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.
91 #include "libavutil/ffmath.h"
92 #include "libavutil/float_dsp.h"
93 #include "libavutil/intfloat.h"
94 #include "libavutil/intreadwrite.h"
99 #include "wmaprodata.h"
102 #include "wma_common.h"
104 /** current decoder limitations */
105 #define WMAPRO_MAX_CHANNELS 8 ///< max number of handled channels
106 #define MAX_SUBFRAMES 32 ///< max number of subframes per channel
107 #define MAX_BANDS 29 ///< max number of scale factor bands
108 #define MAX_FRAMESIZE 32768 ///< maximum compressed frame size
110 #define WMAPRO_BLOCK_MIN_BITS 6 ///< log2 of min block size
111 #define WMAPRO_BLOCK_MAX_BITS 13 ///< log2 of max block size
112 #define WMAPRO_BLOCK_MIN_SIZE (1 << WMAPRO_BLOCK_MIN_BITS) ///< minimum block size
113 #define WMAPRO_BLOCK_MAX_SIZE (1 << WMAPRO_BLOCK_MAX_BITS) ///< maximum block size
114 #define WMAPRO_BLOCK_SIZES (WMAPRO_BLOCK_MAX_BITS - WMAPRO_BLOCK_MIN_BITS + 1) ///< possible block sizes
118 #define SCALEVLCBITS 8
119 #define VEC4MAXDEPTH ((HUFF_VEC4_MAXBITS+VLCBITS-1)/VLCBITS)
120 #define VEC2MAXDEPTH ((HUFF_VEC2_MAXBITS+VLCBITS-1)/VLCBITS)
121 #define VEC1MAXDEPTH ((HUFF_VEC1_MAXBITS+VLCBITS-1)/VLCBITS)
122 #define SCALEMAXDEPTH ((HUFF_SCALE_MAXBITS+SCALEVLCBITS-1)/SCALEVLCBITS)
123 #define SCALERLMAXDEPTH ((HUFF_SCALE_RL_MAXBITS+VLCBITS-1)/VLCBITS)
125 static VLC sf_vlc; ///< scale factor DPCM vlc
126 static VLC sf_rl_vlc; ///< scale factor run length vlc
127 static VLC vec4_vlc; ///< 4 coefficients per symbol
128 static VLC vec2_vlc; ///< 2 coefficients per symbol
129 static VLC vec1_vlc; ///< 1 coefficient per symbol
130 static VLC coef_vlc[2]; ///< coefficient run length vlc codes
131 static float sin64[33]; ///< sine table for decorrelation
134 * @brief frame specific decoder context for a single channel
136 typedef struct WMAProChannelCtx {
137 int16_t prev_block_len; ///< length of the previous block
138 uint8_t transmit_coefs;
139 uint8_t num_subframes;
140 uint16_t subframe_len[MAX_SUBFRAMES]; ///< subframe length in samples
141 uint16_t subframe_offset[MAX_SUBFRAMES]; ///< subframe positions in the current frame
142 uint8_t cur_subframe; ///< current subframe number
143 uint16_t decoded_samples; ///< number of already processed samples
144 uint8_t grouped; ///< channel is part of a group
145 int quant_step; ///< quantization step for the current subframe
146 int8_t reuse_sf; ///< share scale factors between subframes
147 int8_t scale_factor_step; ///< scaling step for the current subframe
148 int max_scale_factor; ///< maximum scale factor for the current subframe
149 int saved_scale_factors[2][MAX_BANDS]; ///< resampled and (previously) transmitted scale factor values
150 int8_t scale_factor_idx; ///< index for the transmitted scale factor values (used for resampling)
151 int* scale_factors; ///< pointer to the scale factor values used for decoding
152 uint8_t table_idx; ///< index in sf_offsets for the scale factor reference block
153 float* coeffs; ///< pointer to the subframe decode buffer
154 uint16_t num_vec_coeffs; ///< number of vector coded coefficients
155 DECLARE_ALIGNED(32, float, out)[WMAPRO_BLOCK_MAX_SIZE + WMAPRO_BLOCK_MAX_SIZE / 2]; ///< output buffer
159 * @brief channel group for channel transformations
161 typedef struct WMAProChannelGrp {
162 uint8_t num_channels; ///< number of channels in the group
163 int8_t transform; ///< transform on / off
164 int8_t transform_band[MAX_BANDS]; ///< controls if the transform is enabled for a certain band
165 float decorrelation_matrix[WMAPRO_MAX_CHANNELS*WMAPRO_MAX_CHANNELS];
166 float* channel_data[WMAPRO_MAX_CHANNELS]; ///< transformation coefficients
170 * @brief main decoder context
172 typedef struct WMAProDecodeCtx {
173 /* generic decoder variables */
174 AVCodecContext* avctx; ///< codec context for av_log
175 AVFloatDSPContext *fdsp;
176 uint8_t frame_data[MAX_FRAMESIZE +
177 AV_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
178 PutBitContext pb; ///< context for filling the frame_data buffer
179 FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
180 DECLARE_ALIGNED(32, float, tmp)[WMAPRO_BLOCK_MAX_SIZE]; ///< IMDCT output buffer
181 const float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes
183 /* frame size dependent frame information (set during initialization) */
184 uint32_t decode_flags; ///< used compression features
185 uint8_t len_prefix; ///< frame is prefixed with its length
186 uint8_t dynamic_range_compression; ///< frame contains DRC data
187 uint8_t bits_per_sample; ///< integer audio sample size for the unscaled IMDCT output (used to scale to [-1.0, 1.0])
188 uint16_t samples_per_frame; ///< number of samples to output
189 uint16_t log2_frame_size;
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
210 uint8_t skip_packets;
212 /* frame decode state */
213 uint32_t frame_num; ///< current frame number (not used for decoding)
215 GetBitContext gb; ///< bitstream reader context
216 int buf_bit_size; ///< buffer size in bits
217 uint8_t drc_gain; ///< gain for the DRC tool
218 int8_t skip_frame; ///< skip output step
219 int8_t parsed_all_subframes; ///< all subframes decoded?
