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 if (s->avctx->sample_rate <= 0) {
345 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
346 return AVERROR_INVALIDDATA;
349 s->num_channels = avctx->channels;
351 if (s->num_channels < 0) {
352 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
353 return AVERROR_INVALIDDATA;
354 } else if (s->num_channels > WMAPRO_MAX_CHANNELS) {
355 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
356 return AVERROR_PATCHWELCOME;
359 /** init previous block len */
360 for (i = 0; i < s->num_channels; i++)
361 s->channel[i].prev_block_len = s->samples_per_frame;
363 /** extract lfe channel position */
366 if (channel_mask & 8) {
368 for (mask = 1; mask < 16; mask <<= 1) {
369 if (channel_mask & mask)
374 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
375 scale_huffbits, 1, 1,
376 scale_huffcodes, 2, 2, 616);
378 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
379 scale_rl_huffbits, 1, 1,
380 scale_rl_huffcodes, 4, 4, 1406);
382 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
383 coef0_huffbits, 1, 1,
384 coef0_huffcodes, 4, 4, 2108);
386 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
387 coef1_huffbits, 1, 1,
388 coef1_huffcodes, 4, 4, 3912);
390 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
392 vec4_huffcodes, 2, 2, 604);
394 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
396 vec2_huffcodes, 2, 2, 562);
398 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
400 vec1_huffcodes, 2, 2, 562);
402 /** calculate number of scale factor bands and their offsets
403 for every possible block size */
404 for (i = 0; i < num_possible_block_sizes; i++) {
405 int subframe_len = s->samples_per_frame >> i;
409 s->sfb_offsets[i][0] = 0;
411 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
412 int offset = (subframe_len * 2 * critical_freq[x])
413 / s->avctx->sample_rate + 2;
415 if (offset > s->sfb_offsets[i][band - 1])
416 s->sfb_offsets[i][band++] = offset;
418 s->sfb_offsets[i][band - 1] = subframe_len;
419 s->num_sfb[i] = band - 1;
423 /** Scale factors can be shared between blocks of different size
424 as every block has a different scale factor band layout.
425 The matrix sf_offsets is needed to find the correct scale factor.
428 for (i = 0; i < num_possible_block_sizes; i++) {
430 for (b = 0; b < s->num_sfb[i]; b++) {
432 int offset = ((s->sfb_offsets[i][b]
433 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
434 for (x = 0; x < num_possible_block_sizes; x++) {
436 while (s->sfb_offsets[x][v + 1] << x < offset)
438 s->sf_offsets[i][x][b] = v;
443 /** init MDCT, FIXME: only init needed sizes */
444 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
445 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
446 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
447 / (1 << (s->bits_per_sample - 1)));
449 /** init MDCT windows: simple sinus window */
450 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
451 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
452 ff_init_ff_sine_windows(win_idx);
453 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
456 /** calculate subwoofer cutoff values */
457 for (i = 0; i < num_possible_block_sizes; i++) {
458 int block_size = s->samples_per_frame >> i;
459 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
460 / s->avctx->sample_rate;
461 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
464 /** calculate sine values for the decorrelation matrix */
465 for (i = 0; i < 33; i++)
466 sin64[i] = sin(i*M_PI / 64.0);
468 if (avctx->debug & FF_DEBUG_BITSTREAM)
471 avctx->channel_layout = channel_mask;
473 avcodec_get_frame_defaults(&s->frame);
474 avctx->coded_frame = &s->frame;
480 *@brief Decode the subframe length.
482 *@param offset sample offset in the frame
483 *@return decoded subframe length on success, < 0 in case of an error
485 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
487 int frame_len_shift = 0;
490 /** no need to read from the bitstream when only one length is possible */
491 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
492 return s->min_samples_per_subframe;
494 /** 1 bit indicates if the subframe is of maximum length */
495 if (s->max_subframe_len_bit) {
496 if (get_bits1(&s->gb))
497 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
499 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
501 subframe_len = s->samples_per_frame >> frame_len_shift;
503 /** sanity check the length */
504 if (subframe_len < s->min_samples_per_subframe ||
505 subframe_len > s->samples_per_frame) {
506 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
508 return AVERROR_INVALIDDATA;
514 *@brief Decode how the data in the frame is split into subframes.
