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.
91 #include "libavutil/float_dsp.h"
92 #include "libavutil/intfloat.h"
93 #include "libavutil/intreadwrite.h"
98 #include "wmaprodata.h"
101 #include "wma_common.h"
103 /** current decoder limitations */
104 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
105 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
106 #define MAX_BANDS 29 ///< max number of scale factor bands
107 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
109 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
110 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
111 #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size
112 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
113 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
117 #define SCALEVLCBITS 8
118 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
119 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
120 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
121 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
122 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
124 static VLC sf_vlc; ///< scale factor DPCM vlc
125 static VLC sf_rl_vlc; ///< scale factor run length vlc
126 static VLC vec4_vlc; ///< 4 coefficients per symbol
127 static VLC vec2_vlc; ///< 2 coefficients per symbol
128 static VLC vec1_vlc; ///< 1 coefficient per symbol
129 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
130 static float sin64[33]; ///< sine table for decorrelation
133 * @brief frame specific decoder context for a single channel
136 int16_t prev_block_len; ///< length of the previous block
137 uint8_t transmit_coefs;
138 uint8_t num_subframes;
139 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
140 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
141 uint8_t cur_subframe; ///< current subframe number
142 uint16_t decoded_samples; ///< number of already processed samples
143 uint8_t grouped; ///< channel is part of a group
144 int quant_step; ///< quantization step for the current subframe
145 int8_t reuse_sf; ///< share scale factors between subframes
146 int8_t scale_factor_step; ///< scaling step for the current subframe
147 int max_scale_factor; ///< maximum scale factor for the current subframe
148 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
149 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
150 int* scale_factors; ///< pointer to the scale factor values used for decoding
151 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
152 float* coeffs; ///< pointer to the subframe decode buffer
153 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
154 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
158 * @brief channel group for channel transformations
161 uint8_t num_channels; ///< number of channels in the group
162 int8_t transform; ///< transform on / off
163 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
164 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
165 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
169 * @brief main decoder context
171 typedef struct WMAProDecodeCtx {
172 /* generic decoder variables */
173 AVCodecContext* avctx; ///< codec context for av_log
174 AVFloatDSPContext fdsp;
175 uint8_t frame_data[MAX_FRAMESIZE +
176 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
177 PutBitContext pb; ///< context for filling the frame_data buffer
178 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
179 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
180 const float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
182 /* frame size dependent frame information (set during initialization) */
183 uint32_t decode_flags; ///< used compression features
184 uint8_t len_prefix; ///< frame is prefixed with its length
185 uint8_t dynamic_range_compression; ///< frame contains DRC data
186 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
187 uint16_t samples_per_frame; ///< number of samples to output
188 uint16_t log2_frame_size;
189 int8_t lfe_channel; ///< lfe channel index
190 uint8_t max_num_subframes;
191 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
192 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
193 uint16_t min_samples_per_subframe;
194 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
195 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
196 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
197 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
199 /* packet decode state */
200 GetBitContext pgb; ///< bitstream reader context for the packet
201 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
202 uint8_t packet_offset; ///< frame offset in the packet
203 uint8_t packet_sequence_number; ///< current packet number
204 int num_saved_bits; ///< saved number of bits
205 int frame_offset; ///< frame offset in the bit reservoir
206 int subframe_offset; ///< subframe offset in the bit reservoir
207 uint8_t packet_loss; ///< set in case of bitstream error
208 uint8_t packet_done; ///< set when a packet is fully decoded
210 /* frame decode state */
211 uint32_t frame_num; ///< current frame number (not used for decoding)
212 GetBitContext gb; ///< bitstream reader context
213 int buf_bit_size; ///< buffer size in bits
214 uint8_t drc_gain; ///< gain for the DRC tool
215 int8_t skip_frame; ///< skip output step
216 int8_t parsed_all_subframes; ///< all subframes decoded?
218 /* subframe/block decode state */
219 int16_t subframe_len; ///< current subframe length
220 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
221 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
222 int8_t num_bands; ///< number of scale factor bands
223 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
224 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
225 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
226 int8_t esc_len; ///< length of escaped coefficients
228 uint8_t num_chgroups; ///< number of channel groups
229 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
231 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
236 *@brief helper function to print the most important members of the context
239 static av_cold void dump_context(WMAProDecodeCtx *s)
241 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
242 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b);
244 PRINT("ed sample bit depth", s->bits_per_sample);
245 PRINT_HEX("ed decode flags", s->decode_flags);
246 PRINT("samples per frame", s->samples_per_frame);
247 PRINT("log2 frame size", s->log2_frame_size);
248 PRINT("max num subframes", s->max_num_subframes);
249 PRINT("len prefix", s->len_prefix);
250 PRINT("num channels", s->avctx->channels);
254 *@brief Uninitialize the decoder and free all resources.
255 *@param avctx codec context
256 *@return 0 on success, < 0 otherwise
258 static av_cold int decode_end(AVCodecContext *avctx)
260 WMAProDecodeCtx *s = avctx->priv_data;
263 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
264 ff_mdct_end(&s->mdct_ctx[i]);
270 *@brief Initialize the decoder.
