2 * Wmapro compatible decoder
3 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 * Copyright (c) 2008 - 2011 Sascha Sommer, Benjamin Larsson
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * @brief wmapro decoder implementation
26 * Wmapro is an MDCT based codec comparable to wma standard or AAC.
27 * The decoding therefore consists of the following steps:
28 * - bitstream decoding
29 * - reconstruction of per-channel data
30 * - rescaling and inverse quantization
32 * - windowing and overlapp-add
34 * The compressed wmapro bitstream is split into individual packets.
35 * Every such packet contains one or more wma frames.
36 * The compressed frames may have a variable length and frames may
37 * cross packet boundaries.
38 * Common to all wmapro frames is the number of samples that are stored in
40 * The number of samples and a few other decode flags are stored
41 * as extradata that has to be passed to the decoder.
43 * The wmapro frames themselves are again split into a variable number of
44 * subframes. Every subframe contains the data for 2^N time domain samples
45 * where N varies between 7 and 12.
47 * Example wmapro bitstream (in samples):
49 * || packet 0 || packet 1 || packet 2 packets
50 * ---------------------------------------------------
51 * || frame 0 || frame 1 || frame 2 || frames
52 * ---------------------------------------------------
53 * || | | || | | | || || subframes of channel 0
54 * ---------------------------------------------------
55 * || | | || | | | || || subframes of channel 1
56 * ---------------------------------------------------
58 * The frame layouts for the individual channels of a wma frame does not need
61 * However, if the offsets and lengths of several subframes of a frame are the
62 * same, the subframes of the channels can be grouped.
63 * Every group may then use special coding techniques like M/S stereo coding
64 * to improve the compression ratio. These channel transformations do not
65 * need to be applied to a whole subframe. Instead, they can also work on
66 * individual scale factor bands (see below).
67 * The coefficients that carry the audio signal in the frequency domain
68 * are transmitted as huffman-coded vectors with 4, 2 and 1 elements.
69 * In addition to that, the encoder can switch to a runlevel coding scheme
70 * by transmitting subframe_length / 128 zero coefficients.
72 * Before the audio signal can be converted to the time domain, the
73 * coefficients have to be rescaled and inverse quantized.
74 * A subframe is therefore split into several scale factor bands that get
75 * scaled individually.
76 * Scale factors are submitted for every frame but they might be shared
77 * between the subframes of a channel. Scale factors are initially DPCM-coded.
78 * Once scale factors are shared, the differences are transmitted as runlevel
80 * Every subframe length and offset combination in the frame layout shares a
81 * common quantization factor that can be adjusted for every channel by a
83 * After the inverse quantization, the coefficients get processed by an IMDCT.
84 * The resulting values are then windowed with a sine window and the first half
85 * of the values are added to the second half of the output from the previous
86 * subframe in order to reconstruct the output samples.
89 #include "libavutil/intfloat.h"
90 #include "libavutil/intreadwrite.h"
95 #include "wmaprodata.h"
97 #include "fmtconvert.h"
100 #include "wma_common.h"
102 /** current decoder limitations */
103 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
104 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
105 #define MAX_BANDS 29 ///< max number of scale factor bands
106 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
108 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
109 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
110 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
111 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
115 #define SCALEVLCBITS 8
116 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
117 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
118 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
119 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
120 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
122 static VLC sf_vlc; ///< scale factor DPCM vlc
123 static VLC sf_rl_vlc; ///< scale factor run length vlc
124 static VLC vec4_vlc; ///< 4 coefficients per symbol
125 static VLC vec2_vlc; ///< 2 coefficients per symbol
126 static VLC vec1_vlc; ///< 1 coefficient per symbol
127 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
128 static float sin64[33]; ///< sinus table for decorrelation
131 * @brief frame specific decoder context for a single channel
134 int16_t prev_block_len; ///< length of the previous block
135 uint8_t transmit_coefs;
136 uint8_t num_subframes;
137 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
138 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
139 uint8_t cur_subframe; ///< current subframe number
140 uint16_t decoded_samples; ///< number of already processed samples
141 uint8_t grouped; ///< channel is part of a group
142 int quant_step; ///< quantization step for the current subframe
143 int8_t reuse_sf; ///< share scale factors between subframes
144 int8_t scale_factor_step; ///< scaling step for the current subframe
145 int max_scale_factor; ///< maximum scale factor for the current subframe
146 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
147 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
148 int* scale_factors; ///< pointer to the scale factor values used for decoding
149 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
150 float* coeffs; ///< pointer to the subframe decode buffer
151 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
152 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
156 * @brief channel group for channel transformations
159 uint8_t num_channels; ///< number of channels in the group
160 int8_t transform; ///< transform on / off
161 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
162 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
163 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
167 * @brief main decoder context
169 typedef struct WMAProDecodeCtx {
170 /* generic decoder variables */
171 AVCodecContext* avctx; ///< codec context for av_log
172 AVFrame frame; ///< AVFrame for decoded output
173 DSPContext dsp; ///< accelerated DSP functions
174 FmtConvertContext fmt_conv;
175 uint8_t frame_data[MAX_FRAMESIZE +
176 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
177 PutBitContext pb; ///< context for filling the frame_data buffer
178 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
179 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
180 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
182 /* frame size dependent frame information (set during initialization) */
183 uint32_t decode_flags; ///< used compression features
184 uint8_t len_prefix; ///< frame is prefixed with its length
185 uint8_t dynamic_range_compression; ///< frame contains DRC data
186 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
187 uint16_t samples_per_frame; ///< number of samples to output
188 uint16_t log2_frame_size;
189 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
190 int8_t lfe_channel; ///< lfe channel index
191 uint8_t max_num_subframes;
192 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
193 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
194 uint16_t min_samples_per_subframe;
195 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
196 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
197 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
198 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
200 /* packet decode state */
201 GetBitContext pgb; ///< bitstream reader context for the packet
202 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
203 uint8_t packet_offset; ///< frame offset in the packet
204 uint8_t packet_sequence_number; ///< current packet number
205 int num_saved_bits; ///< saved number of bits
206 int frame_offset; ///< frame offset in the bit reservoir
207 int subframe_offset; ///< subframe offset in the bit reservoir
208 uint8_t packet_loss; ///< set in case of bitstream error
209 uint8_t packet_done; ///< set when a packet is fully decoded
211 /* frame decode state */
212 uint32_t frame_num; ///< current frame number (not used for decoding)
213 GetBitContext gb; ///< bitstream reader context
214 int buf_bit_size; ///< buffer size in bits
215 uint8_t drc_gain; ///< gain for the DRC tool
216 int8_t skip_frame; ///< skip output step
217 int8_t parsed_all_subframes; ///< all subframes decoded?
