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/float_dsp.h"
90 #include "libavutil/intfloat.h"
91 #include "libavutil/intreadwrite.h"
96 #include "wmaprodata.h"
99 #include "wma_common.h"
101 /** current decoder limitations */
102 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
103 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
104 #define MAX_BANDS 29 ///< max number of scale factor bands
105 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
107 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
108 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
109 #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size
110 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
111 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
115 #define SCALEVLCBITS 8
116 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
117 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
118 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
119 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
120 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
122 static VLC sf_vlc; ///< scale factor DPCM vlc
123 static VLC sf_rl_vlc; ///< scale factor run length vlc
124 static VLC vec4_vlc; ///< 4 coefficients per symbol
125 static VLC vec2_vlc; ///< 2 coefficients per symbol
126 static VLC vec1_vlc; ///< 1 coefficient per symbol
127 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
128 static float sin64[33]; ///< sine table for decorrelation
131 * @brief frame specific decoder context for a single channel
134 int16_t prev_block_len; ///< length of the previous block
135 uint8_t transmit_coefs;
136 uint8_t num_subframes;
137 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
138 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
139 uint8_t cur_subframe; ///< current subframe number
140 uint16_t decoded_samples; ///< number of already processed samples
141 uint8_t grouped; ///< channel is part of a group
142 int quant_step; ///< quantization step for the current subframe
143 int8_t reuse_sf; ///< share scale factors between subframes
144 int8_t scale_factor_step; ///< scaling step for the current subframe
145 int max_scale_factor; ///< maximum scale factor for the current subframe
146 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
147 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
148 int* scale_factors; ///< pointer to the scale factor values used for decoding
149 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
150 float* coeffs; ///< pointer to the subframe decode buffer
151 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
152 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
156 * @brief channel group for channel transformations
159 uint8_t num_channels; ///< number of channels in the group
160 int8_t transform; ///< transform on / off
161 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
162 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
163 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
167 * @brief main decoder context
169 typedef struct WMAProDecodeCtx {
170 /* generic decoder variables */
171 AVCodecContext* avctx; ///< codec context for av_log
172 AVFloatDSPContext fdsp;
173 uint8_t frame_data[MAX_FRAMESIZE +
174 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
175 PutBitContext pb; ///< context for filling the frame_data buffer
176 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
177 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
178 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
180 /* frame size dependent frame information (set during initialization) */
181 uint32_t decode_flags; ///< used compression features
182 uint8_t len_prefix; ///< frame is prefixed with its length
183 uint8_t dynamic_range_compression; ///< frame contains DRC data
184 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
185 uint16_t samples_per_frame; ///< number of samples to output
186 uint16_t log2_frame_size;
187 int8_t lfe_channel; ///< lfe channel index
188 uint8_t max_num_subframes;
189 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
190 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
191 uint16_t min_samples_per_subframe;
192 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
193 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
194 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
195 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
197 /* packet decode state */
198 GetBitContext pgb; ///< bitstream reader context for the packet
199 int next_packet_start; ///< start offset of the next wma packet in the demuxer packet
200 uint8_t packet_offset; ///< frame offset in the packet
201 uint8_t packet_sequence_number; ///< current packet number
202 int num_saved_bits; ///< saved number of bits
203 int frame_offset; ///< frame offset in the bit reservoir
204 int subframe_offset; ///< subframe offset in the bit reservoir
205 uint8_t packet_loss; ///< set in case of bitstream error
206 uint8_t packet_done; ///< set when a packet is fully decoded
208 /* frame decode state */
209 uint32_t frame_num; ///< current frame number (not used for decoding)
210 GetBitContext gb; ///< bitstream reader context
211 int buf_bit_size; ///< buffer size in bits
212 uint8_t drc_gain; ///< gain for the DRC tool
213 int8_t skip_frame; ///< skip output step
214 int8_t parsed_all_subframes; ///< all subframes decoded?
216 /* subframe/block decode state */
217 int16_t subframe_len; ///< current subframe length
218 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
219 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
220 int8_t num_bands; ///< number of scale factor bands
221 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
222 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
223 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
224 int8_t esc_len; ///< length of escaped coefficients
226 uint8_t num_chgroups; ///< number of channel groups
227 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
229 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
234 *@brief helper function to print the most important members of the context
237 static av_cold void dump_context(WMAProDecodeCtx *s)
239 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
240 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
242 PRINT("ed sample bit depth", s->bits_per_sample);
243 PRINT_HEX("ed decode flags", s->decode_flags);
244 PRINT("samples per frame", s->samples_per_frame);
245 PRINT("log2 frame size", s->log2_frame_size);
246 PRINT("max num subframes", s->max_num_subframes);
247 PRINT("len prefix", s->len_prefix);
248 PRINT("num channels", s->avctx->channels);
252 *@brief Uninitialize the decoder and free all resources.
253 *@param avctx codec context
254 *@return 0 on success, < 0 otherwise
256 static av_cold int decode_end(AVCodecContext *avctx)
258 WMAProDecodeCtx *s = avctx->priv_data;
261 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
262 ff_mdct_end(&s->mdct_ctx[i]);
268 *@brief Initialize the decoder.