221 /* subframe/block decode state */
222 int16_t subframe_len; ///< current subframe length
223 int8_t channels_for_cur_subframe; ///< number of channels that contain the subframe
224 int8_t channel_indexes_for_cur_subframe[WMAPRO_MAX_CHANNELS];
225 int8_t num_bands; ///< number of scale factor bands
226 int8_t transmit_num_vec_coeffs; ///< number of vector coded coefficients is part of the bitstream
227 int16_t* cur_sfb_offsets; ///< sfb offsets for the current block
228 uint8_t table_idx; ///< index for the num_sfb, sfb_offsets, sf_offsets and subwoofer_cutoffs tables
229 int8_t esc_len; ///< length of escaped coefficients
231 uint8_t num_chgroups; ///< number of channel groups
232 WMAProChannelGrp chgroup[WMAPRO_MAX_CHANNELS]; ///< channel group information
234 WMAProChannelCtx channel[WMAPRO_MAX_CHANNELS]; ///< per channel data
239 *@brief helper function to print the most important members of the context
242 static av_cold void dump_context(WMAProDecodeCtx *s)
244 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
245 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b);
247 PRINT("ed sample bit depth", s->bits_per_sample);
248 PRINT_HEX("ed decode flags", s->decode_flags);
249 PRINT("samples per frame", s->samples_per_frame);
250 PRINT("log2 frame size", s->log2_frame_size);
251 PRINT("max num subframes", s->max_num_subframes);
252 PRINT("len prefix", s->len_prefix);
253 PRINT("num channels", s->avctx->channels);
257 *@brief Uninitialize the decoder and free all resources.
258 *@param avctx codec context
259 *@return 0 on success, < 0 otherwise
261 static av_cold int decode_end(AVCodecContext *avctx)
263 WMAProDecodeCtx *s = avctx->priv_data;
268 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
269 ff_mdct_end(&s->mdct_ctx[i]);
274 static av_cold int get_rate(AVCodecContext *avctx)
276 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) { // XXX: is this really only for XMA?
277 if (avctx->sample_rate > 44100)
279 else if (avctx->sample_rate > 32000)
281 else if (avctx->sample_rate > 24000)
286 return avctx->sample_rate;
290 *@brief Initialize the decoder.
291 *@param avctx codec context
292 *@return 0 on success, -1 otherwise
294 static av_cold int decode_init(AVCodecContext *avctx)
296 WMAProDecodeCtx *s = avctx->priv_data;
297 uint8_t *edata_ptr = avctx->extradata;
298 unsigned int channel_mask;
300 int log2_max_num_subframes;
301 int num_possible_block_sizes;
303 if (avctx->codec_id == AV_CODEC_ID_XMA1 || avctx->codec_id == AV_CODEC_ID_XMA2)
304 avctx->block_align = 2048;
306 if (!avctx->block_align) {
307 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
308 return AVERROR(EINVAL);
312 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
314 return AVERROR(ENOMEM);
316 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
318 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
320 /** dump the extradata */
321 av_log(avctx, AV_LOG_DEBUG, "extradata:\n");
322 for (i = 0; i < avctx->extradata_size; i++)
323 av_log(avctx, AV_LOG_DEBUG, "[%x] ", avctx->extradata[i]);
324 av_log(avctx, AV_LOG_DEBUG, "\n");
325 if (avctx->codec_id == AV_CODEC_ID_XMA2 && (!avctx->extradata || avctx->extradata_size >= 6)) {
326 s->decode_flags = 0x10d6;
327 channel_mask = avctx->extradata ? AV_RL32(edata_ptr+2) : 0;
328 s->bits_per_sample = 16;
330 } else if (avctx->codec_id == AV_CODEC_ID_XMA1) {
331 s->decode_flags = 0x10d6;
332 s->bits_per_sample = 16;
334 } else if (avctx->codec_id == AV_CODEC_ID_WMAPRO && avctx->extradata_size >= 18) {
335 s->decode_flags = AV_RL16(edata_ptr+14);
336 channel_mask = AV_RL32(edata_ptr+2);
337 s->bits_per_sample = AV_RL16(edata_ptr);
339 if (s->bits_per_sample > 32 || s->bits_per_sample < 1) {
340 avpriv_request_sample(avctx, "bits per sample is %d", s->bits_per_sample);
341 return AVERROR_PATCHWELCOME;
344 avpriv_request_sample(avctx, "Unknown extradata size");
345 return AVERROR_PATCHWELCOME;
348 if (avctx->codec_id != AV_CODEC_ID_WMAPRO && avctx->channels > 2) {
349 avpriv_report_missing_feature(avctx, ">2 channels support");
350 return AVERROR_PATCHWELCOME;
354 s->log2_frame_size = av_log2(avctx->block_align) + 4;
355 if (s->log2_frame_size > 25) {
356 avpriv_request_sample(avctx, "Large block align");
357 return AVERROR_PATCHWELCOME;
361 if (avctx->codec_id != AV_CODEC_ID_WMAPRO)
364 s->skip_frame = 1; /* skip first frame */
367 s->len_prefix = (s->decode_flags & 0x40);
370 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
371 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
372 if (bits > WMAPRO_BLOCK_MAX_BITS) {
373 avpriv_request_sample(avctx, "14-bit block sizes");
374 return AVERROR_PATCHWELCOME;
376 s->samples_per_frame = 1 << bits;
378 s->samples_per_frame = 512;
382 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
383 s->max_num_subframes = 1 << log2_max_num_subframes;
384 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
385 s->max_subframe_len_bit = 1;
386 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
388 num_possible_block_sizes = log2_max_num_subframes + 1;
389 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
390 s->dynamic_range_compression = (s->decode_flags & 0x80);
392 if (s->max_num_subframes > MAX_SUBFRAMES) {
393 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n",
394 s->max_num_subframes);
395 return AVERROR_INVALIDDATA;
398 if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) {
399 av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
400 s->min_samples_per_subframe);
401 return AVERROR_INVALIDDATA;
404 if (s->avctx->sample_rate <= 0) {
405 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
406 return AVERROR_INVALIDDATA;
409 if (avctx->channels < 0) {
410 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
412 return AVERROR_INVALIDDATA;
413 } else if (avctx->channels > WMAPRO_MAX_CHANNELS) {
414 avpriv_request_sample(avctx,
415 "More than %d channels", WMAPRO_MAX_CHANNELS);
