2 * Wmapro compatible decoder
3 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson
6 * This file is part of Libav.
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * @brief wmapro decoder implementation
26 * Wmapro is an MDCT based codec comparable to wma standard or AAC.
27 * The decoding therefore consists of the following steps:
28 * - bitstream decoding
29 * - reconstruction of per-channel data
30 * - rescaling and inverse quantization
32 * - windowing and overlapp-add
34 * The compressed wmapro bitstream is split into individual packets.
35 * Every such packet contains one or more wma frames.
36 * The compressed frames may have a variable length and frames may
37 * cross packet boundaries.
38 * Common to all wmapro frames is the number of samples that are stored in
40 * The number of samples and a few other decode flags are stored
41 * as extradata that has to be passed to the decoder.
43 * The wmapro frames themselves are again split into a variable number of
44 * subframes. Every subframe contains the data for 2^N time domain samples
45 * where N varies between 7 and 12.
47 * Example wmapro bitstream (in samples):
49 * || packet 0 || packet 1 || packet 2 packets
50 * ---------------------------------------------------
51 * || frame 0 || frame 1 || frame 2 || frames
52 * ---------------------------------------------------
53 * || | | || | | | || || subframes of channel 0
54 * ---------------------------------------------------
55 * || | | || | | | || || subframes of channel 1
56 * ---------------------------------------------------
58 * The frame layouts for the individual channels of a wma frame does not need
61 * However, if the offsets and lengths of several subframes of a frame are the
62 * same, the subframes of the channels can be grouped.
63 * Every group may then use special coding techniques like M/S stereo coding
64 * to improve the compression ratio. These channel transformations do not
65 * need to be applied to a whole subframe. Instead, they can also work on
66 * individual scale factor bands (see below).
67 * The coefficients that carry the audio signal in the frequency domain
68 * are transmitted as huffman-coded vectors with 4, 2 and 1 elements.
69 * In addition to that, the encoder can switch to a runlevel coding scheme
70 * by transmitting subframe_length / 128 zero coefficients.
72 * Before the audio signal can be converted to the time domain, the
73 * coefficients have to be rescaled and inverse quantized.
74 * A subframe is therefore split into several scale factor bands that get
75 * scaled individually.
76 * Scale factors are submitted for every frame but they might be shared
77 * between the subframes of a channel. Scale factors are initially DPCM-coded.
78 * Once scale factors are shared, the differences are transmitted as runlevel
80 * Every subframe length and offset combination in the frame layout shares a
81 * common quantization factor that can be adjusted for every channel by a
83 * After the inverse quantization, the coefficients get processed by an IMDCT.
84 * The resulting values are then windowed with a sine window and the first half
85 * of the values are added to the second half of the output from the previous
86 * subframe in order to reconstruct the output samples.
89 #include "libavutil/float_dsp.h"
90 #include "libavutil/intfloat.h"
91 #include "libavutil/intreadwrite.h"
96 #include "wmaprodata.h"
99 #include "wma_common.h"
101 /** current decoder limitations */
102 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
103 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
104 #define MAX_BANDS 29 ///< max number of scale factor bands
105 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
107 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
108 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
109 #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size
110 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
111 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
115 #define SCALEVLCBITS 8
116 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
117 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
118 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
119 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
120 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
122 static VLC sf_vlc; ///< scale factor DPCM vlc
123 static VLC sf_rl_vlc; ///< scale factor run length vlc
124 static VLC vec4_vlc; ///< 4 coefficients per symbol
125 static VLC vec2_vlc; ///< 2 coefficients per symbol
126 static VLC vec1_vlc; ///< 1 coefficient per symbol
127 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
128 static float sin64[33]; ///< sine table for decorrelation
131 * @brief frame specific decoder context for a single channel
134 int16_t prev_block_len; ///< length of the previous block
135 uint8_t transmit_coefs;
136 uint8_t num_subframes;
137 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
138 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
139 uint8_t cur_subframe; ///< current subframe number
140 uint16_t decoded_samples; ///< number of already processed samples
141 uint8_t grouped; ///< channel is part of a group
142 int quant_step; ///< quantization step for the current subframe
143 int8_t reuse_sf; ///< share scale factors between subframes
144 int8_t scale_factor_step; ///< scaling step for the current subframe
145 int max_scale_factor; ///< maximum scale factor for the current subframe
146 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
147 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
148 int* scale_factors; ///< pointer to the scale factor values used for decoding
149 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
150 float* coeffs; ///< pointer to the subframe decode buffer
151 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
152 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
156 * @brief channel group for channel transformations
159 uint8_t num_channels; ///< number of channels in the group
160 int8_t transform; ///< transform on / off
161 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
162 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
163 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
167 * @brief main decoder context
169 typedef struct WMAProDecodeCtx {
170 /* generic decoder variables */
171 AVCodecContext* avctx; ///< codec context for av_log
172 AVFloatDSPContext fdsp;
173 uint8_t frame_data[MAX_FRAMESIZE +
174 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
175 PutBitContext pb; ///< context for filling the frame_data buffer
176 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
177 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
178 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
180 /* frame size dependent frame information (set during initialization) */
181 uint32_t decode_flags; ///< used compression features
182 uint8_t len_prefix; ///< frame is prefixed with its length
183 uint8_t dynamic_range_compression; ///< frame contains DRC data
184 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
185 uint16_t samples_per_frame; ///< number of samples to output
186 uint16_t log2_frame_size;
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 uint8_t drc_gain; ///< gain for the DRC tool
213 int8_t skip_frame; ///< skip output step
214 int8_t parsed_all_subframes; ///< all subframes decoded?