219 /* subframe/block decode state */
220 int16_t subframe_len; ///< current subframe length
221 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
222 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
223 int8_t num_bands; ///< number of scale factor bands
224 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
225 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
226 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
227 int8_t esc_len; ///< length of escaped coefficients
229 uint8_t num_chgroups; ///< number of channel groups
230 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
232 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
237 *@brief helper function to print the most important members of the context
240 static av_cold void dump_context(WMAProDecodeCtx *s)
242 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
243 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
245 PRINT("ed sample bit depth", s->bits_per_sample);
246 PRINT_HEX("ed decode flags", s->decode_flags);
247 PRINT("samples per frame", s->samples_per_frame);
248 PRINT("log2 frame size", s->log2_frame_size);
249 PRINT("max num subframes", s->max_num_subframes);
250 PRINT("len prefix", s->len_prefix);
251 PRINT("num channels", s->num_channels);
255 *@brief Uninitialize the decoder and free all resources.
256 *@param avctx codec context
257 *@return 0 on success, < 0 otherwise
259 static av_cold int decode_end(AVCodecContext *avctx)
261 WMAProDecodeCtx *s = avctx->priv_data;
264 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
265 ff_mdct_end(&s->mdct_ctx[i]);
271 *@brief Initialize the decoder.
272 *@param avctx codec context
273 *@return 0 on success, -1 otherwise
275 static av_cold int decode_init(AVCodecContext *avctx)
277 WMAProDecodeCtx *s = avctx->priv_data;
278 uint8_t *edata_ptr = avctx->extradata;
279 unsigned int channel_mask;
281 int log2_max_num_subframes;
282 int num_possible_block_sizes;
285 ff_dsputil_init(&s->dsp, avctx);
286 ff_fmt_convert_init(&s->fmt_conv, avctx);
287 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
289 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
291 if (avctx->extradata_size >= 18) {
292 s->decode_flags = AV_RL16(edata_ptr+14);
293 channel_mask = AV_RL32(edata_ptr+2);
294 s->bits_per_sample = AV_RL16(edata_ptr);
295 /** dump the extradata */
296 for (i = 0; i < avctx->extradata_size; i++)
297 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
298 av_dlog(avctx, "\n");
301 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
302 return AVERROR_INVALIDDATA;
306 s->log2_frame_size = av_log2(avctx->block_align) + 4;
309 s->skip_frame = 1; /* skip first frame */
311 s->len_prefix = (s->decode_flags & 0x40);
314 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
315 if (bits > WMAPRO_BLOCK_MAX_BITS) {
316 av_log_missing_feature(avctx, "14-bits block sizes", 1);
317 return AVERROR_INVALIDDATA;
319 s->samples_per_frame = 1 << bits;
322 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
323 s->max_num_subframes = 1 << log2_max_num_subframes;
324 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
325 s->max_subframe_len_bit = 1;
326 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
328 num_possible_block_sizes = log2_max_num_subframes + 1;
329 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
330 s->dynamic_range_compression = (s->decode_flags & 0x80);
332 if (s->max_num_subframes > MAX_SUBFRAMES) {
333 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
334 s->max_num_subframes);
335 return AVERROR_INVALIDDATA;
338 if (s->min_samples_per_subframe < (1<<WMAPRO_BLOCK_MIN_BITS)) {
339 av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
340 s->min_samples_per_subframe);
341 return AVERROR_INVALIDDATA;
344 s->num_channels = avctx->channels;
346 if (s->num_channels < 0) {
347 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
348 return AVERROR_INVALIDDATA;
349 } else if (s->num_channels > WMAPRO_MAX_CHANNELS) {
350 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
351 return AVERROR_PATCHWELCOME;
354 /** init previous block len */
355 for (i = 0; i < s->num_channels; i++)
356 s->channel[i].prev_block_len = s->samples_per_frame;
358 /** extract lfe channel position */
361 if (channel_mask & 8) {
363 for (mask = 1; mask < 16; mask <<= 1) {
364 if (channel_mask & mask)
369 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
370 scale_huffbits, 1, 1,
371 scale_huffcodes, 2, 2, 616);
373 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
374 scale_rl_huffbits, 1, 1,
375 scale_rl_huffcodes, 4, 4, 1406);
377 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
378 coef0_huffbits, 1, 1,
379 coef0_huffcodes, 4, 4, 2108);
381 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
382 coef1_huffbits, 1, 1,
383 coef1_huffcodes, 4, 4, 3912);
385 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
387 vec4_huffcodes, 2, 2, 604);
389 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
391 vec2_huffcodes, 2, 2, 562);
393 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
395 vec1_huffcodes, 2, 2, 562);
397 /** calculate number of scale factor bands and their offsets
398 for every possible block size */
399 for (i = 0; i < num_possible_block_sizes; i++) {
400 int subframe_len = s->samples_per_frame >> i;
404 s->sfb_offsets[i][0] = 0;
406 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
407 int offset = (subframe_len * 2 * critical_freq[x])
408 / s->avctx->sample_rate + 2;
410 if (offset > s->sfb_offsets[i][band - 1])
411 s->sfb_offsets[i][band++] = offset;
413 s->sfb_offsets[i][band - 1] = subframe_len;
414 s->num_sfb[i] = band - 1;
418 /** Scale factors can be shared between blocks of different size
419 as every block has a different scale factor band layout.