515 * Every WMA frame contains the encoded data for a fixed number of
516 * samples per channel. The data for every channel might be split
517 * into several subframes. This function will reconstruct the list of
518 * subframes for every channel.
520 * If the subframes are not evenly split, the algorithm estimates the
521 * channels with the lowest number of total samples.
522 * Afterwards, for each of these channels a bit is read from the
523 * bitstream that indicates if the channel contains a subframe with the
524 * next subframe size that is going to be read from the bitstream or not.
525 * If a channel contains such a subframe, the subframe size gets added to
526 * the channel's subframe list.
527 * The algorithm repeats these steps until the frame is properly divided
528 * between the individual channels.
531 *@return 0 on success, < 0 in case of an error
533 static int decode_tilehdr(WMAProDecodeCtx *s)
535 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
536 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
537 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
538 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
539 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
542 /* Should never consume more than 3073 bits (256 iterations for the
543 * while loop when always the minimum amount of 128 samples is subtracted
544 * from missing samples in the 8 channel case).
545 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
548 /** reset tiling information */
549 for (c = 0; c < s->num_channels; c++)
550 s->channel[c].num_subframes = 0;
552 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
553 fixed_channel_layout = 1;
555 /** loop until the frame data is split between the subframes */
559 /** check which channels contain the subframe */
560 for (c = 0; c < s->num_channels; c++) {
561 if (num_samples[c] == min_channel_len) {
562 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
563 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
564 contains_subframe[c] = 1;
566 contains_subframe[c] = get_bits1(&s->gb);
568 contains_subframe[c] = 0;
571 /** get subframe length, subframe_len == 0 is not allowed */
572 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
573 return AVERROR_INVALIDDATA;
575 /** add subframes to the individual channels and find new min_channel_len */
576 min_channel_len += subframe_len;
577 for (c = 0; c < s->num_channels; c++) {
578 WMAProChannelCtx* chan = &s->channel[c];
580 if (contains_subframe[c]) {
581 if (chan->num_subframes >= MAX_SUBFRAMES) {
582 av_log(s->avctx, AV_LOG_ERROR,
583 "broken frame: num subframes > 31\n");
584 return AVERROR_INVALIDDATA;
586 chan->subframe_len[chan->num_subframes] = subframe_len;
587 num_samples[c] += subframe_len;
588 ++chan->num_subframes;
589 if (num_samples[c] > s->samples_per_frame) {
590 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
591 "channel len > samples_per_frame\n");
592 return AVERROR_INVALIDDATA;
594 } else if (num_samples[c] <= min_channel_len) {
595 if (num_samples[c] < min_channel_len) {
596 channels_for_cur_subframe = 0;
597 min_channel_len = num_samples[c];
599 ++channels_for_cur_subframe;
602 } while (min_channel_len < s->samples_per_frame);
604 for (c = 0; c < s->num_channels; c++) {
607 for (i = 0; i < s->channel[c].num_subframes; i++) {
608 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
609 " len %i\n", s->frame_num, c, i,
610 s->channel[c].subframe_len[i]);
611 s->channel[c].subframe_offset[i] = offset;
612 offset += s->channel[c].subframe_len[i];
620 *@brief Calculate a decorrelation matrix from the bitstream parameters.