271 *@param avctx codec context
272 *@return 0 on success, -1 otherwise
274 static av_cold int decode_init(AVCodecContext *avctx)
276 WMAProDecodeCtx *s = avctx->priv_data;
277 uint8_t *edata_ptr = avctx->extradata;
278 unsigned int channel_mask;
280 int log2_max_num_subframes;
281 int num_possible_block_sizes;
283 if (!avctx->block_align) {
284 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
285 return AVERROR(EINVAL);
289 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
291 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
293 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
295 if (avctx->extradata_size >= 18) {
296 s->decode_flags = AV_RL16(edata_ptr+14);
297 channel_mask = AV_RL32(edata_ptr+2);
298 s->bits_per_sample = AV_RL16(edata_ptr);
299 /** dump the extradata */
300 for (i = 0; i < avctx->extradata_size; i++)
301 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
302 av_dlog(avctx, "\n");
305 avpriv_request_sample(avctx, "Unknown extradata size");
306 return AVERROR_PATCHWELCOME;
310 s->log2_frame_size = av_log2(avctx->block_align) + 4;
311 if (s->log2_frame_size > 25) {
312 avpriv_request_sample(avctx, "Large block align");
313 return AVERROR_PATCHWELCOME;
317 s->skip_frame = 1; /* skip first frame */
319 s->len_prefix = (s->decode_flags & 0x40);
322 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
323 if (bits > WMAPRO_BLOCK_MAX_BITS) {
324 avpriv_request_sample(avctx, "14-bit block sizes");
325 return AVERROR_PATCHWELCOME;
327 s->samples_per_frame = 1 << bits;
330 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
331 s->max_num_subframes = 1 << log2_max_num_subframes;
332 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
333 s->max_subframe_len_bit = 1;
334 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
336 num_possible_block_sizes = log2_max_num_subframes + 1;
337 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
338 s->dynamic_range_compression = (s->decode_flags & 0x80);
340 if (s->max_num_subframes > MAX_SUBFRAMES) {
341 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n",
342 s->max_num_subframes);
343 return AVERROR_INVALIDDATA;
346 if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) {
347 av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
348 s->min_samples_per_subframe);
349 return AVERROR_INVALIDDATA;
352 if (s->avctx->sample_rate <= 0) {
353 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
354 return AVERROR_INVALIDDATA;
357 if (avctx->channels < 0) {
358 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
360 return AVERROR_INVALIDDATA;
361 } else if (avctx->channels > WMAPRO_MAX_CHANNELS) {
362 avpriv_request_sample(avctx,
363 "More than %d channels", WMAPRO_MAX_CHANNELS);
364 return AVERROR_PATCHWELCOME;
367 /** init previous block len */
368 for (i = 0; i < avctx->channels; i++)
369 s->channel[i].prev_block_len = s->samples_per_frame;
371 /** extract lfe channel position */
374 if (channel_mask & 8) {
376 for (mask = 1; mask < 16; mask <<= 1) {
377 if (channel_mask & mask)
382 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
383 scale_huffbits, 1, 1,
384 scale_huffcodes, 2, 2, 616);
386 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
387 scale_rl_huffbits, 1, 1,
388 scale_rl_huffcodes, 4, 4, 1406);
390 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
391 coef0_huffbits, 1, 1,
392 coef0_huffcodes, 4, 4, 2108);
394 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
395 coef1_huffbits, 1, 1,
396 coef1_huffcodes, 4, 4, 3912);
398 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
400 vec4_huffcodes, 2, 2, 604);
402 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
404 vec2_huffcodes, 2, 2, 562);
406 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
408 vec1_huffcodes, 2, 2, 562);
410 /** calculate number of scale factor bands and their offsets
411 for every possible block size */
412 for (i = 0; i < num_possible_block_sizes; i++) {
413 int subframe_len = s->samples_per_frame >> i;
417 s->sfb_offsets[i][0] = 0;
419 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
420 int offset = (subframe_len * 2 * critical_freq[x])
421 / s->avctx->sample_rate + 2;
423 if (offset > s->sfb_offsets[i][band - 1])
424 s->sfb_offsets[i][band++] = offset;
426 if (offset >= subframe_len)
429 s->sfb_offsets[i][band - 1] = subframe_len;
430 s->num_sfb[i] = band - 1;
431 if (s->num_sfb[i] <= 0) {
432 av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n");
433 return AVERROR_INVALIDDATA;
438 /** Scale factors can be shared between blocks of different size
439 as every block has a different scale factor band layout.
440 The matrix sf_offsets is needed to find the correct scale factor.
443 for (i = 0; i < num_possible_block_sizes; i++) {
445 for (b = 0; b < s->num_sfb[i]; b++) {
447 int offset = ((s->sfb_offsets[i][b]
448 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
449 for (x = 0; x < num_possible_block_sizes; x++) {
451 while (s->sfb_offsets[x][v + 1] << x < offset) {
453 av_assert0(v < MAX_BANDS);
455 s->sf_offsets[i][x][b] = v;
460 /** init MDCT, FIXME: only init needed sizes */
461 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
462 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
463 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
464 / (1 << (s->bits_per_sample - 1)));
466 /** init MDCT windows: simple sine window */
467 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
468 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
469 ff_init_ff_sine_windows(win_idx);
470 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
473 /** calculate subwoofer cutoff values */
474 for (i = 0; i < num_possible_block_sizes; i++) {
475 int block_size = s->samples_per_frame >> i;
476 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
477 / s->avctx->sample_rate;
478 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
481 /** calculate sine values for the decorrelation matrix */
482 for (i = 0; i < 33; i++)
483 sin64[i] = sin(i*M_PI / 64.0);
485 if (avctx->debug & FF_DEBUG_BITSTREAM)
488 avctx->channel_layout = channel_mask;
494 *@brief Decode the subframe length.