216 /* subframe/block decode state */
217 int16_t subframe_len; ///< current subframe length
218 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
219 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
220 int8_t num_bands; ///< number of scale factor bands
221 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
222 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
223 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
224 int8_t esc_len; ///< length of escaped coefficients
226 uint8_t num_chgroups; ///< number of channel groups
227 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
229 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
234 *@brief helper function to print the most important members of the context
237 static av_cold void dump_context(WMAProDecodeCtx *s)
239 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
240 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
242 PRINT("ed sample bit depth", s->bits_per_sample);
243 PRINT_HEX("ed decode flags", s->decode_flags);
244 PRINT("samples per frame", s->samples_per_frame);
245 PRINT("log2 frame size", s->log2_frame_size);
246 PRINT("max num subframes", s->max_num_subframes);
247 PRINT("len prefix", s->len_prefix);
248 PRINT("num channels", s->avctx->channels);
252 *@brief Uninitialize the decoder and free all resources.
253 *@param avctx codec context
254 *@return 0 on success, < 0 otherwise
256 static av_cold int decode_end(AVCodecContext *avctx)
258 WMAProDecodeCtx *s = avctx->priv_data;
261 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
262 ff_mdct_end(&s->mdct_ctx[i]);
268 *@brief Initialize the decoder.
269 *@param avctx codec context
270 *@return 0 on success, -1 otherwise
272 static av_cold int decode_init(AVCodecContext *avctx)
274 WMAProDecodeCtx *s = avctx->priv_data;
275 uint8_t *edata_ptr = avctx->extradata;
276 unsigned int channel_mask;
278 int log2_max_num_subframes;
279 int num_possible_block_sizes;
281 if (!avctx->block_align) {
282 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
283 return AVERROR(EINVAL);
287 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
289 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
291 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
293 if (avctx->extradata_size >= 18) {
294 s->decode_flags = AV_RL16(edata_ptr+14);
295 channel_mask = AV_RL32(edata_ptr+2);
296 s->bits_per_sample = AV_RL16(edata_ptr);
297 /** dump the extradata */
298 for (i = 0; i < avctx->extradata_size; i++)
299 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
300 av_dlog(avctx, "\n");
303 avpriv_request_sample(avctx, "Unknown extradata size");
304 return AVERROR_PATCHWELCOME;
308 s->log2_frame_size = av_log2(avctx->block_align) + 4;
311 s->skip_frame = 1; /* skip first frame */
313 s->len_prefix = (s->decode_flags & 0x40);
316 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
317 if (bits > WMAPRO_BLOCK_MAX_BITS) {
318 avpriv_request_sample(avctx, "14-bit block sizes");
319 return AVERROR_PATCHWELCOME;
321 s->samples_per_frame = 1 << bits;
324 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
325 s->max_num_subframes = 1 << log2_max_num_subframes;
326 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
327 s->max_subframe_len_bit = 1;
328 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
330 num_possible_block_sizes = log2_max_num_subframes + 1;
331 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
332 s->dynamic_range_compression = (s->decode_flags & 0x80);
334 if (s->max_num_subframes > MAX_SUBFRAMES) {
335 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
336 s->max_num_subframes);
337 return AVERROR_INVALIDDATA;
340 if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) {
341 av_log(avctx, AV_LOG_ERROR, "Invalid minimum block size %i\n",
342 s->max_num_subframes);
343 return AVERROR_INVALIDDATA;
346 if (s->avctx->sample_rate <= 0) {
347 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
348 return AVERROR_INVALIDDATA;
351 if (avctx->channels < 0) {
352 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
354 return AVERROR_INVALIDDATA;
355 } else if (avctx->channels > WMAPRO_MAX_CHANNELS) {
356 avpriv_request_sample(avctx,
357 "More than %d channels", WMAPRO_MAX_CHANNELS);
358 return AVERROR_PATCHWELCOME;
361 /** init previous block len */
362 for (i = 0; i < avctx->channels; i++)
363 s->channel[i].prev_block_len = s->samples_per_frame;
365 /** extract lfe channel position */
368 if (channel_mask & 8) {
370 for (mask = 1; mask < 16; mask <<= 1) {
371 if (channel_mask & mask)
376 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
377 scale_huffbits, 1, 1,
378 scale_huffcodes, 2, 2, 616);
380 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
381 scale_rl_huffbits, 1, 1,
382 scale_rl_huffcodes, 4, 4, 1406);
384 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
385 coef0_huffbits, 1, 1,
386 coef0_huffcodes, 4, 4, 2108);
388 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
389 coef1_huffbits, 1, 1,
390 coef1_huffcodes, 4, 4, 3912);
392 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
394 vec4_huffcodes, 2, 2, 604);
396 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
398 vec2_huffcodes, 2, 2, 562);
400 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
402 vec1_huffcodes, 2, 2, 562);
404 /** calculate number of scale factor bands and their offsets
405 for every possible block size */
406 for (i = 0; i < num_possible_block_sizes; i++) {
407 int subframe_len = s->samples_per_frame >> i;
411 s->sfb_offsets[i][0] = 0;
413 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
414 int offset = (subframe_len * 2 * critical_freq[x])
415 / s->avctx->sample_rate + 2;
417 if (offset > s->sfb_offsets[i][band - 1])
418 s->sfb_offsets[i][band++] = offset;
420 s->sfb_offsets[i][band - 1] = subframe_len;
421 s->num_sfb[i] = band - 1;
425 /** Scale factors can be shared between blocks of different size
426 as every block has a different scale factor band layout.