621 *@param s codec context
622 *@param chgroup channel group for which the matrix needs to be calculated
624 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
625 WMAProChannelGrp *chgroup)
629 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
630 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
631 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
633 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
634 rotation_offset[i] = get_bits(&s->gb, 6);
636 for (i = 0; i < chgroup->num_channels; i++)
637 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
638 get_bits1(&s->gb) ? 1.0 : -1.0;
640 for (i = 1; i < chgroup->num_channels; i++) {
642 for (x = 0; x < i; x++) {
644 for (y = 0; y < i + 1; y++) {
645 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
646 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
647 int n = rotation_offset[offset + x];
653 cosv = sin64[32 - n];
655 sinv = sin64[64 - n];
656 cosv = -sin64[n - 32];
659 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
660 (v1 * sinv) - (v2 * cosv);
661 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
662 (v1 * cosv) + (v2 * sinv);
670 *@brief Decode channel transformation parameters
671 *@param s codec context
672 *@return 0 in case of success, < 0 in case of bitstream errors
674 static int decode_channel_transform(WMAProDecodeCtx* s)
677 /* should never consume more than 1921 bits for the 8 channel case
678 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
679 * + MAX_CHANNELS + MAX_BANDS + 1)
682 /** in the one channel case channel transforms are pointless */
684 if (s->num_channels > 1) {
685 int remaining_channels = s->channels_for_cur_subframe;
687 if (get_bits1(&s->gb)) {
688 av_log_ask_for_sample(s->avctx,
689 "unsupported channel transform bit\n");
690 return AVERROR_INVALIDDATA;
693 for (s->num_chgroups = 0; remaining_channels &&
694 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
695 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
696 float** channel_data = chgroup->channel_data;
697 chgroup->num_channels = 0;
698 chgroup->transform = 0;
700 /** decode channel mask */
701 if (remaining_channels > 2) {
702 for (i = 0; i < s->channels_for_cur_subframe; i++) {
703 int channel_idx = s->channel_indexes_for_cur_subframe[i];
704 if (!s->channel[channel_idx].grouped
705 && get_bits1(&s->gb)) {
706 ++chgroup->num_channels;
707 s->channel[channel_idx].grouped = 1;
708 *channel_data++ = s->channel[channel_idx].coeffs;
712 chgroup->num_channels = remaining_channels;
713 for (i = 0; i < s->channels_for_cur_subframe; i++) {
714 int channel_idx = s->channel_indexes_for_cur_subframe[i];
715 if (!s->channel[channel_idx].grouped)
716 *channel_data++ = s->channel[channel_idx].coeffs;
717 s->channel[channel_idx].grouped = 1;
721 /** decode transform type */
722 if (chgroup->num_channels == 2) {
723 if (get_bits1(&s->gb)) {
724 if (get_bits1(&s->gb)) {
725 av_log_ask_for_sample(s->avctx,
726 "unsupported channel transform type\n");
729 chgroup->transform = 1;
730 if (s->num_channels == 2) {
731 chgroup->decorrelation_matrix[0] = 1.0;
732 chgroup->decorrelation_matrix[1] = -1.0;
733 chgroup->decorrelation_matrix[2] = 1.0;
734 chgroup->decorrelation_matrix[3] = 1.0;
737 chgroup->decorrelation_matrix[0] = 0.70703125;
738 chgroup->decorrelation_matrix[1] = -0.70703125;
739 chgroup->decorrelation_matrix[2] = 0.70703125;
740 chgroup->decorrelation_matrix[3] = 0.70703125;
743 } else if (chgroup->num_channels > 2) {
744 if (get_bits1(&s->gb)) {
745 chgroup->transform = 1;
746 if (get_bits1(&s->gb)) {
747 decode_decorrelation_matrix(s, chgroup);
749 /** FIXME: more than 6 coupled channels not supported */
750 if (chgroup->num_channels > 6) {
751 av_log_ask_for_sample(s->avctx,
752 "coupled channels > 6\n");
754 memcpy(chgroup->decorrelation_matrix,
755 default_decorrelation[chgroup->num_channels],
756 chgroup->num_channels * chgroup->num_channels *
757 sizeof(*chgroup->decorrelation_matrix));
763 /** decode transform on / off */
764 if (chgroup->transform) {
765 if (!get_bits1(&s->gb)) {
767 /** transform can be enabled for individual bands */
768 for (i = 0; i < s->num_bands; i++) {
769 chgroup->transform_band[i] = get_bits1(&s->gb);
772 memset(chgroup->transform_band, 1, s->num_bands);
775 remaining_channels -= chgroup->num_channels;
782 *@brief Extract the coefficients from the bitstream.