420 The matrix sf_offsets is needed to find the correct scale factor.
423 for (i = 0; i < num_possible_block_sizes; i++) {
425 for (b = 0; b < s->num_sfb[i]; b++) {
427 int offset = ((s->sfb_offsets[i][b]
428 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
429 for (x = 0; x < num_possible_block_sizes; x++) {
431 while (s->sfb_offsets[x][v + 1] << x < offset)
433 s->sf_offsets[i][x][b] = v;
438 /** init MDCT, FIXME: only init needed sizes */
439 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
440 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
441 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
442 / (1 << (s->bits_per_sample - 1)));
444 /** init MDCT windows: simple sinus window */
445 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
446 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
447 ff_init_ff_sine_windows(win_idx);
448 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
451 /** calculate subwoofer cutoff values */
452 for (i = 0; i < num_possible_block_sizes; i++) {
453 int block_size = s->samples_per_frame >> i;
454 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
455 / s->avctx->sample_rate;
456 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
459 /** calculate sine values for the decorrelation matrix */
460 for (i = 0; i < 33; i++)
461 sin64[i] = sin(i*M_PI / 64.0);
463 if (avctx->debug & FF_DEBUG_BITSTREAM)
466 avctx->channel_layout = channel_mask;
468 avcodec_get_frame_defaults(&s->frame);
469 avctx->coded_frame = &s->frame;
475 *@brief Decode the subframe length.
477 *@param offset sample offset in the frame
478 *@return decoded subframe length on success, < 0 in case of an error
480 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
482 int frame_len_shift = 0;
485 /** no need to read from the bitstream when only one length is possible */
486 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
487 return s->min_samples_per_subframe;
489 /** 1 bit indicates if the subframe is of maximum length */
490 if (s->max_subframe_len_bit) {
491 if (get_bits1(&s->gb))
492 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
494 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
496 subframe_len = s->samples_per_frame >> frame_len_shift;
498 /** sanity check the length */
499 if (subframe_len < s->min_samples_per_subframe ||
500 subframe_len > s->samples_per_frame) {
501 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
503 return AVERROR_INVALIDDATA;
509 *@brief Decode how the data in the frame is split into subframes.
510 * Every WMA frame contains the encoded data for a fixed number of
511 * samples per channel. The data for every channel might be split
512 * into several subframes. This function will reconstruct the list of
513 * subframes for every channel.
515 * If the subframes are not evenly split, the algorithm estimates the
516 * channels with the lowest number of total samples.
517 * Afterwards, for each of these channels a bit is read from the
518 * bitstream that indicates if the channel contains a subframe with the
519 * next subframe size that is going to be read from the bitstream or not.
520 * If a channel contains such a subframe, the subframe size gets added to
521 * the channel's subframe list.
522 * The algorithm repeats these steps until the frame is properly divided
523 * between the individual channels.
526 *@return 0 on success, < 0 in case of an error
528 static int decode_tilehdr(WMAProDecodeCtx *s)
530 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
531 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
532 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
533 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
534 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
537 /* Should never consume more than 3073 bits (256 iterations for the
538 * while loop when always the minimum amount of 128 samples is subtracted
539 * from missing samples in the 8 channel case).
540 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
543 /** reset tiling information */
544 for (c = 0; c < s->num_channels; c++)
545 s->channel[c].num_subframes = 0;
547 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
548 fixed_channel_layout = 1;
550 /** loop until the frame data is split between the subframes */
554 /** check which channels contain the subframe */
555 for (c = 0; c < s->num_channels; c++) {
556 if (num_samples[c] == min_channel_len) {
557 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
558 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
559 contains_subframe[c] = 1;
561 contains_subframe[c] = get_bits1(&s->gb);
563 contains_subframe[c] = 0;
566 /** get subframe length, subframe_len == 0 is not allowed */
567 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
568 return AVERROR_INVALIDDATA;
570 /** add subframes to the individual channels and find new min_channel_len */
571 min_channel_len += subframe_len;
572 for (c = 0; c < s->num_channels; c++) {
573 WMAProChannelCtx* chan = &s->channel[c];
575 if (contains_subframe[c]) {
576 if (chan->num_subframes >= MAX_SUBFRAMES) {
577 av_log(s->avctx, AV_LOG_ERROR,
578 "broken frame: num subframes > 31\n");
579 return AVERROR_INVALIDDATA;
581 chan->subframe_len[chan->num_subframes] = subframe_len;
582 num_samples[c] += subframe_len;
583 ++chan->num_subframes;
584 if (num_samples[c] > s->samples_per_frame) {
585 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
586 "channel len > samples_per_frame\n");
587 return AVERROR_INVALIDDATA;
589 } else if (num_samples[c] <= min_channel_len) {
590 if (num_samples[c] < min_channel_len) {
591 channels_for_cur_subframe = 0;
592 min_channel_len = num_samples[c];
594 ++channels_for_cur_subframe;
597 } while (min_channel_len < s->samples_per_frame);
599 for (c = 0; c < s->num_channels; c++) {
602 for (i = 0; i < s->channel[c].num_subframes; i++) {
603 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
604 " len %i\n", s->frame_num, c, i,
605 s->channel[c].subframe_len[i]);
606 s->channel[c].subframe_offset[i] = offset;
607 offset += s->channel[c].subframe_len[i];
615 *@brief Calculate a decorrelation matrix from the bitstream parameters.