269 *@param avctx codec context
270 *@return 0 on success, -1 otherwise
272 static av_cold int decode_init(AVCodecContext *avctx)
274 WMAProDecodeCtx *s = avctx->priv_data;
275 uint8_t *edata_ptr = avctx->extradata;
276 unsigned int channel_mask;
278 int log2_max_num_subframes;
279 int num_possible_block_sizes;
281 if (!avctx->block_align) {
282 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
283 return AVERROR(EINVAL);
287 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
289 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
291 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
293 if (avctx->extradata_size >= 18) {
294 s->decode_flags = AV_RL16(edata_ptr+14);
295 channel_mask = AV_RL32(edata_ptr+2);
296 s->bits_per_sample = AV_RL16(edata_ptr);
297 /** dump the extradata */
298 for (i = 0; i < avctx->extradata_size; i++)
299 av_dlog(avctx, "[%x] ", avctx->extradata[i]);
300 av_dlog(avctx, "\n");
303 avpriv_request_sample(avctx, "Unknown extradata size");
304 return AVERROR_PATCHWELCOME;
308 s->log2_frame_size = av_log2(avctx->block_align) + 4;
309 if (s->log2_frame_size > 25) {
310 avpriv_request_sample(avctx, "Large block align");
311 return AVERROR_PATCHWELCOME;
315 s->skip_frame = 1; /* skip first frame */
317 s->len_prefix = (s->decode_flags & 0x40);
320 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
321 if (bits > WMAPRO_BLOCK_MAX_BITS) {
322 avpriv_request_sample(avctx, "14-bit block sizes");
323 return AVERROR_PATCHWELCOME;
325 s->samples_per_frame = 1 << bits;
328 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
329 s->max_num_subframes = 1 << log2_max_num_subframes;
330 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
331 s->max_subframe_len_bit = 1;
332 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
334 num_possible_block_sizes = log2_max_num_subframes + 1;
335 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
336 s->dynamic_range_compression = (s->decode_flags & 0x80);
338 if (s->max_num_subframes > MAX_SUBFRAMES) {
339 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
340 s->max_num_subframes);
341 return AVERROR_INVALIDDATA;
344 if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) {
345 av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
346 s->min_samples_per_subframe);
347 return AVERROR_INVALIDDATA;
350 if (s->avctx->sample_rate <= 0) {
351 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
352 return AVERROR_INVALIDDATA;
355 if (avctx->channels < 0) {
356 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
358 return AVERROR_INVALIDDATA;
359 } else if (avctx->channels > WMAPRO_MAX_CHANNELS) {
360 avpriv_request_sample(avctx,
361 "More than %d channels", WMAPRO_MAX_CHANNELS);
362 return AVERROR_PATCHWELCOME;
365 /** init previous block len */
366 for (i = 0; i < avctx->channels; i++)
367 s->channel[i].prev_block_len = s->samples_per_frame;
369 /** extract lfe channel position */
372 if (channel_mask & 8) {
374 for (mask = 1; mask < 16; mask <<= 1) {
375 if (channel_mask & mask)
380 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
381 scale_huffbits, 1, 1,
382 scale_huffcodes, 2, 2, 616);
384 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
385 scale_rl_huffbits, 1, 1,
386 scale_rl_huffcodes, 4, 4, 1406);
388 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
389 coef0_huffbits, 1, 1,
390 coef0_huffcodes, 4, 4, 2108);
392 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
393 coef1_huffbits, 1, 1,
394 coef1_huffcodes, 4, 4, 3912);
396 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
398 vec4_huffcodes, 2, 2, 604);
400 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
402 vec2_huffcodes, 2, 2, 562);
404 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
406 vec1_huffcodes, 2, 2, 562);
408 /** calculate number of scale factor bands and their offsets
409 for every possible block size */
410 for (i = 0; i < num_possible_block_sizes; i++) {
411 int subframe_len = s->samples_per_frame >> i;
415 s->sfb_offsets[i][0] = 0;
417 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
418 int offset = (subframe_len * 2 * critical_freq[x])
419 / s->avctx->sample_rate + 2;
421 if (offset > s->sfb_offsets[i][band - 1])
422 s->sfb_offsets[i][band++] = offset;
424 s->sfb_offsets[i][band - 1] = subframe_len;
425 s->num_sfb[i] = band - 1;
426 if (s->num_sfb[i] <= 0) {
427 av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n");
428 return AVERROR_INVALIDDATA;
433 /** Scale factors can be shared between blocks of different size
434 as every block has a different scale factor band layout.
435 The matrix sf_offsets is needed to find the correct scale factor.
438 for (i = 0; i < num_possible_block_sizes; i++) {
440 for (b = 0; b < s->num_sfb[i]; b++) {
442 int offset = ((s->sfb_offsets[i][b]
443 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
444 for (x = 0; x < num_possible_block_sizes; x++) {
446 while (s->sfb_offsets[x][v + 1] << x < offset) {
448 av_assert0(v < MAX_BANDS);
450 s->sf_offsets[i][x][b] = v;
455 /** init MDCT, FIXME: only init needed sizes */
456 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
457 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
458 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
459 / (1 << (s->bits_per_sample - 1)));
461 /** init MDCT windows: simple sine window */
462 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
463 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
464 ff_init_ff_sine_windows(win_idx);
465 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
468 /** calculate subwoofer cutoff values */
469 for (i = 0; i < num_possible_block_sizes; i++) {
470 int block_size = s->samples_per_frame >> i;
471 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
472 / s->avctx->sample_rate;
473 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
476 /** calculate sine values for the decorrelation matrix */
477 for (i = 0; i < 33; i++)
478 sin64[i] = sin(i*M_PI / 64.0);
480 if (avctx->debug & FF_DEBUG_BITSTREAM)
483 avctx->channel_layout = channel_mask;
489 *@brief Decode the subframe length.
