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.
93 #include "wmaprodata.h"
97 /** current decoder limitations */
98 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
99 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
100 #define MAX_BANDS 29 ///< max number of scale factor bands
101 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
103 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
104 #define WMAPRO_BLOCK_MAX_BITS 12 ///< log2 of max block size
105 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
106 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
110 #define SCALEVLCBITS 8
111 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
112 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
113 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
114 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
115 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
117 static VLC sf_vlc; ///< scale factor DPCM vlc
118 static VLC sf_rl_vlc; ///< scale factor run length vlc
119 static VLC vec4_vlc; ///< 4 coefficients per symbol
120 static VLC vec2_vlc; ///< 2 coefficients per symbol
121 static VLC vec1_vlc; ///< 1 coefficient per symbol
122 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
123 static float sin64[33]; ///< sinus table for decorrelation
126 * @brief frame specific decoder context for a single channel
129 int16_t prev_block_len; ///< length of the previous block
130 uint8_t transmit_coefs;
131 uint8_t num_subframes;
132 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
133 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
134 uint8_t cur_subframe; ///< current subframe number
135 uint16_t decoded_samples; ///< number of already processed samples
136 uint8_t grouped; ///< channel is part of a group
137 int quant_step; ///< quantization step for the current subframe
138 int8_t reuse_sf; ///< share scale factors between subframes
139 int8_t scale_factor_step; ///< scaling step for the current subframe
140 int max_scale_factor; ///< maximum scale factor for the current subframe
141 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
142 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
143 int* scale_factors; ///< pointer to the scale factor values used for decoding
144 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
145 float* coeffs; ///< pointer to the subframe decode buffer
146 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
147 DECLARE_ALIGNED(16, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
151 * @brief channel group for channel transformations
154 uint8_t num_channels; ///< number of channels in the group
155 int8_t transform; ///< transform on / off
156 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
157 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
158 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
162 * @brief main decoder context
164 typedef struct WMAProDecodeCtx {
165 /* generic decoder variables */
166 AVCodecContext* avctx; ///< codec context for av_log
167 DSPContext dsp; ///< accelerated DSP functions
168 uint8_t frame_data[MAX_FRAMESIZE +
169 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
170 PutBitContext pb; ///< context for filling the frame_data buffer
171 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
172 DECLARE_ALIGNED(16, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
173 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
175 /* frame size dependent frame information (set during initialization) */
176 uint32_t decode_flags; ///< used compression features
177 uint8_t len_prefix; ///< frame is prefixed with its length
178 uint8_t dynamic_range_compression; ///< frame contains DRC data
179 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
180 uint16_t samples_per_frame; ///< number of samples to output
181 uint16_t log2_frame_size;
182 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
183 int8_t lfe_channel; ///< lfe channel index
184 uint8_t max_num_subframes;
185 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
186 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
187 uint16_t min_samples_per_subframe;
188 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
189 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
190 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
191 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
193 /* packet decode state */
194 GetBitContext pgb; ///< bitstream reader context for the packet
195 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
196 uint8_t packet_offset; ///< frame offset in the packet
197 uint8_t packet_sequence_number; ///< current packet number
198 int num_saved_bits; ///< saved number of bits
199 int frame_offset; ///< frame offset in the bit reservoir
200 int subframe_offset; ///< subframe offset in the bit reservoir
201 uint8_t packet_loss; ///< set in case of bitstream error
202 uint8_t packet_done; ///< set when a packet is fully decoded
204 /* frame decode state */
205 uint32_t frame_num; ///< current frame number (not used for decoding)
206 GetBitContext gb; ///< bitstream reader context
207 int buf_bit_size; ///< buffer size in bits
208 float* samples; ///< current samplebuffer pointer
209 float* samples_end; ///< maximum samplebuffer pointer
210 uint8_t drc_gain; ///< gain for the DRC tool
211 int8_t skip_frame; ///< skip output step
212 int8_t parsed_all_subframes; ///< all subframes decoded?