616 *@param s codec context
617 *@param chgroup channel group for which the matrix needs to be calculated
619 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
620 WMAProChannelGrp *chgroup)
624 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
625 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
626 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
628 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
629 rotation_offset[i] = get_bits(&s->gb, 6);
631 for (i = 0; i < chgroup->num_channels; i++)
632 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
633 get_bits1(&s->gb) ? 1.0 : -1.0;
635 for (i = 1; i < chgroup->num_channels; i++) {
637 for (x = 0; x < i; x++) {
639 for (y = 0; y < i + 1; y++) {
640 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
641 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
642 int n = rotation_offset[offset + x];
648 cosv = sin64[32 - n];
650 sinv = sin64[64 - n];
651 cosv = -sin64[n - 32];
654 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
655 (v1 * sinv) - (v2 * cosv);
656 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
657 (v1 * cosv) + (v2 * sinv);
665 *@brief Decode channel transformation parameters
666 *@param s codec context
667 *@return 0 in case of success, < 0 in case of bitstream errors
669 static int decode_channel_transform(WMAProDecodeCtx* s)
672 /* should never consume more than 1921 bits for the 8 channel case
673 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
674 * + MAX_CHANNELS + MAX_BANDS + 1)
677 /** in the one channel case channel transforms are pointless */
679 if (s->num_channels > 1) {
680 int remaining_channels = s->channels_for_cur_subframe;
682 if (get_bits1(&s->gb)) {
683 av_log_ask_for_sample(s->avctx,
684 "unsupported channel transform bit\n");
685 return AVERROR_INVALIDDATA;
688 for (s->num_chgroups = 0; remaining_channels &&
689 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
690 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
691 float** channel_data = chgroup->channel_data;
692 chgroup->num_channels = 0;
693 chgroup->transform = 0;
695 /** decode channel mask */
696 if (remaining_channels > 2) {
697 for (i = 0; i < s->channels_for_cur_subframe; i++) {
698 int channel_idx = s->channel_indexes_for_cur_subframe[i];
699 if (!s->channel[channel_idx].grouped
700 && get_bits1(&s->gb)) {
701 ++chgroup->num_channels;
702 s->channel[channel_idx].grouped = 1;
703 *channel_data++ = s->channel[channel_idx].coeffs;
707 chgroup->num_channels = remaining_channels;
708 for (i = 0; i < s->channels_for_cur_subframe; i++) {
709 int channel_idx = s->channel_indexes_for_cur_subframe[i];
710 if (!s->channel[channel_idx].grouped)
711 *channel_data++ = s->channel[channel_idx].coeffs;
712 s->channel[channel_idx].grouped = 1;
716 /** decode transform type */
717 if (chgroup->num_channels == 2) {
718 if (get_bits1(&s->gb)) {
719 if (get_bits1(&s->gb)) {
720 av_log_ask_for_sample(s->avctx,
721 "unsupported channel transform type\n");
724 chgroup->transform = 1;
725 if (s->num_channels == 2) {
726 chgroup->decorrelation_matrix[0] = 1.0;
727 chgroup->decorrelation_matrix[1] = -1.0;
728 chgroup->decorrelation_matrix[2] = 1.0;
729 chgroup->decorrelation_matrix[3] = 1.0;
732 chgroup->decorrelation_matrix[0] = 0.70703125;
733 chgroup->decorrelation_matrix[1] = -0.70703125;
734 chgroup->decorrelation_matrix[2] = 0.70703125;
735 chgroup->decorrelation_matrix[3] = 0.70703125;
738 } else if (chgroup->num_channels > 2) {
739 if (get_bits1(&s->gb)) {
740 chgroup->transform = 1;
741 if (get_bits1(&s->gb)) {
742 decode_decorrelation_matrix(s, chgroup);
744 /** FIXME: more than 6 coupled channels not supported */
745 if (chgroup->num_channels > 6) {
746 av_log_ask_for_sample(s->avctx,
747 "coupled channels > 6\n");
749 memcpy(chgroup->decorrelation_matrix,
750 default_decorrelation[chgroup->num_channels],
751 chgroup->num_channels * chgroup->num_channels *
752 sizeof(*chgroup->decorrelation_matrix));
758 /** decode transform on / off */
759 if (chgroup->transform) {
760 if (!get_bits1(&s->gb)) {
762 /** transform can be enabled for individual bands */
763 for (i = 0; i < s->num_bands; i++) {
764 chgroup->transform_band[i] = get_bits1(&s->gb);
767 memset(chgroup->transform_band, 1, s->num_bands);
770 remaining_channels -= chgroup->num_channels;
777 *@brief Extract the coefficients from the bitstream.