427 The matrix sf_offsets is needed to find the correct scale factor.
430 for (i = 0; i < num_possible_block_sizes; i++) {
432 for (b = 0; b < s->num_sfb[i]; b++) {
434 int offset = ((s->sfb_offsets[i][b]
435 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
436 for (x = 0; x < num_possible_block_sizes; x++) {
438 while (s->sfb_offsets[x][v + 1] << x < offset)
439 if (++v >= MAX_BANDS)
440 return AVERROR_INVALIDDATA;
441 s->sf_offsets[i][x][b] = v;
446 /** init MDCT, FIXME: only init needed sizes */
447 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
448 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
449 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
450 / (1 << (s->bits_per_sample - 1)));
452 /** init MDCT windows: simple sine window */
453 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
454 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
455 ff_init_ff_sine_windows(win_idx);
456 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
459 /** calculate subwoofer cutoff values */
460 for (i = 0; i < num_possible_block_sizes; i++) {
461 int block_size = s->samples_per_frame >> i;
462 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
463 / s->avctx->sample_rate;
464 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
467 /** calculate sine values for the decorrelation matrix */
468 for (i = 0; i < 33; i++)
469 sin64[i] = sin(i*M_PI / 64.0);
471 if (avctx->debug & FF_DEBUG_BITSTREAM)
474 avctx->channel_layout = channel_mask;
480 *@brief Decode the subframe length.
482 *@param offset sample offset in the frame
483 *@return decoded subframe length on success, < 0 in case of an error
485 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
487 int frame_len_shift = 0;
490 /** no need to read from the bitstream when only one length is possible */
491 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
492 return s->min_samples_per_subframe;
494 /** 1 bit indicates if the subframe is of maximum length */
495 if (s->max_subframe_len_bit) {
496 if (get_bits1(&s->gb))
497 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
499 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
501 subframe_len = s->samples_per_frame >> frame_len_shift;
503 /** sanity check the length */
504 if (subframe_len < s->min_samples_per_subframe ||
505 subframe_len > s->samples_per_frame) {
506 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
508 return AVERROR_INVALIDDATA;
514 *@brief Decode how the data in the frame is split into subframes.
515 * Every WMA frame contains the encoded data for a fixed number of
516 * samples per channel. The data for every channel might be split
517 * into several subframes. This function will reconstruct the list of
518 * subframes for every channel.
520 * If the subframes are not evenly split, the algorithm estimates the
521 * channels with the lowest number of total samples.
522 * Afterwards, for each of these channels a bit is read from the
523 * bitstream that indicates if the channel contains a subframe with the
524 * next subframe size that is going to be read from the bitstream or not.
525 * If a channel contains such a subframe, the subframe size gets added to
526 * the channel's subframe list.
527 * The algorithm repeats these steps until the frame is properly divided
528 * between the individual channels.
531 *@return 0 on success, < 0 in case of an error
533 static int decode_tilehdr(WMAProDecodeCtx *s)
535 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
536 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
537 int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */
538 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
539 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
542 /* Should never consume more than 3073 bits (256 iterations for the
543 * while loop when always the minimum amount of 128 samples is subtracted
544 * from missing samples in the 8 channel case).
545 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
548 /** reset tiling information */
549 for (c = 0; c < s->avctx->channels; c++)
550 s->channel[c].num_subframes = 0;
552 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
553 fixed_channel_layout = 1;
555 /** loop until the frame data is split between the subframes */
559 /** check which channels contain the subframe */
560 for (c = 0; c < s->avctx->channels; c++) {
561 if (num_samples[c] == min_channel_len) {
562 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
563 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
564 contains_subframe[c] = 1;
566 contains_subframe[c] = get_bits1(&s->gb);
568 contains_subframe[c] = 0;
571 /** get subframe length, subframe_len == 0 is not allowed */
572 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
573 return AVERROR_INVALIDDATA;
575 /** add subframes to the individual channels and find new min_channel_len */
576 min_channel_len += subframe_len;
577 for (c = 0; c < s->avctx->channels; c++) {
578 WMAProChannelCtx* chan = &s->channel[c];
580 if (contains_subframe[c]) {
581 if (chan->num_subframes >= MAX_SUBFRAMES) {
582 av_log(s->avctx, AV_LOG_ERROR,
583 "broken frame: num subframes > 31\n");
584 return AVERROR_INVALIDDATA;
586 chan->subframe_len[chan->num_subframes] = subframe_len;
587 num_samples[c] += subframe_len;
588 ++chan->num_subframes;
589 if (num_samples[c] > s->samples_per_frame) {
590 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
591 "channel len > samples_per_frame\n");
592 return AVERROR_INVALIDDATA;
594 } else if (num_samples[c] <= min_channel_len) {
595 if (num_samples[c] < min_channel_len) {
596 channels_for_cur_subframe = 0;
597 min_channel_len = num_samples[c];
599 ++channels_for_cur_subframe;
602 } while (min_channel_len < s->samples_per_frame);
604 for (c = 0; c < s->avctx->channels; c++) {
607 for (i = 0; i < s->channel[c].num_subframes; i++) {
608 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
609 " len %i\n", s->frame_num, c, i,
610 s->channel[c].subframe_len[i]);
611 s->channel[c].subframe_offset[i] = offset;
612 offset += s->channel[c].subframe_len[i];
620 *@brief Calculate a decorrelation matrix from the bitstream parameters.