491 *@param offset sample offset in the frame
492 *@return decoded subframe length on success, < 0 in case of an error
494 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
496 int frame_len_shift = 0;
499 /** no need to read from the bitstream when only one length is possible */
500 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
501 return s->min_samples_per_subframe;
503 if (get_bits_left(&s->gb) < 1)
504 return AVERROR_INVALIDDATA;
506 /** 1 bit indicates if the subframe is of maximum length */
507 if (s->max_subframe_len_bit) {
508 if (get_bits1(&s->gb))
509 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
511 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
513 subframe_len = s->samples_per_frame >> frame_len_shift;
515 /** sanity check the length */
516 if (subframe_len < s->min_samples_per_subframe ||
517 subframe_len > s->samples_per_frame) {
518 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
520 return AVERROR_INVALIDDATA;
526 *@brief Decode how the data in the frame is split into subframes.
527 * Every WMA frame contains the encoded data for a fixed number of
528 * samples per channel. The data for every channel might be split
529 * into several subframes. This function will reconstruct the list of
530 * subframes for every channel.
532 * If the subframes are not evenly split, the algorithm estimates the
533 * channels with the lowest number of total samples.
534 * Afterwards, for each of these channels a bit is read from the
535 * bitstream that indicates if the channel contains a subframe with the
536 * next subframe size that is going to be read from the bitstream or not.
537 * If a channel contains such a subframe, the subframe size gets added to
538 * the channel's subframe list.
539 * The algorithm repeats these steps until the frame is properly divided
540 * between the individual channels.
543 *@return 0 on success, < 0 in case of an error
545 static int decode_tilehdr(WMAProDecodeCtx *s)
547 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
548 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
549 int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */
550 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
551 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
554 /* Should never consume more than 3073 bits (256 iterations for the
555 * while loop when always the minimum amount of 128 samples is subtracted
556 * from missing samples in the 8 channel case).
557 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
560 /** reset tiling information */
561 for (c = 0; c < s->avctx->channels; c++)
562 s->channel[c].num_subframes = 0;
564 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
565 fixed_channel_layout = 1;
567 /** loop until the frame data is split between the subframes */
571 /** check which channels contain the subframe */
572 for (c = 0; c < s->avctx->channels; c++) {
573 if (num_samples[c] == min_channel_len) {
574 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
575 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
576 contains_subframe[c] = 1;
578 contains_subframe[c] = get_bits1(&s->gb);
580 contains_subframe[c] = 0;
583 /** get subframe length, subframe_len == 0 is not allowed */
584 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
585 return AVERROR_INVALIDDATA;
587 /** add subframes to the individual channels and find new min_channel_len */
588 min_channel_len += subframe_len;
589 for (c = 0; c < s->avctx->channels; c++) {
590 WMAProChannelCtx* chan = &s->channel[c];
592 if (contains_subframe[c]) {
593 if (chan->num_subframes >= MAX_SUBFRAMES) {
594 av_log(s->avctx, AV_LOG_ERROR,
595 "broken frame: num subframes > 31\n");
596 return AVERROR_INVALIDDATA;
598 chan->subframe_len[chan->num_subframes] = subframe_len;
599 num_samples[c] += subframe_len;
600 ++chan->num_subframes;
601 if (num_samples[c] > s->samples_per_frame) {
602 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
603 "channel len > samples_per_frame\n");
604 return AVERROR_INVALIDDATA;
606 } else if (num_samples[c] <= min_channel_len) {
607 if (num_samples[c] < min_channel_len) {
608 channels_for_cur_subframe = 0;
609 min_channel_len = num_samples[c];
611 ++channels_for_cur_subframe;
614 } while (min_channel_len < s->samples_per_frame);
616 for (c = 0; c < s->avctx->channels; c++) {
619 for (i = 0; i < s->channel[c].num_subframes; i++) {
620 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
621 " len %i\n", s->frame_num, c, i,
622 s->channel[c].subframe_len[i]);
623 s->channel[c].subframe_offset[i] = offset;
624 offset += s->channel[c].subframe_len[i];
632 *@brief Calculate a decorrelation matrix from the bitstream parameters.