214 /* subframe/block decode state */
215 int16_t subframe_len; ///< current subframe length
216 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
217 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
218 int8_t num_bands; ///< number of scale factor bands
219 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
220 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
221 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
222 int8_t esc_len; ///< length of escaped coefficients
224 uint8_t num_chgroups; ///< number of channel groups
225 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
227 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
232 *@brief helper function to print the most important members of the context
235 static void av_cold dump_context(WMAProDecodeCtx *s)
237 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
238 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
240 PRINT("ed sample bit depth", s->bits_per_sample);
241 PRINT_HEX("ed decode flags", s->decode_flags);
242 PRINT("samples per frame", s->samples_per_frame);
243 PRINT("log2 frame size", s->log2_frame_size);
244 PRINT("max num subframes", s->max_num_subframes);
245 PRINT("len prefix", s->len_prefix);
246 PRINT("num channels", s->num_channels);
250 *@brief Uninitialize the decoder and free all resources.
251 *@param avctx codec context
252 *@return 0 on success, < 0 otherwise
254 static av_cold int decode_end(AVCodecContext *avctx)
256 WMAProDecodeCtx *s = avctx->priv_data;
259 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
260 ff_mdct_end(&s->mdct_ctx[i]);
266 *@brief Initialize the decoder.
267 *@param avctx codec context
268 *@return 0 on success, -1 otherwise
270 static av_cold int decode_init(AVCodecContext *avctx)
272 WMAProDecodeCtx *s = avctx->priv_data;
273 uint8_t *edata_ptr = avctx->extradata;
274 unsigned int channel_mask;
276 int log2_max_num_subframes;
277 int num_possible_block_sizes;
280 dsputil_init(&s->dsp, avctx);
281 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
283 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
285 if (avctx->extradata_size >= 18) {
286 s->decode_flags = AV_RL16(edata_ptr+14);
287 channel_mask = AV_RL32(edata_ptr+2);
288 s->bits_per_sample = AV_RL16(edata_ptr);
289 /** dump the extradata */
290 for (i = 0; i < avctx->extradata_size; i++)
291 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
292 av_dlog(avctx, "\n");
295 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
296 return AVERROR_INVALIDDATA;
300 s->log2_frame_size = av_log2(avctx->block_align) + 4;
303 s->skip_frame = 1; /* skip first frame */
305 s->len_prefix = (s->decode_flags & 0x40);
308 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
311 /** init previous block len */
312 for (i = 0; i < avctx->channels; i++)
313 s->channel[i].prev_block_len = s->samples_per_frame;
316 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
317 s->max_num_subframes = 1 << log2_max_num_subframes;
318 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
319 s->max_subframe_len_bit = 1;
320 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
322 num_possible_block_sizes = log2_max_num_subframes + 1;
323 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
324 s->dynamic_range_compression = (s->decode_flags & 0x80);
326 if (s->max_num_subframes > MAX_SUBFRAMES) {
327 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
328 s->max_num_subframes);
329 return AVERROR_INVALIDDATA;
332 s->num_channels = avctx->channels;
334 /** extract lfe channel position */
337 if (channel_mask & 8) {
339 for (mask = 1; mask < 16; mask <<= 1) {
340 if (channel_mask & mask)
345 if (s->num_channels < 0) {
346 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
347 return AVERROR_INVALIDDATA;
348 } else if (s->num_channels > WMAPRO_MAX_CHANNELS) {
349 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
350 return AVERROR_PATCHWELCOME;
353 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
354 scale_huffbits, 1, 1,
355 scale_huffcodes, 2, 2, 616);
357 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
358 scale_rl_huffbits, 1, 1,
359 scale_rl_huffcodes, 4, 4, 1406);
361 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
362 coef0_huffbits, 1, 1,
363 coef0_huffcodes, 4, 4, 2108);
365 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
366 coef1_huffbits, 1, 1,
367 coef1_huffcodes, 4, 4, 3912);
369 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
371 vec4_huffcodes, 2, 2, 604);
373 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
375 vec2_huffcodes, 2, 2, 562);
377 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
379 vec1_huffcodes, 2, 2, 562);
381 /** calculate number of scale factor bands and their offsets
382 for every possible block size */
383 for (i = 0; i < num_possible_block_sizes; i++) {
384 int subframe_len = s->samples_per_frame >> i;
388 s->sfb_offsets[i][0] = 0;
390 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
391 int offset = (subframe_len * 2 * critical_freq[x])
392 / s->avctx->sample_rate + 2;
394 if (offset > s->sfb_offsets[i][band - 1])
395 s->sfb_offsets[i][band++] = offset;
397 s->sfb_offsets[i][band - 1] = subframe_len;
398 s->num_sfb[i] = band - 1;
402 /** Scale factors can be shared between blocks of different size
403 as every block has a different scale factor band layout.