496 *@param offset sample offset in the frame
497 *@return decoded subframe length on success, < 0 in case of an error
499 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
501 int frame_len_shift = 0;
504 /** no need to read from the bitstream when only one length is possible */
505 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
506 return s->min_samples_per_subframe;
508 if (get_bits_left(&s->gb) < 1)
509 return AVERROR_INVALIDDATA;
511 /** 1 bit indicates if the subframe is of maximum length */
512 if (s->max_subframe_len_bit) {
513 if (get_bits1(&s->gb))
514 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
516 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
518 subframe_len = s->samples_per_frame >> frame_len_shift;
520 /** sanity check the length */
521 if (subframe_len < s->min_samples_per_subframe ||
522 subframe_len > s->samples_per_frame) {
523 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
525 return AVERROR_INVALIDDATA;
531 *@brief Decode how the data in the frame is split into subframes.
532 * Every WMA frame contains the encoded data for a fixed number of
533 * samples per channel. The data for every channel might be split
534 * into several subframes. This function will reconstruct the list of
535 * subframes for every channel.
537 * If the subframes are not evenly split, the algorithm estimates the
538 * channels with the lowest number of total samples.
539 * Afterwards, for each of these channels a bit is read from the
540 * bitstream that indicates if the channel contains a subframe with the
541 * next subframe size that is going to be read from the bitstream or not.
542 * If a channel contains such a subframe, the subframe size gets added to
543 * the channel's subframe list.
544 * The algorithm repeats these steps until the frame is properly divided
545 * between the individual channels.
548 *@return 0 on success, < 0 in case of an error
550 static int decode_tilehdr(WMAProDecodeCtx *s)
552 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
553 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
554 int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */
555 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
556 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
559 /* Should never consume more than 3073 bits (256 iterations for the
560 * while loop when always the minimum amount of 128 samples is subtracted
561 * from missing samples in the 8 channel case).
562 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
565 /** reset tiling information */
566 for (c = 0; c < s->avctx->channels; c++)
567 s->channel[c].num_subframes = 0;
569 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
570 fixed_channel_layout = 1;
572 /** loop until the frame data is split between the subframes */
576 /** check which channels contain the subframe */
577 for (c = 0; c < s->avctx->channels; c++) {
578 if (num_samples[c] == min_channel_len) {
579 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
580 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
581 contains_subframe[c] = 1;
583 contains_subframe[c] = get_bits1(&s->gb);
585 contains_subframe[c] = 0;
588 /** get subframe length, subframe_len == 0 is not allowed */
589 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
590 return AVERROR_INVALIDDATA;
592 /** add subframes to the individual channels and find new min_channel_len */
593 min_channel_len += subframe_len;
594 for (c = 0; c < s->avctx->channels; c++) {
595 WMAProChannelCtx* chan = &s->channel[c];
597 if (contains_subframe[c]) {
598 if (chan->num_subframes >= MAX_SUBFRAMES) {
599 av_log(s->avctx, AV_LOG_ERROR,
600 "broken frame: num subframes > 31\n");
601 return AVERROR_INVALIDDATA;
603 chan->subframe_len[chan->num_subframes] = subframe_len;
604 num_samples[c] += subframe_len;
605 ++chan->num_subframes;
606 if (num_samples[c] > s->samples_per_frame) {
607 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
608 "channel len > samples_per_frame\n");
609 return AVERROR_INVALIDDATA;
611 } else if (num_samples[c] <= min_channel_len) {
612 if (num_samples[c] < min_channel_len) {
613 channels_for_cur_subframe = 0;
614 min_channel_len = num_samples[c];
616 ++channels_for_cur_subframe;
619 } while (min_channel_len < s->samples_per_frame);
621 for (c = 0; c < s->avctx->channels; c++) {
624 for (i = 0; i < s->channel[c].num_subframes; i++) {
625 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
626 " len %i\n", s->frame_num, c, i,
627 s->channel[c].subframe_len[i]);
628 s->channel[c].subframe_offset[i] = offset;
629 offset += s->channel[c].subframe_len[i];
637 *@brief Calculate a decorrelation matrix from the bitstream parameters.