633 *@param s codec context
634 *@param chgroup channel group for which the matrix needs to be calculated
636 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
637 WMAProChannelGrp *chgroup)
641 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
642 memset(chgroup->decorrelation_matrix, 0, s->avctx->channels *
643 s->avctx->channels * sizeof(*chgroup->decorrelation_matrix));
645 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
646 rotation_offset[i] = get_bits(&s->gb, 6);
648 for (i = 0; i < chgroup->num_channels; i++)
649 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
650 get_bits1(&s->gb) ? 1.0 : -1.0;
652 for (i = 1; i < chgroup->num_channels; i++) {
654 for (x = 0; x < i; x++) {
656 for (y = 0; y < i + 1; y++) {
657 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
658 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
659 int n = rotation_offset[offset + x];
665 cosv = sin64[32 - n];
667 sinv = sin64[64 - n];
668 cosv = -sin64[n - 32];
671 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
672 (v1 * sinv) - (v2 * cosv);
673 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
674 (v1 * cosv) + (v2 * sinv);
682 *@brief Decode channel transformation parameters
683 *@param s codec context
684 *@return >= 0 in case of success, < 0 in case of bitstream errors
686 static int decode_channel_transform(WMAProDecodeCtx* s)
689 /* should never consume more than 1921 bits for the 8 channel case
690 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
691 * + MAX_CHANNELS + MAX_BANDS + 1)
694 /** in the one channel case channel transforms are pointless */
696 if (s->avctx->channels > 1) {
697 int remaining_channels = s->channels_for_cur_subframe;
699 if (get_bits1(&s->gb)) {
700 avpriv_request_sample(s->avctx,
701 "Channel transform bit");
702 return AVERROR_PATCHWELCOME;
705 for (s->num_chgroups = 0; remaining_channels &&
706 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
707 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
708 float** channel_data = chgroup->channel_data;
709 chgroup->num_channels = 0;
710 chgroup->transform = 0;
712 /** decode channel mask */
713 if (remaining_channels > 2) {
714 for (i = 0; i < s->channels_for_cur_subframe; i++) {
715 int channel_idx = s->channel_indexes_for_cur_subframe[i];
716 if (!s->channel[channel_idx].grouped
717 && get_bits1(&s->gb)) {
718 ++chgroup->num_channels;
719 s->channel[channel_idx].grouped = 1;
720 *channel_data++ = s->channel[channel_idx].coeffs;
724 chgroup->num_channels = remaining_channels;
725 for (i = 0; i < s->channels_for_cur_subframe; i++) {
726 int channel_idx = s->channel_indexes_for_cur_subframe[i];
727 if (!s->channel[channel_idx].grouped)
728 *channel_data++ = s->channel[channel_idx].coeffs;
729 s->channel[channel_idx].grouped = 1;
733 /** decode transform type */
734 if (chgroup->num_channels == 2) {
735 if (get_bits1(&s->gb)) {
736 if (get_bits1(&s->gb)) {
737 avpriv_request_sample(s->avctx,
738 "Unknown channel transform type");
739 return AVERROR_PATCHWELCOME;
742 chgroup->transform = 1;
743 if (s->avctx->channels == 2) {
744 chgroup->decorrelation_matrix[0] = 1.0;
745 chgroup->decorrelation_matrix[1] = -1.0;
746 chgroup->decorrelation_matrix[2] = 1.0;
747 chgroup->decorrelation_matrix[3] = 1.0;
750 chgroup->decorrelation_matrix[0] = 0.70703125;
751 chgroup->decorrelation_matrix[1] = -0.70703125;
752 chgroup->decorrelation_matrix[2] = 0.70703125;
753 chgroup->decorrelation_matrix[3] = 0.70703125;
756 } else if (chgroup->num_channels > 2) {
757 if (get_bits1(&s->gb)) {
758 chgroup->transform = 1;
759 if (get_bits1(&s->gb)) {
760 decode_decorrelation_matrix(s, chgroup);
762 /** FIXME: more than 6 coupled channels not supported */
763 if (chgroup->num_channels > 6) {
764 avpriv_request_sample(s->avctx,
765 "Coupled channels > 6");
767 memcpy(chgroup->decorrelation_matrix,
768 default_decorrelation[chgroup->num_channels],
769 chgroup->num_channels * chgroup->num_channels *
770 sizeof(*chgroup->decorrelation_matrix));
776 /** decode transform on / off */
777 if (chgroup->transform) {
778 if (!get_bits1(&s->gb)) {
780 /** transform can be enabled for individual bands */
781 for (i = 0; i < s->num_bands; i++) {
782 chgroup->transform_band[i] = get_bits1(&s->gb);
785 memset(chgroup->transform_band, 1, s->num_bands);
788 remaining_channels -= chgroup->num_channels;
795 *@brief Extract the coefficients from the bitstream.
