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 void av_cold 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 s->num_channels = avctx->channels;
340 if (s->num_channels < 0) {
341 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
342 return AVERROR_INVALIDDATA;
343 } else if (s->num_channels > WMAPRO_MAX_CHANNELS) {
344 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
345 return AVERROR_PATCHWELCOME;
348 /** init previous block len */
349 for (i = 0; i < s->num_channels; i++)
350 s->channel[i].prev_block_len = s->samples_per_frame;
352 /** extract lfe channel position */
355 if (channel_mask & 8) {
357 for (mask = 1; mask < 16; mask <<= 1) {
358 if (channel_mask & mask)
363 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
364 scale_huffbits, 1, 1,
365 scale_huffcodes, 2, 2, 616);
367 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
368 scale_rl_huffbits, 1, 1,
369 scale_rl_huffcodes, 4, 4, 1406);
371 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
372 coef0_huffbits, 1, 1,
373 coef0_huffcodes, 4, 4, 2108);
375 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
376 coef1_huffbits, 1, 1,
377 coef1_huffcodes, 4, 4, 3912);
379 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
381 vec4_huffcodes, 2, 2, 604);
383 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
385 vec2_huffcodes, 2, 2, 562);
387 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
389 vec1_huffcodes, 2, 2, 562);
391 /** calculate number of scale factor bands and their offsets
392 for every possible block size */
393 for (i = 0; i < num_possible_block_sizes; i++) {
394 int subframe_len = s->samples_per_frame >> i;
398 s->sfb_offsets[i][0] = 0;
400 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
401 int offset = (subframe_len * 2 * critical_freq[x])
402 / s->avctx->sample_rate + 2;
404 if (offset > s->sfb_offsets[i][band - 1])
405 s->sfb_offsets[i][band++] = offset;
407 s->sfb_offsets[i][band - 1] = subframe_len;
408 s->num_sfb[i] = band - 1;
412 /** Scale factors can be shared between blocks of different size
413 as every block has a different scale factor band layout.
414 The matrix sf_offsets is needed to find the correct scale factor.
417 for (i = 0; i < num_possible_block_sizes; i++) {
419 for (b = 0; b < s->num_sfb[i]; b++) {
421 int offset = ((s->sfb_offsets[i][b]
422 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
423 for (x = 0; x < num_possible_block_sizes; x++) {
425 while (s->sfb_offsets[x][v + 1] << x < offset)
427 s->sf_offsets[i][x][b] = v;
432 /** init MDCT, FIXME: only init needed sizes */
433 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
434 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
435 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
436 / (1 << (s->bits_per_sample - 1)));
438 /** init MDCT windows: simple sinus window */
439 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
440 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
441 ff_init_ff_sine_windows(win_idx);
442 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
445 /** calculate subwoofer cutoff values */
446 for (i = 0; i < num_possible_block_sizes; i++) {
447 int block_size = s->samples_per_frame >> i;
448 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
449 / s->avctx->sample_rate;
450 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
453 /** calculate sine values for the decorrelation matrix */
454 for (i = 0; i < 33; i++)
455 sin64[i] = sin(i*M_PI / 64.0);
457 if (avctx->debug & FF_DEBUG_BITSTREAM)
460 avctx->channel_layout = channel_mask;
462 avcodec_get_frame_defaults(&s->frame);
463 avctx->coded_frame = &s->frame;
469 *@brief Decode the subframe length.
471 *@param offset sample offset in the frame
472 *@return decoded subframe length on success, < 0 in case of an error
474 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
476 int frame_len_shift = 0;
479 /** no need to read from the bitstream when only one length is possible */
480 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
481 return s->min_samples_per_subframe;
483 /** 1 bit indicates if the subframe is of maximum length */
484 if (s->max_subframe_len_bit) {
485 if (get_bits1(&s->gb))
486 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
488 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
490 subframe_len = s->samples_per_frame >> frame_len_shift;
492 /** sanity check the length */
493 if (subframe_len < s->min_samples_per_subframe ||
494 subframe_len > s->samples_per_frame) {
495 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
497 return AVERROR_INVALIDDATA;
503 *@brief Decode how the data in the frame is split into subframes.