404 The matrix sf_offsets is needed to find the correct scale factor.
407 for (i = 0; i < num_possible_block_sizes; i++) {
409 for (b = 0; b < s->num_sfb[i]; b++) {
411 int offset = ((s->sfb_offsets[i][b]
412 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
413 for (x = 0; x < num_possible_block_sizes; x++) {
415 while (s->sfb_offsets[x][v + 1] << x < offset)
417 s->sf_offsets[i][x][b] = v;
422 /** init MDCT, FIXME: only init needed sizes */
423 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
424 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
425 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
426 / (1 << (s->bits_per_sample - 1)));
428 /** init MDCT windows: simple sinus window */
429 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
430 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
431 ff_init_ff_sine_windows(win_idx);
432 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
435 /** calculate subwoofer cutoff values */
436 for (i = 0; i < num_possible_block_sizes; i++) {
437 int block_size = s->samples_per_frame >> i;
438 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
439 / s->avctx->sample_rate;
440 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
443 /** calculate sine values for the decorrelation matrix */
444 for (i = 0; i < 33; i++)
445 sin64[i] = sin(i*M_PI / 64.0);
447 if (avctx->debug & FF_DEBUG_BITSTREAM)
450 avctx->channel_layout = channel_mask;
455 *@brief Decode the subframe length.
457 *@param offset sample offset in the frame
458 *@return decoded subframe length on success, < 0 in case of an error
460 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
462 int frame_len_shift = 0;
465 /** no need to read from the bitstream when only one length is possible */
466 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
467 return s->min_samples_per_subframe;
469 /** 1 bit indicates if the subframe is of maximum length */
470 if (s->max_subframe_len_bit) {
471 if (get_bits1(&s->gb))
472 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
474 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
476 subframe_len = s->samples_per_frame >> frame_len_shift;
478 /** sanity check the length */
479 if (subframe_len < s->min_samples_per_subframe ||
480 subframe_len > s->samples_per_frame) {
481 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
483 return AVERROR_INVALIDDATA;
489 *@brief Decode how the data in the frame is split into subframes.
490 * Every WMA frame contains the encoded data for a fixed number of
491 * samples per channel. The data for every channel might be split
492 * into several subframes. This function will reconstruct the list of
493 * subframes for every channel.
495 * If the subframes are not evenly split, the algorithm estimates the
496 * channels with the lowest number of total samples.
497 * Afterwards, for each of these channels a bit is read from the
498 * bitstream that indicates if the channel contains a subframe with the
499 * next subframe size that is going to be read from the bitstream or not.
500 * If a channel contains such a subframe, the subframe size gets added to
501 * the channel's subframe list.
502 * The algorithm repeats these steps until the frame is properly divided
503 * between the individual channels.
506 *@return 0 on success, < 0 in case of an error
508 static int decode_tilehdr(WMAProDecodeCtx *s)
510 uint16_t num_samples[WMAPRO_MAX_CHANNELS]; /**< sum of samples for all currently known subframes of a channel */
511 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
512 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
513 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
514 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
517 /* Should never consume more than 3073 bits (256 iterations for the
518 * while loop when always the minimum amount of 128 samples is substracted
519 * from missing samples in the 8 channel case).