416 return AVERROR_PATCHWELCOME;
419 /** init previous block len */
420 for (i = 0; i < avctx->channels; i++)
421 s->channel[i].prev_block_len = s->samples_per_frame;
423 /** extract lfe channel position */
426 if (channel_mask & 8) {
428 for (mask = 1; mask < 16; mask <<= 1) {
429 if (channel_mask & mask)
434 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
435 scale_huffbits, 1, 1,
436 scale_huffcodes, 2, 2, 616);
438 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
439 scale_rl_huffbits, 1, 1,
440 scale_rl_huffcodes, 4, 4, 1406);
442 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
443 coef0_huffbits, 1, 1,
444 coef0_huffcodes, 4, 4, 2108);
446 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
447 coef1_huffbits, 1, 1,
448 coef1_huffcodes, 4, 4, 3912);
450 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
452 vec4_huffcodes, 2, 2, 604);
454 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
456 vec2_huffcodes, 2, 2, 562);
458 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
460 vec1_huffcodes, 2, 2, 562);
462 /** calculate number of scale factor bands and their offsets
463 for every possible block size */
464 for (i = 0; i < num_possible_block_sizes; i++) {
465 int subframe_len = s->samples_per_frame >> i;
468 int rate = get_rate(avctx);
470 s->sfb_offsets[i][0] = 0;
472 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
473 int offset = (subframe_len * 2 * critical_freq[x]) / rate + 2;
475 if (offset > s->sfb_offsets[i][band - 1])
476 s->sfb_offsets[i][band++] = offset;
478 if (offset >= subframe_len)
481 s->sfb_offsets[i][band - 1] = subframe_len;
482 s->num_sfb[i] = band - 1;
483 if (s->num_sfb[i] <= 0) {
484 av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n");
485 return AVERROR_INVALIDDATA;
490 /** Scale factors can be shared between blocks of different size
491 as every block has a different scale factor band layout.
492 The matrix sf_offsets is needed to find the correct scale factor.
495 for (i = 0; i < num_possible_block_sizes; i++) {
497 for (b = 0; b < s->num_sfb[i]; b++) {
499 int offset = ((s->sfb_offsets[i][b]
500 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
501 for (x = 0; x < num_possible_block_sizes; x++) {
503 while (s->sfb_offsets[x][v + 1] << x < offset) {
505 av_assert0(v < MAX_BANDS);
507 s->sf_offsets[i][x][b] = v;
512 /** init MDCT, FIXME: only init needed sizes */
513 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
514 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
515 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
516 / (1 << (s->bits_per_sample - 1)));
518 /** init MDCT windows: simple sine window */
519 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
520 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
521 ff_init_ff_sine_windows(win_idx);
522 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
525 /** calculate subwoofer cutoff values */
526 for (i = 0; i < num_possible_block_sizes; i++) {
527 int block_size = s->samples_per_frame >> i;
528 int cutoff = (440*block_size + 3LL * (s->avctx->sample_rate >> 1) - 1)
529 / s->avctx->sample_rate;
530 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
533 /** calculate sine values for the decorrelation matrix */
534 for (i = 0; i < 33; i++)
535 sin64[i] = sin(i*M_PI / 64.0);
537 if (avctx->debug & FF_DEBUG_BITSTREAM)
540 avctx->channel_layout = channel_mask;
546 *@brief Decode the subframe length.
548 *@param offset sample offset in the frame
549 *@return decoded subframe length on success, < 0 in case of an error
551 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
553 int frame_len_shift = 0;
556 /** no need to read from the bitstream when only one length is possible */
557 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
558 return s->min_samples_per_subframe;
560 if (get_bits_left(&s->gb) < 1)
561 return AVERROR_INVALIDDATA;
563 /** 1 bit indicates if the subframe is of maximum length */
564 if (s->max_subframe_len_bit) {
565 if (get_bits1(&s->gb))
566 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
568 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
570 subframe_len = s->samples_per_frame >> frame_len_shift;
572 /** sanity check the length */
573 if (subframe_len < s->min_samples_per_subframe ||
574 subframe_len > s->samples_per_frame) {
575 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
577 return AVERROR_INVALIDDATA;
583 *@brief Decode how the data in the frame is split into subframes.
584 * Every WMA frame contains the encoded data for a fixed number of
585 * samples per channel. The data for every channel might be split
586 * into several subframes. This function will reconstruct the list of
587 * subframes for every channel.
589 * If the subframes are not evenly split, the algorithm estimates the
590 * channels with the lowest number of total samples.
591 * Afterwards, for each of these channels a bit is read from the
592 * bitstream that indicates if the channel contains a subframe with the
593 * next subframe size that is going to be read from the bitstream or not.
594 * If a channel contains such a subframe, the subframe size gets added to
595 * the channel's subframe list.
596 * The algorithm repeats these steps until the frame is properly divided
597 * between the individual channels.
600 *@return 0 on success, < 0 in case of an error
602 static int decode_tilehdr(WMAProDecodeCtx *s)
604 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
605 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
606 int channels_for_cur_subframe = s->avctx->channels; /**< number of channels that contain the current subframe */
607 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
608 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
611 /* Should never consume more than 3073 bits (256 iterations for the
612 * while loop when always the minimum amount of 128 samples is subtracted
613 * from missing samples in the 8 channel case).