778 *@param s codec context
779 *@param c current channel number
780 *@return 0 on success, < 0 in case of bitstream errors
782 static int decode_coeffs(WMAProDecodeCtx *s, int c)
784 /* Integers 0..15 as single-precision floats. The table saves a
785 costly int to float conversion, and storing the values as
786 integers allows fast sign-flipping. */
787 static const uint32_t fval_tab[16] = {
788 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
789 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
790 0x41000000, 0x41100000, 0x41200000, 0x41300000,
791 0x41400000, 0x41500000, 0x41600000, 0x41700000,
795 WMAProChannelCtx* ci = &s->channel[c];
802 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
804 vlctable = get_bits1(&s->gb);
805 vlc = &coef_vlc[vlctable];
815 /** decode vector coefficients (consumes up to 167 bits per iteration for
816 4 vector coded large values) */
817 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
818 (cur_coeff + 3 < ci->num_vec_coeffs)) {
823 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
825 if (idx == HUFF_VEC4_SIZE - 1) {
826 for (i = 0; i < 4; i += 2) {
827 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
828 if (idx == HUFF_VEC2_SIZE - 1) {
830 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
831 if (v0 == HUFF_VEC1_SIZE - 1)
832 v0 += ff_wma_get_large_val(&s->gb);
833 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
834 if (v1 == HUFF_VEC1_SIZE - 1)
835 v1 += ff_wma_get_large_val(&s->gb);
836 vals[i ] = av_float2int(v0);
837 vals[i+1] = av_float2int(v1);
839 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
840 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
844 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
845 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
846 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
847 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
851 for (i = 0; i < 4; i++) {
853 uint32_t sign = get_bits1(&s->gb) - 1;
854 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
857 ci->coeffs[cur_coeff] = 0;
858 /** switch to run level mode when subframe_len / 128 zeros
859 were found in a row */
860 rl_mode |= (++num_zeros > s->subframe_len >> 8);
866 /** decode run level coded coefficients */
867 if (cur_coeff < s->subframe_len) {
868 memset(&ci->coeffs[cur_coeff], 0,
869 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
870 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
871 level, run, 1, ci->coeffs,
872 cur_coeff, s->subframe_len,
873 s->subframe_len, s->esc_len, 0))
874 return AVERROR_INVALIDDATA;
881 *@brief Extract scale factors from the bitstream.
882 *@param s codec context
883 *@return 0 on success, < 0 in case of bitstream errors
885 static int decode_scale_factors(WMAProDecodeCtx* s)
889 /** should never consume more than 5344 bits
890 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
893 for (i = 0; i < s->channels_for_cur_subframe; i++) {
894 int c = s->channel_indexes_for_cur_subframe[i];
897 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
898 sf_end = s->channel[c].scale_factors + s->num_bands;
900 /** resample scale factors for the new block size
901 * as the scale factors might need to be resampled several times
902 * before some new values are transmitted, a backup of the last
903 * transmitted scale factors is kept in saved_scale_factors
905 if (s->channel[c].reuse_sf) {
906 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
908 for (b = 0; b < s->num_bands; b++)
909 s->channel[c].scale_factors[b] =
910 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
913 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
915 if (!s->channel[c].reuse_sf) {
917 /** decode DPCM coded scale factors */
918 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
919 val = 45 / s->channel[c].scale_factor_step;
920 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
921 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
926 /** run level decode differences to the resampled factors */
927 for (i = 0; i < s->num_bands; i++) {
933 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
936 uint32_t code = get_bits(&s->gb, 14);
938 sign = (code & 1) - 1;
939 skip = (code & 0x3f) >> 1;
940 } else if (idx == 1) {
943 skip = scale_rl_run[idx];
944 val = scale_rl_level[idx];
945 sign = get_bits1(&s->gb)-1;
949 if (i >= s->num_bands) {
950 av_log(s->avctx, AV_LOG_ERROR,
951 "invalid scale factor coding\n");
952 return AVERROR_INVALIDDATA;
954 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
958 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
959 s->channel[c].table_idx = s->table_idx;
960 s->channel[c].reuse_sf = 1;
963 /** calculate new scale factor maximum */
964 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
965 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
966 s->channel[c].max_scale_factor =
967 FFMAX(s->channel[c].max_scale_factor, *sf);
975 *@brief Reconstruct the individual channel data.
976 *@param s codec context
978 static void inverse_channel_transform(WMAProDecodeCtx *s)
982 for (i = 0; i < s->num_chgroups; i++) {
983 if (s->chgroup[i].transform) {
984 float data[WMAPRO_MAX_CHANNELS];
985 const int num_channels = s->chgroup[i].num_channels;
986 float** ch_data = s->chgroup[i].channel_data;
987 float** ch_end = ch_data + num_channels;
988 const int8_t* tb = s->chgroup[i].transform_band;
991 /** multichannel decorrelation */
992 for (sfb = s->cur_sfb_offsets;
993 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
996 /** multiply values with the decorrelation_matrix */
997 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
998 const float* mat = s->chgroup[i].decorrelation_matrix;
999 const float* data_end = data + num_channels;
1000 float* data_ptr = data;
1003 for (ch = ch_data; ch < ch_end; ch++)
1004 *data_ptr++ = (*ch)[y];
1006 for (ch = ch_data; ch < ch_end; ch++) {
1009 while (data_ptr < data_end)
1010 sum += *data_ptr++ * *mat++;
1015 } else if (s->num_channels == 2) {
1016 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1017 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1018 ch_data[0] + sfb[0],
1020 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1021 ch_data[1] + sfb[0],
1030 *@brief Apply sine window and reconstruct the output buffer.
