2 * Wmapro compatible decoder
3 * Copyright (c) 2007 Baptiste Coudurier, Benjamin Larsson, Ulion
4 * Copyright (c) 2008 - 2009 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.
93 #include "wmaprodata.h"
97 /** current decoder limitations */
98 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
99 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
100 #define MAX_BANDS 29 ///< max number of scale factor bands
101 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
103 #define WMAPRO_BLOCK_MAX_BITS 12 ///< log2 of max block size
104 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
105 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - BLOCK_MIN_BITS + 1) ///< possible block sizes
109 #define SCALEVLCBITS 8
110 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
111 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
112 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
113 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
114 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
116 static VLC sf_vlc; ///< scale factor DPCM vlc
117 static VLC sf_rl_vlc; ///< scale factor run length vlc
118 static VLC vec4_vlc; ///< 4 coefficients per symbol
119 static VLC vec2_vlc; ///< 2 coefficients per symbol
120 static VLC vec1_vlc; ///< 1 coefficient per symbol
121 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
122 static float sin64[33]; ///< sinus table for decorrelation
125 * @brief frame specific decoder context for a single channel
128 int16_t prev_block_len; ///< length of the previous block
129 uint8_t transmit_coefs;
130 uint8_t num_subframes;
131 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
132 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
133 uint8_t cur_subframe; ///< current subframe number
134 uint16_t decoded_samples; ///< number of already processed samples
135 uint8_t grouped; ///< channel is part of a group
136 int quant_step; ///< quantization step for the current subframe
137 int8_t reuse_sf; ///< share scale factors between subframes
138 int8_t scale_factor_step; ///< scaling step for the current subframe
139 int max_scale_factor; ///< maximum scale factor for the current subframe
140 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
141 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
142 int* scale_factors; ///< pointer to the scale factor values used for decoding
143 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
144 float* coeffs; ///< pointer to the subframe decode buffer
145 DECLARE_ALIGNED(16, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
149 * @brief channel group for channel transformations
152 uint8_t num_channels; ///< number of channels in the group
153 int8_t transform; ///< transform on / off
154 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
155 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
156 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
160 * @brief main decoder context
162 typedef struct WMAProDecodeCtx {
163 /* generic decoder variables */
164 AVCodecContext* avctx; ///< codec context for av_log
165 DSPContext dsp; ///< accelerated DSP functions
166 uint8_t frame_data[MAX_FRAMESIZE +
167 FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
168 PutBitContext pb; ///< context for filling the frame_data buffer
169 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
170 DECLARE_ALIGNED(16, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
171 float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
173 /* frame size dependent frame information (set during initialization) */
174 uint32_t decode_flags; ///< used compression features
175 uint8_t len_prefix; ///< frame is prefixed with its length
176 uint8_t dynamic_range_compression; ///< frame contains DRC data
177 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
178 uint16_t samples_per_frame; ///< number of samples to output
179 uint16_t log2_frame_size;
180 int8_t num_channels; ///< number of channels in the stream (same as AVCodecContext.num_channels)
181 int8_t lfe_channel; ///< lfe channel index
182 uint8_t max_num_subframes;
183 uint8_t subframe_len_bits; ///< number of bits used for the subframe length
184 uint8_t max_subframe_len_bit; ///< flag indicating that the subframe is of maximum size when the first subframe length bit is 1
185 uint16_t min_samples_per_subframe;
186 int8_t num_sfb[WMAPRO_BLOCK_SIZES]; ///< scale factor bands per block size
187 int16_t sfb_offsets[WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor band offsets (multiples of 4)
188 int8_t sf_offsets[WMAPRO_BLOCK_SIZES][WMAPRO_BLOCK_SIZES][MAX_BANDS]; ///< scale factor resample matrix
189 int16_t subwoofer_cutoffs[WMAPRO_BLOCK_SIZES]; ///< subwoofer cutoff values
191 /* packet decode state */
192 GetBitContext pgb; ///< bitstream reader context for the packet
193 uint8_t packet_offset; ///< frame offset in the packet
194 uint8_t packet_sequence_number; ///< current packet number
195 int num_saved_bits; ///< saved number of bits
196 int frame_offset; ///< frame offset in the bit reservoir
197 int subframe_offset; ///< subframe offset in the bit reservoir
198 uint8_t packet_loss; ///< set in case of bitstream error
199 uint8_t packet_done; ///< set when a packet is fully decoded
201 /* frame decode state */
202 uint32_t frame_num; ///< current frame number (not used for decoding)
203 GetBitContext gb; ///< bitstream reader context
204 int buf_bit_size; ///< buffer size in bits
205 float* samples; ///< current samplebuffer pointer
206 float* samples_end; ///< maximum samplebuffer pointer
207 uint8_t drc_gain; ///< gain for the DRC tool
208 int8_t skip_frame; ///< skip output step
209 int8_t parsed_all_subframes; ///< all subframes decoded?