783 *@param s codec context
784 *@param c current channel number
785 *@return 0 on success, < 0 in case of bitstream errors
787 static int decode_coeffs(WMAProDecodeCtx *s, int c)
789 /* Integers 0..15 as single-precision floats. The table saves a
790 costly int to float conversion, and storing the values as
791 integers allows fast sign-flipping. */
792 static const uint32_t fval_tab[16] = {
793 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
794 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
795 0x41000000, 0x41100000, 0x41200000, 0x41300000,
796 0x41400000, 0x41500000, 0x41600000, 0x41700000,
800 WMAProChannelCtx* ci = &s->channel[c];
807 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
809 vlctable = get_bits1(&s->gb);
810 vlc = &coef_vlc[vlctable];
820 /** decode vector coefficients (consumes up to 167 bits per iteration for
821 4 vector coded large values) */
822 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
823 (cur_coeff + 3 < ci->num_vec_coeffs)) {
828 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
830 if (idx == HUFF_VEC4_SIZE - 1) {
831 for (i = 0; i < 4; i += 2) {
832 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
833 if (idx == HUFF_VEC2_SIZE - 1) {
835 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
836 if (v0 == HUFF_VEC1_SIZE - 1)
837 v0 += ff_wma_get_large_val(&s->gb);
838 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
839 if (v1 == HUFF_VEC1_SIZE - 1)
840 v1 += ff_wma_get_large_val(&s->gb);
841 vals[i ] = av_float2int(v0);
842 vals[i+1] = av_float2int(v1);
844 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
845 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
849 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
850 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
851 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
852 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
856 for (i = 0; i < 4; i++) {
858 uint32_t sign = get_bits1(&s->gb) - 1;
859 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
862 ci->coeffs[cur_coeff] = 0;
863 /** switch to run level mode when subframe_len / 128 zeros
864 were found in a row */
865 rl_mode |= (++num_zeros > s->subframe_len >> 8);
871 /** decode run level coded coefficients */
872 if (cur_coeff < s->subframe_len) {
873 memset(&ci->coeffs[cur_coeff], 0,
874 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
875 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
876 level, run, 1, ci->coeffs,
877 cur_coeff, s->subframe_len,
878 s->subframe_len, s->esc_len, 0))
879 return AVERROR_INVALIDDATA;
886 *@brief Extract scale factors from the bitstream.
887 *@param s codec context
888 *@return 0 on success, < 0 in case of bitstream errors
890 static int decode_scale_factors(WMAProDecodeCtx* s)
894 /** should never consume more than 5344 bits
895 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
898 for (i = 0; i < s->channels_for_cur_subframe; i++) {
899 int c = s->channel_indexes_for_cur_subframe[i];
902 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
903 sf_end = s->channel[c].scale_factors + s->num_bands;
905 /** resample scale factors for the new block size
906 * as the scale factors might need to be resampled several times
907 * before some new values are transmitted, a backup of the last
908 * transmitted scale factors is kept in saved_scale_factors
910 if (s->channel[c].reuse_sf) {
911 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
913 for (b = 0; b < s->num_bands; b++)
914 s->channel[c].scale_factors[b] =
915 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
918 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
920 if (!s->channel[c].reuse_sf) {
922 /** decode DPCM coded scale factors */
923 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
924 val = 45 / s->channel[c].scale_factor_step;
925 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
926 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
931 /** run level decode differences to the resampled factors */
932 for (i = 0; i < s->num_bands; i++) {
938 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
941 uint32_t code = get_bits(&s->gb, 14);
943 sign = (code & 1) - 1;
944 skip = (code & 0x3f) >> 1;
945 } else if (idx == 1) {
948 skip = scale_rl_run[idx];
949 val = scale_rl_level[idx];
950 sign = get_bits1(&s->gb)-1;
954 if (i >= s->num_bands) {
955 av_log(s->avctx, AV_LOG_ERROR,
956 "invalid scale factor coding\n");
957 return AVERROR_INVALIDDATA;
959 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
963 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
964 s->channel[c].table_idx = s->table_idx;
965 s->channel[c].reuse_sf = 1;
968 /** calculate new scale factor maximum */
969 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
970 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
971 s->channel[c].max_scale_factor =
972 FFMAX(s->channel[c].max_scale_factor, *sf);
980 *@brief Reconstruct the individual channel data.