638 *@param s codec context
639 *@param chgroup channel group for which the matrix needs to be calculated
641 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
642 WMAProChannelGrp *chgroup)
646 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
647 memset(chgroup->decorrelation_matrix, 0, s->avctx->channels *
648 s->avctx->channels * sizeof(*chgroup->decorrelation_matrix));
650 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
651 rotation_offset[i] = get_bits(&s->gb, 6);
653 for (i = 0; i < chgroup->num_channels; i++)
654 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
655 get_bits1(&s->gb) ? 1.0 : -1.0;
657 for (i = 1; i < chgroup->num_channels; i++) {
659 for (x = 0; x < i; x++) {
661 for (y = 0; y < i + 1; y++) {
662 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
663 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
664 int n = rotation_offset[offset + x];
670 cosv = sin64[32 - n];
672 sinv = sin64[64 - n];
673 cosv = -sin64[n - 32];
676 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
677 (v1 * sinv) - (v2 * cosv);
678 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
679 (v1 * cosv) + (v2 * sinv);
687 *@brief Decode channel transformation parameters
688 *@param s codec context
689 *@return >= 0 in case of success, < 0 in case of bitstream errors
691 static int decode_channel_transform(WMAProDecodeCtx* s)
694 /* should never consume more than 1921 bits for the 8 channel case
695 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
696 * + MAX_CHANNELS + MAX_BANDS + 1)
699 /** in the one channel case channel transforms are pointless */
701 if (s->avctx->channels > 1) {
702 int remaining_channels = s->channels_for_cur_subframe;
704 if (get_bits1(&s->gb)) {
705 avpriv_request_sample(s->avctx,
706 "Channel transform bit");
707 return AVERROR_PATCHWELCOME;
710 for (s->num_chgroups = 0; remaining_channels &&
711 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
712 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
713 float** channel_data = chgroup->channel_data;
714 chgroup->num_channels = 0;
715 chgroup->transform = 0;
717 /** decode channel mask */
718 if (remaining_channels > 2) {
719 for (i = 0; i < s->channels_for_cur_subframe; i++) {
720 int channel_idx = s->channel_indexes_for_cur_subframe[i];
721 if (!s->channel[channel_idx].grouped
722 && get_bits1(&s->gb)) {
723 ++chgroup->num_channels;
724 s->channel[channel_idx].grouped = 1;
725 *channel_data++ = s->channel[channel_idx].coeffs;
729 chgroup->num_channels = remaining_channels;
730 for (i = 0; i < s->channels_for_cur_subframe; i++) {
731 int channel_idx = s->channel_indexes_for_cur_subframe[i];
732 if (!s->channel[channel_idx].grouped)
733 *channel_data++ = s->channel[channel_idx].coeffs;
734 s->channel[channel_idx].grouped = 1;
738 /** decode transform type */
739 if (chgroup->num_channels == 2) {
740 if (get_bits1(&s->gb)) {
741 if (get_bits1(&s->gb)) {
742 avpriv_request_sample(s->avctx,
743 "Unknown channel transform type");
744 return AVERROR_PATCHWELCOME;
747 chgroup->transform = 1;
748 if (s->avctx->channels == 2) {
749 chgroup->decorrelation_matrix[0] = 1.0;
750 chgroup->decorrelation_matrix[1] = -1.0;
751 chgroup->decorrelation_matrix[2] = 1.0;
752 chgroup->decorrelation_matrix[3] = 1.0;
755 chgroup->decorrelation_matrix[0] = 0.70703125;
756 chgroup->decorrelation_matrix[1] = -0.70703125;
757 chgroup->decorrelation_matrix[2] = 0.70703125;
758 chgroup->decorrelation_matrix[3] = 0.70703125;
761 } else if (chgroup->num_channels > 2) {
762 if (get_bits1(&s->gb)) {
763 chgroup->transform = 1;
764 if (get_bits1(&s->gb)) {
765 decode_decorrelation_matrix(s, chgroup);
767 /** FIXME: more than 6 coupled channels not supported */
768 if (chgroup->num_channels > 6) {
769 avpriv_request_sample(s->avctx,
770 "Coupled channels > 6");
772 memcpy(chgroup->decorrelation_matrix,
773 default_decorrelation[chgroup->num_channels],
774 chgroup->num_channels * chgroup->num_channels *
775 sizeof(*chgroup->decorrelation_matrix));
781 /** decode transform on / off */
782 if (chgroup->transform) {
783 if (!get_bits1(&s->gb)) {
785 /** transform can be enabled for individual bands */
786 for (i = 0; i < s->num_bands; i++) {
787 chgroup->transform_band[i] = get_bits1(&s->gb);
790 memset(chgroup->transform_band, 1, s->num_bands);
793 remaining_channels -= chgroup->num_channels;
800 *@brief Extract the coefficients from the bitstream.