1031 *@param s codec context
1033 static void wmapro_window(WMAProDecodeCtx *s)
1036 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1037 int c = s->channel_indexes_for_cur_subframe[i];
1039 int winlen = s->channel[c].prev_block_len;
1040 float* start = s->channel[c].coeffs - (winlen >> 1);
1042 if (s->subframe_len < winlen) {
1043 start += (winlen - s->subframe_len) >> 1;
1044 winlen = s->subframe_len;
1047 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1051 s->dsp.vector_fmul_window(start, start, start + winlen,
1054 s->channel[c].prev_block_len = s->subframe_len;
1059 *@brief Decode a single subframe (block).
1060 *@param s codec context
1061 *@return 0 on success, < 0 when decoding failed
1063 static int decode_subframe(WMAProDecodeCtx *s)
1065 int offset = s->samples_per_frame;
1066 int subframe_len = s->samples_per_frame;
1068 int total_samples = s->samples_per_frame * s->num_channels;
1069 int transmit_coeffs = 0;
1070 int cur_subwoofer_cutoff;
1072 s->subframe_offset = get_bits_count(&s->gb);
1074 /** reset channel context and find the next block offset and size
1075 == the next block of the channel with the smallest number of
1078 for (i = 0; i < s->num_channels; i++) {
1079 s->channel[i].grouped = 0;
1080 if (offset > s->channel[i].decoded_samples) {
1081 offset = s->channel[i].decoded_samples;
1083 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1088 "processing subframe with offset %i len %i\n", offset, subframe_len);
1090 /** get a list of all channels that contain the estimated block */
1091 s->channels_for_cur_subframe = 0;
1092 for (i = 0; i < s->num_channels; i++) {
1093 const int cur_subframe = s->channel[i].cur_subframe;
1094 /** substract already processed samples */
1095 total_samples -= s->channel[i].decoded_samples;
1097 /** and count if there are multiple subframes that match our profile */
1098 if (offset == s->channel[i].decoded_samples &&
1099 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1100 total_samples -= s->channel[i].subframe_len[cur_subframe];
1101 s->channel[i].decoded_samples +=
1102 s->channel[i].subframe_len[cur_subframe];
1103 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1104 ++s->channels_for_cur_subframe;
1108 /** check if the frame will be complete after processing the
1111 s->parsed_all_subframes = 1;
1114 av_dlog(s->avctx, "subframe is part of %i channels\n",
1115 s->channels_for_cur_subframe);
1117 /** calculate number of scale factor bands and their offsets */
1118 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1119 s->num_bands = s->num_sfb[s->table_idx];
1120 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1121 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1123 /** configure the decoder for the current subframe */
1124 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1125 int c = s->channel_indexes_for_cur_subframe[i];
1127 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1131 s->subframe_len = subframe_len;
1132 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1134 /** skip extended header if any */
1135 if (get_bits1(&s->gb)) {
1137 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1138 int len = get_bits(&s->gb, 4);
1139 num_fill_bits = (len ? get_bits(&s->gb, len) : 0) + 1;
1142 if (num_fill_bits >= 0) {
1143 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1144 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1145 return AVERROR_INVALIDDATA;
1148 skip_bits_long(&s->gb, num_fill_bits);
1152 /** no idea for what the following bit is used */
1153 if (get_bits1(&s->gb)) {
1154 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1155 return AVERROR_INVALIDDATA;
1159 if (decode_channel_transform(s) < 0)
1160 return AVERROR_INVALIDDATA;
1163 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1164 int c = s->channel_indexes_for_cur_subframe[i];
1165 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1166 transmit_coeffs = 1;
1169 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1170 if (transmit_coeffs) {
1172 int quant_step = 90 * s->bits_per_sample >> 4;
1174 /** decode number of vector coded coefficients */
1175 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1176 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1177 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1178 int c = s->channel_indexes_for_cur_subframe[i];
1179 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1180 if (num_vec_coeffs > s->subframe_len) {
1181 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1182 return AVERROR_INVALIDDATA;
1184 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1187 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1188 int c = s->channel_indexes_for_cur_subframe[i];
1189 s->channel[c].num_vec_coeffs = s->subframe_len;
1192 /** decode quantization step */
1193 step = get_sbits(&s->gb, 6);
1195 if (step == -32 || step == 31) {
1196 const int sign = (step == 31) - 1;
1198 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1199 (step = get_bits(&s->gb, 5)) == 31) {
1202 quant_step += ((quant + step) ^ sign) - sign;
1204 if (quant_step < 0) {
1205 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1208 /** decode quantization step modifiers for every channel */
1210 if (s->channels_for_cur_subframe == 1) {
1211 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1213 int modifier_len = get_bits(&s->gb, 3);
1214 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1215 int c = s->channel_indexes_for_cur_subframe[i];
1216 s->channel[c].quant_step = quant_step;
1217 if (get_bits1(&s->gb)) {
1219 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1221 ++s->channel[c].quant_step;