614 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
617 /** reset tiling information */
618 for (c = 0; c < s->avctx->channels; c++)
619 s->channel[c].num_subframes = 0;
621 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
622 fixed_channel_layout = 1;
624 /** loop until the frame data is split between the subframes */
628 /** check which channels contain the subframe */
629 for (c = 0; c < s->avctx->channels; c++) {
630 if (num_samples[c] == min_channel_len) {
631 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
632 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
633 contains_subframe[c] = 1;
635 contains_subframe[c] = get_bits1(&s->gb);
637 contains_subframe[c] = 0;
640 /** get subframe length, subframe_len == 0 is not allowed */
641 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
642 return AVERROR_INVALIDDATA;
644 /** add subframes to the individual channels and find new min_channel_len */
645 min_channel_len += subframe_len;
646 for (c = 0; c < s->avctx->channels; c++) {
647 WMAProChannelCtx* chan = &s->channel[c];
649 if (contains_subframe[c]) {
650 if (chan->num_subframes >= MAX_SUBFRAMES) {
651 av_log(s->avctx, AV_LOG_ERROR,
652 "broken frame: num subframes > 31\n");
653 return AVERROR_INVALIDDATA;
655 chan->subframe_len[chan->num_subframes] = subframe_len;
656 num_samples[c] += subframe_len;
657 ++chan->num_subframes;
658 if (num_samples[c] > s->samples_per_frame) {
659 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
660 "channel len > samples_per_frame\n");
661 return AVERROR_INVALIDDATA;
663 } else if (num_samples[c] <= min_channel_len) {
664 if (num_samples[c] < min_channel_len) {
665 channels_for_cur_subframe = 0;
666 min_channel_len = num_samples[c];
668 ++channels_for_cur_subframe;
671 } while (min_channel_len < s->samples_per_frame);
673 for (c = 0; c < s->avctx->channels; c++) {
676 for (i = 0; i < s->channel[c].num_subframes; i++) {
677 ff_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
678 " len %i\n", s->frame_num, c, i,
679 s->channel[c].subframe_len[i]);
680 s->channel[c].subframe_offset[i] = offset;
681 offset += s->channel[c].subframe_len[i];
689 *@brief Calculate a decorrelation matrix from the bitstream parameters.
690 *@param s codec context
691 *@param chgroup channel group for which the matrix needs to be calculated
693 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
694 WMAProChannelGrp *chgroup)
698 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
699 memset(chgroup->decorrelation_matrix, 0, s->avctx->channels *
700 s->avctx->channels * sizeof(*chgroup->decorrelation_matrix));
702 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
703 rotation_offset[i] = get_bits(&s->gb, 6);
705 for (i = 0; i < chgroup->num_channels; i++)
706 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
707 get_bits1(&s->gb) ? 1.0 : -1.0;
709 for (i = 1; i < chgroup->num_channels; i++) {
711 for (x = 0; x < i; x++) {
713 for (y = 0; y < i + 1; y++) {
714 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
715 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
716 int n = rotation_offset[offset + x];
722 cosv = sin64[32 - n];
724 sinv = sin64[64 - n];
725 cosv = -sin64[n - 32];
728 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
729 (v1 * sinv) - (v2 * cosv);
730 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
731 (v1 * cosv) + (v2 * sinv);
739 *@brief Decode channel transformation parameters
740 *@param s codec context
741 *@return >= 0 in case of success, < 0 in case of bitstream errors
743 static int decode_channel_transform(WMAProDecodeCtx* s)
746 /* should never consume more than 1921 bits for the 8 channel case
747 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
748 * + MAX_CHANNELS + MAX_BANDS + 1)
751 /** in the one channel case channel transforms are pointless */
753 if (s->avctx->channels > 1) {
754 int remaining_channels = s->channels_for_cur_subframe;
756 if (get_bits1(&s->gb)) {
757 avpriv_request_sample(s->avctx,
758 "Channel transform bit");
759 return AVERROR_PATCHWELCOME;
762 for (s->num_chgroups = 0; remaining_channels &&
763 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
764 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
765 float** channel_data = chgroup->channel_data;
766 chgroup->num_channels = 0;
767 chgroup->transform = 0;
769 /** decode channel mask */
770 if (remaining_channels > 2) {
771 for (i = 0; i < s->channels_for_cur_subframe; i++) {
772 int channel_idx = s->channel_indexes_for_cur_subframe[i];
773 if (!s->channel[channel_idx].grouped
774 && get_bits1(&s->gb)) {
775 ++chgroup->num_channels;
776 s->channel[channel_idx].grouped = 1;
777 *channel_data++ = s->channel[channel_idx].coeffs;
781 chgroup->num_channels = remaining_channels;
782 for (i = 0; i < s->channels_for_cur_subframe; i++) {
783 int channel_idx = s->channel_indexes_for_cur_subframe[i];
784 if (!s->channel[channel_idx].grouped)
785 *channel_data++ = s->channel[channel_idx].coeffs;
786 s->channel[channel_idx].grouped = 1;
790 /** decode transform type */
791 if (chgroup->num_channels == 2) {
792 if (get_bits1(&s->gb)) {
793 if (get_bits1(&s->gb)) {
794 avpriv_request_sample(s->avctx,
795 "Unknown channel transform type");
796 return AVERROR_PATCHWELCOME;
799 chgroup->transform = 1;
800 if (s->avctx->channels == 2) {
801 chgroup->decorrelation_matrix[0] = 1.0;
802 chgroup->decorrelation_matrix[1] = -1.0;
803 chgroup->decorrelation_matrix[2] = 1.0;
804 chgroup->decorrelation_matrix[3] = 1.0;
807 chgroup->decorrelation_matrix[0] = 0.70703125;
808 chgroup->decorrelation_matrix[1] = -0.70703125;
809 chgroup->decorrelation_matrix[2] = 0.70703125;
810 chgroup->decorrelation_matrix[3] = 0.70703125;
813 } else if (chgroup->num_channels > 2) {
814 if (get_bits1(&s->gb)) {
815 chgroup->transform = 1;
816 if (get_bits1(&s->gb)) {
817 decode_decorrelation_matrix(s, chgroup);
819 /** FIXME: more than 6 coupled channels not supported */
820 if (chgroup->num_channels > 6) {
821 avpriv_request_sample(s->avctx,
822 "Coupled channels > 6");
824 memcpy(chgroup->decorrelation_matrix,
825 default_decorrelation[chgroup->num_channels],
826 chgroup->num_channels * chgroup->num_channels *
827 sizeof(*chgroup->decorrelation_matrix));
833 /** decode transform on / off */
834 if (chgroup->transform) {
835 if (!get_bits1(&s->gb)) {
837 /** transform can be enabled for individual bands */
838 for (i = 0; i < s->num_bands; i++) {
839 chgroup->transform_band[i] = get_bits1(&s->gb);
842 memset(chgroup->transform_band, 1, s->num_bands);
845 remaining_channels -= chgroup->num_channels;
852 *@brief Extract the coefficients from the bitstream.