796 *@param s codec context
797 *@param c current channel number
798 *@return 0 on success, < 0 in case of bitstream errors
800 static int decode_coeffs(WMAProDecodeCtx *s, int c)
802 /* Integers 0..15 as single-precision floats. The table saves a
803 costly int to float conversion, and storing the values as
804 integers allows fast sign-flipping. */
805 static const uint32_t fval_tab[16] = {
806 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
807 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
808 0x41000000, 0x41100000, 0x41200000, 0x41300000,
809 0x41400000, 0x41500000, 0x41600000, 0x41700000,
813 WMAProChannelCtx* ci = &s->channel[c];
820 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
822 vlctable = get_bits1(&s->gb);
823 vlc = &coef_vlc[vlctable];
833 /** decode vector coefficients (consumes up to 167 bits per iteration for
834 4 vector coded large values) */
835 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
836 (cur_coeff + 3 < ci->num_vec_coeffs)) {
841 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
843 if (idx == HUFF_VEC4_SIZE - 1) {
844 for (i = 0; i < 4; i += 2) {
845 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
846 if (idx == HUFF_VEC2_SIZE - 1) {
848 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
849 if (v0 == HUFF_VEC1_SIZE - 1)
850 v0 += ff_wma_get_large_val(&s->gb);
851 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
852 if (v1 == HUFF_VEC1_SIZE - 1)
853 v1 += ff_wma_get_large_val(&s->gb);
854 vals[i ] = av_float2int(v0);
855 vals[i+1] = av_float2int(v1);
857 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
858 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
862 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
863 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
864 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
865 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
869 for (i = 0; i < 4; i++) {
871 uint32_t sign = get_bits1(&s->gb) - 1;
872 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
875 ci->coeffs[cur_coeff] = 0;
876 /** switch to run level mode when subframe_len / 128 zeros
877 were found in a row */
878 rl_mode |= (++num_zeros > s->subframe_len >> 8);
884 /** decode run level coded coefficients */
885 if (cur_coeff < s->subframe_len) {
886 memset(&ci->coeffs[cur_coeff], 0,
887 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
888 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
889 level, run, 1, ci->coeffs,
890 cur_coeff, s->subframe_len,
891 s->subframe_len, s->esc_len, 0))
892 return AVERROR_INVALIDDATA;
899 *@brief Extract scale factors from the bitstream.
900 *@param s codec context
901 *@return 0 on success, < 0 in case of bitstream errors
903 static int decode_scale_factors(WMAProDecodeCtx* s)
907 /** should never consume more than 5344 bits
908 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
911 for (i = 0; i < s->channels_for_cur_subframe; i++) {
912 int c = s->channel_indexes_for_cur_subframe[i];
915 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
916 sf_end = s->channel[c].scale_factors + s->num_bands;
918 /** resample scale factors for the new block size
919 * as the scale factors might need to be resampled several times
920 * before some new values are transmitted, a backup of the last
921 * transmitted scale factors is kept in saved_scale_factors
923 if (s->channel[c].reuse_sf) {
924 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
926 for (b = 0; b < s->num_bands; b++)
927 s->channel[c].scale_factors[b] =
928 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
931 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
933 if (!s->channel[c].reuse_sf) {
935 /** decode DPCM coded scale factors */
936 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
937 val = 45 / s->channel[c].scale_factor_step;
938 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
939 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
944 /** run level decode differences to the resampled factors */
945 for (i = 0; i < s->num_bands; i++) {
951 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
954 uint32_t code = get_bits(&s->gb, 14);
956 sign = (code & 1) - 1;
957 skip = (code & 0x3f) >> 1;
958 } else if (idx == 1) {
961 skip = scale_rl_run[idx];
962 val = scale_rl_level[idx];
963 sign = get_bits1(&s->gb)-1;
967 if (i >= s->num_bands) {
968 av_log(s->avctx, AV_LOG_ERROR,
969 "invalid scale factor coding\n");
970 return AVERROR_INVALIDDATA;
972 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
976 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
977 s->channel[c].table_idx = s->table_idx;
978 s->channel[c].reuse_sf = 1;
981 /** calculate new scale factor maximum */
982 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
983 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
984 s->channel[c].max_scale_factor =
985 FFMAX(s->channel[c].max_scale_factor, *sf);
993 *@brief Reconstruct the individual channel data.
994 *@param s codec context
996 static void inverse_channel_transform(WMAProDecodeCtx *s)
1000 for (i = 0; i < s->num_chgroups; i++) {
1001 if (s->chgroup[i].transform) {
1002 float data[WMAPRO_MAX_CHANNELS];
1003 const int num_channels = s->chgroup[i].num_channels;
1004 float** ch_data = s->chgroup[i].channel_data;
1005 float** ch_end = ch_data + num_channels;
1006 const int8_t* tb = s->chgroup[i].transform_band;
1009 /** multichannel decorrelation */
1010 for (sfb = s->cur_sfb_offsets;
1011 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1014 /** multiply values with the decorrelation_matrix */
1015 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1016 const float* mat = s->chgroup[i].decorrelation_matrix;
1017 const float* data_end = data + num_channels;
1018 float* data_ptr = data;
1021 for (ch = ch_data; ch < ch_end; ch++)
1022 *data_ptr++ = (*ch)[y];
1024 for (ch = ch_data; ch < ch_end; ch++) {
1027 while (data_ptr < data_end)
1028 sum += *data_ptr++ * *mat++;
1033 } else if (s->avctx->channels == 2) {
1034 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1035 s->fdsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1036 ch_data[0] + sfb[0],
1038 s->fdsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1039 ch_data[1] + sfb[0],
1048 *@brief Apply sine window and reconstruct the output buffer.
1049 *@param s codec context
1051 static void wmapro_window(WMAProDecodeCtx *s)
1054 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1055 int c = s->channel_indexes_for_cur_subframe[i];
1057 int winlen = s->channel[c].prev_block_len;
1058 float* start = s->channel[c].coeffs - (winlen >> 1);
1060 if (s->subframe_len < winlen) {
1061 start += (winlen - s->subframe_len) >> 1;
1062 winlen = s->subframe_len;
1065 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1069 s->fdsp.vector_fmul_window(start, start, start + winlen,
1072 s->channel[c].prev_block_len = s->subframe_len;
1077 *@brief Decode a single subframe (block).
