decode_tilehdr(WMAProDecodeCtx *s)
515 uint16_t num_samples[WMAPRO_MAX_CHANNELS]; /**< sum of samples for all currently known subframes of a channel */
516 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
517 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
518 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
519 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
522 /* Should never consume more than 3073 bits (256 iterations for the
523 * while loop when always the minimum amount of 128 samples is substracted
524 * from missing samples in the 8 channel case).
525 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
528 /** reset tiling information */
529 for (c = 0; c < s->num_channels; c++)
530 s->channel[c].num_subframes = 0;
532 memset(num_samples, 0, sizeof(num_samples));
534 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
535 fixed_channel_layout = 1;
537 /** loop until the frame data is split between the subframes */
541 /** check which channels contain the subframe */
542 for (c = 0; c < s->num_channels; c++) {
543 if (num_samples[c] == min_channel_len) {
544 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
545 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
546 contains_subframe[c] = 1;
548 contains_subframe[c] = get_bits1(&s->gb);
550 contains_subframe[c] = 0;
553 /** get subframe length, subframe_len == 0 is not allowed */
554 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
555 return AVERROR_INVALIDDATA;
557 /** add subframes to the individual channels and find new min_channel_len */
558 min_channel_len += subframe_len;
559 for (c = 0; c < s->num_channels; c++) {
560 WMAProChannelCtx* chan = &s->channel[c];
562 if (contains_subframe[c]) {
563 if (chan->num_subframes >= MAX_SUBFRAMES) {
564 av_log(s->avctx, AV_LOG_ERROR,
565 "broken frame: num subframes > 31\n");
566 return AVERROR_INVALIDDATA;
568 chan->subframe_len[chan->num_subframes] = subframe_len;
569 num_samples[c] += subframe_len;
570 ++chan->num_subframes;
571 if (num_samples[c] > s->samples_per_frame) {
572 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
573 "channel len > samples_per_frame\n");
574 return AVERROR_INVALIDDATA;
576 } else if (num_samples[c] <= min_channel_len) {
577 if (num_samples[c] < min_channel_len) {
578 channels_for_cur_subframe = 0;
579 min_channel_len = num_samples[c];
581 ++channels_for_cur_subframe;
584 } while (min_channel_len < s->samples_per_frame);
586 for (c = 0; c < s->num_channels; c++) {
589 for (i = 0; i < s->channel[c].num_subframes; i++) {
590 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
591 " len %i\n", s->frame_num, c, i,
592 s->channel[c].subframe_len[i]);
593 s->channel[c].subframe_offset[i] = offset;
594 offset += s->channel[c].subframe_len[i];
602 *@brief Calculate a decorrelation matrix from the bitstream parameters.
603 *@param s codec context
604 *@param chgroup channel group for which the matrix needs to be calculated
606 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
607 WMAProChannelGrp *chgroup)
611 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
612 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
613 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
615 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
616 rotation_offset[i] = get_bits(&s->gb, 6);
618 for (i = 0; i < chgroup->num_channels; i++)
619 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
620 get_bits1(&s->gb) ? 1.0 : -1.0;
622 for (i = 1; i < chgroup->num_channels; i++) {
624 for (x = 0; x < i; x++) {
626 for (y = 0; y < i + 1; y++) {
627 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