504 * Every WMA frame contains the encoded data for a fixed number of
505 * samples per channel. The data for every channel might be split
506 * into several subframes. This function will reconstruct the list of
507 * subframes for every channel.
509 * If the subframes are not evenly split, the algorithm estimates the
510 * channels with the lowest number of total samples.
511 * Afterwards, for each of these channels a bit is read from the
512 * bitstream that indicates if the channel contains a subframe with the
513 * next subframe size that is going to be read from the bitstream or not.
514 * If a channel contains such a subframe, the subframe size gets added to
515 * the channel's subframe list.
516 * The algorithm repeats these steps until the frame is properly divided
517 * between the individual channels.
520 *@return 0 on success, < 0 in case of an error
522 static int decode_tilehdr(WMAProDecodeCtx *s)
524 uint16_t num_samples[WMAPRO_MAX_CHANNELS]; /**< sum of samples for all currently known subframes of a channel */
525 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
526 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
527 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
528 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
531 /* Should never consume more than 3073 bits (256 iterations for the
532 * while loop when always the minimum amount of 128 samples is substracted
533 * from missing samples in the 8 channel case).
534 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
537 /** reset tiling information */
538 for (c = 0; c < s->num_channels; c++)
539 s->channel[c].num_subframes = 0;
541 memset(num_samples, 0, sizeof(num_samples));
543 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
544 fixed_channel_layout = 1;
546 /** loop until the frame data is split between the subframes */
550 /** check which channels contain the subframe */
551 for (c = 0; c < s->num_channels; c++) {
552 if (num_samples[c] == min_channel_len) {
553 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
554 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
555 contains_subframe[c] = 1;
557 contains_subframe[c] = get_bits1(&s->gb);
559 contains_subframe[c] = 0;
562 /** get subframe length, subframe_len == 0 is not allowed */
563 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
564 return AVERROR_INVALIDDATA;
566 /** add subframes to the individual channels and find new min_channel_len */
567 min_channel_len += subframe_len;
568 for (c = 0; c < s->num_channels; c++) {
569 WMAProChannelCtx* chan = &s->channel[c];
571 if (contains_subframe[c]) {
572 if (chan->num_subframes >= MAX_SUBFRAMES) {
573 av_log(s->avctx, AV_LOG_ERROR,
574 "broken frame: num subframes > 31\n");
575 return AVERROR_INVALIDDATA;
577 chan->subframe_len[chan->num_subframes] = subframe_len;
578 num_samples[c] += subframe_len;
579 ++chan->num_subframes;
580 if (num_samples[c] > s->samples_per_frame) {
581 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
582 "channel len > samples_per_frame\n");
583 return AVERROR_INVALIDDATA;
585 } else if (num_samples[c] <= min_channel_len) {
586 if (num_samples[c] < min_channel_len) {
587 channels_for_cur_subframe = 0;
588 min_channel_len = num_samples[c];
590 ++channels_for_cur_subframe;
593 } while (min_channel_len < s->samples_per_frame);
595 for (c = 0; c < s->num_channels; c++) {
598 for (i = 0; i < s->channel[c].num_subframes; i++) {
599 av_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
600 " len %i\n", s->frame_num, c, i,
601 s->channel[c].subframe_len[i]);
602 s->channel[c].subframe_offset[i] = offset;
603 offset += s->channel[c].subframe_len[i];
611 *@brief Calculate a decorrelation matrix from the bitstream parameters.