1226 /** decode scale factors */
1227 if (decode_scale_factors(s) < 0)
1228 return AVERROR_INVALIDDATA;
1231 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1232 get_bits_count(&s->gb) - s->subframe_offset);
1234 /** parse coefficients */
1235 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1236 int c = s->channel_indexes_for_cur_subframe[i];
1237 if (s->channel[c].transmit_coefs &&
1238 get_bits_count(&s->gb) < s->num_saved_bits) {
1239 decode_coeffs(s, c);
1241 memset(s->channel[c].coeffs, 0,
1242 sizeof(*s->channel[c].coeffs) * subframe_len);
1245 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1246 get_bits_count(&s->gb) - s->subframe_offset);
1248 if (transmit_coeffs) {
1249 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1250 /** reconstruct the per channel data */
1251 inverse_channel_transform(s);
1252 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1253 int c = s->channel_indexes_for_cur_subframe[i];
1254 const int* sf = s->channel[c].scale_factors;
1257 if (c == s->lfe_channel)
1258 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1259 (subframe_len - cur_subwoofer_cutoff));
1261 /** inverse quantization and rescaling */
1262 for (b = 0; b < s->num_bands; b++) {
1263 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1264 const int exp = s->channel[c].quant_step -
1265 (s->channel[c].max_scale_factor - *sf++) *
1266 s->channel[c].scale_factor_step;
1267 const float quant = pow(10.0, exp / 20.0);
1268 int start = s->cur_sfb_offsets[b];
1269 s->dsp.vector_fmul_scalar(s->tmp + start,
1270 s->channel[c].coeffs + start,
1271 quant, end - start);
1274 /** apply imdct (imdct_half == DCTIV with reverse) */
1275 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1279 /** window and overlapp-add */
1282 /** handled one subframe */
1283 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1284 int c = s->channel_indexes_for_cur_subframe[i];
1285 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1286 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1287 return AVERROR_INVALIDDATA;
1289 ++s->channel[c].cur_subframe;
1296 *@brief Decode one WMA frame.
1297 *@param s codec context
1298 *@return 0 if the trailer bit indicates that this is the last frame,
1299 * 1 if there are additional frames
1301 static int decode_frame(WMAProDecodeCtx *s, int *got_frame_ptr)
1303 AVCodecContext *avctx = s->avctx;
1304 GetBitContext* gb = &s->gb;
1305 int more_frames = 0;
1308 const float *out_ptr[WMAPRO_MAX_CHANNELS];
1311 /** get frame length */
1313 len = get_bits(gb, s->log2_frame_size);
1315 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1317 /** decode tile information */
1318 if (decode_tilehdr(s)) {
1323 /** read postproc transform */
1324 if (s->num_channels > 1 && get_bits1(gb)) {
1325 if (get_bits1(gb)) {
1326 for (i = 0; i < s->num_channels * s->num_channels; i++)
1331 /** read drc info */
1332 if (s->dynamic_range_compression) {
1333 s->drc_gain = get_bits(gb, 8);
1334 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1337 /** no idea what these are for, might be the number of samples
1338 that need to be skipped at the beginning or end of a stream */
1339 if (get_bits1(gb)) {
1342 /** usually true for the first frame */
1343 if (get_bits1(gb)) {
1344 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1345 av_dlog(s->avctx, "start skip: %i\n", skip);
1348 /** sometimes true for the last frame */
1349 if (get_bits1(gb)) {
1350 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1351 av_dlog(s->avctx, "end skip: %i\n", skip);
1356 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1357 get_bits_count(gb) - s->frame_offset);
1359 /** reset subframe states */
1360 s->parsed_all_subframes = 0;
1361 for (i = 0; i < s->num_channels; i++) {
1362 s->channel[i].decoded_samples = 0;
1363 s->channel[i].cur_subframe = 0;
1364 s->channel[i].reuse_sf = 0;
1367 /** decode all subframes */
1368 while (!s->parsed_all_subframes) {
1369 if (decode_subframe(s) < 0) {
1375 /* get output buffer */
1376 s->frame.nb_samples = s->samples_per_frame;
1377 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
1378 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1382 samples = (float *)s->frame.data[0];
1384 /** interleave samples and write them to the output buffer */
1385 for (i = 0; i < s->num_channels; i++)
1386 out_ptr[i] = s->channel[i].out;
1387 s->fmt_conv.float_interleave(samples, out_ptr, s->samples_per_frame,
1390 for (i = 0; i < s->num_channels; i++) {
1391 /** reuse second half of the IMDCT output for the next frame */
1392 memcpy(&s->channel[i].out[0],
1393 &s->channel[i].out[s->samples_per_frame],
1394 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1397 if (s->skip_frame) {
1404 if (s->len_prefix) {
1405 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1406 /** FIXME: not sure if this is always an error */
1407 av_log(s->avctx, AV_LOG_ERROR,
1408 "frame[%i] would have to skip %i bits\n", s->frame_num,
1409 len - (get_bits_count(gb) - s->frame_offset) - 1);
1414 /** skip the rest of the frame data */
1415 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1417 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1421 /** decode trailer bit */
1422 more_frames = get_bits1(gb);
1429 *@brief Calculate remaining input buffer length.
1430 *@param s codec context
1431 *@param gb bitstream reader context
1432 *@return remaining size in bits
1434 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1436 return s->buf_bit_size - get_bits_count(gb);
1440 *@brief Fill the bit reservoir with a (partial) frame.