628 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
629 int n = rotation_offset[offset + x];
635 cosv = sin64[32 - n];
637 sinv = sin64[64 - n];
638 cosv = -sin64[n - 32];
641 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
642 (v1 * sinv) - (v2 * cosv);
643 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
644 (v1 * cosv) + (v2 * sinv);
652 *@brief Decode channel transformation parameters
653 *@param s codec context
654 *@return 0 in case of success, < 0 in case of bitstream errors
656 static int decode_channel_transform(WMAProDecodeCtx* s)
659 /* should never consume more than 1921 bits for the 8 channel case
660 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
661 * + MAX_CHANNELS + MAX_BANDS + 1)
664 /** in the one channel case channel transforms are pointless */
666 if (s->num_channels > 1) {
667 int remaining_channels = s->channels_for_cur_subframe;
669 if (get_bits1(&s->gb)) {
670 av_log_ask_for_sample(s->avctx,
671 "unsupported channel transform bit\n");
672 return AVERROR_INVALIDDATA;
675 for (s->num_chgroups = 0; remaining_channels &&
676 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
677 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
678 float** channel_data = chgroup->channel_data;
679 chgroup->num_channels = 0;
680 chgroup->transform = 0;
682 /** decode channel mask */
683 if (remaining_channels > 2) {
684 for (i = 0; i < s->channels_for_cur_subframe; i++) {
685 int channel_idx = s->channel_indexes_for_cur_subframe[i];
686 if (!s->channel[channel_idx].grouped
687 && get_bits1(&s->gb)) {
688 ++chgroup->num_channels;
689 s->channel[channel_idx].grouped = 1;
690 *channel_data++ = s->channel[channel_idx].coeffs;
694 chgroup->num_channels = remaining_channels;
695 for (i = 0; i < s->channels_for_cur_subframe; i++) {
696 int channel_idx = s->channel_indexes_for_cur_subframe[i];
697 if (!s->channel[channel_idx].grouped)
698 *channel_data++ = s->channel[channel_idx].coeffs;
699 s->channel[channel_idx].grouped = 1;
703 /** decode transform type */
704 if (chgroup->num_channels == 2) {
705 if (get_bits1(&s->gb)) {
706 if (get_bits1(&s->gb)) {
707 av_log_ask_for_sample(s->avctx,
708 "unsupported channel transform type\n");
711 chgroup->transform = 1;
712 if (s->num_channels == 2) {
713 chgroup->decorrelation_matrix[0] = 1.0;
714 chgroup->decorrelation_matrix[1] = -1.0;
715 chgroup->decorrelation_matrix[2] = 1.0;
716 chgroup->decorrelation_matrix[3] = 1.0;
719 chgroup->decorrelation_matrix[0] = 0.70703125;
720 chgroup->decorrelation_matrix[1] = -0.70703125;
721 chgroup->decorrelation_matrix[2] = 0.70703125;
722 chgroup->decorrelation_matrix[3] = 0.70703125;
725 } else if (chgroup->num_channels > 2) {
726 if (get_bits1(&s->gb)) {
727 chgroup->transform = 1;
728 if (get_bits1(&s->gb)) {
729 decode_decorrelation_matrix(s, chgroup);
731 /** FIXME: more than 6 coupled channels not supported */
732 if (chgroup->num_channels > 6) {
733 av_log_ask_for_sample(s->avctx,
734 "coupled channels > 6\n");
736 memcpy(chgroup->decorrelation_matrix,
737 default_decorrelation[chgroup->num_channels],
738 chgroup->num_channels * chgroup->num_channels *
739 sizeof(*chgroup->decorrelation_matrix));
745 /** decode transform on / off */
746 if (chgroup->transform) {
747 if (!get_bits1(&s->gb)) {
749 /** transform can be enabled for individual bands */
750 for (i = 0; i < s->num_bands; i++) {
751 chgroup->transform_band[i] = get_bits1(&s->gb);
754 memset(chgroup->transform_band, 1, s->num_bands);
757 remaining_channels -= chgroup->num_channels;
764 *@brief Extract the coefficients from the bitstream.