612 *@param s codec context
613 *@param chgroup channel group for which the matrix needs to be calculated
615 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
616 WMAProChannelGrp *chgroup)
620 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
621 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
622 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
624 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
625 rotation_offset[i] = get_bits(&s->gb, 6);
627 for (i = 0; i < chgroup->num_channels; i++)
628 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
629 get_bits1(&s->gb) ? 1.0 : -1.0;
631 for (i = 1; i < chgroup->num_channels; i++) {
633 for (x = 0; x < i; x++) {
635 for (y = 0; y < i + 1; y++) {
636 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
637 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
638 int n = rotation_offset[offset + x];
644 cosv = sin64[32 - n];
646 sinv = sin64[64 - n];
647 cosv = -sin64[n - 32];
650 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
651 (v1 * sinv) - (v2 * cosv);
652 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
653 (v1 * cosv) + (v2 * sinv);
661 *@brief Decode channel transformation parameters
662 *@param s codec context
663 *@return 0 in case of success, < 0 in case of bitstream errors
665 static int decode_channel_transform(WMAProDecodeCtx* s)
668 /* should never consume more than 1921 bits for the 8 channel case
669 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
670 * + MAX_CHANNELS + MAX_BANDS + 1)
673 /** in the one channel case channel transforms are pointless */
675 if (s->num_channels > 1) {
676 int remaining_channels = s->channels_for_cur_subframe;
678 if (get_bits1(&s->gb)) {
679 av_log_ask_for_sample(s->avctx,
680 "unsupported channel transform bit\n");
681 return AVERROR_INVALIDDATA;
684 for (s->num_chgroups = 0; remaining_channels &&
685 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
686 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
687 float** channel_data = chgroup->channel_data;
688 chgroup->num_channels = 0;
689 chgroup->transform = 0;
691 /** decode channel mask */
692 if (remaining_channels > 2) {
693 for (i = 0; i < s->channels_for_cur_subframe; i++) {
694 int channel_idx = s->channel_indexes_for_cur_subframe[i];
695 if (!s->channel[channel_idx].grouped
696 && get_bits1(&s->gb)) {
697 ++chgroup->num_channels;
698 s->channel[channel_idx].grouped = 1;
699 *channel_data++ = s->channel[channel_idx].coeffs;
703 chgroup->num_channels = remaining_channels;
704 for (i = 0; i < s->channels_for_cur_subframe; i++) {
705 int channel_idx = s->channel_indexes_for_cur_subframe[i];
706 if (!s->channel[channel_idx].grouped)
707 *channel_data++ = s->channel[channel_idx].coeffs;
708 s->channel[channel_idx].grouped = 1;
712 /** decode transform type */
713 if (chgroup->num_channels == 2) {
714 if (get_bits1(&s->gb)) {
715 if (get_bits1(&s->gb)) {
716 av_log_ask_for_sample(s->avctx,
717 "unsupported channel transform type\n");
720 chgroup->transform = 1;
721 if (s->num_channels == 2) {
722 chgroup->decorrelation_matrix[0] = 1.0;
723 chgroup->decorrelation_matrix[1] = -1.0;
724 chgroup->decorrelation_matrix[2] = 1.0;
725 chgroup->decorrelation_matrix[3] = 1.0;
728 chgroup->decorrelation_matrix[0] = 0.70703125;
729 chgroup->decorrelation_matrix[1] = -0.70703125;
730 chgroup->decorrelation_matrix[2] = 0.70703125;
731 chgroup->decorrelation_matrix[3] = 0.70703125;
734 } else if (chgroup->num_channels > 2) {
735 if (get_bits1(&s->gb)) {
736 chgroup->transform = 1;
737 if (get_bits1(&s->gb)) {
738 decode_decorrelation_matrix(s, chgroup);
740 /** FIXME: more than 6 coupled channels not supported */
741 if (chgroup->num_channels > 6) {
742 av_log_ask_for_sample(s->avctx,
743 "coupled channels > 6\n");
745 memcpy(chgroup->decorrelation_matrix,
746 default_decorrelation[chgroup->num_channels],
747 chgroup->num_channels * chgroup->num_channels *
748 sizeof(*chgroup->decorrelation_matrix));
754 /** decode transform on / off */
755 if (chgroup->transform) {
756 if (!get_bits1(&s->gb)) {
758 /** transform can be enabled for individual bands */
759 for (i = 0; i < s->num_bands; i++) {
760 chgroup->transform_band[i] = get_bits1(&s->gb);
763 memset(chgroup->transform_band, 1, s->num_bands);
766 remaining_channels -= chgroup->num_channels;
773 *@brief Extract the coefficients from the bitstream.