621 *@param s codec context
622 *@param chgroup channel group for which the matrix needs to be calculated
624 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
625 WMAProChannelGrp *chgroup)
629 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
630 memset(chgroup->decorrelation_matrix, 0, s->avctx->channels *
631 s->avctx->channels * sizeof(*chgroup->decorrelation_matrix));
633 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
634 rotation_offset[i] = get_bits(&s->gb, 6);
636 for (i = 0; i < chgroup->num_channels; i++)
637 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
638 get_bits1(&s->gb) ? 1.0 : -1.0;
640 for (i = 1; i < chgroup->num_channels; i++) {
642 for (x = 0; x < i; x++) {
644 for (y = 0; y < i + 1; y++) {
645 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
646 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
647 int n = rotation_offset[offset + x];
653 cosv = sin64[32 - n];
655 sinv = sin64[64 - n];
656 cosv = -sin64[n - 32];
659 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
660 (v1 * sinv) - (v2 * cosv);
661 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
662 (v1 * cosv) + (v2 * sinv);
670 *@brief Decode channel transformation parameters
671 *@param s codec context
672 *@return 0 in case of success, < 0 in case of bitstream errors
674 static int decode_channel_transform(WMAProDecodeCtx* s)
677 /* should never consume more than 1921 bits for the 8 channel case
678 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
679 * + MAX_CHANNELS + MAX_BANDS + 1)
682 /** in the one channel case channel transforms are pointless */
684 if (s->avctx->channels > 1) {
685 int remaining_channels = s->channels_for_cur_subframe;
687 if (get_bits1(&s->gb)) {
688 avpriv_request_sample(s->avctx,
689 "Channel transform bit");
690 return AVERROR_PATCHWELCOME;
693 for (s->num_chgroups = 0; remaining_channels &&
694 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
695 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
696 float** channel_data = chgroup->channel_data;
697 chgroup->num_channels = 0;
698 chgroup->transform = 0;
700 /** decode channel mask */
701 if (remaining_channels > 2) {
702 for (i = 0; i < s->channels_for_cur_subframe; i++) {
703 int channel_idx = s->channel_indexes_for_cur_subframe[i];
704 if (!s->channel[channel_idx].grouped
705 && get_bits1(&s->gb)) {
706 ++chgroup->num_channels;
707 s->channel[channel_idx].grouped = 1;
708 *channel_data++ = s->channel[channel_idx].coeffs;
712 chgroup->num_channels = remaining_channels;
713 for (i = 0; i < s->channels_for_cur_subframe; i++) {
714 int channel_idx = s->channel_indexes_for_cur_subframe[i];
715 if (!s->channel[channel_idx].grouped)
716 *channel_data++ = s->channel[channel_idx].coeffs;
717 s->channel[channel_idx].grouped = 1;
721 /** decode transform type */
722 if (chgroup->num_channels == 2) {
723 if (get_bits1(&s->gb)) {
724 if (get_bits1(&s->gb)) {
725 avpriv_request_sample(s->avctx,
726 "Unknown channel transform type");
727 return AVERROR_PATCHWELCOME;
730 chgroup->transform = 1;
731 if (s->avctx->channels == 2) {
732 chgroup->decorrelation_matrix[0] = 1.0;
733 chgroup->decorrelation_matrix[1] = -1.0;
734 chgroup->decorrelation_matrix[2] = 1.0;
735 chgroup->decorrelation_matrix[3] = 1.0;
738 chgroup->decorrelation_matrix[0] = 0.70703125;
739 chgroup->decorrelation_matrix[1] = -0.70703125;
740 chgroup->decorrelation_matrix[2] = 0.70703125;
741 chgroup->decorrelation_matrix[3] = 0.70703125;
744 } else if (chgroup->num_channels > 2) {
745 if (get_bits1(&s->gb)) {
746 chgroup->transform = 1;
747 if (get_bits1(&s->gb)) {
748 decode_decorrelation_matrix(s, chgroup);
750 /** FIXME: more than 6 coupled channels not supported */
751 if (chgroup->num_channels > 6) {
752 avpriv_request_sample(s->avctx,
753 "Coupled channels > 6");
755 memcpy(chgroup->decorrelation_matrix,
756 default_decorrelation[chgroup->num_channels],
757 chgroup->num_channels * chgroup->num_channels *
758 sizeof(*chgroup->decorrelation_matrix));
764 /** decode transform on / off */
765 if (chgroup->transform) {
766 if (!get_bits1(&s->gb)) {
768 /** transform can be enabled for individual bands */
769 for (i = 0; i < s->num_bands; i++) {
770 chgroup->transform_band[i] = get_bits1(&s->gb);
773 memset(chgroup->transform_band, 1, s->num_bands);
776 remaining_channels -= chgroup->num_channels;
783 *@brief Extract the coefficients from the bitstream.