211 /* subframe/block decode state */
212 int16_t subframe_len; ///< current subframe length
213 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
214 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
215 int8_t num_bands; ///< number of scale factor bands
216 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
217 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
218 int8_t esc_len; ///< length of escaped coefficients
220 uint8_t num_chgroups; ///< number of channel groups
221 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
223 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
228 *@brief helper function to print the most important members of the context
231 static void av_cold dump_context(WMAProDecodeCtx *s)
233 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
234 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %x\n", a, b);
236 PRINT("ed sample bit depth", s->bits_per_sample);
237 PRINT_HEX("ed decode flags", s->decode_flags);
238 PRINT("samples per frame", s->samples_per_frame);
239 PRINT("log2 frame size", s->log2_frame_size);
240 PRINT("max num subframes", s->max_num_subframes);
241 PRINT("len prefix", s->len_prefix);
242 PRINT("num channels", s->num_channels);
246 *@brief Uninitialize the decoder and free all resources.
247 *@param avctx codec context
248 *@return 0 on success, < 0 otherwise
250 static av_cold int decode_end(AVCodecContext *avctx)
252 WMAProDecodeCtx *s = avctx->priv_data;
255 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
256 ff_mdct_end(&s->mdct_ctx[i]);
262 *@brief Initialize the decoder.
263 *@param avctx codec context
264 *@return 0 on success, -1 otherwise
266 static av_cold int decode_init(AVCodecContext *avctx)
268 WMAProDecodeCtx *s = avctx->priv_data;
269 uint8_t *edata_ptr = avctx->extradata;
270 unsigned int channel_mask;
272 int log2_max_num_subframes;
273 int num_possible_block_sizes;
276 dsputil_init(&s->dsp, avctx);
277 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
279 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
281 if (avctx->extradata_size >= 18) {
282 s->decode_flags = AV_RL16(edata_ptr+14);
283 channel_mask = AV_RL32(edata_ptr+2);
284 s->bits_per_sample = AV_RL16(edata_ptr);
285 /** dump the extradata */
286 for (i = 0; i < avctx->extradata_size; i++)
287 dprintf(avctx, "[%x] ", avctx->extradata[i]);
288 dprintf(avctx, "\n");
291 av_log_ask_for_sample(avctx, "Unknown extradata size\n");
292 return AVERROR_INVALIDDATA;
296 s->log2_frame_size = av_log2(avctx->block_align) + 4;
299 s->skip_frame = 1; /* skip first frame */
301 s->len_prefix = (s->decode_flags & 0x40);
304 s->samples_per_frame = 1 << ff_wma_get_frame_len_bits(avctx->sample_rate,
307 /** init previous block len */
308 for (i = 0; i < avctx->channels; i++)
309 s->channel[i].prev_block_len = s->samples_per_frame;
312 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
313 s->max_num_subframes = 1 << log2_max_num_subframes;
314 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
315 s->max_subframe_len_bit = 1;
316 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
318 num_possible_block_sizes = log2_max_num_subframes + 1;
319 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
320 s->dynamic_range_compression = (s->decode_flags & 0x80);
322 if (s->max_num_subframes > MAX_SUBFRAMES) {
323 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %i\n",
324 s->max_num_subframes);
325 return AVERROR_INVALIDDATA;
328 s->num_channels = avctx->channels;
330 /** extract lfe channel position */
333 if (channel_mask & 8) {
335 for (mask = 1; mask < 16; mask <<= 1) {
336 if (channel_mask & mask)
341 if (s->num_channels < 0) {
342 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n", s->num_channels);
343 return AVERROR_INVALIDDATA;
344 } else if (s->num_channels > WMAPRO_MAX_CHANNELS) {
345 av_log_ask_for_sample(avctx, "unsupported number of channels\n");
346 return AVERROR_PATCHWELCOME;
349 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
350 scale_huffbits, 1, 1,
351 scale_huffcodes, 2, 2, 616);
353 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
354 scale_rl_huffbits, 1, 1,
355 scale_rl_huffcodes, 4, 4, 1406);
357 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
358 coef0_huffbits, 1, 1,
359 coef0_huffcodes, 4, 4, 2108);
361 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
362 coef1_huffbits, 1, 1,
363 coef1_huffcodes, 4, 4, 3912);
365 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
367 vec4_huffcodes, 2, 2, 604);
369 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
371 vec2_huffcodes, 2, 2, 562);
373 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
375 vec1_huffcodes, 2, 2, 562);
377 /** calculate number of scale factor bands and their offsets
378 for every possible block size */
379 for (i = 0; i < num_possible_block_sizes; i++) {
380 int subframe_len = s->samples_per_frame >> i;
384 s->sfb_offsets[i][0] = 0;
386 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
387 int offset = (subframe_len * 2 * critical_freq[x])
388 / s->avctx->sample_rate + 2;
390 if (offset > s->sfb_offsets[i][band - 1])
391 s->sfb_offsets[i][band++] = offset;
393 s->sfb_offsets[i][band - 1] = subframe_len;
394 s->num_sfb[i] = band - 1;
398 /** Scale factors can be shared between blocks of different size
399 as every block has a different scale factor band layout.