765 *@param s codec context
766 *@param c current channel number
767 *@return 0 on success, < 0 in case of bitstream errors
769 static int decode_coeffs(WMAProDecodeCtx *s, int c)
771 /* Integers 0..15 as single-precision floats. The table saves a
772 costly int to float conversion, and storing the values as
773 integers allows fast sign-flipping. */
774 static const uint32_t fval_tab[16] = {
775 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
776 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
777 0x41000000, 0x41100000, 0x41200000, 0x41300000,
778 0x41400000, 0x41500000, 0x41600000, 0x41700000,
782 WMAProChannelCtx* ci = &s->channel[c];
789 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
791 vlctable = get_bits1(&s->gb);
792 vlc = &coef_vlc[vlctable];
802 /** decode vector coefficients (consumes up to 167 bits per iteration for
803 4 vector coded large values) */
804 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
805 (cur_coeff + 3 < ci->num_vec_coeffs)) {
810 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
812 if (idx == HUFF_VEC4_SIZE - 1) {
813 for (i = 0; i < 4; i += 2) {
814 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
815 if (idx == HUFF_VEC2_SIZE - 1) {
817 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
818 if (v0 == HUFF_VEC1_SIZE - 1)
819 v0 += ff_wma_get_large_val(&s->gb);
820 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
821 if (v1 == HUFF_VEC1_SIZE - 1)
822 v1 += ff_wma_get_large_val(&s->gb);
823 vals[i ] = ((av_alias32){ .f32 = v0 }).u32;
824 vals[i+1] = ((av_alias32){ .f32 = v1 }).u32;
826 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
827 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
831 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
832 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
833 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
834 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
838 for (i = 0; i < 4; i++) {
840 uint32_t sign = get_bits1(&s->gb) - 1;
841 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
844 ci->coeffs[cur_coeff] = 0;
845 /** switch to run level mode when subframe_len / 128 zeros
846 were found in a row */
847 rl_mode |= (++num_zeros > s->subframe_len >> 8);
853 /** decode run level coded coefficients */
854 if (cur_coeff < s->subframe_len) {
855 memset(&ci->coeffs[cur_coeff], 0,
856 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
857 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
858 level, run, 1, ci->coeffs,
859 cur_coeff, s->subframe_len,
860 s->subframe_len, s->esc_len, 0))
861 return AVERROR_INVALIDDATA;
868 *@brief Extract scale factors from the bitstream.
869 *@param s codec context
870 *@return 0 on success, < 0 in case of bitstream errors
872 static int decode_scale_factors(WMAProDecodeCtx* s)
876 /** should never consume more than 5344 bits
877 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
880 for (i = 0; i < s->channels_for_cur_subframe; i++) {
881 int c = s->channel_indexes_for_cur_subframe[i];
884 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
885 sf_end = s->channel[c].scale_factors + s->num_bands;
887 /** resample scale factors for the new block size
888 * as the scale factors might need to be resampled several times
889 * before some new values are transmitted, a backup of the last
890 * transmitted scale factors is kept in saved_scale_factors
892 if (s->channel[c].reuse_sf) {
893 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
895 for (b = 0; b < s->num_bands; b++)
896 s->channel[c].scale_factors[b] =
897 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
900 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
902 if (!s->channel[c].reuse_sf) {
904 /** decode DPCM coded scale factors */
905 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
906 val = 45 / s->channel[c].scale_factor_step;
907 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
908 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
913 /** run level decode differences to the resampled factors */
914 for (i = 0; i < s->num_bands; i++) {
920 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
923 uint32_t code = get_bits(&s->gb, 14);
925 sign = (code & 1) - 1;
926 skip = (code & 0x3f) >> 1;
927 } else if (idx == 1) {
930 skip = scale_rl_run[idx];
931 val = scale_rl_level[idx];
932 sign = get_bits1(&s->gb)-1;
936 if (i >= s->num_bands) {
937 av_log(s->avctx, AV_LOG_ERROR,
938 "invalid scale factor coding\n");
939 return AVERROR_INVALIDDATA;
941 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
945 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
946 s->channel[c].table_idx = s->table_idx;
947 s->channel[c].reuse_sf = 1;
950 /** calculate new scale factor maximum */
951 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
952 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
953 s->channel[c].max_scale_factor =
954 FFMAX(s->channel[c].max_scale_factor, *sf);
962 *@brief Reconstruct the individual channel data.