801 *@param s codec context
802 *@param c current channel number
803 *@return 0 on success, < 0 in case of bitstream errors
805 static int decode_coeffs(WMAProDecodeCtx *s, int c)
807 /* Integers 0..15 as single-precision floats. The table saves a
808 costly int to float conversion, and storing the values as
809 integers allows fast sign-flipping. */
810 static const uint32_t fval_tab[16] = {
811 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
812 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
813 0x41000000, 0x41100000, 0x41200000, 0x41300000,
814 0x41400000, 0x41500000, 0x41600000, 0x41700000,
818 WMAProChannelCtx* ci = &s->channel[c];
825 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
827 vlctable = get_bits1(&s->gb);
828 vlc = &coef_vlc[vlctable];
838 /** decode vector coefficients (consumes up to 167 bits per iteration for
839 4 vector coded large values) */
840 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
841 (cur_coeff + 3 < ci->num_vec_coeffs)) {
846 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
848 if (idx == HUFF_VEC4_SIZE - 1) {
849 for (i = 0; i < 4; i += 2) {
850 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
851 if (idx == HUFF_VEC2_SIZE - 1) {
853 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
854 if (v0 == HUFF_VEC1_SIZE - 1)
855 v0 += ff_wma_get_large_val(&s->gb);
856 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
857 if (v1 == HUFF_VEC1_SIZE - 1)
858 v1 += ff_wma_get_large_val(&s->gb);
859 vals[i ] = av_float2int(v0);
860 vals[i+1] = av_float2int(v1);
862 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
863 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
867 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
868 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
869 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
870 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
874 for (i = 0; i < 4; i++) {
876 uint32_t sign = get_bits1(&s->gb) - 1;
877 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
880 ci->coeffs[cur_coeff] = 0;
881 /** switch to run level mode when subframe_len / 128 zeros
882 were found in a row */
883 rl_mode |= (++num_zeros > s->subframe_len >> 8);
889 /** decode run level coded coefficients */
890 if (cur_coeff < s->subframe_len) {
891 memset(&ci->coeffs[cur_coeff], 0,
892 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
893 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
894 level, run, 1, ci->coeffs,
895 cur_coeff, s->subframe_len,
896 s->subframe_len, s->esc_len, 0))
897 return AVERROR_INVALIDDATA;
904 *@brief Extract scale factors from the bitstream.
905 *@param s codec context
906 *@return 0 on success, < 0 in case of bitstream errors
908 static int decode_scale_factors(WMAProDecodeCtx* s)
912 /** should never consume more than 5344 bits
913 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
916 for (i = 0; i < s->channels_for_cur_subframe; i++) {
917 int c = s->channel_indexes_for_cur_subframe[i];
920 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
921 sf_end = s->channel[c].scale_factors + s->num_bands;
923 /** resample scale factors for the new block size
924 * as the scale factors might need to be resampled several times
925 * before some new values are transmitted, a backup of the last
926 * transmitted scale factors is kept in saved_scale_factors
928 if (s->channel[c].reuse_sf) {
929 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
931 for (b = 0; b < s->num_bands; b++)
932 s->channel[c].scale_factors[b] =
933 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
936 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
938 if (!s->channel[c].reuse_sf) {
940 /** decode DPCM coded scale factors */
941 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
942 val = 45 / s->channel[c].scale_factor_step;
943 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
944 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
949 /** run level decode differences to the resampled factors */
950 for (i = 0; i < s->num_bands; i++) {
956 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
959 uint32_t code = get_bits(&s->gb, 14);
961 sign = (code & 1) - 1;
962 skip = (code & 0x3f) >> 1;
963 } else if (idx == 1) {
966 skip = scale_rl_run[idx];
967 val = scale_rl_level[idx];
968 sign = get_bits1(&s->gb)-1;
972 if (i >= s->num_bands) {
973 av_log(s->avctx, AV_LOG_ERROR,
974 "invalid scale factor coding\n");
975 return AVERROR_INVALIDDATA;
977 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
981 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
982 s->channel[c].table_idx = s->table_idx;
983 s->channel[c].reuse_sf = 1;
986 /** calculate new scale factor maximum */
987 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
988 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
989 s->channel[c].max_scale_factor =
990 FFMAX(s->channel[c].max_scale_factor, *sf);
998 *@brief Reconstruct the individual channel data.
999 *@param s codec context
1001 static void inverse_channel_transform(WMAProDecodeCtx *s)
1005 for (i = 0; i < s->num_chgroups; i++) {
1006 if (s->chgroup[i].transform) {
1007 float data[WMAPRO_MAX_CHANNELS];
1008 const int num_channels = s->chgroup[i].num_channels;
1009 float** ch_data = s->chgroup[i].channel_data;
1010 float** ch_end = ch_data + num_channels;
1011 const int8_t* tb = s->chgroup[i].transform_band;
1014 /** multichannel decorrelation */
1015 for (sfb = s->cur_sfb_offsets;
1016 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1019 /** multiply values with the decorrelation_matrix */
1020 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1021 const float* mat = s->chgroup[i].decorrelation_matrix;
1022 const float* data_end = data + num_channels;
1023 float* data_ptr = data;
1026 for (ch = ch_data; ch < ch_end; ch++)
1027 *data_ptr++ = (*ch)[y];
1029 for (ch = ch_data; ch < ch_end; ch++) {
1032 while (data_ptr < data_end)
1033 sum += *data_ptr++ * *mat++;
1038 } else if (s->avctx->channels == 2) {
1039 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1040 s->fdsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1041 ch_data[0] + sfb[0],
1043 s->fdsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1044 ch_data[1] + sfb[0],
1053 *@brief Apply sine window and reconstruct the output buffer.
1054 *@param s codec context
1056 static void wmapro_window(WMAProDecodeCtx *s)
1059 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1060 int c = s->channel_indexes_for_cur_subframe[i];
1061 const float* window;
1062 int winlen = s->channel[c].prev_block_len;
1063 float* start = s->channel[c].coeffs - (winlen >> 1);
1065 if (s->subframe_len < winlen) {
1066 start += (winlen - s->subframe_len) >> 1;
1067 winlen = s->subframe_len;
1070 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1074 s->fdsp.vector_fmul_window(start, start, start + winlen,
1077 s->channel[c].prev_block_len = s->subframe_len;
1082 *@brief Decode a single subframe (block).