981 *@param s codec context
983 static void inverse_channel_transform(WMAProDecodeCtx *s)
987 for (i = 0; i < s->num_chgroups; i++) {
988 if (s->chgroup[i].transform) {
989 float data[WMAPRO_MAX_CHANNELS];
990 const int num_channels = s->chgroup[i].num_channels;
991 float** ch_data = s->chgroup[i].channel_data;
992 float** ch_end = ch_data + num_channels;
993 const int8_t* tb = s->chgroup[i].transform_band;
996 /** multichannel decorrelation */
997 for (sfb = s->cur_sfb_offsets;
998 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1001 /** multiply values with the decorrelation_matrix */
1002 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1003 const float* mat = s->chgroup[i].decorrelation_matrix;
1004 const float* data_end = data + num_channels;
1005 float* data_ptr = data;
1008 for (ch = ch_data; ch < ch_end; ch++)
1009 *data_ptr++ = (*ch)[y];
1011 for (ch = ch_data; ch < ch_end; ch++) {
1014 while (data_ptr < data_end)
1015 sum += *data_ptr++ * *mat++;
1020 } else if (s->num_channels == 2) {
1021 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1022 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1023 ch_data[0] + sfb[0],
1025 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1026 ch_data[1] + sfb[0],
1035 *@brief Apply sine window and reconstruct the output buffer.
1036 *@param s codec context
1038 static void wmapro_window(WMAProDecodeCtx *s)
1041 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1042 int c = s->channel_indexes_for_cur_subframe[i];
1044 int winlen = s->channel[c].prev_block_len;
1045 float* start = s->channel[c].coeffs - (winlen >> 1);
1047 if (s->subframe_len < winlen) {
1048 start += (winlen - s->subframe_len) >> 1;
1049 winlen = s->subframe_len;
1052 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1056 s->dsp.vector_fmul_window(start, start, start + winlen,
1059 s->channel[c].prev_block_len = s->subframe_len;
1064 *@brief Decode a single subframe (block).
1065 *@param s codec context
1066 *@return 0 on success, < 0 when decoding failed
1068 static int decode_subframe(WMAProDecodeCtx *s)
1070 int offset = s->samples_per_frame;
1071 int subframe_len = s->samples_per_frame;
1073 int total_samples = s->samples_per_frame * s->num_channels;
1074 int transmit_coeffs = 0;
1075 int cur_subwoofer_cutoff;
1077 s->subframe_offset = get_bits_count(&s->gb);
1079 /** reset channel context and find the next block offset and size
1080 == the next block of the channel with the smallest number of
1083 for (i = 0; i < s->num_channels; i++) {
1084 s->channel[i].grouped = 0;
1085 if (offset > s->channel[i].decoded_samples) {
1086 offset = s->channel[i].decoded_samples;
1088 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1093 "processing subframe with offset %i len %i\n", offset, subframe_len);
1095 /** get a list of all channels that contain the estimated block */
1096 s->channels_for_cur_subframe = 0;
1097 for (i = 0; i < s->num_channels; i++) {
1098 const int cur_subframe = s->channel[i].cur_subframe;
1099 /** substract already processed samples */
1100 total_samples -= s->channel[i].decoded_samples;
1102 /** and count if there are multiple subframes that match our profile */
1103 if (offset == s->channel[i].decoded_samples &&
1104 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1105 total_samples -= s->channel[i].subframe_len[cur_subframe];
1106 s->channel[i].decoded_samples +=
1107 s->channel[i].subframe_len[cur_subframe];
1108 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1109 ++s->channels_for_cur_subframe;
1113 /** check if the frame will be complete after processing the
1116 s->parsed_all_subframes = 1;
1119 av_dlog(s->avctx, "subframe is part of %i channels\n",
1120 s->channels_for_cur_subframe);
1122 /** calculate number of scale factor bands and their offsets */
1123 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1124 s->num_bands = s->num_sfb[s->table_idx];
1125 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1126 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1128 /** configure the decoder for the current subframe */
1129 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1130 int c = s->channel_indexes_for_cur_subframe[i];
1132 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1136 s->subframe_len = subframe_len;
1137 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1139 /** skip extended header if any */
1140 if (get_bits1(&s->gb)) {
1142 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1143 int len = get_bits(&s->gb, 4);
1144 num_fill_bits = (len ? get_bits(&s->gb, len) : 0) + 1;