784 *@param s codec context
785 *@param c current channel number
786 *@return 0 on success, < 0 in case of bitstream errors
788 static int decode_coeffs(WMAProDecodeCtx *s, int c)
790 /* Integers 0..15 as single-precision floats. The table saves a
791 costly int to float conversion, and storing the values as
792 integers allows fast sign-flipping. */
793 static const uint32_t fval_tab[16] = {
794 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
795 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
796 0x41000000, 0x41100000, 0x41200000, 0x41300000,
797 0x41400000, 0x41500000, 0x41600000, 0x41700000,
801 WMAProChannelCtx* ci = &s->channel[c];
808 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
810 vlctable = get_bits1(&s->gb);
811 vlc = &coef_vlc[vlctable];
821 /** decode vector coefficients (consumes up to 167 bits per iteration for
822 4 vector coded large values) */
823 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
824 (cur_coeff + 3 < ci->num_vec_coeffs)) {
829 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
831 if (idx == HUFF_VEC4_SIZE - 1) {
832 for (i = 0; i < 4; i += 2) {
833 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
834 if (idx == HUFF_VEC2_SIZE - 1) {
836 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
837 if (v0 == HUFF_VEC1_SIZE - 1)
838 v0 += ff_wma_get_large_val(&s->gb);
839 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
840 if (v1 == HUFF_VEC1_SIZE - 1)
841 v1 += ff_wma_get_large_val(&s->gb);
842 vals[i ] = av_float2int(v0);
843 vals[i+1] = av_float2int(v1);
845 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
846 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
850 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
851 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
852 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
853 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
857 for (i = 0; i < 4; i++) {
859 uint32_t sign = get_bits1(&s->gb) - 1;
860 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
863 ci->coeffs[cur_coeff] = 0;
864 /** switch to run level mode when subframe_len / 128 zeros
865 were found in a row */
866 rl_mode |= (++num_zeros > s->subframe_len >> 8);
872 /** decode run level coded coefficients */
873 if (cur_coeff < s->subframe_len) {
874 memset(&ci->coeffs[cur_coeff], 0,
875 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
876 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
877 level, run, 1, ci->coeffs,
878 cur_coeff, s->subframe_len,
879 s->subframe_len, s->esc_len, 0))
880 return AVERROR_INVALIDDATA;
887 *@brief Extract scale factors from the bitstream.
888 *@param s codec context
889 *@return 0 on success, < 0 in case of bitstream errors
891 static int decode_scale_factors(WMAProDecodeCtx* s)
895 /** should never consume more than 5344 bits
896 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
899 for (i = 0; i < s->channels_for_cur_subframe; i++) {
900 int c = s->channel_indexes_for_cur_subframe[i];
903 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
904 sf_end = s->channel[c].scale_factors + s->num_bands;
906 /** resample scale factors for the new block size
907 * as the scale factors might need to be resampled several times
908 * before some new values are transmitted, a backup of the last
909 * transmitted scale factors is kept in saved_scale_factors
911 if (s->channel[c].reuse_sf) {
912 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
914 for (b = 0; b < s->num_bands; b++)
915 s->channel[c].scale_factors[b] =
916 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
919 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
921 if (!s->channel[c].reuse_sf) {
923 /** decode DPCM coded scale factors */
924 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
925 val = 45 / s->channel[c].scale_factor_step;
926 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
927 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
932 /** run level decode differences to the resampled factors */
933 for (i = 0; i < s->num_bands; i++) {
939 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
942 uint32_t code = get_bits(&s->gb, 14);
944 sign = (code & 1) - 1;
945 skip = (code & 0x3f) >> 1;
946 } else if (idx == 1) {
949 skip = scale_rl_run[idx];
950 val = scale_rl_level[idx];
951 sign = get_bits1(&s->gb)-1;
955 if (i >= s->num_bands) {
956 av_log(s->avctx, AV_LOG_ERROR,
957 "invalid scale factor coding\n");
958 return AVERROR_INVALIDDATA;
960 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
964 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
965 s->channel[c].table_idx = s->table_idx;
966 s->channel[c].reuse_sf = 1;
969 /** calculate new scale factor maximum */
970 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
971 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
972 s->channel[c].max_scale_factor =
973 FFMAX(s->channel[c].max_scale_factor, *sf);
981 *@brief Reconstruct the individual channel data.
982 *@param s codec context
984 static void inverse_channel_transform(WMAProDecodeCtx *s)
988 for (i = 0; i < s->num_chgroups; i++) {
989 if (s->chgroup[i].transform) {
990 float data[WMAPRO_MAX_CHANNELS];
991 const int num_channels = s->chgroup[i].num_channels;
992 float** ch_data = s->chgroup[i].channel_data;
993 float** ch_end = ch_data + num_channels;
994 const int8_t* tb = s->chgroup[i].transform_band;
997 /** multichannel decorrelation */
998 for (sfb = s->cur_sfb_offsets;
999 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1002 /** multiply values with the decorrelation_matrix */
1003 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1004 const float* mat = s->chgroup[i].decorrelation_matrix;
1005 const float* data_end = data + num_channels;
1006 float* data_ptr = data;
1009 for (ch = ch_data; ch < ch_end; ch++)
1010 *data_ptr++ = (*ch)[y];
1012 for (ch = ch_data; ch < ch_end; ch++) {
1015 while (data_ptr < data_end)
1016 sum += *data_ptr++ * *mat++;
1021 } else if (s->avctx->channels == 2) {
1022 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1023 s->fdsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1024 ch_data[0] + sfb[0],
1026 s->fdsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1027 ch_data[1] + sfb[0],
1036 *@brief Apply sine window and reconstruct the output buffer.