1441 *@param s codec context
1442 *@param gb bitstream reader context
1443 *@param len length of the partial frame
1444 *@param append decides whether to reset the buffer or not
1446 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1451 /** when the frame data does not need to be concatenated, the input buffer
1452 is reset and additional bits from the previous frame are copied
1453 and skipped later so that a fast byte copy is possible */
1456 s->frame_offset = get_bits_count(gb) & 7;
1457 s->num_saved_bits = s->frame_offset;
1458 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1461 buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1463 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1464 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1469 s->num_saved_bits += len;
1471 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1474 int align = 8 - (get_bits_count(gb) & 7);
1475 align = FFMIN(align, len);
1476 put_bits(&s->pb, align, get_bits(gb, align));
1478 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1480 skip_bits_long(gb, len);
1483 PutBitContext tmp = s->pb;
1484 flush_put_bits(&tmp);
1487 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1488 skip_bits(&s->gb, s->frame_offset);
1492 *@brief Decode a single WMA packet.
1493 *@param avctx codec context
1494 *@param data the output buffer
1495 *@param avpkt input packet
1496 *@return number of bytes that were read from the input buffer
1498 static int decode_packet(AVCodecContext *avctx, void *data,
1499 int *got_frame_ptr, AVPacket* avpkt)
1501 WMAProDecodeCtx *s = avctx->priv_data;
1502 GetBitContext* gb = &s->pgb;
1503 const uint8_t* buf = avpkt->data;
1504 int buf_size = avpkt->size;
1505 int num_bits_prev_frame;
1506 int packet_sequence_number;
1510 if (s->packet_done || s->packet_loss) {
1513 /** sanity check for the buffer length */
1514 if (buf_size < avctx->block_align)
1517 s->next_packet_start = buf_size - avctx->block_align;
1518 buf_size = avctx->block_align;
1519 s->buf_bit_size = buf_size << 3;
1521 /** parse packet header */
1522 init_get_bits(gb, buf, s->buf_bit_size);
1523 packet_sequence_number = get_bits(gb, 4);
1526 /** get number of bits that need to be added to the previous frame */
1527 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1528 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1529 num_bits_prev_frame);
1531 /** check for packet loss */
1532 if (!s->packet_loss &&
1533 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1535 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1536 s->packet_sequence_number, packet_sequence_number);
1538 s->packet_sequence_number = packet_sequence_number;
1540 if (num_bits_prev_frame > 0) {
1541 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1542 if (num_bits_prev_frame >= remaining_packet_bits) {
1543 num_bits_prev_frame = remaining_packet_bits;
1547 /** append the previous frame data to the remaining data from the
1548 previous packet to create a full frame */
1549 save_bits(s, gb, num_bits_prev_frame, 1);
1550 av_dlog(avctx, "accumulated %x bits of frame data\n",
1551 s->num_saved_bits - s->frame_offset);
1553 /** decode the cross packet frame if it is valid */
1554 if (!s->packet_loss)
1555 decode_frame(s, got_frame_ptr);
1556 } else if (s->num_saved_bits - s->frame_offset) {
1557 av_dlog(avctx, "ignoring %x previously saved bits\n",
1558 s->num_saved_bits - s->frame_offset);
1561 if (s->packet_loss) {
1562 /** reset number of saved bits so that the decoder
1563 does not start to decode incomplete frames in the
1564 s->len_prefix == 0 case */
1565 s->num_saved_bits = 0;
1571 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1572 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1573 skip_bits(gb, s->packet_offset);
1574 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1575 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1576 frame_size <= remaining_bits(s, gb)) {
1577 save_bits(s, gb, frame_size, 0);
1578 s->packet_done = !decode_frame(s, got_frame_ptr);
1579 } else if (!s->len_prefix
1580 && s->num_saved_bits > get_bits_count(&s->gb)) {
1581 /** when the frames do not have a length prefix, we don't know
1582 the compressed length of the individual frames
1583 however, we know what part of a new packet belongs to the
1585 therefore we save the incoming packet first, then we append
1586 the "previous frame" data from the next packet so that
1587 we get a buffer that only contains full frames */
1588 s->packet_done = !decode_frame(s, got_frame_ptr);
1593 if (s->packet_done && !s->packet_loss &&
1594 remaining_bits(s, gb) > 0) {
1595 /** save the rest of the data so that it can be decoded
1596 with the next packet */
1597 save_bits(s, gb, remaining_bits(s, gb), 0);
1600 s->packet_offset = get_bits_count(gb) & 7;
1602 return AVERROR_INVALIDDATA;
1605 *(AVFrame *)data = s->frame;
1607 return get_bits_count(gb) >> 3;
1611 *@brief Clear decoder buffers (for seeking).
1612 *@param avctx codec context
1614 static void flush(AVCodecContext *avctx)
1616 WMAProDecodeCtx *s = avctx->priv_data;
1618 /** reset output buffer as a part of it is used during the windowing of a
1620 for (i = 0; i < s->num_channels; i++)
1621 memset(s->channel[i].out, 0, s->samples_per_frame *
1622 sizeof(*s->channel[i].out));
1628 *@brief wmapro decoder
1630 AVCodec ff_wmapro_decoder = {
1632 .type = AVMEDIA_TYPE_AUDIO,
1633 .id = CODEC_ID_WMAPRO,
1634 .priv_data_size = sizeof(WMAProDecodeCtx),
1635 .init = decode_init,
1636 .close = decode_end,
1637 .decode = decode_packet,
1638 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1640 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),