400 The matrix sf_offsets is needed to find the correct scale factor.
403 for (i = 0; i < num_possible_block_sizes; i++) {
405 for (b = 0; b < s->num_sfb[i]; b++) {
407 int offset = ((s->sfb_offsets[i][b]
408 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
409 for (x = 0; x < num_possible_block_sizes; x++) {
411 while (s->sfb_offsets[x][v + 1] << x < offset)
413 s->sf_offsets[i][x][b] = v;
418 /** init MDCT, FIXME: only init needed sizes */
419 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
420 ff_mdct_init(&s->mdct_ctx[i], BLOCK_MIN_BITS+1+i, 1,
421 1.0 / (1 << (BLOCK_MIN_BITS + i - 1))
422 / (1 << (s->bits_per_sample - 1)));
424 /** init MDCT windows: simple sinus window */
425 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
426 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
427 ff_init_ff_sine_windows(win_idx);
428 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
431 /** calculate subwoofer cutoff values */
432 for (i = 0; i < num_possible_block_sizes; i++) {
433 int block_size = s->samples_per_frame >> i;
434 int cutoff = (440*block_size + 3 * (s->avctx->sample_rate >> 1) - 1)
435 / s->avctx->sample_rate;
436 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
439 /** calculate sine values for the decorrelation matrix */
440 for (i = 0; i < 33; i++)
441 sin64[i] = sin(i*M_PI / 64.0);
443 if (avctx->debug & FF_DEBUG_BITSTREAM)
446 avctx->channel_layout = channel_mask;
451 *@brief Decode the subframe length.
453 *@param offset sample offset in the frame
454 *@return decoded subframe length on success, < 0 in case of an error
456 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
458 int frame_len_shift = 0;
461 /** no need to read from the bitstream when only one length is possible */
462 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
463 return s->min_samples_per_subframe;
465 /** 1 bit indicates if the subframe is of maximum length */
466 if (s->max_subframe_len_bit) {
467 if (get_bits1(&s->gb))
468 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
470 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
472 subframe_len = s->samples_per_frame >> frame_len_shift;
474 /** sanity check the length */
475 if (subframe_len < s->min_samples_per_subframe ||
476 subframe_len > s->samples_per_frame) {
477 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
479 return AVERROR_INVALIDDATA;
485 *@brief Decode how the data in the frame is split into subframes.
486 * Every WMA frame contains the encoded data for a fixed number of
487 * samples per channel. The data for every channel might be split
488 * into several subframes. This function will reconstruct the list of
489 * subframes for every channel.
491 * If the subframes are not evenly split, the algorithm estimates the
492 * channels with the lowest number of total samples.
493 * Afterwards, for each of these channels a bit is read from the
494 * bitstream that indicates if the channel contains a subframe with the
495 * next subframe size that is going to be read from the bitstream or not.
496 * If a channel contains such a subframe, the subframe size gets added to
497 * the channel's subframe list.
498 * The algorithm repeats these steps until the frame is properly divided
499 * between the individual channels.
502 *@return 0 on success, < 0 in case of an error
504 static int decode_tilehdr(WMAProDecodeCtx *s)
506 uint16_t num_samples[WMAPRO_MAX_CHANNELS]; /**< sum of samples for all currently known subframes of a channel */
507 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
508 int channels_for_cur_subframe = s->num_channels; /**< number of channels that contain the current subframe */
509 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
510 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
513 /* Should never consume more than 3073 bits (256 iterations for the
514 * while loop when always the minimum amount of 128 samples is substracted
515 * from missing samples in the 8 channel case).
516 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
519 /** reset tiling information */
520 for (c = 0; c < s->num_channels; c++)
521 s->channel[c].num_subframes = 0;
523 memset(num_samples, 0, sizeof(num_samples));
525 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
526 fixed_channel_layout = 1;
528 /** loop until the frame data is split between the subframes */
532 /** check which channels contain the subframe */
533 for (c = 0; c < s->num_channels; c++) {
534 if (num_samples[c] == min_channel_len) {
535 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
536 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
537 contains_subframe[c] = 1;
539 contains_subframe[c] = get_bits1(&s->gb);
541 contains_subframe[c] = 0;
544 /** get subframe length, subframe_len == 0 is not allowed */
545 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
546 return AVERROR_INVALIDDATA;
548 /** add subframes to the individual channels and find new min_channel_len */
549 min_channel_len += subframe_len;
550 for (c = 0; c < s->num_channels; c++) {
551 WMAProChannelCtx* chan = &s->channel[c];
553 if (contains_subframe[c]) {
554 if (chan->num_subframes >= MAX_SUBFRAMES) {
555 av_log(s->avctx, AV_LOG_ERROR,
556 "broken frame: num subframes > 31\n");
557 return AVERROR_INVALIDDATA;
559 chan->subframe_len[chan->num_subframes] = subframe_len;
560 num_samples[c] += subframe_len;
561 ++chan->num_subframes;
562 if (num_samples[c] > s->samples_per_frame) {
563 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
564 "channel len > samples_per_frame\n");
565 return AVERROR_INVALIDDATA;
567 } else if (num_samples[c] <= min_channel_len) {
568 if (num_samples[c] < min_channel_len) {
569 channels_for_cur_subframe = 0;
570 min_channel_len = num_samples[c];
572 ++channels_for_cur_subframe;
575 } while (min_channel_len < s->samples_per_frame);
577 for (c = 0; c < s->num_channels; c++) {
580 for (i = 0; i < s->channel[c].num_subframes; i++) {
581 dprintf(s->avctx, "frame[%i] channel[%i] subframe[%i]"
582 " len %i\n", s->frame_num, c, i,
583 s->channel[c].subframe_len[i]);
584 s->channel[c].subframe_offset[i] = offset;
585 offset += s->channel[c].subframe_len[i];
593 *@brief Calculate a decorrelation matrix from the bitstream parameters.