1037 *@param s codec context
1039 static void wmapro_window(WMAProDecodeCtx *s)
1042 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1043 int c = s->channel_indexes_for_cur_subframe[i];
1045 int winlen = s->channel[c].prev_block_len;
1046 float* start = s->channel[c].coeffs - (winlen >> 1);
1048 if (s->subframe_len < winlen) {
1049 start += (winlen - s->subframe_len) >> 1;
1050 winlen = s->subframe_len;
1053 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1057 s->fdsp.vector_fmul_window(start, start, start + winlen,
1060 s->channel[c].prev_block_len = s->subframe_len;
1065 *@brief Decode a single subframe (block).
1066 *@param s codec context
1067 *@return 0 on success, < 0 when decoding failed
1069 static int decode_subframe(WMAProDecodeCtx *s)
1071 int offset = s->samples_per_frame;
1072 int subframe_len = s->samples_per_frame;
1074 int total_samples = s->samples_per_frame * s->avctx->channels;
1075 int transmit_coeffs = 0;
1076 int cur_subwoofer_cutoff;
1078 s->subframe_offset = get_bits_count(&s->gb);
1080 /** reset channel context and find the next block offset and size
1081 == the next block of the channel with the smallest number of
1084 for (i = 0; i < s->avctx->channels; i++) {
1085 s->channel[i].grouped = 0;
1086 if (offset > s->channel[i].decoded_samples) {
1087 offset = s->channel[i].decoded_samples;
1089 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1094 "processing subframe with offset %i len %i\n", offset, subframe_len);
1096 /** get a list of all channels that contain the estimated block */
1097 s->channels_for_cur_subframe = 0;
1098 for (i = 0; i < s->avctx->channels; i++) {
1099 const int cur_subframe = s->channel[i].cur_subframe;
1100 /** subtract already processed samples */
1101 total_samples -= s->channel[i].decoded_samples;
1103 /** and count if there are multiple subframes that match our profile */
1104 if (offset == s->channel[i].decoded_samples &&
1105 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1106 total_samples -= s->channel[i].subframe_len[cur_subframe];
1107 s->channel[i].decoded_samples +=
1108 s->channel[i].subframe_len[cur_subframe];
1109 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1110 ++s->channels_for_cur_subframe;
1114 /** check if the frame will be complete after processing the
1117 s->parsed_all_subframes = 1;
1120 av_dlog(s->avctx, "subframe is part of %i channels\n",
1121 s->channels_for_cur_subframe);
1123 /** calculate number of scale factor bands and their offsets */
1124 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1125 s->num_bands = s->num_sfb[s->table_idx];
1126 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1127 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1129 /** configure the decoder for the current subframe */
1130 offset += s->samples_per_frame >> 1;
1132 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1133 int c = s->channel_indexes_for_cur_subframe[i];
1135 s->channel[c].coeffs = &s->channel[c].out[offset];
1138 s->subframe_len = subframe_len;
1139 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1141 /** skip extended header if any */
1142 if (get_bits1(&s->gb)) {
1144 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1145 int len = get_bits(&s->gb, 4);
1146 num_fill_bits = get_bits(&s->gb, len) + 1;
1149 if (num_fill_bits >= 0) {
1150 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1151 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1152 return AVERROR_INVALIDDATA;
1155 skip_bits_long(&s->gb, num_fill_bits);
1159 /** no idea for what the following bit is used */
1160 if (get_bits1(&s->gb)) {
1161 avpriv_request_sample(s->avctx, "Reserved bit");
1162 return AVERROR_PATCHWELCOME;
1166 if (decode_channel_transform(s) < 0)
1167 return AVERROR_INVALIDDATA;
1170 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1171 int c = s->channel_indexes_for_cur_subframe[i];
1172 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1173 transmit_coeffs = 1;
1176 if (transmit_coeffs) {
1178 int quant_step = 90 * s->bits_per_sample >> 4;
1180 /** decode number of vector coded coefficients */
1181 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1182 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1183 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1184 int c = s->channel_indexes_for_cur_subframe[i];
1185 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1186 if (num_vec_coeffs + offset > FF_ARRAY_ELEMS(s->channel[c].out)) {
1187 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1188 return AVERROR_INVALIDDATA;
1190 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1193 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1194 int c = s->channel_indexes_for_cur_subframe[i];
1195 s->channel[c].num_vec_coeffs = s->subframe_len;
1198 /** decode quantization step */
1199 step = get_sbits(&s->gb, 6);
1201 if (step == -32 || step == 31) {
1202 const int sign = (step == 31) - 1;
1204 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1205 (step = get_bits(&s->gb, 5)) == 31) {
1208 quant_step += ((quant + step) ^ sign) - sign;
1210 if (quant_step < 0) {
1211 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1214 /** decode quantization step modifiers for every channel */
1216 if (s->channels_for_cur_subframe == 1) {
1217 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1219 int modifier_len = get_bits(&s->gb, 3);
1220 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1221 int c = s->channel_indexes_for_cur_subframe[i];
1222 s->channel[c].quant_step = quant_step;
1223 if (get_bits1(&s->gb)) {