774 *@param s codec context
775 *@param c current channel number
776 *@return 0 on success, < 0 in case of bitstream errors
778 static int decode_coeffs(WMAProDecodeCtx *s, int c)
780 /* Integers 0..15 as single-precision floats. The table saves a
781 costly int to float conversion, and storing the values as
782 integers allows fast sign-flipping. */
783 static const uint32_t fval_tab[16] = {
784 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
785 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
786 0x41000000, 0x41100000, 0x41200000, 0x41300000,
787 0x41400000, 0x41500000, 0x41600000, 0x41700000,
791 WMAProChannelCtx* ci = &s->channel[c];
798 av_dlog(s->avctx, "decode coefficients for channel %i\n", c);
800 vlctable = get_bits1(&s->gb);
801 vlc = &coef_vlc[vlctable];
811 /** decode vector coefficients (consumes up to 167 bits per iteration for
812 4 vector coded large values) */
813 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
814 (cur_coeff + 3 < ci->num_vec_coeffs)) {
819 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
821 if (idx == HUFF_VEC4_SIZE - 1) {
822 for (i = 0; i < 4; i += 2) {
823 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
824 if (idx == HUFF_VEC2_SIZE - 1) {
826 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
827 if (v0 == HUFF_VEC1_SIZE - 1)
828 v0 += ff_wma_get_large_val(&s->gb);
829 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
830 if (v1 == HUFF_VEC1_SIZE - 1)
831 v1 += ff_wma_get_large_val(&s->gb);
832 vals[i ] = av_float2int(v0);
833 vals[i+1] = av_float2int(v1);
835 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
836 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
840 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
841 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
842 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
843 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
847 for (i = 0; i < 4; i++) {
849 uint32_t sign = get_bits1(&s->gb) - 1;
850 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
853 ci->coeffs[cur_coeff] = 0;
854 /** switch to run level mode when subframe_len / 128 zeros
855 were found in a row */
856 rl_mode |= (++num_zeros > s->subframe_len >> 8);
862 /** decode run level coded coefficients */
863 if (cur_coeff < s->subframe_len) {
864 memset(&ci->coeffs[cur_coeff], 0,
865 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
866 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
867 level, run, 1, ci->coeffs,
868 cur_coeff, s->subframe_len,
869 s->subframe_len, s->esc_len, 0))
870 return AVERROR_INVALIDDATA;
877 *@brief Extract scale factors from the bitstream.
878 *@param s codec context
879 *@return 0 on success, < 0 in case of bitstream errors
881 static int decode_scale_factors(WMAProDecodeCtx* s)
885 /** should never consume more than 5344 bits
886 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
889 for (i = 0; i < s->channels_for_cur_subframe; i++) {
890 int c = s->channel_indexes_for_cur_subframe[i];
893 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
894 sf_end = s->channel[c].scale_factors + s->num_bands;
896 /** resample scale factors for the new block size
897 * as the scale factors might need to be resampled several times
898 * before some new values are transmitted, a backup of the last
899 * transmitted scale factors is kept in saved_scale_factors
901 if (s->channel[c].reuse_sf) {
902 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
904 for (b = 0; b < s->num_bands; b++)
905 s->channel[c].scale_factors[b] =
906 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
909 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
911 if (!s->channel[c].reuse_sf) {
913 /** decode DPCM coded scale factors */
914 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
915 val = 45 / s->channel[c].scale_factor_step;
916 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
917 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
922 /** run level decode differences to the resampled factors */
923 for (i = 0; i < s->num_bands; i++) {
929 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
932 uint32_t code = get_bits(&s->gb, 14);
934 sign = (code & 1) - 1;
935 skip = (code & 0x3f) >> 1;
936 } else if (idx == 1) {
939 skip = scale_rl_run[idx];
940 val = scale_rl_level[idx];
941 sign = get_bits1(&s->gb)-1;
945 if (i >= s->num_bands) {
946 av_log(s->avctx, AV_LOG_ERROR,
947 "invalid scale factor coding\n");
948 return AVERROR_INVALIDDATA;
950 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
954 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
955 s->channel[c].table_idx = s->table_idx;
956 s->channel[c].reuse_sf = 1;
959 /** calculate new scale factor maximum */
960 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
961 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
962 s->channel[c].max_scale_factor =
963 FFMAX(s->channel[c].max_scale_factor, *sf);
971 *@brief Reconstruct the individual channel data.