853 *@param s codec context
854 *@param c current channel number
855 *@return 0 on success, < 0 in case of bitstream errors
857 static int decode_coeffs(WMAProDecodeCtx *s, int c)
859 /* Integers 0..15 as single-precision floats. The table saves a
860 costly int to float conversion, and storing the values as
861 integers allows fast sign-flipping. */
862 static const uint32_t fval_tab[16] = {
863 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
864 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
865 0x41000000, 0x41100000, 0x41200000, 0x41300000,
866 0x41400000, 0x41500000, 0x41600000, 0x41700000,
870 WMAProChannelCtx* ci = &s->channel[c];
877 ff_dlog(s->avctx, "decode coefficients for channel %i\n", c);
879 vlctable = get_bits1(&s->gb);
880 vlc = &coef_vlc[vlctable];
890 /** decode vector coefficients (consumes up to 167 bits per iteration for
891 4 vector coded large values) */
892 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
893 (cur_coeff + 3 < ci->num_vec_coeffs)) {
898 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
900 if (idx == HUFF_VEC4_SIZE - 1) {
901 for (i = 0; i < 4; i += 2) {
902 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
903 if (idx == HUFF_VEC2_SIZE - 1) {
905 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
906 if (v0 == HUFF_VEC1_SIZE - 1)
907 v0 += ff_wma_get_large_val(&s->gb);
908 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
909 if (v1 == HUFF_VEC1_SIZE - 1)
910 v1 += ff_wma_get_large_val(&s->gb);
911 vals[i ] = av_float2int(v0);
912 vals[i+1] = av_float2int(v1);
914 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
915 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
919 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
920 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
921 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
922 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
926 for (i = 0; i < 4; i++) {
928 uint32_t sign = get_bits1(&s->gb) - 1;
929 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
932 ci->coeffs[cur_coeff] = 0;
933 /** switch to run level mode when subframe_len / 128 zeros
934 were found in a row */
935 rl_mode |= (++num_zeros > s->subframe_len >> 8);
941 /** decode run level coded coefficients */
942 if (cur_coeff < s->subframe_len) {
943 memset(&ci->coeffs[cur_coeff], 0,
944 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
945 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
946 level, run, 1, ci->coeffs,
947 cur_coeff, s->subframe_len,
948 s->subframe_len, s->esc_len, 0))
949 return AVERROR_INVALIDDATA;
956 *@brief Extract scale factors from the bitstream.
957 *@param s codec context
958 *@return 0 on success, < 0 in case of bitstream errors
960 static int decode_scale_factors(WMAProDecodeCtx* s)
964 /** should never consume more than 5344 bits
965 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
968 for (i = 0; i < s->channels_for_cur_subframe; i++) {
969 int c = s->channel_indexes_for_cur_subframe[i];
972 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
973 sf_end = s->channel[c].scale_factors + s->num_bands;
975 /** resample scale factors for the new block size
976 * as the scale factors might need to be resampled several times
977 * before some new values are transmitted, a backup of the last
978 * transmitted scale factors is kept in saved_scale_factors
980 if (s->channel[c].reuse_sf) {
981 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
983 for (b = 0; b < s->num_bands; b++)
984 s->channel[c].scale_factors[b] =
985 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
988 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
990 if (!s->channel[c].reuse_sf) {
992 /** decode DPCM coded scale factors */
993 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
994 val = 45 / s->channel[c].scale_factor_step;
995 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
996 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
1001 /** run level decode differences to the resampled factors */
1002 for (i = 0; i < s->num_bands; i++) {
1008 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
1011 uint32_t code = get_bits(&s->gb, 14);
1013 sign = (code & 1) - 1;
1014 skip = (code & 0x3f) >> 1;
1015 } else if (idx == 1) {
1018 skip = scale_rl_run[idx];
1019 val = scale_rl_level[idx];
1020 sign = get_bits1(&s->gb)-1;
1024 if (i >= s->num_bands) {
1025 av_log(s->avctx, AV_LOG_ERROR,
1026 "invalid scale factor coding\n");
1027 return AVERROR_INVALIDDATA;
1029 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
1033 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
1034 s->channel[c].table_idx = s->table_idx;
1035 s->channel[c].reuse_sf = 1;
1038 /** calculate new scale factor maximum */
1039 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
1040 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
1041 s->channel[c].max_scale_factor =
1042 FFMAX(s->channel[c].max_scale_factor, *sf);
1050 *@brief Reconstruct the individual channel data.
1051 *@param s codec context
1053 static void inverse_channel_transform(WMAProDecodeCtx *s)
1057 for (i = 0; i < s->num_chgroups; i++) {
1058 if (s->chgroup[i].transform) {
1059 float data[WMAPRO_MAX_CHANNELS];
1060 const int num_channels = s->chgroup[i].num_channels;
1061 float** ch_data = s->chgroup[i].channel_data;
1062 float** ch_end = ch_data + num_channels;
1063 const int8_t* tb = s->chgroup[i].transform_band;
1066 /** multichannel decorrelation */
1067 for (sfb = s->cur_sfb_offsets;
1068 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1071 /** multiply values with the decorrelation_matrix */
1072 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1073 const float* mat = s->chgroup[i].decorrelation_matrix;
1074 const float* data_end = data + num_channels;
1075 float* data_ptr = data;
1078 for (ch = ch_data; ch < ch_end; ch++)
1079 *data_ptr++ = (*ch)[y];
1081 for (ch = ch_data; ch < ch_end; ch++) {
1084 while (data_ptr < data_end)
1085 sum += *data_ptr++ * *mat++;
1090 } else if (s->avctx->channels == 2) {
1091 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1092 s->fdsp->vector_fmul_scalar(ch_data[0] + sfb[0],
1093 ch_data[0] + sfb[0],
1095 s->fdsp->vector_fmul_scalar(ch_data[1] + sfb[0],
1096 ch_data[1] + sfb[0],
1105 *@brief Apply sine window and reconstruct the output buffer.