1147 if (num_fill_bits >= 0) {
1148 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1149 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1150 return AVERROR_INVALIDDATA;
1153 skip_bits_long(&s->gb, num_fill_bits);
1157 /** no idea for what the following bit is used */
1158 if (get_bits1(&s->gb)) {
1159 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1160 return AVERROR_INVALIDDATA;
1164 if (decode_channel_transform(s) < 0)
1165 return AVERROR_INVALIDDATA;
1168 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1169 int c = s->channel_indexes_for_cur_subframe[i];
1170 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1171 transmit_coeffs = 1;
1174 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1175 if (transmit_coeffs) {
1177 int quant_step = 90 * s->bits_per_sample >> 4;
1179 /** decode number of vector coded coefficients */
1180 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1181 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1182 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1183 int c = s->channel_indexes_for_cur_subframe[i];
1184 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1185 if (num_vec_coeffs > s->subframe_len) {
1186 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1187 return AVERROR_INVALIDDATA;
1189 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1192 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1193 int c = s->channel_indexes_for_cur_subframe[i];
1194 s->channel[c].num_vec_coeffs = s->subframe_len;
1197 /** decode quantization step */
1198 step = get_sbits(&s->gb, 6);
1200 if (step == -32 || step == 31) {
1201 const int sign = (step == 31) - 1;
1203 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1204 (step = get_bits(&s->gb, 5)) == 31) {
1207 quant_step += ((quant + step) ^ sign) - sign;
1209 if (quant_step < 0) {
1210 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1213 /** decode quantization step modifiers for every channel */
1215 if (s->channels_for_cur_subframe == 1) {
1216 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1218 int modifier_len = get_bits(&s->gb, 3);
1219 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1220 int c = s->channel_indexes_for_cur_subframe[i];
1221 s->channel[c].quant_step = quant_step;
1222 if (get_bits1(&s->gb)) {
1224 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1226 ++s->channel[c].quant_step;
1231 /** decode scale factors */
1232 if (decode_scale_factors(s) < 0)
1233 return AVERROR_INVALIDDATA;
1236 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1237 get_bits_count(&s->gb) - s->subframe_offset);
1239 /** parse coefficients */
1240 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1241 int c = s->channel_indexes_for_cur_subframe[i];
1242 if (s->channel[c].transmit_coefs &&
1243 get_bits_count(&s->gb) < s->num_saved_bits) {
1244 decode_coeffs(s, c);
1246 memset(s->channel[c].coeffs, 0,
1247 sizeof(*s->channel[c].coeffs) * subframe_len);
1250 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1251 get_bits_count(&s->gb) - s->subframe_offset);
1253 if (transmit_coeffs) {
1254 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1255 /** reconstruct the per channel data */
1256 inverse_channel_transform(s);
1257 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1258 int c = s->channel_indexes_for_cur_subframe[i];
1259 const int* sf = s->channel[c].scale_factors;
1262 if (c == s->lfe_channel)
1263 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1264 (subframe_len - cur_subwoofer_cutoff));
1266 /** inverse quantization and rescaling */
1267 for (b = 0; b < s->num_bands; b++) {
1268 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1269 const int exp = s->channel[c].quant_step -
1270 (s->channel[c].max_scale_factor - *sf++) *
1271 s->channel[c].scale_factor_step;
1272 const float quant = pow(10.0, exp / 20.0);
1273 int start = s->cur_sfb_offsets[b];
1274 s->dsp.vector_fmul_scalar(s->tmp + start,
1275 s->channel[c].coeffs + start,
1276 quant, end - start);
1279 /** apply imdct (imdct_half == DCTIV with reverse) */
1280 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1284 /** window and overlapp-add */
1287 /** handled one subframe */
1288 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1289 int c = s->channel_indexes_for_cur_subframe[i];
1290 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1291 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1292 return AVERROR_INVALIDDATA;
1294 ++s->channel[c].cur_subframe;
1301 *@brief Decode one WMA frame.