1083 *@param s codec context
1084 *@return 0 on success, < 0 when decoding failed
1086 static int decode_subframe(WMAProDecodeCtx *s)
1088 int offset = s->samples_per_frame;
1089 int subframe_len = s->samples_per_frame;
1091 int total_samples = s->samples_per_frame * s->avctx->channels;
1092 int transmit_coeffs = 0;
1093 int cur_subwoofer_cutoff;
1095 s->subframe_offset = get_bits_count(&s->gb);
1097 /** reset channel context and find the next block offset and size
1098 == the next block of the channel with the smallest number of
1101 for (i = 0; i < s->avctx->channels; i++) {
1102 s->channel[i].grouped = 0;
1103 if (offset > s->channel[i].decoded_samples) {
1104 offset = s->channel[i].decoded_samples;
1106 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1111 "processing subframe with offset %i len %i\n", offset, subframe_len);
1113 /** get a list of all channels that contain the estimated block */
1114 s->channels_for_cur_subframe = 0;
1115 for (i = 0; i < s->avctx->channels; i++) {
1116 const int cur_subframe = s->channel[i].cur_subframe;
1117 /** subtract already processed samples */
1118 total_samples -= s->channel[i].decoded_samples;
1120 /** and count if there are multiple subframes that match our profile */
1121 if (offset == s->channel[i].decoded_samples &&
1122 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1123 total_samples -= s->channel[i].subframe_len[cur_subframe];
1124 s->channel[i].decoded_samples +=
1125 s->channel[i].subframe_len[cur_subframe];
1126 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1127 ++s->channels_for_cur_subframe;
1131 /** check if the frame will be complete after processing the
1134 s->parsed_all_subframes = 1;
1137 av_dlog(s->avctx, "subframe is part of %i channels\n",
1138 s->channels_for_cur_subframe);
1140 /** calculate number of scale factor bands and their offsets */
1141 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1142 s->num_bands = s->num_sfb[s->table_idx];
1143 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1144 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1146 /** configure the decoder for the current subframe */
1147 offset += s->samples_per_frame >> 1;
1149 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1150 int c = s->channel_indexes_for_cur_subframe[i];
1152 s->channel[c].coeffs = &s->channel[c].out[offset];
1155 s->subframe_len = subframe_len;
1156 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1158 /** skip extended header if any */
1159 if (get_bits1(&s->gb)) {
1161 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1162 int len = get_bits(&s->gb, 4);
1163 num_fill_bits = (len ? get_bits(&s->gb, len) : 0) + 1;
1166 if (num_fill_bits >= 0) {
1167 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1168 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1169 return AVERROR_INVALIDDATA;
1172 skip_bits_long(&s->gb, num_fill_bits);
1176 /** no idea for what the following bit is used */
1177 if (get_bits1(&s->gb)) {
1178 avpriv_request_sample(s->avctx, "Reserved bit");
1179 return AVERROR_PATCHWELCOME;
1183 if (decode_channel_transform(s) < 0)
1184 return AVERROR_INVALIDDATA;
1187 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1188 int c = s->channel_indexes_for_cur_subframe[i];
1189 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1190 transmit_coeffs = 1;
1193 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1194 if (transmit_coeffs) {
1196 int quant_step = 90 * s->bits_per_sample >> 4;
1198 /** decode number of vector coded coefficients */
1199 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1200 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1201 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1202 int c = s->channel_indexes_for_cur_subframe[i];
1203 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1204 if (num_vec_coeffs > s->subframe_len) {
1205 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1206 return AVERROR_INVALIDDATA;
1208 av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out));
1209 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1212 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1213 int c = s->channel_indexes_for_cur_subframe[i];
1214 s->channel[c].num_vec_coeffs = s->subframe_len;
1217 /** decode quantization step */
1218 step = get_sbits(&s->gb, 6);
1220 if (step == -32 || step == 31) {
1221 const int sign = (step == 31) - 1;
1223 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1224 (step = get_bits(&s->gb, 5)) == 31) {
1227 quant_step += ((quant + step) ^ sign) - sign;
1229 if (quant_step < 0) {
1230 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1233 /** decode quantization step modifiers for every channel */
1235 if (s->channels_for_cur_subframe == 1) {
1236 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1238 int modifier_len = get_bits(&s->gb, 3);
1239 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1240 int c = s->channel_indexes_for_cur_subframe[i];
1241 s->channel[c].quant_step = quant_step;
1242 if (get_bits1(&s->gb)) {
1244 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1246 ++s->channel[c].quant_step;
1251 /** decode scale factors */
1252 if (decode_scale_factors(s) < 0)
1253 return AVERROR_INVALIDDATA;
1256 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1257 get_bits_count(&s->gb) - s->subframe_offset);
1259 /** parse coefficients */
1260 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1261 int c = s->channel_indexes_for_cur_subframe[i];
1262 if (s->channel[c].transmit_coefs &&
1263 get_bits_count(&s->gb) < s->num_saved_bits) {
1264 decode_coeffs(s, c);
1266 memset(s->channel[c].coeffs, 0,
1267 sizeof(*s->channel[c].coeffs) * subframe_len);
1270 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1271 get_bits_count(&s->gb) - s->subframe_offset);
1273 if (transmit_coeffs) {
1274 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1275 /** reconstruct the per channel data */
1276 inverse_channel_transform(s);
1277 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1278 int c = s->channel_indexes_for_cur_subframe[i];
1279 const int* sf = s->channel[c].scale_factors;
1282 if (c == s->lfe_channel)
1283 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1284 (subframe_len - cur_subwoofer_cutoff));
1286 /** inverse quantization and rescaling */
1287 for (b = 0; b < s->num_bands; b++) {
1288 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1289 const int exp = s->channel[c].quant_step -
1290 (s->channel[c].max_scale_factor - *sf++) *
1291 s->channel[c].scale_factor_step;
1292 const float quant = pow(10.0, exp / 20.0);
1293 int start = s->cur_sfb_offsets[b];
1294 s->fdsp.vector_fmul_scalar(s->tmp + start,
1295 s->channel[c].coeffs + start,
1296 quant, end - start);
1299 /** apply imdct (imdct_half == DCTIV with reverse) */
1300 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1304 /** window and overlapp-add */
1307 /** handled one subframe */
1308 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1309 int c = s->channel_indexes_for_cur_subframe[i];
1310 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1311 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1312 return AVERROR_INVALIDDATA;
1314 ++s->channel[c].cur_subframe;
1321 *@brief Decode one WMA frame.