1078 *@param s codec context
1079 *@return 0 on success, < 0 when decoding failed
1081 static int decode_subframe(WMAProDecodeCtx *s)
1083 int offset = s->samples_per_frame;
1084 int subframe_len = s->samples_per_frame;
1086 int total_samples = s->samples_per_frame * s->avctx->channels;
1087 int transmit_coeffs = 0;
1088 int cur_subwoofer_cutoff;
1090 s->subframe_offset = get_bits_count(&s->gb);
1092 /** reset channel context and find the next block offset and size
1093 == the next block of the channel with the smallest number of
1096 for (i = 0; i < s->avctx->channels; i++) {
1097 s->channel[i].grouped = 0;
1098 if (offset > s->channel[i].decoded_samples) {
1099 offset = s->channel[i].decoded_samples;
1101 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1106 "processing subframe with offset %i len %i\n", offset, subframe_len);
1108 /** get a list of all channels that contain the estimated block */
1109 s->channels_for_cur_subframe = 0;
1110 for (i = 0; i < s->avctx->channels; i++) {
1111 const int cur_subframe = s->channel[i].cur_subframe;
1112 /** subtract already processed samples */
1113 total_samples -= s->channel[i].decoded_samples;
1115 /** and count if there are multiple subframes that match our profile */
1116 if (offset == s->channel[i].decoded_samples &&
1117 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1118 total_samples -= s->channel[i].subframe_len[cur_subframe];
1119 s->channel[i].decoded_samples +=
1120 s->channel[i].subframe_len[cur_subframe];
1121 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1122 ++s->channels_for_cur_subframe;
1126 /** check if the frame will be complete after processing the
1129 s->parsed_all_subframes = 1;
1132 av_dlog(s->avctx, "subframe is part of %i channels\n",
1133 s->channels_for_cur_subframe);
1135 /** calculate number of scale factor bands and their offsets */
1136 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1137 s->num_bands = s->num_sfb[s->table_idx];
1138 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1139 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1141 /** configure the decoder for the current subframe */
1142 offset += s->samples_per_frame >> 1;
1144 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1145 int c = s->channel_indexes_for_cur_subframe[i];
1147 s->channel[c].coeffs = &s->channel[c].out[offset];
1150 s->subframe_len = subframe_len;
1151 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1153 /** skip extended header if any */
1154 if (get_bits1(&s->gb)) {
1156 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1157 int len = get_bits(&s->gb, 4);
1158 num_fill_bits = (len ? get_bits(&s->gb, len) : 0) + 1;
1161 if (num_fill_bits >= 0) {
1162 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1163 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1164 return AVERROR_INVALIDDATA;
1167 skip_bits_long(&s->gb, num_fill_bits);
1171 /** no idea for what the following bit is used */
1172 if (get_bits1(&s->gb)) {
1173 avpriv_request_sample(s->avctx, "Reserved bit");
1174 return AVERROR_PATCHWELCOME;
1178 if (decode_channel_transform(s) < 0)
1179 return AVERROR_INVALIDDATA;
1182 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1183 int c = s->channel_indexes_for_cur_subframe[i];
1184 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1185 transmit_coeffs = 1;
1188 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1189 if (transmit_coeffs) {
1191 int quant_step = 90 * s->bits_per_sample >> 4;
1193 /** decode number of vector coded coefficients */
1194 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1195 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1196 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1197 int c = s->channel_indexes_for_cur_subframe[i];
1198 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1199 if (num_vec_coeffs > s->subframe_len) {
1200 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1201 return AVERROR_INVALIDDATA;
1203 av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out));
1204 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1207 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1208 int c = s->channel_indexes_for_cur_subframe[i];
1209 s->channel[c].num_vec_coeffs = s->subframe_len;
1212 /** decode quantization step */
1213 step = get_sbits(&s->gb, 6);
1215 if (step == -32 || step == 31) {
1216 const int sign = (step == 31) - 1;
1218 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1219 (step = get_bits(&s->gb, 5)) == 31) {
1222 quant_step += ((quant + step) ^ sign) - sign;
1224 if (quant_step < 0) {
1225 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1228 /** decode quantization step modifiers for every channel */
1230 if (s->channels_for_cur_subframe == 1) {
1231 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1233 int modifier_len = get_bits(&s->gb, 3);
1234 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1235 int c = s->channel_indexes_for_cur_subframe[i];
1236 s->channel[c].quant_step = quant_step;
1237 if (get_bits1(&s->gb)) {
1239 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1241 ++s->channel[c].quant_step;
1246 /** decode scale factors */
1247 if (decode_scale_factors(s) < 0)
1248 return AVERROR_INVALIDDATA;
1251 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1252 get_bits_count(&s->gb) - s->subframe_offset);
1254 /** parse coefficients */
1255 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1256 int c = s->channel_indexes_for_cur_subframe[i];
1257 if (s->channel[c].transmit_coefs &&
1258 get_bits_count(&s->gb) < s->num_saved_bits) {
1259 decode_coeffs(s, c);
1261 memset(s->channel[c].coeffs, 0,
1262 sizeof(*s->channel[c].coeffs) * subframe_len);
1265 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1266 get_bits_count(&s->gb) - s->subframe_offset);
1268 if (transmit_coeffs) {
1269 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1270 /** reconstruct the per channel data */
1271 inverse_channel_transform(s);
1272 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1273 int c = s->channel_indexes_for_cur_subframe[i];
1274 const int* sf = s->channel[c].scale_factors;
1277 if (c == s->lfe_channel)
1278 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1279 (subframe_len - cur_subwoofer_cutoff));
1281 /** inverse quantization and rescaling */
1282 for (b = 0; b < s->num_bands; b++) {
1283 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1284 const int exp = s->channel[c].quant_step -
1285 (s->channel[c].max_scale_factor - *sf++) *
1286 s->channel[c].scale_factor_step;
1287 const float quant = pow(10.0, exp / 20.0);
1288 int start = s->cur_sfb_offsets[b];
1289 s->fdsp.vector_fmul_scalar(s->tmp + start,
1290 s->channel[c].coeffs + start,
1291 quant, end - start);
1294 /** apply imdct (imdct_half == DCTIV with reverse) */
1295 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1299 /** window and overlapp-add */
1302 /** handled one subframe */
1303 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1304 int c = s->channel_indexes_for_cur_subframe[i];
1305 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1306 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1307 return AVERROR_INVALIDDATA;
1309 ++s->channel[c].cur_subframe;
1316 *@brief Decode one WMA frame.