594 *@param s codec context
595 *@param chgroup channel group for which the matrix needs to be calculated
597 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
598 WMAProChannelGrp *chgroup)
602 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
603 memset(chgroup->decorrelation_matrix, 0, s->num_channels *
604 s->num_channels * sizeof(*chgroup->decorrelation_matrix));
606 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
607 rotation_offset[i] = get_bits(&s->gb, 6);
609 for (i = 0; i < chgroup->num_channels; i++)
610 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
611 get_bits1(&s->gb) ? 1.0 : -1.0;
613 for (i = 1; i < chgroup->num_channels; i++) {
615 for (x = 0; x < i; x++) {
617 for (y = 0; y < i + 1; y++) {
618 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
619 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
620 int n = rotation_offset[offset + x];
626 cosv = sin64[32 - n];
628 sinv = sin64[64 - n];
629 cosv = -sin64[n - 32];
632 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
633 (v1 * sinv) - (v2 * cosv);
634 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
635 (v1 * cosv) + (v2 * sinv);
643 *@brief Decode channel transformation parameters
644 *@param s codec context
645 *@return 0 in case of success, < 0 in case of bitstream errors
647 static int decode_channel_transform(WMAProDecodeCtx* s)
650 /* should never consume more than 1921 bits for the 8 channel case
651 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
652 * + MAX_CHANNELS + MAX_BANDS + 1)
655 /** in the one channel case channel transforms are pointless */
657 if (s->num_channels > 1) {
658 int remaining_channels = s->channels_for_cur_subframe;
660 if (get_bits1(&s->gb)) {
661 av_log_ask_for_sample(s->avctx,
662 "unsupported channel transform bit\n");
663 return AVERROR_INVALIDDATA;
666 for (s->num_chgroups = 0; remaining_channels &&
667 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
668 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
669 float** channel_data = chgroup->channel_data;
670 chgroup->num_channels = 0;
671 chgroup->transform = 0;
673 /** decode channel mask */
674 if (remaining_channels > 2) {
675 for (i = 0; i < s->channels_for_cur_subframe; i++) {
676 int channel_idx = s->channel_indexes_for_cur_subframe[i];
677 if (!s->channel[channel_idx].grouped
678 && get_bits1(&s->gb)) {
679 ++chgroup->num_channels;
680 s->channel[channel_idx].grouped = 1;
681 *channel_data++ = s->channel[channel_idx].coeffs;
685 chgroup->num_channels = remaining_channels;
686 for (i = 0; i < s->channels_for_cur_subframe; i++) {
687 int channel_idx = s->channel_indexes_for_cur_subframe[i];
688 if (!s->channel[channel_idx].grouped)
689 *channel_data++ = s->channel[channel_idx].coeffs;
690 s->channel[channel_idx].grouped = 1;
694 /** decode transform type */
695 if (chgroup->num_channels == 2) {
696 if (get_bits1(&s->gb)) {
697 if (get_bits1(&s->gb)) {
698 av_log_ask_for_sample(s->avctx,
699 "unsupported channel transform type\n");
702 chgroup->transform = 1;
703 if (s->num_channels == 2) {
704 chgroup->decorrelation_matrix[0] = 1.0;
705 chgroup->decorrelation_matrix[1] = -1.0;
706 chgroup->decorrelation_matrix[2] = 1.0;
707 chgroup->decorrelation_matrix[3] = 1.0;
710 chgroup->decorrelation_matrix[0] = 0.70703125;
711 chgroup->decorrelation_matrix[1] = -0.70703125;
712 chgroup->decorrelation_matrix[2] = 0.70703125;
713 chgroup->decorrelation_matrix[3] = 0.70703125;
716 } else if (chgroup->num_channels > 2) {
717 if (get_bits1(&s->gb)) {
718 chgroup->transform = 1;
719 if (get_bits1(&s->gb)) {
720 decode_decorrelation_matrix(s, chgroup);
722 /** FIXME: more than 6 coupled channels not supported */
723 if (chgroup->num_channels > 6) {
724 av_log_ask_for_sample(s->avctx,
725 "coupled channels > 6\n");
727 memcpy(chgroup->decorrelation_matrix,
728 default_decorrelation[chgroup->num_channels],
729 chgroup->num_channels * chgroup->num_channels *
730 sizeof(*chgroup->decorrelation_matrix));
736 /** decode transform on / off */
737 if (chgroup->transform) {
738 if (!get_bits1(&s->gb)) {
740 /** transform can be enabled for individual bands */
741 for (i = 0; i < s->num_bands; i++) {
742 chgroup->transform_band[i] = get_bits1(&s->gb);
745 memset(chgroup->transform_band, 1, s->num_bands);
748 remaining_channels -= chgroup->num_channels;
755 *@brief Extract the coefficients from the bitstream.