1322 *@param s codec context
1323 *@return 0 if the trailer bit indicates that this is the last frame,
1324 * 1 if there are additional frames
1326 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1328 AVCodecContext *avctx = s->avctx;
1329 GetBitContext* gb = &s->gb;
1330 int more_frames = 0;
1334 /** get frame length */
1336 len = get_bits(gb, s->log2_frame_size);
1338 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1340 /** decode tile information */
1341 if (decode_tilehdr(s)) {
1346 /** read postproc transform */
1347 if (s->avctx->channels > 1 && get_bits1(gb)) {
1348 if (get_bits1(gb)) {
1349 for (i = 0; i < avctx->channels * avctx->channels; i++)
1354 /** read drc info */
1355 if (s->dynamic_range_compression) {
1356 s->drc_gain = get_bits(gb, 8);
1357 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1360 /** no idea what these are for, might be the number of samples
1361 that need to be skipped at the beginning or end of a stream */
1362 if (get_bits1(gb)) {
1365 /** usually true for the first frame */
1366 if (get_bits1(gb)) {
1367 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1368 av_dlog(s->avctx, "start skip: %i\n", skip);
1371 /** sometimes true for the last frame */
1372 if (get_bits1(gb)) {
1373 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1374 av_dlog(s->avctx, "end skip: %i\n", skip);
1379 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1380 get_bits_count(gb) - s->frame_offset);
1382 /** reset subframe states */
1383 s->parsed_all_subframes = 0;
1384 for (i = 0; i < avctx->channels; i++) {
1385 s->channel[i].decoded_samples = 0;
1386 s->channel[i].cur_subframe = 0;
1387 s->channel[i].reuse_sf = 0;
1390 /** decode all subframes */
1391 while (!s->parsed_all_subframes) {
1392 if (decode_subframe(s) < 0) {
1398 /* get output buffer */
1399 frame->nb_samples = s->samples_per_frame;
1400 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1405 /** copy samples to the output buffer */
1406 for (i = 0; i < avctx->channels; i++)
1407 memcpy(frame->extended_data[i], s->channel[i].out,
1408 s->samples_per_frame * sizeof(*s->channel[i].out));
1410 for (i = 0; i < avctx->channels; i++) {
1411 /** reuse second half of the IMDCT output for the next frame */
1412 memcpy(&s->channel[i].out[0],
1413 &s->channel[i].out[s->samples_per_frame],
1414 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1417 if (s->skip_frame) {
1420 av_frame_unref(frame);
1425 if (s->len_prefix) {
1426 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1427 /** FIXME: not sure if this is always an error */
1428 av_log(s->avctx, AV_LOG_ERROR,
1429 "frame[%"PRIu32"] would have to skip %i bits\n",
1431 len - (get_bits_count(gb) - s->frame_offset) - 1);
1436 /** skip the rest of the frame data */
1437 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1439 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1443 /** decode trailer bit */
1444 more_frames = get_bits1(gb);
1451 *@brief Calculate remaining input buffer length.
1452 *@param s codec context
1453 *@param gb bitstream reader context
1454 *@return remaining size in bits
1456 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1458 return s->buf_bit_size - get_bits_count(gb);
1462 *@brief Fill the bit reservoir with a (partial) frame.
1463 *@param s codec context
1464 *@param gb bitstream reader context
1465 *@param len length of the partial frame
1466 *@param append decides whether to reset the buffer or not
1468 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1473 /** when the frame data does not need to be concatenated, the input buffer
1474 is reset and additional bits from the previous frame are copied
1475 and skipped later so that a fast byte copy is possible */
1478 s->frame_offset = get_bits_count(gb) & 7;
1479 s->num_saved_bits = s->frame_offset;
1480 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1483 buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1485 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1486 avpriv_request_sample(s->avctx, "Too small input buffer");
1491 av_assert0(len <= put_bits_left(&s->pb));
1493 s->num_saved_bits += len;
1495 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1498 int align = 8 - (get_bits_count(gb) & 7);
1499 align = FFMIN(align, len);
1500 put_bits(&s->pb, align, get_bits(gb, align));
1502 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1504 skip_bits_long(gb, len);
1507 PutBitContext tmp = s->pb;
1508 flush_put_bits(&tmp);
1511 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1512 skip_bits(&s->gb, s->frame_offset);
1516 *@brief Decode a single WMA packet.