1317 *@param s codec context
1318 *@return 0 if the trailer bit indicates that this is the last frame,
1319 * 1 if there are additional frames
1321 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1323 AVCodecContext *avctx = s->avctx;
1324 GetBitContext* gb = &s->gb;
1325 int more_frames = 0;
1329 /** get frame length */
1331 len = get_bits(gb, s->log2_frame_size);
1333 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1335 /** decode tile information */
1336 if (decode_tilehdr(s)) {
1341 /** read postproc transform */
1342 if (s->avctx->channels > 1 && get_bits1(gb)) {
1343 if (get_bits1(gb)) {
1344 for (i = 0; i < avctx->channels * avctx->channels; i++)
1349 /** read drc info */
1350 if (s->dynamic_range_compression) {
1351 s->drc_gain = get_bits(gb, 8);
1352 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1355 /** no idea what these are for, might be the number of samples
1356 that need to be skipped at the beginning or end of a stream */
1357 if (get_bits1(gb)) {
1360 /** usually true for the first frame */
1361 if (get_bits1(gb)) {
1362 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1363 av_dlog(s->avctx, "start skip: %i\n", skip);
1366 /** sometimes true for the last frame */
1367 if (get_bits1(gb)) {
1368 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1369 av_dlog(s->avctx, "end skip: %i\n", skip);
1374 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1375 get_bits_count(gb) - s->frame_offset);
1377 /** reset subframe states */
1378 s->parsed_all_subframes = 0;
1379 for (i = 0; i < avctx->channels; i++) {
1380 s->channel[i].decoded_samples = 0;
1381 s->channel[i].cur_subframe = 0;
1382 s->channel[i].reuse_sf = 0;
1385 /** decode all subframes */
1386 while (!s->parsed_all_subframes) {
1387 if (decode_subframe(s) < 0) {
1393 /* get output buffer */
1394 frame->nb_samples = s->samples_per_frame;
1395 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1400 /** copy samples to the output buffer */
1401 for (i = 0; i < avctx->channels; i++)
1402 memcpy(frame->extended_data[i], s->channel[i].out,
1403 s->samples_per_frame * sizeof(*s->channel[i].out));
1405 for (i = 0; i < avctx->channels; i++) {
1406 /** reuse second half of the IMDCT output for the next frame */
1407 memcpy(&s->channel[i].out[0],
1408 &s->channel[i].out[s->samples_per_frame],
1409 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1412 if (s->skip_frame) {
1415 av_frame_unref(frame);
1420 if (s->len_prefix) {
1421 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1422 /** FIXME: not sure if this is always an error */
1423 av_log(s->avctx, AV_LOG_ERROR,
1424 "frame[%i] would have to skip %i bits\n", s->frame_num,
1425 len - (get_bits_count(gb) - s->frame_offset) - 1);
1430 /** skip the rest of the frame data */
1431 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1433 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1437 /** decode trailer bit */
1438 more_frames = get_bits1(gb);
1445 *@brief Calculate remaining input buffer length.
1446 *@param s codec context
1447 *@param gb bitstream reader context
1448 *@return remaining size in bits
1450 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1452 return s->buf_bit_size - get_bits_count(gb);
1456 *@brief Fill the bit reservoir with a (partial) frame.