520 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
523 /** reset tiling information */
524 for (c = 0; c < s->num_channels; c++)
525 s->channel[c].num_subframes = 0;
527 memset(num_samples, 0, sizeof(num_samples));
529 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
530 fixed_channel_layout = 1;
532 /** loop until the frame data is split between the subframes */
536 /** check which channels contain the subframe */
537 for (c = 0; c < s->num_channels; c++) {
538 if (num_samples[c] == min_channel_len) {
539 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
540 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
541 contains_subframe[c] = 1;
543 contains_subframe[c] = get_bits1(&s->gb);
545 contains_subframe[c] = 0;
548 /** get subframe length, subframe_len == 0 is not allowed */
549 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
550 return AVERROR_INVALIDDATA;
552 /** add subframes to the individual channels and find new min_channel_len */
553 min_channel_len += subframe_len;
554 for (c = 0; c < s->num_channels; c++) {
555 WMAProChannelCtx* chan = &s->channel[c];
557 if (contains_subframe[c]) {
558 if (chan->num_subframes >= MAX_SUBFRAMES) {
559 av_log(s->avctx, AV_LOG_ERROR,
560 "broken frame: num subframes > 31\n");
561 return AVERROR_INVALIDDATA;
563 chan->subframe_len[chan->num_subframes] = subframe_len;
564 num_samples[c] += subframe_len;
565 ++chan->num_subframes;
566 if (num_samples[c] > s->samples_per_frame) {
567 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
568 "channel len > samples_per_frame\n");
569 return AVERROR_INVALIDDATA;
571 } else if (num_samples[c] <= min_channel_len) {
572 if (num_samples[c] < min_channel_len) {
573 channels_for_cur_subframe = 0;
574 min_channel_len = num_samples[c];
576 ++channels_for_cur_subframe;
579 } while (min_channel_len < s->samples_per_frame);
581 for (c = 0; c < s->num_channels; c++) {
584 for (i = 0; i < s->channel[c].num_subframes; i++) {
585 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
586 " len %i\n", s->frame_num, c, i,
587 s->channel[c].subframe_len[i]);
588 s->channel[c].subframe_offset[i] = offset;
589 offset += s->channel[c].subframe_len[i];
597 *@brief Calculate a decorrelation matrix from the bitstream parameters.
598 *@param s codec context
599 *@param chgroup channel group for which the matrix needs to be calculated
601 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
602 WMAProChannelGrp *chgroup)
606 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
607 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
608 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
610 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
611 rotation_offset[i] = get_bits(&s->gb, 6);
613 for (i = 0; i < chgroup->num_channels; i++)
614 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
615 get_bits1(&s->gb) ? 1.0 : -1.0;
617 for (i = 1; i < chgroup->num_channels; i++) {
619 for (x = 0; x < i; x++) {
621 for (y = 0; y < i + 1; y++) {
622 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
623 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
624 int n = rotation_offset[offset + x];
630 cosv = sin64[32 - n];
632 sinv = sin64[64 - n];
633 cosv = -sin64[n - 32];
636 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
637 (v1 * sinv) - (v2 * cosv);
638 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
639 (v1 * cosv) + (v2 * sinv);
647 *@brief Decode channel transformation parameters
648 *@param s codec context
649 *@return 0 in case of success, < 0 in case of bitstream errors
651 static int decode_channel_transform(WMAProDecodeCtx* s)
654 /* should never consume more than 1921 bits for the 8 channel case
655 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
656 * + MAX_CHANNELS + MAX_BANDS + 1)
659 /** in the one channel case channel transforms are pointless */
661 if (s->num_channels > 1) {
662 int remaining_channels = s->channels_for_cur_subframe;
664 if (get_bits1(&s->gb)) {
665 av_log_ask_for_sample(s->avctx,
666 "unsupported channel transform bit\n");
667 return AVERROR_INVALIDDATA;
670 for (s->num_chgroups = 0; remaining_channels &&
671 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
672 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
673 float** channel_data = chgroup->channel_data;
674 chgroup->num_channels = 0;
675 chgroup->transform = 0;
677 /** decode channel mask */
678 if (remaining_channels > 2) {
679 for (i = 0; i < s->channels_for_cur_subframe; i++) {
680 int channel_idx = s->channel_indexes_for_cur_subframe[i];
681 if (!s->channel[channel_idx].grouped
682 && get_bits1(&s->gb)) {
683 ++chgroup->num_channels;