1517 *@param avctx codec context
1518 *@param data the output buffer
1519 *@param avpkt input packet
1520 *@return number of bytes that were read from the input buffer
1522 static int decode_packet(AVCodecContext *avctx, void *data,
1523 int *got_frame_ptr, AVPacket* avpkt)
1525 WMAProDecodeCtx *s = avctx->priv_data;
1526 GetBitContext* gb = &s->pgb;
1527 const uint8_t* buf = avpkt->data;
1528 int buf_size = avpkt->size;
1529 int num_bits_prev_frame;
1530 int packet_sequence_number;
1534 if (s->packet_done || s->packet_loss) {
1537 /** sanity check for the buffer length */
1538 if (buf_size < avctx->block_align) {
1539 av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1540 buf_size, avctx->block_align);
1541 return AVERROR_INVALIDDATA;
1544 s->next_packet_start = buf_size - avctx->block_align;
1545 buf_size = avctx->block_align;
1546 s->buf_bit_size = buf_size << 3;
1548 /** parse packet header */
1549 init_get_bits(gb, buf, s->buf_bit_size);
1550 packet_sequence_number = get_bits(gb, 4);
1553 /** get number of bits that need to be added to the previous frame */
1554 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1555 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1556 num_bits_prev_frame);
1558 /** check for packet loss */
1559 if (!s->packet_loss &&
1560 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1562 av_log(avctx, AV_LOG_ERROR,
1563 "Packet loss detected! seq %"PRIx8" vs %x\n",
1564 s->packet_sequence_number, packet_sequence_number);
1566 s->packet_sequence_number = packet_sequence_number;
1568 if (num_bits_prev_frame > 0) {
1569 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1570 if (num_bits_prev_frame >= remaining_packet_bits) {
1571 num_bits_prev_frame = remaining_packet_bits;
1575 /** append the previous frame data to the remaining data from the
1576 previous packet to create a full frame */
1577 save_bits(s, gb, num_bits_prev_frame, 1);
1578 av_dlog(avctx, "accumulated %x bits of frame data\n",
1579 s->num_saved_bits - s->frame_offset);
1581 /** decode the cross packet frame if it is valid */
1582 if (!s->packet_loss)
1583 decode_frame(s, data, got_frame_ptr);
1584 } else if (s->num_saved_bits - s->frame_offset) {
1585 av_dlog(avctx, "ignoring %x previously saved bits\n",
1586 s->num_saved_bits - s->frame_offset);
1589 if (s->packet_loss) {
1590 /** reset number of saved bits so that the decoder
1591 does not start to decode incomplete frames in the
1592 s->len_prefix == 0 case */
1593 s->num_saved_bits = 0;
1599 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1600 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1601 skip_bits(gb, s->packet_offset);
1602 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1603 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1604 frame_size <= remaining_bits(s, gb)) {
1605 save_bits(s, gb, frame_size, 0);
1606 if (!s->packet_loss)
1607 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1608 } else if (!s->len_prefix
1609 && s->num_saved_bits > get_bits_count(&s->gb)) {
1610 /** when the frames do not have a length prefix, we don't know
1611 the compressed length of the individual frames
1612 however, we know what part of a new packet belongs to the
1614 therefore we save the incoming packet first, then we append
1615 the "previous frame" data from the next packet so that
1616 we get a buffer that only contains full frames */
1617 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1622 if (s->packet_done && !s->packet_loss &&
1623 remaining_bits(s, gb) > 0) {
1624 /** save the rest of the data so that it can be decoded
1625 with the next packet */
1626 save_bits(s, gb, remaining_bits(s, gb), 0);
1629 s->packet_offset = get_bits_count(gb) & 7;
1631 return AVERROR_INVALIDDATA;
1633 return get_bits_count(gb) >> 3;
1637 *@brief Clear decoder buffers (for seeking).
1638 *@param avctx codec context
1640 static void flush(AVCodecContext *avctx)
1642 WMAProDecodeCtx *s = avctx->priv_data;
1644 /** reset output buffer as a part of it is used during the windowing of a
1646 for (i = 0; i < avctx->channels; i++)
1647 memset(s->channel[i].out, 0, s->samples_per_frame *
1648 sizeof(*s->channel[i].out));
1654 *@brief wmapro decoder
1656 AVCodec ff_wmapro_decoder = {
1658 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1659 .type = AVMEDIA_TYPE_AUDIO,
1660 .id = AV_CODEC_ID_WMAPRO,
1661 .priv_data_size = sizeof(WMAProDecodeCtx),
1662 .init = decode_init,
1663 .close = decode_end,
1664 .decode = decode_packet,
1665 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1667 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1668 AV_SAMPLE_FMT_NONE },