1457 *@param s codec context
1458 *@param gb bitstream reader context
1459 *@param len length of the partial frame
1460 *@param append decides whether to reset the buffer or not
1462 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1467 /** when the frame data does not need to be concatenated, the input buffer
1468 is reset and additional bits from the previous frame are copied
1469 and skipped later so that a fast byte copy is possible */
1472 s->frame_offset = get_bits_count(gb) & 7;
1473 s->num_saved_bits = s->frame_offset;
1474 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1477 buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1479 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1480 avpriv_request_sample(s->avctx, "Too small input buffer");
1485 av_assert0(len <= put_bits_left(&s->pb));
1487 s->num_saved_bits += len;
1489 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1492 int align = 8 - (get_bits_count(gb) & 7);
1493 align = FFMIN(align, len);
1494 put_bits(&s->pb, align, get_bits(gb, align));
1496 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1498 skip_bits_long(gb, len);
1501 PutBitContext tmp = s->pb;
1502 flush_put_bits(&tmp);
1505 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1506 skip_bits(&s->gb, s->frame_offset);
1510 *@brief Decode a single WMA packet.
1511 *@param avctx codec context
1512 *@param data the output buffer
1513 *@param avpkt input packet
1514 *@return number of bytes that were read from the input buffer
1516 static int decode_packet(AVCodecContext *avctx, void *data,
1517 int *got_frame_ptr, AVPacket* avpkt)
1519 WMAProDecodeCtx *s = avctx->priv_data;
1520 GetBitContext* gb = &s->pgb;
1521 const uint8_t* buf = avpkt->data;
1522 int buf_size = avpkt->size;
1523 int num_bits_prev_frame;
1524 int packet_sequence_number;
1528 if (s->packet_done || s->packet_loss) {
1531 /** sanity check for the buffer length */
1532 if (buf_size < avctx->block_align) {
1533 av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1534 buf_size, avctx->block_align);
1535 return AVERROR_INVALIDDATA;
1538 s->next_packet_start = buf_size - avctx->block_align;
1539 buf_size = avctx->block_align;
1540 s->buf_bit_size = buf_size << 3;
1542 /** parse packet header */
1543 init_get_bits(gb, buf, s->buf_bit_size);
1544 packet_sequence_number = get_bits(gb, 4);
1547 /** get number of bits that need to be added to the previous frame */
1548 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1549 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1550 num_bits_prev_frame);
1552 /** check for packet loss */
1553 if (!s->packet_loss &&
1554 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1556 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1557 s->packet_sequence_number, packet_sequence_number);
1559 s->packet_sequence_number = packet_sequence_number;
1561 if (num_bits_prev_frame > 0) {
1562 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1563 if (num_bits_prev_frame >= remaining_packet_bits) {
1564 num_bits_prev_frame = remaining_packet_bits;
1568 /** append the previous frame data to the remaining data from the
1569 previous packet to create a full frame */
1570 save_bits(s, gb, num_bits_prev_frame, 1);
1571 av_dlog(avctx, "accumulated %x bits of frame data\n",
1572 s->num_saved_bits - s->frame_offset);
1574 /** decode the cross packet frame if it is valid */
1575 if (!s->packet_loss)
1576 decode_frame(s, data, got_frame_ptr);
1577 } else if (s->num_saved_bits - s->frame_offset) {
1578 av_dlog(avctx, "ignoring %x previously saved bits\n",
1579 s->num_saved_bits - s->frame_offset);
1582 if (s->packet_loss) {
1583 /** reset number of saved bits so that the decoder
1584 does not start to decode incomplete frames in the
1585 s->len_prefix == 0 case */
1586 s->num_saved_bits = 0;
1592 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1593 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1594 skip_bits(gb, s->packet_offset);
1595 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1596 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1597 frame_size <= remaining_bits(s, gb)) {
1598 save_bits(s, gb, frame_size, 0);
1599 if (!s->packet_loss)
1600 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1601 } else if (!s->len_prefix
1602 && s->num_saved_bits > get_bits_count(&s->gb)) {
1603 /** when the frames do not have a length prefix, we don't know
1604 the compressed length of the individual frames
1605 however, we know what part of a new packet belongs to the
1607 therefore we save the incoming packet first, then we append
1608 the "previous frame" data from the next packet so that
1609 we get a buffer that only contains full frames */
1610 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1615 if (s->packet_done && !s->packet_loss &&
1616 remaining_bits(s, gb) > 0) {
1617 /** save the rest of the data so that it can be decoded
1618 with the next packet */
1619 save_bits(s, gb, remaining_bits(s, gb), 0);
1622 s->packet_offset = get_bits_count(gb) & 7;
1624 return AVERROR_INVALIDDATA;
1626 return get_bits_count(gb) >> 3;
1630 *@brief Clear decoder buffers (for seeking).
1631 *@param avctx codec context
1633 static void flush(AVCodecContext *avctx)
1635 WMAProDecodeCtx *s = avctx->priv_data;
1637 /** reset output buffer as a part of it is used during the windowing of a
1639 for (i = 0; i < avctx->channels; i++)
1640 memset(s->channel[i].out, 0, s->samples_per_frame *
1641 sizeof(*s->channel[i].out));
1647 *@brief wmapro decoder
1649 AVCodec ff_wmapro_decoder = {
1651 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1652 .type = AVMEDIA_TYPE_AUDIO,
1653 .id = AV_CODEC_ID_WMAPRO,
1654 .priv_data_size = sizeof(WMAProDecodeCtx),
1655 .init = decode_init,
1656 .close = decode_end,
1657 .decode = decode_packet,
1658 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1660 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1661 AV_SAMPLE_FMT_NONE },