1302 *@param s codec context
1303 *@return 0 if the trailer bit indicates that this is the last frame,
1304 * 1 if there are additional frames
1306 static int decode_frame(WMAProDecodeCtx *s, int *got_frame_ptr)
1308 AVCodecContext *avctx = s->avctx;
1309 GetBitContext* gb = &s->gb;
1310 int more_frames = 0;
1313 const float *out_ptr[WMAPRO_MAX_CHANNELS];
1316 /** get frame length */
1318 len = get_bits(gb, s->log2_frame_size);
1320 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1322 /** decode tile information */
1323 if (decode_tilehdr(s)) {
1328 /** read postproc transform */
1329 if (s->num_channels > 1 && get_bits1(gb)) {
1330 if (get_bits1(gb)) {
1331 for (i = 0; i < s->num_channels * s->num_channels; i++)
1336 /** read drc info */
1337 if (s->dynamic_range_compression) {
1338 s->drc_gain = get_bits(gb, 8);
1339 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1342 /** no idea what these are for, might be the number of samples
1343 that need to be skipped at the beginning or end of a stream */
1344 if (get_bits1(gb)) {
1347 /** usually true for the first frame */
1348 if (get_bits1(gb)) {
1349 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1350 av_dlog(s->avctx, "start skip: %i\n", skip);
1353 /** sometimes true for the last frame */
1354 if (get_bits1(gb)) {
1355 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1356 av_dlog(s->avctx, "end skip: %i\n", skip);
1361 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1362 get_bits_count(gb) - s->frame_offset);
1364 /** reset subframe states */
1365 s->parsed_all_subframes = 0;
1366 for (i = 0; i < s->num_channels; i++) {
1367 s->channel[i].decoded_samples = 0;
1368 s->channel[i].cur_subframe = 0;
1369 s->channel[i].reuse_sf = 0;
1372 /** decode all subframes */
1373 while (!s->parsed_all_subframes) {
1374 if (decode_subframe(s) < 0) {
1380 /* get output buffer */
1381 s->frame.nb_samples = s->samples_per_frame;
1382 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
1383 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1387 samples = (float *)s->frame.data[0];
1389 /** interleave samples and write them to the output buffer */
1390 for (i = 0; i < s->num_channels; i++)
1391 out_ptr[i] = s->channel[i].out;
1392 s->fmt_conv.float_interleave(samples, out_ptr, s->samples_per_frame,
1395 for (i = 0; i < s->num_channels; i++) {
1396 /** reuse second half of the IMDCT output for the next frame */
1397 memcpy(&s->channel[i].out[0],
1398 &s->channel[i].out[s->samples_per_frame],
1399 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1402 if (s->skip_frame) {
1409 if (s->len_prefix) {
1410 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1411 /** FIXME: not sure if this is always an error */
1412 av_log(s->avctx, AV_LOG_ERROR,
1413 "frame[%i] would have to skip %i bits\n", s->frame_num,
1414 len - (get_bits_count(gb) - s->frame_offset) - 1);
1419 /** skip the rest of the frame data */
1420 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1422 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1426 /** decode trailer bit */
1427 more_frames = get_bits1(gb);
1434 *@brief Calculate remaining input buffer length.
1435 *@param s codec context
1436 *@param gb bitstream reader context
1437 *@return remaining size in bits
1439 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1441 return s->buf_bit_size - get_bits_count(gb);
1445 *@brief Fill the bit reservoir with a (partial) frame.
1446 *@param s codec context
1447 *@param gb bitstream reader context
1448 *@param len length of the partial frame
1449 *@param append decides whether to reset the buffer or not
1451 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1456 /** when the frame data does not need to be concatenated, the input buffer
1457 is reset and additional bits from the previous frame are copied
1458 and skipped later so that a fast byte copy is possible */
1461 s->frame_offset = get_bits_count(gb) & 7;
1462 s->num_saved_bits = s->frame_offset;
1463 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1466 buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1468 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1469 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1474 s->num_saved_bits += len;
1476 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1479 int align = 8 - (get_bits_count(gb) & 7);
1480 align = FFMIN(align, len);
1481 put_bits(&s->pb, align, get_bits(gb, align));
1483 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1485 skip_bits_long(gb, len);
1488 PutBitContext tmp = s->pb;
1489 flush_put_bits(&tmp);
1492 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1493 skip_bits(&s->gb, s->frame_offset);
1497 *@brief Decode a single WMA packet.