756 *@param s codec context
757 *@param c current channel number
758 *@return 0 on success, < 0 in case of bitstream errors
760 static int decode_coeffs(WMAProDecodeCtx *s, int c)
762 /* Integers 0..15 as single-precision floats. The table saves a
763 costly int to float conversion, and storing the values as
764 integers allows fast sign-flipping. */
765 static const int fval_tab[16] = {
766 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
767 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
768 0x41000000, 0x41100000, 0x41200000, 0x41300000,
769 0x41400000, 0x41500000, 0x41600000, 0x41700000,
773 WMAProChannelCtx* ci = &s->channel[c];
780 dprintf(s->avctx, "decode coefficients for channel %i\n", c);
782 vlctable = get_bits1(&s->gb);
783 vlc = &coef_vlc[vlctable];
793 /** decode vector coefficients (consumes up to 167 bits per iteration for
794 4 vector coded large values) */
795 while (!rl_mode && cur_coeff + 3 < s->subframe_len) {
800 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
802 if (idx == HUFF_VEC4_SIZE - 1) {
803 for (i = 0; i < 4; i += 2) {
804 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
805 if (idx == HUFF_VEC2_SIZE - 1) {
807 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
808 if (v0 == HUFF_VEC1_SIZE - 1)
809 v0 += ff_wma_get_large_val(&s->gb);
810 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
811 if (v1 == HUFF_VEC1_SIZE - 1)
812 v1 += ff_wma_get_large_val(&s->gb);
813 ((float*)vals)[i ] = v0;
814 ((float*)vals)[i+1] = v1;
816 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
817 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
821 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
822 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
823 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
824 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
828 for (i = 0; i < 4; i++) {
830 int sign = get_bits1(&s->gb) - 1;
831 *(uint32_t*)&ci->coeffs[cur_coeff] = vals[i] ^ sign<<31;
834 ci->coeffs[cur_coeff] = 0;
835 /** switch to run level mode when subframe_len / 128 zeros
836 were found in a row */
837 rl_mode |= (++num_zeros > s->subframe_len >> 8);
843 /** decode run level coded coefficients */
845 memset(&ci->coeffs[cur_coeff], 0,
846 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
847 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
848 level, run, 1, ci->coeffs,
849 cur_coeff, s->subframe_len,
850 s->subframe_len, s->esc_len, 0))
851 return AVERROR_INVALIDDATA;
858 *@brief Extract scale factors from the bitstream.
859 *@param s codec context
860 *@return 0 on success, < 0 in case of bitstream errors
862 static int decode_scale_factors(WMAProDecodeCtx* s)
866 /** should never consume more than 5344 bits
867 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
870 for (i = 0; i < s->channels_for_cur_subframe; i++) {
871 int c = s->channel_indexes_for_cur_subframe[i];
874 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
875 sf_end = s->channel[c].scale_factors + s->num_bands;
877 /** resample scale factors for the new block size
878 * as the scale factors might need to be resampled several times
879 * before some new values are transmitted, a backup of the last
880 * transmitted scale factors is kept in saved_scale_factors
882 if (s->channel[c].reuse_sf) {
883 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
885 for (b = 0; b < s->num_bands; b++)
886 s->channel[c].scale_factors[b] =
887 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
890 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
892 if (!s->channel[c].reuse_sf) {
894 /** decode DPCM coded scale factors */
895 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
896 val = 45 / s->channel[c].scale_factor_step;
897 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
898 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
903 /** run level decode differences to the resampled factors */
904 for (i = 0; i < s->num_bands; i++) {
910 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
913 uint32_t code = get_bits(&s->gb, 14);
915 sign = (code & 1) - 1;
916 skip = (code & 0x3f) >> 1;
917 } else if (idx == 1) {
920 skip = scale_rl_run[idx];
921 val = scale_rl_level[idx];
922 sign = get_bits1(&s->gb)-1;
926 if (i >= s->num_bands) {
927 av_log(s->avctx, AV_LOG_ERROR,
928 "invalid scale factor coding\n");
929 return AVERROR_INVALIDDATA;
931 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
935 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
936 s->channel[c].table_idx = s->table_idx;
937 s->channel[c].reuse_sf = 1;
940 /** calculate new scale factor maximum */
941 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
942 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
943 s->channel[c].max_scale_factor =
944 FFMAX(s->channel[c].max_scale_factor, *sf);
952 *@brief Reconstruct the individual channel data.
953 *@param s codec context
955 static void inverse_channel_transform(WMAProDecodeCtx *s)
959 for (i = 0; i < s->num_chgroups; i++) {
960 if (s->chgroup[i].transform) {
961 float data[WMAPRO_MAX_CHANNELS];
962 const int num_channels = s->chgroup[i].num_channels;
963 float** ch_data = s->chgroup[i].channel_data;
964 float** ch_end = ch_data + num_channels;
965 const int8_t* tb = s->chgroup[i].transform_band;
968 /** multichannel decorrelation */
969 for (sfb = s->cur_sfb_offsets;
970 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
973 /** multiply values with the decorrelation_matrix */
974 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
975 const float* mat = s->chgroup[i].decorrelation_matrix;
976 const float* data_end = data + num_channels;
977 float* data_ptr = data;
980 for (ch = ch_data; ch < ch_end; ch++)
981 *data_ptr++ = (*ch)[y];
983 for (ch = ch_data; ch < ch_end; ch++) {
986 while (data_ptr < data_end)
987 sum += *data_ptr++ * *mat++;
992 } else if (s->num_channels == 2) {
993 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
994 s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
997 s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
1007 *@brief Apply sine window and reconstruct the output buffer.
1008 *@param s codec context
1010 static void wmapro_window(WMAProDecodeCtx *s)
1013 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1014 int c = s->channel_indexes_for_cur_subframe[i];
1016 int winlen = s->channel[c].prev_block_len;
1017 float* start = s->channel[c].coeffs - (winlen >> 1);
1019 if (s->subframe_len < winlen) {
1020 start += (winlen - s->subframe_len) >> 1;
1021 winlen = s->subframe_len;
1024 window = s->windows[av_log2(winlen) - BLOCK_MIN_BITS];
1028 s->dsp.vector_fmul_window(start, start, start + winlen,
1031 s->channel[c].prev_block_len = s->subframe_len;
1036 *@brief Decode a single subframe (block).