972 *@param s codec context
974 static void inverse_channel_transform(WMAProDecodeCtx *s)
978 for (i = 0; i < s->num_chgroups; i++) {
979 if (s->chgroup[i].transform) {
980 float data[WMAPRO_MAX_CHANNELS];
981 const int num_channels = s->chgroup[i].num_channels;
982 float** ch_data = s->chgroup[i].channel_data;
983 float** ch_end = ch_data + num_channels;
984 const int8_t* tb = s->chgroup[i].transform_band;
987 /** multichannel decorrelation */
988 for (sfb = s->cur_sfb_offsets;
989 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
992 /** multiply values with the decorrelation_matrix */
993 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
994 const float* mat = s->chgroup[i].decorrelation_matrix;
995 const float* data_end = data + num_channels;
996 float* data_ptr = data;
999 for (ch = ch_data; ch < ch_end; ch++)
1000 *data_ptr++ = (*ch)[y];
1002 for (ch = ch_data; ch < ch_end; ch++) {
1005 while (data_ptr < data_end)
1006 sum += *data_ptr++ * *mat++;
1011 } else if (s->num_channels == 2) {
1012 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1013 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
1014 ch_data[0] + sfb[0],
1016 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1017 ch_data[1] + sfb[0],
1026 *@brief Apply sine window and reconstruct the output buffer.
1027 *@param s codec context
1029 static void wmapro_window(WMAProDecodeCtx *s)
1032 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1033 int c = s->channel_indexes_for_cur_subframe[i];
1035 int winlen = s->channel[c].prev_block_len;
1036 float* start = s->channel[c].coeffs - (winlen >> 1);
1038 if (s->subframe_len < winlen) {
1039 start += (winlen - s->subframe_len) >> 1;
1040 winlen = s->subframe_len;
1043 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1047 s->dsp.vector_fmul_window(start, start, start + winlen,
1050 s->channel[c].prev_block_len = s->subframe_len;
1055 *@brief Decode a single subframe (block).
1056 *@param s codec context
1057 *@return 0 on success, < 0 when decoding failed
1059 static int decode_subframe(WMAProDecodeCtx *s)
1061 int offset = s->samples_per_frame;
1062 int subframe_len = s->samples_per_frame;
1064 int total_samples = s->samples_per_frame * s->num_channels;
1065 int transmit_coeffs = 0;
1066 int cur_subwoofer_cutoff;
1068 s->subframe_offset = get_bits_count(&s->gb);
1070 /** reset channel context and find the next block offset and size
1071 == the next block of the channel with the smallest number of
1074 for (i = 0; i < s->num_channels; i++) {
1075 s->channel[i].grouped = 0;
1076 if (offset > s->channel[i].decoded_samples) {
1077 offset = s->channel[i].decoded_samples;
1079 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1084 "processing subframe with offset %i len %i\n", offset, subframe_len);
1086 /** get a list of all channels that contain the estimated block */
1087 s->channels_for_cur_subframe = 0;
1088 for (i = 0; i < s->num_channels; i++) {
1089 const int cur_subframe = s->channel[i].cur_subframe;
1090 /** substract already processed samples */
1091 total_samples -= s->channel[i].decoded_samples;
1093 /** and count if there are multiple subframes that match our profile */
1094 if (offset == s->channel[i].decoded_samples &&
1095 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1096 total_samples -= s->channel[i].subframe_len[cur_subframe];
1097 s->channel[i].decoded_samples +=
1098 s->channel[i].subframe_len[cur_subframe];
1099 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1100 ++s->channels_for_cur_subframe;
1104 /** check if the frame will be complete after processing the
1107 s->parsed_all_subframes = 1;
1110 av_dlog(s->avctx, "subframe is part of %i channels\n",
1111 s->channels_for_cur_subframe);
1113 /** calculate number of scale factor bands and their offsets */
1114 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1115 s->num_bands = s->num_sfb[s->table_idx];
1116 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1117 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1119 /** configure the decoder for the current subframe */
1120 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1121 int c = s->channel_indexes_for_cur_subframe[i];
1123 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1127 s->subframe_len = subframe_len;
1128 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1130 /** skip extended header if any */
1131 if (get_bits1(&s->gb)) {