963 *@param s codec context
965 static void inverse_channel_transform(WMAProDecodeCtx *s)
969 for (i = 0; i < s->num_chgroups; i++) {
970 if (s->chgroup[i].transform) {
971 float data[WMAPRO_MAX_CHANNELS];
972 const int num_channels = s->chgroup[i].num_channels;
973 float** ch_data = s->chgroup[i].channel_data;
974 float** ch_end = ch_data + num_channels;
975 const int8_t* tb = s->chgroup[i].transform_band;
978 /** multichannel decorrelation */
979 for (sfb = s->cur_sfb_offsets;
980 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
983 /** multiply values with the decorrelation_matrix */
984 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
985 const float* mat = s->chgroup[i].decorrelation_matrix;
986 const float* data_end = data + num_channels;
987 float* data_ptr = data;
990 for (ch = ch_data; ch < ch_end; ch++)
991 *data_ptr++ = (*ch)[y];
993 for (ch = ch_data; ch < ch_end; ch++) {
996 while (data_ptr < data_end)
997 sum += *data_ptr++ * *mat++;
1002 } else if (s->num_channels == 2) {
1003 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1004 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1005 ch_data[0] + sfb[0],
1007 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1008 ch_data[1] + sfb[0],
1017 *@brief Apply sine window and reconstruct the output buffer.
1018 *@param s codec context
1020 static void wmapro_window(WMAProDecodeCtx *s)
1023 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1024 int c = s->channel_indexes_for_cur_subframe[i];
1026 int winlen = s->channel[c].prev_block_len;
1027 float* start = s->channel[c].coeffs - (winlen >> 1);
1029 if (s->subframe_len < winlen) {
1030 start += (winlen - s->subframe_len) >> 1;
1031 winlen = s->subframe_len;
1034 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1038 s->dsp.vector_fmul_window(start, start, start + winlen,
1041 s->channel[c].prev_block_len = s->subframe_len;
1046 *@brief Decode a single subframe (block).
1047 *@param s codec context
1048 *@return 0 on success, < 0 when decoding failed
1050 static int decode_subframe(WMAProDecodeCtx *s)
1052 int offset = s->samples_per_frame;
1053 int subframe_len = s->samples_per_frame;
1055 int total_samples = s->samples_per_frame * s->num_channels;
1056 int transmit_coeffs = 0;
1057 int cur_subwoofer_cutoff;
1059 s->subframe_offset = get_bits_count(&s->gb);
1061 /** reset channel context and find the next block offset and size
1062 == the next block of the channel with the smallest number of
1065 for (i = 0; i < s->num_channels; i++) {
1066 s->channel[i].grouped = 0;
1067 if (offset > s->channel[i].decoded_samples) {
1068 offset = s->channel[i].decoded_samples;
1070 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1075 "processing subframe with offset %i len %i\n", offset, subframe_len);
1077 /** get a list of all channels that contain the estimated block */
1078 s->channels_for_cur_subframe = 0;
1079 for (i = 0; i < s->num_channels; i++) {
1080 const int cur_subframe = s->channel[i].cur_subframe;
1081 /** substract already processed samples */
1082 total_samples -= s->channel[i].decoded_samples;
1084 /** and count if there are multiple subframes that match our profile */
1085 if (offset == s->channel[i].decoded_samples &&
1086 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1087 total_samples -= s->channel[i].subframe_len[cur_subframe];
1088 s->channel[i].decoded_samples +=
1089 s->channel[i].subframe_len[cur_subframe];
1090 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1091 ++s->channels_for_cur_subframe;
1095 /** check if the frame will be complete after processing the
1098 s->parsed_all_subframes = 1;
1101 av_dlog(s->avctx, "subframe is part of %i channels\n",
1102 s->channels_for_cur_subframe);
1104 /** calculate number of scale factor bands and their offsets */
1105 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1106 s->num_bands = s->num_sfb[s->table_idx];
1107 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1108 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1110 /** configure the decoder for the current subframe */
1111 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1112 int c = s->channel_indexes_for_cur_subframe[i];
1114 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1118 s->subframe_len = subframe_len;
1119 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1121 /** skip extended header if any */
1122 if (get_bits1(&s->gb)) {