1037 *@param s codec context
1038 *@return 0 on success, < 0 when decoding failed
1040 static int decode_subframe(WMAProDecodeCtx *s)
1042 int offset = s->samples_per_frame;
1043 int subframe_len = s->samples_per_frame;
1045 int total_samples = s->samples_per_frame * s->num_channels;
1046 int transmit_coeffs = 0;
1047 int cur_subwoofer_cutoff;
1049 s->subframe_offset = get_bits_count(&s->gb);
1051 /** reset channel context and find the next block offset and size
1052 == the next block of the channel with the smallest number of
1055 for (i = 0; i < s->num_channels; i++) {
1056 s->channel[i].grouped = 0;
1057 if (offset > s->channel[i].decoded_samples) {
1058 offset = s->channel[i].decoded_samples;
1060 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1065 "processing subframe with offset %i len %i\n", offset, subframe_len);
1067 /** get a list of all channels that contain the estimated block */
1068 s->channels_for_cur_subframe = 0;
1069 for (i = 0; i < s->num_channels; i++) {
1070 const int cur_subframe = s->channel[i].cur_subframe;
1071 /** substract already processed samples */
1072 total_samples -= s->channel[i].decoded_samples;
1074 /** and count if there are multiple subframes that match our profile */
1075 if (offset == s->channel[i].decoded_samples &&
1076 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1077 total_samples -= s->channel[i].subframe_len[cur_subframe];
1078 s->channel[i].decoded_samples +=
1079 s->channel[i].subframe_len[cur_subframe];
1080 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1081 ++s->channels_for_cur_subframe;
1085 /** check if the frame will be complete after processing the
1088 s->parsed_all_subframes = 1;
1091 dprintf(s->avctx, "subframe is part of %i channels\n",
1092 s->channels_for_cur_subframe);
1094 /** calculate number of scale factor bands and their offsets */
1095 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1096 s->num_bands = s->num_sfb[s->table_idx];
1097 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1098 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1100 /** configure the decoder for the current subframe */
1101 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1102 int c = s->channel_indexes_for_cur_subframe[i];
1104 s->channel[c].coeffs = &s->channel[c].out[(s->samples_per_frame >> 1)
1108 s->subframe_len = subframe_len;
1109 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1111 /** skip extended header if any */
1112 if (get_bits1(&s->gb)) {
1114 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1115 int len = get_bits(&s->gb, 4);
1116 num_fill_bits = get_bits(&s->gb, len) + 1;
1119 if (num_fill_bits >= 0) {
1120 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1121 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1122 return AVERROR_INVALIDDATA;
1125 skip_bits_long(&s->gb, num_fill_bits);
1129 /** no idea for what the following bit is used */
1130 if (get_bits1(&s->gb)) {
1131 av_log_ask_for_sample(s->avctx, "reserved bit set\n");
1132 return AVERROR_INVALIDDATA;
1136 if (decode_channel_transform(s) < 0)
1137 return AVERROR_INVALIDDATA;
1140 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1141 int c = s->channel_indexes_for_cur_subframe[i];
1142 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1143 transmit_coeffs = 1;
1146 if (transmit_coeffs) {
1148 int quant_step = 90 * s->bits_per_sample >> 4;
1149 if ((get_bits1(&s->gb))) {
1150 /** FIXME: might change run level mode decision */
1151 av_log_ask_for_sample(s->avctx, "unsupported quant step coding\n");
1152 return AVERROR_INVALIDDATA;
1154 /** decode quantization step */
1155 step = get_sbits(&s->gb, 6);
1157 if (step == -32 || step == 31) {
1158 const int sign = (step == 31) - 1;
1160 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1161 (step = get_bits(&s->gb, 5)) == 31) {
1164 quant_step += ((quant + step) ^ sign) - sign;
1166 if (quant_step < 0) {
1167 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1170 /** decode quantization step modifiers for every channel */
1172 if (s->channels_for_cur_subframe == 1) {
1173 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1175 int modifier_len = get_bits(&s->gb, 3);
1176 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1177 int c = s->channel_indexes_for_cur_subframe[i];
1178 s->channel[c].quant_step = quant_step;
1179 if (get_bits1(&s->gb)) {
1181 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1183 ++s->channel[c].quant_step;
1188 /** decode scale factors */
1189 if (decode_scale_factors(s) < 0)
1190 return AVERROR_INVALIDDATA;
1193 dprintf(s->avctx, "BITSTREAM: subframe header length was %i\n",
1194 get_bits_count(&s->gb) - s->subframe_offset);
1196 /** parse coefficients */
1197 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1198 int c = s->channel_indexes_for_cur_subframe[i];
1199 if (s->channel[c].transmit_coefs &&
1200 get_bits_count(&s->gb) < s->num_saved_bits) {
1201 decode_coeffs(s, c);
1203 memset(s->channel[c].coeffs, 0,
1204 sizeof(*s->channel[c].coeffs) * subframe_len);
1207 dprintf(s->avctx, "BITSTREAM: subframe length was %i\n",
1208 get_bits_count(&s->gb) - s->subframe_offset);
1210 if (transmit_coeffs) {
1211 /** reconstruct the per channel data */
1212 inverse_channel_transform(s);
1213 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1214 int c = s->channel_indexes_for_cur_subframe[i];
1215 const int* sf = s->channel[c].scale_factors;
1218 if (c == s->lfe_channel)
1219 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1220 (subframe_len - cur_subwoofer_cutoff));
1222 /** inverse quantization and rescaling */
1223 for (b = 0; b < s->num_bands; b++) {
1224 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1225 const int exp = s->channel[c].quant_step -
1226 (s->channel[c].max_scale_factor - *sf++) *
1227 s->channel[c].scale_factor_step;
1228 const float quant = pow(10.0, exp / 20.0);
1229 int start = s->cur_sfb_offsets[b];
1230 s->dsp.vector_fmul_scalar(s->tmp + start,
1231 s->channel[c].coeffs + start,
1232 quant, end - start);
1235 /** apply imdct (ff_imdct_half == DCTIV with reverse) */
1236 ff_imdct_half(&s->mdct_ctx[av_log2(subframe_len) - BLOCK_MIN_BITS],
1237 s->channel[c].coeffs, s->tmp);
1241 /** window and overlapp-add */
1244 /** handled one subframe */
1245 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1246 int c = s->channel_indexes_for_cur_subframe[i];
1247 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1248 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1249 return AVERROR_INVALIDDATA;
1251 ++s->channel[c].cur_subframe;
1258 *@brief Decode one WMA frame.