1133 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1134 int len = get_bits(&s->gb, 4);
1135 num_fill_bits = get_bits(&s->gb, len) + 1;
1138 if (num_fill_bits >= 0) {
1139 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1140 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1141 return AVERROR_INVALIDDATA;
1144 skip_bits_long(&s->gb, num_fill_bits);
1148 /** no idea for what the following bit is used */
1149 if (get_bits1(&s->gb)) {
1150 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1151 return AVERROR_INVALIDDATA;
1155 if (decode_channel_transform(s) < 0)
1156 return AVERROR_INVALIDDATA;
1159 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1160 int c = s->channel_indexes_for_cur_subframe[i];
1161 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1162 transmit_coeffs = 1;
1165 if (transmit_coeffs) {
1167 int quant_step = 90 * s->bits_per_sample >> 4;
1169 /** decode number of vector coded coefficients */
1170 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1171 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1172 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1173 int c = s->channel_indexes_for_cur_subframe[i];
1174 s->channel[c].num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1177 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1178 int c = s->channel_indexes_for_cur_subframe[i];
1179 s->channel[c].num_vec_coeffs = s->subframe_len;
1182 /** decode quantization step */
1183 step = get_sbits(&s->gb, 6);
1185 if (step == -32 || step == 31) {
1186 const int sign = (step == 31) - 1;
1188 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1189 (step = get_bits(&s->gb, 5)) == 31) {
1192 quant_step += ((quant + step) ^ sign) - sign;
1194 if (quant_step < 0) {
1195 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1198 /** decode quantization step modifiers for every channel */
1200 if (s->channels_for_cur_subframe == 1) {
1201 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1203 int modifier_len = get_bits(&s->gb, 3);
1204 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1205 int c = s->channel_indexes_for_cur_subframe[i];
1206 s->channel[c].quant_step = quant_step;
1207 if (get_bits1(&s->gb)) {
1209 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1211 ++s->channel[c].quant_step;
1216 /** decode scale factors */
1217 if (decode_scale_factors(s) < 0)
1218 return AVERROR_INVALIDDATA;
1221 av_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1222 get_bits_count(&s->gb) - s->subframe_offset);
1224 /** parse coefficients */
1225 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1226 int c = s->channel_indexes_for_cur_subframe[i];
1227 if (s->channel[c].transmit_coefs &&
1228 get_bits_count(&s->gb) < s->num_saved_bits) {
1229 decode_coeffs(s, c);
1231 memset(s->channel[c].coeffs, 0,
1232 sizeof(*s->channel[c].coeffs) * subframe_len);
1235 av_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1236 get_bits_count(&s->gb) - s->subframe_offset);
1238 if (transmit_coeffs) {
1239 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1240 /** reconstruct the per channel data */
1241 inverse_channel_transform(s);
1242 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1243 int c = s->channel_indexes_for_cur_subframe[i];
1244 const int* sf = s->channel[c].scale_factors;
1247 if (c == s->lfe_channel)
1248 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1249 (subframe_len - cur_subwoofer_cutoff));
1251 /** inverse quantization and rescaling */
1252 for (b = 0; b < s->num_bands; b++) {
1253 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1254 const int exp = s->channel[c].quant_step -
1255 (s->channel[c].max_scale_factor - *sf++) *
1256 s->channel[c].scale_factor_step;
1257 const float quant = pow(10.0, exp / 20.0);
1258 int start = s->cur_sfb_offsets[b];
1259 s->dsp.vector_fmul_scalar(s->tmp + start,
1260 s->channel[c].coeffs + start,
1261 quant, end - start);
1264 /** apply imdct (imdct_half == DCTIV with reverse) */
1265 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1269 /** window and overlapp-add */
1272 /** handled one subframe */
1273 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1274 int c = s->channel_indexes_for_cur_subframe[i];
1275 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1276 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1277 return AVERROR_INVALIDDATA;
1279 ++s->channel[c].cur_subframe;
1286 *@brief Decode one WMA frame.