1124 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1125 int len = get_bits(&s->gb, 4);
1126 num_fill_bits = get_bits(&s->gb, len) + 1;
1129 if (num_fill_bits >= 0) {
1130 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1131 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1132 return AVERROR_INVALIDDATA;
1135 skip_bits_long(&s->gb, num_fill_bits);
1139 /** no idea for what the following bit is used */
1140 if (get_bits1(&s->gb)) {
1141 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1142 return AVERROR_INVALIDDATA;
1146 if (decode_channel_transform(s) < 0)
1147 return AVERROR_INVALIDDATA;
1150 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1151 int c = s->channel_indexes_for_cur_subframe[i];
1152 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1153 transmit_coeffs = 1;
1156 if (transmit_coeffs) {
1158 int quant_step = 90 * s->bits_per_sample >> 4;
1160 /** decode number of vector coded coefficients */
1161 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1162 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1163 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1164 int c = s->channel_indexes_for_cur_subframe[i];
1165 s->channel[c].num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1168 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1169 int c = s->channel_indexes_for_cur_subframe[i];
1170 s->channel[c].num_vec_coeffs = s->subframe_len;
1173 /** decode quantization step */
1174 step = get_sbits(&s->gb, 6);
1176 if (step == -32 || step == 31) {
1177 const int sign = (step == 31) - 1;
1179 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1180 (step = get_bits(&s->gb, 5)) == 31) {
1183 quant_step += ((quant + step) ^ sign) - sign;
1185 if (quant_step < 0) {
1186 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1189 /** decode quantization step modifiers for every channel */
1191 if (s->channels_for_cur_subframe == 1) {
1192 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1194 int modifier_len = get_bits(&s->gb, 3);
1195 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1196 int c = s->channel_indexes_for_cur_subframe[i];
1197 s->channel[c].quant_step = quant_step;
1198 if (get_bits1(&s->gb)) {
1200 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1202 ++s->channel[c].quant_step;
1207 /** decode scale factors */
1208 if (decode_scale_factors(s) < 0)
1209 return AVERROR_INVALIDDATA;
1212 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1213 get_bits_count(&s->gb) - s->subframe_offset);
1215 /** parse coefficients */
1216 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1217 int c = s->channel_indexes_for_cur_subframe[i];
1218 if (s->channel[c].transmit_coefs &&
1219 get_bits_count(&s->gb) < s->num_saved_bits) {
1220 decode_coeffs(s, c);
1222 memset(s->channel[c].coeffs, 0,
1223 sizeof(*s->channel[c].coeffs) * subframe_len);
1226 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1227 get_bits_count(&s->gb) - s->subframe_offset);
1229 if (transmit_coeffs) {
1230 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1231 /** reconstruct the per channel data */
1232 inverse_channel_transform(s);
1233 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1234 int c = s->channel_indexes_for_cur_subframe[i];
1235 const int* sf = s->channel[c].scale_factors;
1238 if (c == s->lfe_channel)
1239 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1240 (subframe_len - cur_subwoofer_cutoff));
1242 /** inverse quantization and rescaling */
1243 for (b = 0; b < s->num_bands; b++) {
1244 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1245 const int exp = s->channel[c].quant_step -
1246 (s->channel[c].max_scale_factor - *sf++) *
1247 s->channel[c].scale_factor_step;
1248 const float quant = pow(10.0, exp / 20.0);
1249 int start = s->cur_sfb_offsets[b];
1250 s->dsp.vector_fmul_scalar(s->tmp + start,
1251 s->channel[c].coeffs + start,
1252 quant, end - start);
1255 /** apply imdct (imdct_half == DCTIV with reverse) */
1256 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1260 /** window and overlapp-add */
1263 /** handled one subframe */
1264 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1265 int c = s->channel_indexes_for_cur_subframe[i];
1266 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1267 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1268 return AVERROR_INVALIDDATA;
1270 ++s->channel[c].cur_subframe;
1277 *@brief Decode one WMA frame.