1259 *@param s codec context
1260 *@return 0 if the trailer bit indicates that this is the last frame,
1261 * 1 if there are additional frames
1263 static int decode_frame(WMAProDecodeCtx *s)
1265 GetBitContext* gb = &s->gb;
1266 int more_frames = 0;
1270 /** check for potential output buffer overflow */
1271 if (s->num_channels * s->samples_per_frame > s->samples_end - s->samples) {
1272 /** return an error if no frame could be decoded at all */
1273 av_log(s->avctx, AV_LOG_ERROR,
1274 "not enough space for the output samples\n");
1279 /** get frame length */
1281 len = get_bits(gb, s->log2_frame_size);
1283 dprintf(s->avctx, "decoding frame with length %x\n", len);
1285 /** decode tile information */
1286 if (decode_tilehdr(s)) {
1291 /** read postproc transform */
1292 if (s->num_channels > 1 && get_bits1(gb)) {
1293 if (get_bits1(gb)) {
1294 for (i = 0; i < s->num_channels * s->num_channels; i++)
1299 /** read drc info */
1300 if (s->dynamic_range_compression) {
1301 s->drc_gain = get_bits(gb, 8);
1302 dprintf(s->avctx, "drc_gain %i\n", s->drc_gain);
1305 /** no idea what these are for, might be the number of samples
1306 that need to be skipped at the beginning or end of a stream */
1307 if (get_bits1(gb)) {
1310 /** usually true for the first frame */
1311 if (get_bits1(gb)) {
1312 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1313 dprintf(s->avctx, "start skip: %i\n", skip);
1316 /** sometimes true for the last frame */
1317 if (get_bits1(gb)) {
1318 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1319 dprintf(s->avctx, "end skip: %i\n", skip);
1324 dprintf(s->avctx, "BITSTREAM: frame header length was %i\n",
1325 get_bits_count(gb) - s->frame_offset);
1327 /** reset subframe states */
1328 s->parsed_all_subframes = 0;
1329 for (i = 0; i < s->num_channels; i++) {
1330 s->channel[i].decoded_samples = 0;
1331 s->channel[i].cur_subframe = 0;
1332 s->channel[i].reuse_sf = 0;
1335 /** decode all subframes */
1336 while (!s->parsed_all_subframes) {
1337 if (decode_subframe(s) < 0) {
1343 /** interleave samples and write them to the output buffer */
1344 for (i = 0; i < s->num_channels; i++) {
1345 float* ptr = s->samples + i;
1346 int incr = s->num_channels;
1347 float* iptr = s->channel[i].out;
1348 float* iend = iptr + s->samples_per_frame;
1350 // FIXME should create/use a DSP function here
1351 while (iptr < iend) {
1356 /** reuse second half of the IMDCT output for the next frame */
1357 memcpy(&s->channel[i].out[0],
1358 &s->channel[i].out[s->samples_per_frame],
1359 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1362 if (s->skip_frame) {
1365 s->samples += s->num_channels * s->samples_per_frame;
1367 if (s->len_prefix) {
1368 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1369 /** FIXME: not sure if this is always an error */
1370 av_log(s->avctx, AV_LOG_ERROR,
1371 "frame[%i] would have to skip %i bits\n", s->frame_num,
1372 len - (get_bits_count(gb) - s->frame_offset) - 1);
1377 /** skip the rest of the frame data */
1378 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1380 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1384 /** decode trailer bit */
1385 more_frames = get_bits1(gb);
1392 *@brief Calculate remaining input buffer length.
1393 *@param s codec context
1394 *@param gb bitstream reader context
1395 *@return remaining size in bits
1397 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1399 return s->buf_bit_size - get_bits_count(gb);
1403 *@brief Fill the bit reservoir with a (partial) frame.