1287 *@param s codec context
1288 *@return 0 if the trailer bit indicates that this is the last frame,
1289 * 1 if there are additional frames
1291 static int decode_frame(WMAProDecodeCtx *s, int *got_frame_ptr)
1293 AVCodecContext *avctx = s->avctx;
1294 GetBitContext* gb = &s->gb;
1295 int more_frames = 0;
1298 const float *out_ptr[WMAPRO_MAX_CHANNELS];
1301 /** get frame length */
1303 len = get_bits(gb, s->log2_frame_size);
1305 av_dlog(s->avctx, "decoding frame with length %x\n", len);
1307 /** decode tile information */
1308 if (decode_tilehdr(s)) {
1313 /** read postproc transform */
1314 if (s->num_channels > 1 && get_bits1(gb)) {
1315 if (get_bits1(gb)) {
1316 for (i = 0; i < s->num_channels * s->num_channels; i++)
1321 /** read drc info */
1322 if (s->dynamic_range_compression) {
1323 s->drc_gain = get_bits(gb, 8);
1324 av_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1327 /** no idea what these are for, might be the number of samples
1328 that need to be skipped at the beginning or end of a stream */
1329 if (get_bits1(gb)) {
1332 /** usually true for the first frame */
1333 if (get_bits1(gb)) {
1334 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1335 av_dlog(s->avctx, "start skip: %i\n", skip);
1338 /** sometimes true for the last frame */
1339 if (get_bits1(gb)) {
1340 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1341 av_dlog(s->avctx, "end skip: %i\n", skip);
1346 av_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1347 get_bits_count(gb) - s->frame_offset);
1349 /** reset subframe states */
1350 s->parsed_all_subframes = 0;
1351 for (i = 0; i < s->num_channels; i++) {
1352 s->channel[i].decoded_samples = 0;
1353 s->channel[i].cur_subframe = 0;
1354 s->channel[i].reuse_sf = 0;
1357 /** decode all subframes */
1358 while (!s->parsed_all_subframes) {
1359 if (decode_subframe(s) < 0) {
1365 /* get output buffer */
1366 s->frame.nb_samples = s->samples_per_frame;
1367 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
1368 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1372 samples = (float *)s->frame.data[0];
1374 /** interleave samples and write them to the output buffer */
1375 for (i = 0; i < s->num_channels; i++)
1376 out_ptr[i] = s->channel[i].out;
1377 s->fmt_conv.float_interleave(samples, out_ptr, s->samples_per_frame,
1380 for (i = 0; i < s->num_channels; i++) {
1381 /** reuse second half of the IMDCT output for the next frame */
1382 memcpy(&s->channel[i].out[0],
1383 &s->channel[i].out[s->samples_per_frame],
1384 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1387 if (s->skip_frame) {
1394 if (s->len_prefix) {
1395 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1396 /** FIXME: not sure if this is always an error */
1397 av_log(s->avctx, AV_LOG_ERROR,
1398 "frame[%i] would have to skip %i bits\n", s->frame_num,
1399 len - (get_bits_count(gb) - s->frame_offset) - 1);
1404 /** skip the rest of the frame data */
1405 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1407 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1411 /** decode trailer bit */
1412 more_frames = get_bits1(gb);
1419 *@brief Calculate remaining input buffer length.
1420 *@param s codec context
1421 *@param gb bitstream reader context
1422 *@return remaining size in bits
1424 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1426 return s->buf_bit_size - get_bits_count(gb);
1430 *@brief Fill the bit reservoir with a (partial) frame.
1431 *@param s codec context
1432 *@param gb bitstream reader context
1433 *@param len length of the partial frame
1434 *@param append decides whether to reset the buffer or not
1436 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1441 /** when the frame data does not need to be concatenated, the input buffer
1442 is resetted and additional bits from the previous frame are copyed
1443 and skipped later so that a fast byte copy is possible */
1446 s->frame_offset = get_bits_count(gb) & 7;
1447 s->num_saved_bits = s->frame_offset;
1448 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1451 buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1453 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1454 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1459 s->num_saved_bits += len;
1461 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1464 int align = 8 - (get_bits_count(gb) & 7);
1465 align = FFMIN(align, len);
1466 put_bits(&s->pb, align, get_bits(gb, align));
1468 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1470 skip_bits_long(gb, len);
1473 PutBitContext tmp = s->pb;
1474 flush_put_bits(&tmp);
1477 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1478 skip_bits(&s->gb, s->frame_offset);
1482 *@brief Decode a single WMA packet.