684 s->channel[channel_idx].grouped = 1;
685 *channel_data++ = s->channel[channel_idx].coeffs;
689 chgroup->num_channels = remaining_channels;
690 for (i = 0; i < s->channels_for_cur_subframe; i++) {
691 int channel_idx = s->channel_indexes_for_cur_subframe[i];
692 if (!s->channel[channel_idx].grouped)
693 *channel_data++ = s->channel[channel_idx].coeffs;
694 s->channel[channel_idx].grouped = 1;
698 /** decode transform type */
699 if (chgroup->num_channels == 2) {
700 if (get_bits1(&s->gb)) {
701 if (get_bits1(&s->gb)) {
702 av_log_ask_for_sample(s->avctx,
703 "unsupported channel transform type\n");
706 chgroup->transform = 1;
707 if (s->num_channels == 2) {
708 chgroup->decorrelation_matrix[0] = 1.0;
709 chgroup->decorrelation_matrix[1] = -1.0;
710 chgroup->decorrelation_matrix[2] = 1.0;
711 chgroup->decorrelation_matrix[3] = 1.0;
714 chgroup->decorrelation_matrix[0] = 0.70703125;
715 chgroup->decorrelation_matrix[1] = -0.70703125;
716 chgroup->decorrelation_matrix[2] = 0.70703125;
717 chgroup->decorrelation_matrix[3] = 0.70703125;
720 } else if (chgroup->num_channels > 2) {
721 if (get_bits1(&s->gb)) {
722 chgroup->transform = 1;
723 if (get_bits1(&s->gb)) {
724 decode_decorrelation_matrix(s, chgroup);
726 /** FIXME: more than 6 coupled channels not supported */
727 if (chgroup->num_channels > 6) {
728 av_log_ask_for_sample(s->avctx,
729 "coupled channels > 6\n");
731 memcpy(chgroup->decorrelation_matrix,
732 default_decorrelation[chgroup->num_channels],
733 chgroup->num_channels * chgroup->num_channels *
734 sizeof(*chgroup->decorrelation_matrix));
740 /** decode transform on / off */
741 if (chgroup->transform) {
742 if (!get_bits1(&s->gb)) {
744 /** transform can be enabled for individual bands */
745 for (i = 0; i < s->num_bands; i++) {
746 chgroup->transform_band[i] = get_bits1(&s->gb);
749 memset(chgroup->transform_band, 1, s->num_bands);
752 remaining_channels -= chgroup->num_channels;
759 *@brief Extract the coefficients from the bitstream.
760 *@param s codec context
761 *@param c current channel number
762 *@return 0 on success, < 0 in case of bitstream errors
764 static int decode_coeffs(WMAProDecodeCtx *s, int c)
766 /* Integers 0..15 as single-precision floats. The table saves a
767 costly int to float conversion, and storing the values as
768 integers allows fast sign-flipping. */
769 static const int fval_tab[16] = {
770 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
771 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
772 0x41000000, 0x41100000, 0x41200000, 0x41300000,
773 0x41400000, 0x41500000, 0x41600000, 0x41700000,
777 WMAProChannelCtx* ci = &s->channel[c];
784 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
786 vlctable = get_bits1(&s->gb);
787 vlc = &coef_vlc[vlctable];
797 /** decode vector coefficients (consumes up to 167 bits per iteration for
798 4 vector coded large values) */
799 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
800 (cur_coeff + 3 < ci->num_vec_coeffs)) {
805 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
807 if (idx == HUFF_VEC4_SIZE - 1) {
808 for (i = 0; i < 4; i += 2) {
809 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
810 if (idx == HUFF_VEC2_SIZE - 1) {
812 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
813 if (v0 == HUFF_VEC1_SIZE - 1)
814 v0 += ff_wma_get_large_val(&s->gb);
815 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
816 if (v1 == HUFF_VEC1_SIZE - 1)
817 v1 += ff_wma_get_large_val(&s->gb);
818 ((float*)vals)[i ] = v0;
819 ((float*)vals)[i+1] = v1;
821 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
822 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
826 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
827 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
828 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
829 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
833 for (i = 0; i < 4; i++) {
835 int sign = get_bits1(&s->gb) - 1;
836 *(uint32_t*)&ci->coeffs[cur_coeff] = vals[i] ^ sign<<31;
839 ci->coeffs[cur_coeff] = 0;
840 /** switch to run level mode when subframe_len / 128 zeros
841 were found in a row */
842 rl_mode |= (++num_zeros > s->subframe_len >> 8);
848 /** decode run level coded coefficients */
849 if (cur_coeff < s->subframe_len) {
850 memset(&ci->coeffs[cur_coeff], 0,
851 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
852 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
853 level, run, 1, ci->coeffs,
854 cur_coeff, s->subframe_len,
855 s->subframe_len, s->esc_len, 0))
856 return AVERROR_INVALIDDATA;
863 *@brief Extract scale factors from the bitstream.