1106 *@param s codec context
1108 static void wmapro_window(WMAProDecodeCtx *s)
1111 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1112 int c = s->channel_indexes_for_cur_subframe[i];
1113 const float* window;
1114 int winlen = s->channel[c].prev_block_len;
1115 float* start = s->channel[c].coeffs - (winlen >> 1);
1117 if (s->subframe_len < winlen) {
1118 start += (winlen - s->subframe_len) >> 1;
1119 winlen = s->subframe_len;
1122 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1126 s->fdsp->vector_fmul_window(start, start, start + winlen,
1129 s->channel[c].prev_block_len = s->subframe_len;
1134 *@brief Decode a single subframe (block).
1135 *@param s codec context
1136 *@return 0 on success, < 0 when decoding failed
1138 static int decode_subframe(WMAProDecodeCtx *s)
1140 int offset = s->samples_per_frame;
1141 int subframe_len = s->samples_per_frame;
1143 int total_samples = s->samples_per_frame * s->avctx->channels;
1144 int transmit_coeffs = 0;
1145 int cur_subwoofer_cutoff;
1147 s->subframe_offset = get_bits_count(&s->gb);
1149 /** reset channel context and find the next block offset and size
1150 == the next block of the channel with the smallest number of
1153 for (i = 0; i < s->avctx->channels; i++) {
1154 s->channel[i].grouped = 0;
1155 if (offset > s->channel[i].decoded_samples) {
1156 offset = s->channel[i].decoded_samples;
1158 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1163 "processing subframe with offset %i len %i\n", offset, subframe_len);
1165 /** get a list of all channels that contain the estimated block */
1166 s->channels_for_cur_subframe = 0;
1167 for (i = 0; i < s->avctx->channels; i++) {
1168 const int cur_subframe = s->channel[i].cur_subframe;
1169 /** subtract already processed samples */
1170 total_samples -= s->channel[i].decoded_samples;
1172 /** and count if there are multiple subframes that match our profile */
1173 if (offset == s->channel[i].decoded_samples &&
1174 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1175 total_samples -= s->channel[i].subframe_len[cur_subframe];
1176 s->channel[i].decoded_samples +=
1177 s->channel[i].subframe_len[cur_subframe];
1178 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1179 ++s->channels_for_cur_subframe;
1183 /** check if the frame will be complete after processing the
1186 s->parsed_all_subframes = 1;
1189 ff_dlog(s->avctx, "subframe is part of %i channels\n",
1190 s->channels_for_cur_subframe);
1192 /** calculate number of scale factor bands and their offsets */
1193 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1194 s->num_bands = s->num_sfb[s->table_idx];
1195 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1196 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1198 /** configure the decoder for the current subframe */
1199 offset += s->samples_per_frame >> 1;
1201 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1202 int c = s->channel_indexes_for_cur_subframe[i];
1204 s->channel[c].coeffs = &s->channel[c].out[offset];
1207 s->subframe_len = subframe_len;
1208 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1210 /** skip extended header if any */
1211 if (get_bits1(&s->gb)) {
1213 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1214 int len = get_bits(&s->gb, 4);
1215 num_fill_bits = get_bitsz(&s->gb, len) + 1;
1218 if (num_fill_bits >= 0) {
1219 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1220 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1221 return AVERROR_INVALIDDATA;
1224 skip_bits_long(&s->gb, num_fill_bits);
1228 /** no idea for what the following bit is used */
1229 if (get_bits1(&s->gb)) {
1230 avpriv_request_sample(s->avctx, "Reserved bit");
1231 return AVERROR_PATCHWELCOME;
1235 if (decode_channel_transform(s) < 0)
1236 return AVERROR_INVALIDDATA;
1239 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1240 int c = s->channel_indexes_for_cur_subframe[i];
1241 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1242 transmit_coeffs = 1;
1245 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1246 if (transmit_coeffs) {
1248 int quant_step = 90 * s->bits_per_sample >> 4;
1250 /** decode number of vector coded coefficients */
1251 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1252 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1253 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1254 int c = s->channel_indexes_for_cur_subframe[i];
1255 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1256 if (num_vec_coeffs > s->subframe_len) {
1257 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1258 return AVERROR_INVALIDDATA;
1260 av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out));
1261 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1264 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1265 int c = s->channel_indexes_for_cur_subframe[i];
1266 s->channel[c].num_vec_coeffs = s->subframe_len;
1269 /** decode quantization step */
1270 step = get_sbits(&s->gb, 6);
1272 if (step == -32 || step == 31) {
1273 const int sign = (step == 31) - 1;
1275 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1276 (step = get_bits(&s->gb, 5)) == 31) {
1279 quant_step += ((quant + step) ^ sign) - sign;
1281 if (quant_step < 0) {
1282 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1285 /** decode quantization step modifiers for every channel */
1287 if (s->channels_for_cur_subframe == 1) {
1288 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1290 int modifier_len = get_bits(&s->gb, 3);
1291 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1292 int c = s->channel_indexes_for_cur_subframe[i];
1293 s->channel[c].quant_step = quant_step;
1294 if (get_bits1(&s->gb)) {
1296 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1298 ++s->channel[c].quant_step;
1303 /** decode scale factors */
1304 if (decode_scale_factors(s) < 0)
1305 return AVERROR_INVALIDDATA;
1308 ff_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1309 get_bits_count(&s->gb) - s->subframe_offset);
1311 /** parse coefficients */
1312 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1313 int c = s->channel_indexes_for_cur_subframe[i];
1314 if (s->channel[c].transmit_coefs &&
1315 get_bits_count(&s->gb) < s->num_saved_bits) {
1316 decode_coeffs(s, c);
1318 memset(s->channel[c].coeffs, 0,
1319 sizeof(*s->channel[c].coeffs) * subframe_len);
1322 ff_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1323 get_bits_count(&s->gb) - s->subframe_offset);
1325 if (transmit_coeffs) {
1326 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1327 /** reconstruct the per channel data */
1328 inverse_channel_transform(s);
1329 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1330 int c = s->channel_indexes_for_cur_subframe[i];
1331 const int* sf = s->channel[c].scale_factors;
1334 if (c == s->lfe_channel)
1335 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1336 (subframe_len - cur_subwoofer_cutoff));
1338 /** inverse quantization and rescaling */
1339 for (b = 0; b < s->num_bands; b++) {
1340 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1341 const int exp = s->channel[c].quant_step -
1342 (s->channel[c].max_scale_factor - *sf++) *
1343 s->channel[c].scale_factor_step;
1344 const float quant = ff_exp10(exp / 20.0);
1345 int start = s->cur_sfb_offsets[b];
1346 s->fdsp->vector_fmul_scalar(s->tmp + start,
1347 s->channel[c].coeffs + start,
1348 quant, end - start);
1351 /** apply imdct (imdct_half == DCTIV with reverse) */
1352 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1356 /** window and overlapp-add */
1359 /** handled one subframe */
1360 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1361 int c = s->channel_indexes_for_cur_subframe[i];
1362 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1363 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1364 return AVERROR_INVALIDDATA;
1366 ++s->channel[c].cur_subframe;
1373 *@brief Decode one WMA frame.