1483 *@param avctx codec context
1484 *@param data the output buffer
1485 *@param avpkt input packet
1486 *@return number of bytes that were read from the input buffer
1488 static int decode_packet(AVCodecContext *avctx, void *data,
1489 int *got_frame_ptr, AVPacket* avpkt)
1491 WMAProDecodeCtx *s = avctx->priv_data;
1492 GetBitContext* gb = &s->pgb;
1493 const uint8_t* buf = avpkt->data;
1494 int buf_size = avpkt->size;
1495 int num_bits_prev_frame;
1496 int packet_sequence_number;
1500 if (s->packet_done || s->packet_loss) {
1503 /** sanity check for the buffer length */
1504 if (buf_size < avctx->block_align)
1507 s->next_packet_start = buf_size - avctx->block_align;
1508 buf_size = avctx->block_align;
1509 s->buf_bit_size = buf_size << 3;
1511 /** parse packet header */
1512 init_get_bits(gb, buf, s->buf_bit_size);
1513 packet_sequence_number = get_bits(gb, 4);
1516 /** get number of bits that need to be added to the previous frame */
1517 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1518 av_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1519 num_bits_prev_frame);
1521 /** check for packet loss */
1522 if (!s->packet_loss &&
1523 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1525 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1526 s->packet_sequence_number, packet_sequence_number);
1528 s->packet_sequence_number = packet_sequence_number;
1530 if (num_bits_prev_frame > 0) {
1531 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1532 if (num_bits_prev_frame >= remaining_packet_bits) {
1533 num_bits_prev_frame = remaining_packet_bits;
1537 /** append the previous frame data to the remaining data from the
1538 previous packet to create a full frame */
1539 save_bits(s, gb, num_bits_prev_frame, 1);
1540 av_dlog(avctx, "accumulated %x bits of frame data\n",
1541 s->num_saved_bits - s->frame_offset);
1543 /** decode the cross packet frame if it is valid */
1544 if (!s->packet_loss)
1545 decode_frame(s, got_frame_ptr);
1546 } else if (s->num_saved_bits - s->frame_offset) {
1547 av_dlog(avctx, "ignoring %x previously saved bits\n",
1548 s->num_saved_bits - s->frame_offset);
1551 if (s->packet_loss) {
1552 /** reset number of saved bits so that the decoder
1553 does not start to decode incomplete frames in the
1554 s->len_prefix == 0 case */
1555 s->num_saved_bits = 0;
1561 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1562 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1563 skip_bits(gb, s->packet_offset);
1564 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1565 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1566 frame_size <= remaining_bits(s, gb)) {
1567 save_bits(s, gb, frame_size, 0);
1568 s->packet_done = !decode_frame(s, got_frame_ptr);
1569 } else if (!s->len_prefix
1570 && s->num_saved_bits > get_bits_count(&s->gb)) {
1571 /** when the frames do not have a length prefix, we don't know
1572 the compressed length of the individual frames
1573 however, we know what part of a new packet belongs to the
1575 therefore we save the incoming packet first, then we append
1576 the "previous frame" data from the next packet so that
1577 we get a buffer that only contains full frames */
1578 s->packet_done = !decode_frame(s, got_frame_ptr);
1583 if (s->packet_done && !s->packet_loss &&
1584 remaining_bits(s, gb) > 0) {
1585 /** save the rest of the data so that it can be decoded
1586 with the next packet */
1587 save_bits(s, gb, remaining_bits(s, gb), 0);
1590 s->packet_offset = get_bits_count(gb) & 7;
1592 return AVERROR_INVALIDDATA;
1595 *(AVFrame *)data = s->frame;
1597 return get_bits_count(gb) >> 3;
1601 *@brief Clear decoder buffers (for seeking).
1602 *@param avctx codec context
1604 static void flush(AVCodecContext *avctx)
1606 WMAProDecodeCtx *s = avctx->priv_data;
1608 /** reset output buffer as a part of it is used during the windowing of a
1610 for (i = 0; i < s->num_channels; i++)
1611 memset(s->channel[i].out, 0, s->samples_per_frame *
1612 sizeof(*s->channel[i].out));
1618 *@brief wmapro decoder
1620 AVCodec ff_wmapro_decoder = {
1622 .type = AVMEDIA_TYPE_AUDIO,
1623 .id = CODEC_ID_WMAPRO,
1624 .priv_data_size = sizeof(WMAProDecodeCtx),
1625 .init = decode_init,
1626 .close = decode_end,
1627 .decode = decode_packet,
1628 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1630 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),