864 *@param s codec context
865 *@return 0 on success, < 0 in case of bitstream errors
867 static int decode_scale_factors(WMAProDecodeCtx* s)
871 /** should never consume more than 5344 bits
872 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
875 for (i = 0; i < s->channels_for_cur_subframe; i++) {
876 int c = s->channel_indexes_for_cur_subframe[i];
879 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
880 sf_end = s->channel[c].scale_factors + s->num_bands;
882 /** resample scale factors for the new block size
883 * as the scale factors might need to be resampled several times
884 * before some new values are transmitted, a backup of the last
885 * transmitted scale factors is kept in saved_scale_factors
887 if (s->channel[c].reuse_sf) {
888 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
890 for (b = 0; b < s->num_bands; b++)
891 s->channel[c].scale_factors[b] =
892 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
895 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
897 if (!s->channel[c].reuse_sf) {
899 /** decode DPCM coded scale factors */
900 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
901 val = 45 / s->channel[c].scale_factor_step;
902 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
903 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
908 /** run level decode differences to the resampled factors */
909 for (i = 0; i < s->num_bands; i++) {
915 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
918 uint32_t code = get_bits(&s->gb, 14);
920 sign = (code & 1) - 1;
921 skip = (code & 0x3f) >> 1;
922 } else if (idx == 1) {
925 skip = scale_rl_run[idx];
926 val = scale_rl_level[idx];
927 sign = get_bits1(&s->gb)-1;
931 if (i >= s->num_bands) {
932 av_log(s->avctx, AV_LOG_ERROR,
933 "invalid scale factor coding\n");
934 return AVERROR_INVALIDDATA;
936 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
940 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
941 s->channel[c].table_idx = s->table_idx;
942 s->channel[c].reuse_sf = 1;
945 /** calculate new scale factor maximum */
946 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
947 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
948 s->channel[c].max_scale_factor =
949 FFMAX(s->channel[c].max_scale_factor, *sf);
957 *@brief Reconstruct the individual channel data.
958 *@param s codec context
960 static void inverse_channel_transform(WMAProDecodeCtx *s)
964 for (i = 0; i < s->num_chgroups; i++) {
965 if (s->chgroup[i].transform) {
966 float data[WMAPRO_MAX_CHANNELS];
967 const int num_channels = s->chgroup[i].num_channels;
968 float** ch_data = s->chgroup[i].channel_data;
969 float** ch_end = ch_data + num_channels;
970 const int8_t* tb = s->chgroup[i].transform_band;
973 /** multichannel decorrelation */
974 for (sfb = s->cur_sfb_offsets;
975 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
978 /** multiply values with the decorrelation_matrix */
979 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
980 const float* mat = s->chgroup[i].decorrelation_matrix;
981 const float* data_end = data + num_channels;
982 float* data_ptr = data;
985 for (ch = ch_data; ch < ch_end; ch++)
986 *data_ptr++ = (*ch)[y];
988 for (ch = ch_data; ch < ch_end; ch++) {
991 while (data_ptr < data_end)
992 sum += *data_ptr++ * *mat++;
997 } else if (s->num_channels == 2) {
998 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
999 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1000 ch_data[0] + sfb[0],
1002 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1003 ch_data[1] + sfb[0],
1012 *@brief Apply sine window and reconstruct the output buffer.
1013 *@param s codec context
1015 static void wmapro_window(WMAProDecodeCtx *s)
1018 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1019 int c = s->channel_indexes_for_cur_subframe[i];
1021 int winlen = s->channel[c].prev_block_len;
1022 float* start = s->channel[c].coeffs - (winlen >> 1);
1024 if (s->subframe_len < winlen) {
1025 start += (winlen - s->subframe_len) >> 1;
1026 winlen = s->subframe_len;
1029 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1033 s->dsp.vector_fmul_window(start, start, start + winlen,
1036 s->channel[c].prev_block_len = s->subframe_len;
1041 *@brief Decode a single subframe (block).