1404 *@param s codec context
1405 *@param gb bitstream reader context
1406 *@param len length of the partial frame
1407 *@param append decides wether to reset the buffer or not
1409 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1414 /** when the frame data does not need to be concatenated, the input buffer
1415 is resetted and additional bits from the previous frame are copyed
1416 and skipped later so that a fast byte copy is possible */
1419 s->frame_offset = get_bits_count(gb) & 7;
1420 s->num_saved_bits = s->frame_offset;
1421 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1424 buflen = (s->num_saved_bits + len + 8) >> 3;
1426 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1427 av_log_ask_for_sample(s->avctx, "input buffer too small\n");
1432 s->num_saved_bits += len;
1434 ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1437 int align = 8 - (get_bits_count(gb) & 7);
1438 align = FFMIN(align, len);
1439 put_bits(&s->pb, align, get_bits(gb, align));
1441 ff_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1443 skip_bits_long(gb, len);
1446 PutBitContext tmp = s->pb;
1447 flush_put_bits(&tmp);
1450 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1451 skip_bits(&s->gb, s->frame_offset);
1455 *@brief Decode a single WMA packet.
1456 *@param avctx codec context
1457 *@param data the output buffer
1458 *@param data_size number of bytes that were written to the output buffer
1459 *@param avpkt input packet
1460 *@return number of bytes that were read from the input buffer
1462 static int decode_packet(AVCodecContext *avctx,
1463 void *data, int *data_size, AVPacket* avpkt)
1465 WMAProDecodeCtx *s = avctx->priv_data;
1466 GetBitContext* gb = &s->pgb;
1467 const uint8_t* buf = avpkt->data;
1468 int buf_size = avpkt->size;
1469 int num_bits_prev_frame;
1470 int packet_sequence_number;
1473 s->samples_end = (float*)((int8_t*)data + *data_size);
1476 if (s->packet_done || s->packet_loss) {
1478 s->buf_bit_size = buf_size << 3;
1480 /** sanity check for the buffer length */
1481 if (buf_size < avctx->block_align)
1484 buf_size = avctx->block_align;
1486 /** parse packet header */
1487 init_get_bits(gb, buf, s->buf_bit_size);
1488 packet_sequence_number = get_bits(gb, 4);
1491 /** get number of bits that need to be added to the previous frame */
1492 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1493 dprintf(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1494 num_bits_prev_frame);
1496 /** check for packet loss */
1497 if (!s->packet_loss &&
1498 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1500 av_log(avctx, AV_LOG_ERROR, "Packet loss detected! seq %x vs %x\n",
1501 s->packet_sequence_number, packet_sequence_number);
1503 s->packet_sequence_number = packet_sequence_number;
1505 if (num_bits_prev_frame > 0) {
1506 /** append the previous frame data to the remaining data from the
1507 previous packet to create a full frame */
1508 save_bits(s, gb, num_bits_prev_frame, 1);
1509 dprintf(avctx, "accumulated %x bits of frame data\n",
1510 s->num_saved_bits - s->frame_offset);
1512 /** decode the cross packet frame if it is valid */
1513 if (!s->packet_loss)
1515 } else if (s->num_saved_bits - s->frame_offset) {
1516 dprintf(avctx, "ignoring %x previously saved bits\n",
1517 s->num_saved_bits - s->frame_offset);
1520 if (s->packet_loss) {
1521 /** reset number of saved bits so that the decoder
1522 does not start to decode incomplete frames in the
1523 s->len_prefix == 0 case */
1524 s->num_saved_bits = 0;
1530 s->buf_bit_size = avpkt->size << 3;
1531 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1532 skip_bits(gb, s->packet_offset);
1533 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1534 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1535 frame_size <= remaining_bits(s, gb)) {
1536 save_bits(s, gb, frame_size, 0);
1537 s->packet_done = !decode_frame(s);
1538 } else if (!s->len_prefix
1539 && s->num_saved_bits > get_bits_count(&s->gb)) {
1540 /** when the frames do not have a length prefix, we don't know
1541 the compressed length of the individual frames
1542 however, we know what part of a new packet belongs to the
1544 therefore we save the incoming packet first, then we append
1545 the "previous frame" data from the next packet so that
1546 we get a buffer that only contains full frames */
1547 s->packet_done = !decode_frame(s);
1552 if (s->packet_done && !s->packet_loss &&
1553 remaining_bits(s, gb) > 0) {
1554 /** save the rest of the data so that it can be decoded
1555 with the next packet */
1556 save_bits(s, gb, remaining_bits(s, gb), 0);
1559 *data_size = (int8_t *)s->samples - (int8_t *)data;
1560 s->packet_offset = get_bits_count(gb) & 7;
1562 return (s->packet_loss) ? AVERROR_INVALIDDATA : get_bits_count(gb) >> 3;
1566 *@brief Clear decoder buffers (for seeking).
1567 *@param avctx codec context
1569 static void flush(AVCodecContext *avctx)
1571 WMAProDecodeCtx *s = avctx->priv_data;
1573 /** reset output buffer as a part of it is used during the windowing of a
1575 for (i = 0; i < s->num_channels; i++)
1576 memset(s->channel[i].out, 0, s->samples_per_frame *
1577 sizeof(*s->channel[i].out));
1583 *@brief wmapro decoder
1585 AVCodec wmapro_decoder = {
1589 sizeof(WMAProDecodeCtx),
1594 .capabilities = CODEC_CAP_SUBFRAMES,
1596 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),