1225 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1227 ++s->channel[c].quant_step;
1232 /** decode scale factors */
1233 if (decode_scale_factors(s) < 0)
1234 return AVERROR_INVALIDDATA;
1237 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1238 get_bits_count(&s->gb) - s->subframe_offset);
1240 /** parse coefficients */
1241 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1242 int c = s->channel_indexes_for_cur_subframe[i];
1243 if (s->channel[c].transmit_coefs &&
1244 get_bits_count(&s->gb) < s->num_saved_bits) {
1245 decode_coeffs(s, c);
1247 memset(s->channel[c].coeffs, 0,
1248 sizeof(*s->channel[c].coeffs) * subframe_len);
1251 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1252 get_bits_count(&s->gb) - s->subframe_offset);
1254 if (transmit_coeffs) {
1255 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1256 /** reconstruct the per channel data */
1257 inverse_channel_transform(s);
1258 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1259 int c = s->channel_indexes_for_cur_subframe[i];
1260 const int* sf = s->channel[c].scale_factors;
1263 if (c == s->lfe_channel)
1264 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1265 (subframe_len - cur_subwoofer_cutoff));
1267 /** inverse quantization and rescaling */
1268 for (b = 0; b < s->num_bands; b++) {
1269 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1270 const int exp = s->channel[c].quant_step -
1271 (s->channel[c].max_scale_factor - *sf++) *
1272 s->channel[c].scale_factor_step;
1273 const float quant = pow(10.0, exp / 20.0);
1274 int start = s->cur_sfb_offsets[b];
1275 s->fdsp.vector_fmul_scalar(s->tmp + start,
1276 s->channel[c].coeffs + start,
1277 quant, end - start);
1280 /** apply imdct (imdct_half == DCTIV with reverse) */
1281 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1285 /** window and overlapp-add */
1288 /** handled one subframe */
1289 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1290 int c = s->channel_indexes_for_cur_subframe[i];
1291 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1292 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1293 return AVERROR_INVALIDDATA;
1295 ++s->channel[c].cur_subframe;
1302 *@brief Decode one WMA frame.
1303 *@param s codec context
1304 *@return 0 if the trailer bit indicates that this is the last frame,
1305 * 1 if there are additional frames
1307 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1309 AVCodecContext *avctx = s->avctx;
1310 GetBitContext* gb = &s->gb;
1311 int more_frames = 0;
1315 /** get frame length */
1317 len = get_bits(gb, s->log2_frame_size);
1319 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1321 /** decode tile information */
1322 if (decode_tilehdr(s)) {
1327 /** read postproc transform */
1328 if (s->avctx->channels > 1 && get_bits1(gb)) {
1329 if (get_bits1(gb)) {
1330 for (i = 0; i < avctx->channels * avctx->channels; i++)
1335 /** read drc info */
1336 if (s->dynamic_range_compression) {
1337 s->drc_gain = get_bits(gb, 8);
1338 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1341 /** no idea what these are for, might be the number of samples
1342 that need to be skipped at the beginning or end of a stream */
1343 if (get_bits1(gb)) {
1346 /** usually true for the first frame */
1347 if (get_bits1(gb)) {
1348 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1349 av_dlog(s->avctx, "start skip: %i\n", skip);
1352 /** sometimes true for the last frame */
1353 if (get_bits1(gb)) {
1354 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1355 av_dlog(s->avctx, "end skip: %i\n", skip);
1360 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1361 get_bits_count(gb) - s->frame_offset);
1363 /** reset subframe states */
1364 s->parsed_all_subframes = 0;
1365 for (i = 0; i < avctx->channels; i++) {
1366 s->channel[i].decoded_samples = 0;
1367 s->channel[i].cur_subframe = 0;
1368 s->channel[i].reuse_sf = 0;
1371 /** decode all subframes */
1372 while (!s->parsed_all_subframes) {
1373 if (decode_subframe(s) < 0) {
1379 /* get output buffer */
1380 frame->nb_samples = s->samples_per_frame;
1381 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1382 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1387 /** copy samples to the output buffer */
1388 for (i = 0; i < avctx->channels; i++)
1389 memcpy(frame->extended_data[i], s->channel[i].out,
1390 s->samples_per_frame * sizeof(*s->channel[i].out));
1392 for (i = 0; i < avctx->channels; i++) {
1393 /** reuse second half of the IMDCT output for the next frame */
1394 memcpy(&s->channel[i].out[0],
1395 &s->channel[i].out[s->samples_per_frame],
1396 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1399 if (s->skip_frame) {
1402 av_frame_unref(frame);
1407 if (s->len_prefix) {
1408 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1409 /** FIXME: not sure if this is always an error */
1410 av_log(s->avctx, AV_LOG_ERROR,
1411 "frame[%i] would have to skip %i bits\n", s->frame_num,
1412 len - (get_bits_count(gb) - s->frame_offset) - 1);
1417 /** skip the rest of the frame data */
1418 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1420 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1424 /** decode trailer bit */
1425 more_frames = get_bits1(gb);
1432 *@brief Calculate remaining input buffer length.
1433 *@param s codec context
1434 *@param gb bitstream reader context
1435 *@return remaining size in bits
1437 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1439 return s->buf_bit_size - get_bits_count(gb);
1443 *@brief Fill the bit reservoir with a (partial) frame.