1042 *@param s codec context
1043 *@return 0 on success, < 0 when decoding failed
1045 static int decode_subframe(WMAProDecodeCtx *s)
1047 int offset = s->samples_per_frame;
1048 int subframe_len = s->samples_per_frame;
1050 int total_samples = s->samples_per_frame * s->num_channels;
1051 int transmit_coeffs = 0;
1052 int cur_subwoofer_cutoff;
1054 s->subframe_offset = get_bits_count(&s->gb);
1056 /** reset channel context and find the next block offset and size
1057 == the next block of the channel with the smallest number of
1060 for (i = 0; i < s->num_channels; i++) {
1061 s->channel[i].grouped = 0;
1062 if (offset > s->channel[i].decoded_samples) {
1063 offset = s->channel[i].decoded_samples;
1065 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1070 "processing subframe with offset %i len %i\n", offset, subframe_len);
1072 /** get a list of all channels that contain the estimated block */
1073 s->channels_for_cur_subframe = 0;
1074 for (i = 0; i < s->num_channels; i++) {
1075 const int cur_subframe = s->channel[i].cur_subframe;
1076 /** substract already processed samples */
1077 total_samples -= s->channel[i].decoded_samples;
1079 /** and count if there are multiple subframes that match our profile */
1080 if (offset == s->channel[i].decoded_samples &&
1081 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1082 total_samples -= s->channel[i].subframe_len[cur_subframe];
1083 s->channel[i].decoded_samples +=
1084 s->channel[i].subframe_len[cur_subframe];
1085 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1086 ++s->channels_for_cur_subframe;
1090 /** check if the frame will be complete after processing the
1093 s->parsed_all_subframes = 1;
1096 av_dlog(s->avctx, "subframe is part of %i channels\n",
1097 s->channels_for_cur_subframe);
1099 /** calculate number of scale factor bands and their offsets */
1100 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1101 s->num_bands = s->num_sfb[s->table_idx];
1102 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1103 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1105 /** configure the decoder for the current subframe */
1106 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1107 int c = s->channel_indexes_for_cur_subframe[i];
1109 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1113 s->subframe_len = subframe_len;
1114 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1116 /** skip extended header if any */
1117 if (get_bits1(&s->gb)) {
1119 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1120 int len = get_bits(&s->gb, 4);
1121 num_fill_bits = get_bits(&s->gb, len) + 1;
1124 if (num_fill_bits >= 0) {
1125 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1126 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1127 return AVERROR_INVALIDDATA;
1130 skip_bits_long(&s->gb, num_fill_bits);
1134 /** no idea for what the following bit is used */
1135 if (get_bits1(&s->gb)) {
1136 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1137 return AVERROR_INVALIDDATA;
1141 if (decode_channel_transform(s) < 0)
1142 return AVERROR_INVALIDDATA;
1145 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1146 int c = s->channel_indexes_for_cur_subframe[i];
1147 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1148 transmit_coeffs = 1;
1151 if (transmit_coeffs) {
1153 int quant_step = 90 * s->bits_per_sample >> 4;
1155 /** decode number of vector coded coefficients */
1156 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1157 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1158 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1159 int c = s->channel_indexes_for_cur_subframe[i];
1160 s->channel[c].num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
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 = s->subframe_len;
1168 /** decode quantization step */
1169 step = get_sbits(&s->gb, 6);
1171 if (step == -32 || step == 31) {
1172 const int sign = (step == 31) - 1;
1174 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1175 (step = get_bits(&s->gb, 5)) == 31) {
1178 quant_step += ((quant + step) ^ sign) - sign;
1180 if (quant_step < 0) {
1181 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1184 /** decode quantization step modifiers for every channel */
1186 if (s->channels_for_cur_subframe == 1) {
1187 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1189 int modifier_len = get_bits(&s->gb, 3);
1190 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1191 int c = s->channel_indexes_for_cur_subframe[i];
1192 s->channel[c].quant_step = quant_step;
1193 if (get_bits1(&s->gb)) {
1195 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1197 ++s->channel[c].quant_step;
1202 /** decode scale factors */
1203 if (decode_scale_factors(s) < 0)
1204 return AVERROR_INVALIDDATA;