1278 *@param s codec context
1279 *@return 0 if the trailer bit indicates that this is the last frame,
1280 * 1 if there are additional frames
1282 static int decode_frame(WMAProDecodeCtx *s)
1284 GetBitContext* gb = &s->gb;
1285 int more_frames = 0;
1288 const float *out_ptr[WMAPRO_MAX_CHANNELS];
1290 /** check for potential output buffer overflow */
1291 if (s->num_channels * s->samples_per_frame > s->samples_end - s->samples) {
1292 /** return an error if no frame could be decoded at all */
1293 av_log(s->avctx, AV_LOG_ERROR,
1294 "not enough space for the output samples\n");
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 /** interleave samples and write them to the output buffer */
1364 for (i = 0; i < s->num_channels; i++)
1365 out_ptr[i] = s->channel[i].out;
1366 s->fmt_conv.float_interleave(s->samples, out_ptr, s->samples_per_frame,
1369 for (i = 0; i < s->num_channels; i++) {
1370 /** reuse second half of the IMDCT output for the next frame */
1371 memcpy(&s->channel[i].out[0],
1372 &s->channel[i].out[s->samples_per_frame],
1373 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1376 if (s->skip_frame) {
1379 s->samples += s->num_channels * s->samples_per_frame;
1381 if (s->len_prefix) {
1382 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1383 /** FIXME: not sure if this is always an error */
1384 av_log(s->avctx, AV_LOG_ERROR,
1385 "frame[%i] would have to skip %i bits\n", s->frame_num,
1386 len - (get_bits_count(gb) - s->frame_offset) - 1);
1391 /** skip the rest of the frame data */
1392 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1394 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1398 /** decode trailer bit */
1399 more_frames = get_bits1(gb);
1406 *@brief Calculate remaining input buffer length.
1407 *@param s codec context
1408 *@param gb bitstream reader context
1409 *@return remaining size in bits
1411 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1413 return s->buf_bit_size - get_bits_count(gb);
1417 *@brief Fill the bit reservoir with a (partial) frame.
1418 *@param s codec context
1419 *@param gb bitstream reader context
1420 *@param len length of the partial frame
1421 *@param append decides wether to reset the buffer or not
1423 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1428 /** when the frame data does not need to be concatenated, the input buffer
1429 is resetted and additional bits from the previous frame are copyed
1430 and skipped later so that a fast byte copy is possible */
1433 s->frame_offset = get_bits_count(gb) & 7;
1434 s->num_saved_bits = s->frame_offset;
1435 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1438 buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1440 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1441 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1446 s->num_saved_bits += len;
1448 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1451 int align = 8 - (get_bits_count(gb) & 7);
1452 align = FFMIN(align, len);
1453 put_bits(&s->pb, align, get_bits(gb, align));
1455 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1457 skip_bits_long(gb, len);
1460 PutBitContext tmp = s->pb;
1461 flush_put_bits(&tmp);
1464 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1465 skip_bits(&s->gb, s->frame_offset);
1469 *@brief Decode a single WMA packet.