1498 *@param avctx codec context
1499 *@param data the output buffer
1500 *@param avpkt input packet
1501 *@return number of bytes that were read from the input buffer
1503 static int decode_packet(AVCodecContext *avctx, void *data,
1504 int *got_frame_ptr, AVPacket* avpkt)
1506 WMAProDecodeCtx *s = avctx->priv_data;
1507 GetBitContext* gb = &s->pgb;
1508 const uint8_t* buf = avpkt->data;
1509 int buf_size = avpkt->size;
1510 int num_bits_prev_frame;
1511 int packet_sequence_number;
1515 if (s->packet_done || s->packet_loss) {
1518 /** sanity check for the buffer length */
1519 if (buf_size < avctx->block_align)
1522 s->next_packet_start = buf_size - avctx->block_align;
1523 buf_size = avctx->block_align;
1524 s->buf_bit_size = buf_size << 3;
1526 /** parse packet header */
1527 init_get_bits(gb, buf, s->buf_bit_size);
1528 packet_sequence_number = get_bits(gb, 4);
1531 /** get number of bits that need to be added to the previous frame */
1532 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1533 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1534 num_bits_prev_frame);
1536 /** check for packet loss */
1537 if (!s->packet_loss &&
1538 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1540 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1541 s->packet_sequence_number, packet_sequence_number);
1543 s->packet_sequence_number = packet_sequence_number;
1545 if (num_bits_prev_frame > 0) {
1546 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1547 if (num_bits_prev_frame >= remaining_packet_bits) {
1548 num_bits_prev_frame = remaining_packet_bits;
1552 /** append the previous frame data to the remaining data from the
1553 previous packet to create a full frame */
1554 save_bits(s, gb, num_bits_prev_frame, 1);
1555 av_dlog(avctx, "accumulated %x bits of frame data\n",
1556 s->num_saved_bits - s->frame_offset);
1558 /** decode the cross packet frame if it is valid */
1559 if (!s->packet_loss)
1560 decode_frame(s, got_frame_ptr);
1561 } else if (s->num_saved_bits - s->frame_offset) {
1562 av_dlog(avctx, "ignoring %x previously saved bits\n",
1563 s->num_saved_bits - s->frame_offset);
1566 if (s->packet_loss) {
1567 /** reset number of saved bits so that the decoder
1568 does not start to decode incomplete frames in the
1569 s->len_prefix == 0 case */
1570 s->num_saved_bits = 0;
1576 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1577 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1578 skip_bits(gb, s->packet_offset);
1579 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1580 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1581 frame_size <= remaining_bits(s, gb)) {
1582 save_bits(s, gb, frame_size, 0);
1583 s->packet_done = !decode_frame(s, got_frame_ptr);
1584 } else if (!s->len_prefix
1585 && s->num_saved_bits > get_bits_count(&s->gb)) {
1586 /** when the frames do not have a length prefix, we don't know
1587 the compressed length of the individual frames
1588 however, we know what part of a new packet belongs to the
1590 therefore we save the incoming packet first, then we append
1591 the "previous frame" data from the next packet so that
1592 we get a buffer that only contains full frames */
1593 s->packet_done = !decode_frame(s, got_frame_ptr);
1598 if (s->packet_done && !s->packet_loss &&
1599 remaining_bits(s, gb) > 0) {
1600 /** save the rest of the data so that it can be decoded
1601 with the next packet */
1602 save_bits(s, gb, remaining_bits(s, gb), 0);
1605 s->packet_offset = get_bits_count(gb) & 7;
1607 return AVERROR_INVALIDDATA;
1610 *(AVFrame *)data = s->frame;
1612 return get_bits_count(gb) >> 3;
1616 *@brief Clear decoder buffers (for seeking).
1617 *@param avctx codec context
1619 static void flush(AVCodecContext *avctx)
1621 WMAProDecodeCtx *s = avctx->priv_data;
1623 /** reset output buffer as a part of it is used during the windowing of a
1625 for (i = 0; i < s->num_channels; i++)
1626 memset(s->channel[i].out, 0, s->samples_per_frame *
1627 sizeof(*s->channel[i].out));
1633 *@brief wmapro decoder
1635 AVCodec ff_wmapro_decoder = {
1637 .type = AVMEDIA_TYPE_AUDIO,
1638 .id = AV_CODEC_ID_WMAPRO,
1639 .priv_data_size = sizeof(WMAProDecodeCtx),
1640 .init = decode_init,
1641 .close = decode_end,
1642 .decode = decode_packet,
1643 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1645 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),