1207 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1208 get_bits_count(&s->gb) - s->subframe_offset);
1210 /** parse coefficients */
1211 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1212 int c = s->channel_indexes_for_cur_subframe[i];
1213 if (s->channel[c].transmit_coefs &&
1214 get_bits_count(&s->gb) < s->num_saved_bits) {
1215 decode_coeffs(s, c);
1217 memset(s->channel[c].coeffs, 0,
1218 sizeof(*s->channel[c].coeffs) * subframe_len);
1221 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1222 get_bits_count(&s->gb) - s->subframe_offset);
1224 if (transmit_coeffs) {
1225 /** reconstruct the per channel data */
1226 inverse_channel_transform(s);
1227 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1228 int c = s->channel_indexes_for_cur_subframe[i];
1229 const int* sf = s->channel[c].scale_factors;
1232 if (c == s->lfe_channel)
1233 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1234 (subframe_len - cur_subwoofer_cutoff));
1236 /** inverse quantization and rescaling */
1237 for (b = 0; b < s->num_bands; b++) {
1238 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1239 const int exp = s->channel[c].quant_step -
1240 (s->channel[c].max_scale_factor - *sf++) *
1241 s->channel[c].scale_factor_step;
1242 const float quant = pow(10.0, exp / 20.0);
1243 int start = s->cur_sfb_offsets[b];
1244 s->dsp.vector_fmul_scalar(s->tmp + start,
1245 s->channel[c].coeffs + start,
1246 quant, end - start);
1249 /** apply imdct (ff_imdct_half == DCTIV with reverse) */
1250 ff_imdct_half(&s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS],
1251 s->channel[c].coeffs, s->tmp);
1255 /** window and overlapp-add */
1258 /** handled one subframe */
1259 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1260 int c = s->channel_indexes_for_cur_subframe[i];
1261 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1262 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1263 return AVERROR_INVALIDDATA;
1265 ++s->channel[c].cur_subframe;
1272 *@brief Decode one WMA frame.
1273 *@param s codec context
1274 *@return 0 if the trailer bit indicates that this is the last frame,
1275 * 1 if there are additional frames
1277 static int decode_frame(WMAProDecodeCtx *s)
1279 GetBitContext* gb = &s->gb;
1280 int more_frames = 0;
1284 /** check for potential output buffer overflow */
1285 if (s->num_channels * s->samples_per_frame > s->samples_end - s->samples) {
1286 /** return an error if no frame could be decoded at all */
1287 av_log(s->avctx, AV_LOG_ERROR,
1288 "not enough space for the output samples\n");
1293 /** get frame length */
1295 len = get_bits(gb, s->log2_frame_size);
1297 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1299 /** decode tile information */
1300 if (decode_tilehdr(s)) {
1305 /** read postproc transform */
1306 if (s->num_channels > 1 && get_bits1(gb)) {
1307 if (get_bits1(gb)) {
1308 for (i = 0; i < s->num_channels * s->num_channels; i++)
1313 /** read drc info */
1314 if (s->dynamic_range_compression) {
1315 s->drc_gain = get_bits(gb, 8);
1316 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1319 /** no idea what these are for, might be the number of samples
1320 that need to be skipped at the beginning or end of a stream */
1321 if (get_bits1(gb)) {
1324 /** usually true for the first frame */
1325 if (get_bits1(gb)) {
1326 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1327 av_dlog(s->avctx, "start skip: %i\n", skip);
1330 /** sometimes true for the last frame */
1331 if (get_bits1(gb)) {
1332 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1333 av_dlog(s->avctx, "end skip: %i\n", skip);
1338 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1339 get_bits_count(gb) - s->frame_offset);
1341 /** reset subframe states */
1342 s->parsed_all_subframes = 0;
1343 for (i = 0; i < s->num_channels; i++) {
1344 s->channel[i].decoded_samples = 0;
1345 s->channel[i].cur_subframe = 0;
1346 s->channel[i].reuse_sf = 0;
1349 /** decode all subframes */
1350 while (!s->parsed_all_subframes) {
1351 if (decode_subframe(s) < 0) {
1357 /** interleave samples and write them to the output buffer */
1358 for (i = 0; i < s->num_channels; i++) {
1359 float* ptr = s->samples + i;
1360 int incr = s->num_channels;
1361 float* iptr = s->channel[i].out;
1362 float* iend = iptr + s->samples_per_frame;
1364 // FIXME should create/use a DSP function here
1365 while (iptr < iend) {
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 = (s->num_saved_bits + 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 ff_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 ff_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 = {
1610 sizeof(WMAProDecodeCtx),
1615 .capabilities = CODEC_CAP_SUBFRAMES,
1617 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),