1470 *@param avctx codec context
1471 *@param data the output buffer
1472 *@param data_size number of bytes that were written to the output buffer
1473 *@param avpkt input packet
1474 *@return number of bytes that were read from the input buffer
1476 static int decode_packet(AVCodecContext *avctx,
1477 void *data, int *data_size, AVPacket* avpkt)
1479 WMAProDecodeCtx *s = avctx->priv_data;
1480 GetBitContext* gb = &s->pgb;
1481 const uint8_t* buf = avpkt->data;
1482 int buf_size = avpkt->size;
1483 int num_bits_prev_frame;
1484 int packet_sequence_number;
1487 s->samples_end = (float*)((int8_t*)data + *data_size);
1490 if (s->packet_done || s->packet_loss) {
1493 /** sanity check for the buffer length */
1494 if (buf_size < avctx->block_align)
1497 s->next_packet_start = buf_size - avctx->block_align;
1498 buf_size = avctx->block_align;
1499 s->buf_bit_size = buf_size << 3;
1501 /** parse packet header */
1502 init_get_bits(gb, buf, s->buf_bit_size);
1503 packet_sequence_number = get_bits(gb, 4);
1506 /** get number of bits that need to be added to the previous frame */
1507 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1508 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1509 num_bits_prev_frame);
1511 /** check for packet loss */
1512 if (!s->packet_loss &&
1513 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1515 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1516 s->packet_sequence_number, packet_sequence_number);
1518 s->packet_sequence_number = packet_sequence_number;
1520 if (num_bits_prev_frame > 0) {
1521 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1522 if (num_bits_prev_frame >= remaining_packet_bits) {
1523 num_bits_prev_frame = remaining_packet_bits;
1527 /** append the previous frame data to the remaining data from the
1528 previous packet to create a full frame */
1529 save_bits(s, gb, num_bits_prev_frame, 1);
1530 av_dlog(avctx, "accumulated %x bits of frame data\n",
1531 s->num_saved_bits - s->frame_offset);
1533 /** decode the cross packet frame if it is valid */
1534 if (!s->packet_loss)
1536 } else if (s->num_saved_bits - s->frame_offset) {
1537 av_dlog(avctx, "ignoring %x previously saved bits\n",
1538 s->num_saved_bits - s->frame_offset);
1541 if (s->packet_loss) {
1542 /** reset number of saved bits so that the decoder
1543 does not start to decode incomplete frames in the
1544 s->len_prefix == 0 case */
1545 s->num_saved_bits = 0;
1551 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1552 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1553 skip_bits(gb, s->packet_offset);
1554 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1555 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1556 frame_size <= remaining_bits(s, gb)) {
1557 save_bits(s, gb, frame_size, 0);
1558 s->packet_done = !decode_frame(s);
1559 } else if (!s->len_prefix
1560 && s->num_saved_bits > get_bits_count(&s->gb)) {
1561 /** when the frames do not have a length prefix, we don't know
1562 the compressed length of the individual frames
1563 however, we know what part of a new packet belongs to the
1565 therefore we save the incoming packet first, then we append
1566 the "previous frame" data from the next packet so that
1567 we get a buffer that only contains full frames */
1568 s->packet_done = !decode_frame(s);
1573 if (s->packet_done && !s->packet_loss &&
1574 remaining_bits(s, gb) > 0) {
1575 /** save the rest of the data so that it can be decoded
1576 with the next packet */
1577 save_bits(s, gb, remaining_bits(s, gb), 0);
1580 *data_size = (int8_t *)s->samples - (int8_t *)data;
1581 s->packet_offset = get_bits_count(gb) & 7;
1583 return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3;
1587 *@brief Clear decoder buffers (for seeking).
1588 *@param avctx codec context
1590 static void flush(AVCodecContext *avctx)
1592 WMAProDecodeCtx *s = avctx->priv_data;
1594 /** reset output buffer as a part of it is used during the windowing of a
1596 for (i = 0; i < s->num_channels; i++)
1597 memset(s->channel[i].out, 0, s->samples_per_frame *
1598 sizeof(*s->channel[i].out));
1604 *@brief wmapro decoder
1606 AVCodec ff_wmapro_decoder = {
1608 .type = AVMEDIA_TYPE_AUDIO,
1609 .id = CODEC_ID_WMAPRO,
1610 .priv_data_size = sizeof(WMAProDecodeCtx),
1611 .init = decode_init,
1612 .close = decode_end,
1613 .decode = decode_packet,
1614 .capabilities = CODEC_CAP_SUBFRAMES,
1616 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),