1444 *@param s codec context
1445 *@param gb bitstream reader context
1446 *@param len length of the partial frame
1447 *@param append decides whether to reset the buffer or not
1449 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1454 /** when the frame data does not need to be concatenated, the input buffer
1455 is resetted and additional bits from the previous frame are copyed
1456 and skipped later so that a fast byte copy is possible */
1459 s->frame_offset = get_bits_count(gb) & 7;
1460 s->num_saved_bits = s->frame_offset;
1461 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1464 buflen = (s->num_saved_bits + len + 8) >> 3;
1466 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1467 avpriv_request_sample(s->avctx, "Too small input buffer");
1472 if (len > put_bits_left(&s->pb)) {
1473 av_log(s->avctx, AV_LOG_ERROR,
1474 "Cannot append %d bits, only %d bits available.\n",
1475 len, put_bits_left(&s->pb));
1480 s->num_saved_bits += len;
1482 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1485 int align = 8 - (get_bits_count(gb) & 7);
1486 align = FFMIN(align, len);
1487 put_bits(&s->pb, align, get_bits(gb, align));
1489 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1491 skip_bits_long(gb, len);
1494 PutBitContext tmp = s->pb;
1495 flush_put_bits(&tmp);
1498 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1499 skip_bits(&s->gb, s->frame_offset);
1503 *@brief Decode a single WMA packet.
1504 *@param avctx codec context
1505 *@param data the output buffer
1506 *@param avpkt input packet
1507 *@return number of bytes that were read from the input buffer
1509 static int decode_packet(AVCodecContext *avctx, void *data,
1510 int *got_frame_ptr, AVPacket* avpkt)
1512 WMAProDecodeCtx *s = avctx->priv_data;
1513 GetBitContext* gb = &s->pgb;
1514 const uint8_t* buf = avpkt->data;
1515 int buf_size = avpkt->size;
1516 int num_bits_prev_frame;
1517 int packet_sequence_number;
1521 if (s->packet_done || s->packet_loss) {
1524 /** sanity check for the buffer length */
1525 if (buf_size < avctx->block_align) {
1526 av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1527 buf_size, avctx->block_align);
1528 return AVERROR_INVALIDDATA;
1531 s->next_packet_start = buf_size - avctx->block_align;
1532 buf_size = avctx->block_align;
1533 s->buf_bit_size = buf_size << 3;
1535 /** parse packet header */
1536 init_get_bits(gb, buf, s->buf_bit_size);
1537 packet_sequence_number = get_bits(gb, 4);
1540 /** get number of bits that need to be added to the previous frame */
1541 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1542 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1543 num_bits_prev_frame);
1545 /** check for packet loss */
1546 if (!s->packet_loss &&
1547 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1549 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1550 s->packet_sequence_number, packet_sequence_number);
1552 s->packet_sequence_number = packet_sequence_number;
1554 if (num_bits_prev_frame > 0) {
1555 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1556 if (num_bits_prev_frame >= remaining_packet_bits) {
1557 num_bits_prev_frame = remaining_packet_bits;
1561 /** append the previous frame data to the remaining data from the
1562 previous packet to create a full frame */
1563 save_bits(s, gb, num_bits_prev_frame, 1);
1564 av_dlog(avctx, "accumulated %x bits of frame data\n",
1565 s->num_saved_bits - s->frame_offset);
1567 /** decode the cross packet frame if it is valid */
1568 if (!s->packet_loss)
1569 decode_frame(s, data, got_frame_ptr);
1570 } else if (s->num_saved_bits - s->frame_offset) {
1571 av_dlog(avctx, "ignoring %x previously saved bits\n",
1572 s->num_saved_bits - s->frame_offset);
1575 if (s->packet_loss) {
1576 /** reset number of saved bits so that the decoder
1577 does not start to decode incomplete frames in the
1578 s->len_prefix == 0 case */
1579 s->num_saved_bits = 0;
1585 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1586 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1587 skip_bits(gb, s->packet_offset);
1588 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1589 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1590 frame_size <= remaining_bits(s, gb)) {
1591 save_bits(s, gb, frame_size, 0);
1592 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1593 } else if (!s->len_prefix
1594 && s->num_saved_bits > get_bits_count(&s->gb)) {
1595 /** when the frames do not have a length prefix, we don't know
1596 the compressed length of the individual frames
1597 however, we know what part of a new packet belongs to the
1599 therefore we save the incoming packet first, then we append
1600 the "previous frame" data from the next packet so that
1601 we get a buffer that only contains full frames */
1602 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1607 if (s->packet_done && !s->packet_loss &&
1608 remaining_bits(s, gb) > 0) {
1609 /** save the rest of the data so that it can be decoded
1610 with the next packet */
1611 save_bits(s, gb, remaining_bits(s, gb), 0);
1614 s->packet_offset = get_bits_count(gb) & 7;
1616 return AVERROR_INVALIDDATA;
1618 return get_bits_count(gb) >> 3;
1622 *@brief Clear decoder buffers (for seeking).
1623 *@param avctx codec context
1625 static void flush(AVCodecContext *avctx)
1627 WMAProDecodeCtx *s = avctx->priv_data;
1629 /** reset output buffer as a part of it is used during the windowing of a
1631 for (i = 0; i < avctx->channels; i++)
1632 memset(s->channel[i].out, 0, s->samples_per_frame *
1633 sizeof(*s->channel[i].out));
1639 *@brief wmapro decoder
1641 AVCodec ff_wmapro_decoder = {
1643 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1644 .type = AVMEDIA_TYPE_AUDIO,
1645 .id = AV_CODEC_ID_WMAPRO,
1646 .priv_data_size = sizeof(WMAProDecodeCtx),
1647 .init = decode_init,
1648 .close = decode_end,
1649 .decode = decode_packet,
1650 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1652 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1653 AV_SAMPLE_FMT_NONE },