1374 *@param s codec context
1375 *@return 0 if the trailer bit indicates that this is the last frame,
1376 * 1 if there are additional frames
1378 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1380 AVCodecContext *avctx = s->avctx;
1381 GetBitContext* gb = &s->gb;
1382 int more_frames = 0;
1386 /** get frame length */
1388 len = get_bits(gb, s->log2_frame_size);
1390 ff_dlog(s->avctx, "decoding frame with length %x\n", len);
1392 /** decode tile information */
1393 if (decode_tilehdr(s)) {
1398 /** read postproc transform */
1399 if (s->avctx->channels > 1 && get_bits1(gb)) {
1400 if (get_bits1(gb)) {
1401 for (i = 0; i < avctx->channels * avctx->channels; i++)
1406 /** read drc info */
1407 if (s->dynamic_range_compression) {
1408 s->drc_gain = get_bits(gb, 8);
1409 ff_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1412 /** no idea what these are for, might be the number of samples
1413 that need to be skipped at the beginning or end of a stream */
1414 if (get_bits1(gb)) {
1417 /** usually true for the first frame */
1418 if (get_bits1(gb)) {
1419 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1420 ff_dlog(s->avctx, "start skip: %i\n", skip);
1423 /** sometimes true for the last frame */
1424 if (get_bits1(gb)) {
1425 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1426 ff_dlog(s->avctx, "end skip: %i\n", skip);
1431 ff_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1432 get_bits_count(gb) - s->frame_offset);
1434 /** reset subframe states */
1435 s->parsed_all_subframes = 0;
1436 for (i = 0; i < avctx->channels; i++) {
1437 s->channel[i].decoded_samples = 0;
1438 s->channel[i].cur_subframe = 0;
1439 s->channel[i].reuse_sf = 0;
1442 /** decode all subframes */
1443 while (!s->parsed_all_subframes) {
1444 if (decode_subframe(s) < 0) {
1450 /* get output buffer */
1451 frame->nb_samples = s->samples_per_frame;
1452 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1457 /** copy samples to the output buffer */
1458 for (i = 0; i < avctx->channels; i++)
1459 memcpy(frame->extended_data[i], s->channel[i].out,
1460 s->samples_per_frame * sizeof(*s->channel[i].out));
1462 for (i = 0; i < avctx->channels; i++) {
1463 /** reuse second half of the IMDCT output for the next frame */
1464 memcpy(&s->channel[i].out[0],
1465 &s->channel[i].out[s->samples_per_frame],
1466 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1469 if (s->skip_frame) {
1472 av_frame_unref(frame);
1477 if (s->len_prefix) {
1478 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1479 /** FIXME: not sure if this is always an error */
1480 av_log(s->avctx, AV_LOG_ERROR,
1481 "frame[%"PRIu32"] would have to skip %i bits\n",
1483 len - (get_bits_count(gb) - s->frame_offset) - 1);
1488 /** skip the rest of the frame data */
1489 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1491 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1495 /** decode trailer bit */
1496 more_frames = get_bits1(gb);
1503 *@brief Calculate remaining input buffer length.
1504 *@param s codec context
1505 *@param gb bitstream reader context
1506 *@return remaining size in bits
1508 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1510 return s->buf_bit_size - get_bits_count(gb);
1514 *@brief Fill the bit reservoir with a (partial) frame.
1515 *@param s codec context
1516 *@param gb bitstream reader context
1517 *@param len length of the partial frame
1518 *@param append decides whether to reset the buffer or not
1520 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1525 /** when the frame data does not need to be concatenated, the input buffer
1526 is reset and additional bits from the previous frame are copied
1527 and skipped later so that a fast byte copy is possible */
1530 s->frame_offset = get_bits_count(gb) & 7;
1531 s->num_saved_bits = s->frame_offset;
1532 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1535 buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1537 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1538 avpriv_request_sample(s->avctx, "Too small input buffer");
1543 av_assert0(len <= put_bits_left(&s->pb));
1545 s->num_saved_bits += len;
1547 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1550 int align = 8 - (get_bits_count(gb) & 7);
1551 align = FFMIN(align, len);
1552 put_bits(&s->pb, align, get_bits(gb, align));
1554 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1556 skip_bits_long(gb, len);
1559 PutBitContext tmp = s->pb;
1560 flush_put_bits(&tmp);
1563 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1564 skip_bits(&s->gb, s->frame_offset);
1568 *@brief Decode a single WMA packet.
1569 *@param avctx codec context
1570 *@param data the output buffer
1571 *@param avpkt input packet
1572 *@return number of bytes that were read from the input buffer
1574 static int decode_packet(AVCodecContext *avctx, void *data,
1575 int *got_frame_ptr, AVPacket* avpkt)
1577 WMAProDecodeCtx *s = avctx->priv_data;
1578 GetBitContext* gb = &s->pgb;
1579 const uint8_t* buf = avpkt->data;
1580 int buf_size = avpkt->size;
1581 int num_bits_prev_frame;
1582 int packet_sequence_number;
1586 if (s->skip_packets > 0) {
1588 return FFMIN(avpkt->size, avctx->block_align);
1591 if (s->packet_done || s->packet_loss) {
1594 /** sanity check for the buffer length */
1595 if (avctx->codec_id == AV_CODEC_ID_WMAPRO && buf_size < avctx->block_align) {
1596 av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1597 buf_size, avctx->block_align);
1598 return AVERROR_INVALIDDATA;
1601 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
1602 s->next_packet_start = buf_size - avctx->block_align;
1603 buf_size = avctx->block_align;
1605 s->next_packet_start = buf_size - FFMIN(buf_size, avctx->block_align);
1606 buf_size = FFMIN(buf_size, avctx->block_align);
1608 s->buf_bit_size = buf_size << 3;
1610 /** parse packet header */
1611 init_get_bits(gb, buf, s->buf_bit_size);
1612 if (avctx->codec_id != AV_CODEC_ID_XMA2) {
1613 packet_sequence_number = get_bits(gb, 4);
1616 s->num_frames = get_bits(gb, 6);
1617 packet_sequence_number = 0;
1620 /** get number of bits that need to be added to the previous frame */
1621 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1622 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) {
1624 s->skip_packets = get_bits(gb, 8);
1627 ff_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1628 num_bits_prev_frame);
1630 /** check for packet loss */
1631 if (avctx->codec_id != AV_CODEC_ID_XMA2 && !s->packet_loss &&
1632 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1634 av_log(avctx, AV_LOG_ERROR,
1635 "Packet loss detected! seq %"PRIx8" vs %x\n",
1636 s->packet_sequence_number, packet_sequence_number);
1638 s->packet_sequence_number = packet_sequence_number;
1640 if (num_bits_prev_frame > 0) {
1641 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1642 if (num_bits_prev_frame >= remaining_packet_bits) {
1643 num_bits_prev_frame = remaining_packet_bits;
1647 /** append the previous frame data to the remaining data from the
1648 previous packet to create a full frame */
1649 save_bits(s, gb, num_bits_prev_frame, 1);
1650 ff_dlog(avctx, "accumulated %x bits of frame data\n",
1651 s->num_saved_bits - s->frame_offset);
1653 /** decode the cross packet frame if it is valid */
1654 if (!s->packet_loss)
1655 decode_frame(s, data, got_frame_ptr);
1656 } else if (s->num_saved_bits - s->frame_offset) {
1657 ff_dlog(avctx, "ignoring %x previously saved bits\n",
1658 s->num_saved_bits - s->frame_offset);
1661 if (s->packet_loss) {
1662 /** reset number of saved bits so that the decoder
1663 does not start to decode incomplete frames in the
1664 s->len_prefix == 0 case */
1665 s->num_saved_bits = 0;
1671 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1672 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1673 skip_bits(gb, s->packet_offset);
1674 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1675 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1676 frame_size <= remaining_bits(s, gb)) {
1677 save_bits(s, gb, frame_size, 0);
1678 if (!s->packet_loss)
1679 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1680 } else if (!s->len_prefix
1681 && s->num_saved_bits > get_bits_count(&s->gb)) {
1682 /** when the frames do not have a length prefix, we don't know
1683 the compressed length of the individual frames
1684 however, we know what part of a new packet belongs to the
1686 therefore we save the incoming packet first, then we append
1687 the "previous frame" data from the next packet so that
1688 we get a buffer that only contains full frames */
1689 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1694 if (remaining_bits(s, gb) < 0) {
1695 av_log(avctx, AV_LOG_ERROR, "Overread %d\n", -remaining_bits(s, gb));
1699 if (s->packet_done && !s->packet_loss &&
1700 remaining_bits(s, gb) > 0) {
1701 /** save the rest of the data so that it can be decoded
1702 with the next packet */
1703 save_bits(s, gb, remaining_bits(s, gb), 0);
1706 s->packet_offset = get_bits_count(gb) & 7;
1708 return AVERROR_INVALIDDATA;
1710 return get_bits_count(gb) >> 3;
1714 *@brief Clear decoder buffers (for seeking).
1715 *@param avctx codec context
1717 static void flush(AVCodecContext *avctx)
1719 WMAProDecodeCtx *s = avctx->priv_data;
1721 /** reset output buffer as a part of it is used during the windowing of a
1723 for (i = 0; i < avctx->channels; i++)
1724 memset(s->channel[i].out, 0, s->samples_per_frame *
1725 sizeof(*s->channel[i].out));
1731 *@brief wmapro decoder
1733 AVCodec ff_wmapro_decoder = {
1735 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1736 .type = AVMEDIA_TYPE_AUDIO,
1737 .id = AV_CODEC_ID_WMAPRO,
1738 .priv_data_size = sizeof(WMAProDecodeCtx),
1739 .init = decode_init,
1740 .close = decode_end,
1741 .decode = decode_packet,
1742 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1744 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1745 AV_SAMPLE_FMT_NONE },
1748 AVCodec ff_xma1_decoder = {
1750 .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 1"),
1751 .type = AVMEDIA_TYPE_AUDIO,
1752 .id = AV_CODEC_ID_XMA1,
1753 .priv_data_size = sizeof(WMAProDecodeCtx),
1754 .init = decode_init,
1755 .close = decode_end,
1756 .decode = decode_packet,
1757 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1759 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1760 AV_SAMPLE_FMT_NONE },
1763 AVCodec ff_xma2_decoder = {
1765 .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 2"),
1766 .type = AVMEDIA_TYPE_AUDIO,
1767 .id = AV_CODEC_ID_XMA2,
1768 .priv_data_size = sizeof(WMAProDecodeCtx),
1769 .init = decode_init,
1770 .close = decode_end,
1771 .decode = decode_packet,
1772 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1774 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1775 AV_SAMPLE_FMT_NONE },