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
211 /* frame decode state */
212 uint32_t frame_num; ///< current frame number (not used for decoding)
213 GetBitContext gb; ///< bitstream reader context
214 int buf_bit_size; ///< buffer size in bits
215 uint8_t drc_gain; ///< gain for the DRC tool
216 int8_t skip_frame; ///< skip output step
217 int8_t parsed_all_subframes; ///< all subframes decoded?
218 uint8_t skip_packets;
220 /* subframe/block decode state */
221 int16_t subframe_len; ///< current subframe length
222 int8_t nb_channels; ///< number of channels in stream (XMA1/2)
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
237 typedef struct XMADecodeCtx {
238 WMAProDecodeCtx xma[4];
241 float samples[8][512 * 64];
246 *@brief helper function to print the most important members of the context
249 static av_cold void dump_context(WMAProDecodeCtx *s)
251 #define PRINT(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %d\n", a, b);
252 #define PRINT_HEX(a, b) av_log(s->avctx, AV_LOG_DEBUG, " %s = %"PRIx32"\n", a, b);
254 PRINT("ed sample bit depth", s->bits_per_sample);
255 PRINT_HEX("ed decode flags", s->decode_flags);
256 PRINT("samples per frame", s->samples_per_frame);
257 PRINT("log2 frame size", s->log2_frame_size);
258 PRINT("max num subframes", s->max_num_subframes);
259 PRINT("len prefix", s->len_prefix);
260 PRINT("num channels", s->nb_channels);
264 *@brief Uninitialize the decoder and free all resources.
265 *@param avctx codec context
266 *@return 0 on success, < 0 otherwise
268 static av_cold int decode_end(WMAProDecodeCtx *s)
274 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
275 ff_mdct_end(&s->mdct_ctx[i]);
280 static av_cold int wmapro_decode_end(AVCodecContext *avctx)
282 WMAProDecodeCtx *s = avctx->priv_data;
289 static av_cold int get_rate(AVCodecContext *avctx)
291 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) { // XXX: is this really only for XMA?
292 if (avctx->sample_rate > 44100)
294 else if (avctx->sample_rate > 32000)
296 else if (avctx->sample_rate > 24000)
301 return avctx->sample_rate;
305 *@brief Initialize the decoder.
306 *@param avctx codec context
307 *@return 0 on success, -1 otherwise
309 static av_cold int decode_init(WMAProDecodeCtx *s, AVCodecContext *avctx)
311 uint8_t *edata_ptr = avctx->extradata;
312 unsigned int channel_mask;
314 int log2_max_num_subframes;
315 int num_possible_block_sizes;
317 if (avctx->codec_id == AV_CODEC_ID_XMA1 || avctx->codec_id == AV_CODEC_ID_XMA2)
318 avctx->block_align = 2048;
320 if (!avctx->block_align) {
321 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n");
322 return AVERROR(EINVAL);
326 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
328 return AVERROR(ENOMEM);
330 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
332 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
334 /** dump the extradata */
335 av_log(avctx, AV_LOG_DEBUG, "extradata:\n");
336 for (i = 0; i < avctx->extradata_size; i++)
337 av_log(avctx, AV_LOG_DEBUG, "[%x] ", avctx->extradata[i]);
338 av_log(avctx, AV_LOG_DEBUG, "\n");
339 if (avctx->codec_id == AV_CODEC_ID_XMA2 && (!avctx->extradata || avctx->extradata_size >= 6)) {
340 s->decode_flags = 0x10d6;
341 channel_mask = avctx->extradata ? AV_RL32(edata_ptr+2) : 0;
342 s->bits_per_sample = 16;
343 } else if (avctx->codec_id == AV_CODEC_ID_XMA1) {
344 s->decode_flags = 0x10d6;
345 s->bits_per_sample = 16;
347 } else if (avctx->codec_id == AV_CODEC_ID_WMAPRO && avctx->extradata_size >= 18) {
348 s->decode_flags = AV_RL16(edata_ptr+14);
349 channel_mask = AV_RL32(edata_ptr+2);
350 s->bits_per_sample = AV_RL16(edata_ptr);
352 if (s->bits_per_sample > 32 || s->bits_per_sample < 1) {
353 avpriv_request_sample(avctx, "bits per sample is %d", s->bits_per_sample);
354 return AVERROR_PATCHWELCOME;
357 avpriv_request_sample(avctx, "Unknown extradata size");
358 return AVERROR_PATCHWELCOME;
361 if (avctx->codec_id != AV_CODEC_ID_WMAPRO && avctx->channels > 2) {
364 s->nb_channels = avctx->channels;
368 s->log2_frame_size = av_log2(avctx->block_align) + 4;
369 if (s->log2_frame_size > 25) {
370 avpriv_request_sample(avctx, "Large block align");
371 return AVERROR_PATCHWELCOME;
375 if (avctx->codec_id != AV_CODEC_ID_WMAPRO)
378 s->skip_frame = 1; /* skip first frame */
381 s->len_prefix = (s->decode_flags & 0x40);
384 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
385 bits = ff_wma_get_frame_len_bits(avctx->sample_rate, 3, s->decode_flags);
386 if (bits > WMAPRO_BLOCK_MAX_BITS) {
387 avpriv_request_sample(avctx, "14-bit block sizes");
388 return AVERROR_PATCHWELCOME;
390 s->samples_per_frame = 1 << bits;
392 s->samples_per_frame = 512;
396 log2_max_num_subframes = ((s->decode_flags & 0x38) >> 3);
397 s->max_num_subframes = 1 << log2_max_num_subframes;
398 if (s->max_num_subframes == 16 || s->max_num_subframes == 4)
399 s->max_subframe_len_bit = 1;
400 s->subframe_len_bits = av_log2(log2_max_num_subframes) + 1;
402 num_possible_block_sizes = log2_max_num_subframes + 1;
403 s->min_samples_per_subframe = s->samples_per_frame / s->max_num_subframes;
404 s->dynamic_range_compression = (s->decode_flags & 0x80);
406 if (s->max_num_subframes > MAX_SUBFRAMES) {
407 av_log(avctx, AV_LOG_ERROR, "invalid number of subframes %"PRId8"\n",
408 s->max_num_subframes);
409 return AVERROR_INVALIDDATA;
412 if (s->min_samples_per_subframe < WMAPRO_BLOCK_MIN_SIZE) {
413 av_log(avctx, AV_LOG_ERROR, "min_samples_per_subframe of %d too small\n",
414 s->min_samples_per_subframe);
415 return AVERROR_INVALIDDATA;
418 if (s->avctx->sample_rate <= 0) {
419 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
420 return AVERROR_INVALIDDATA;
423 if (s->nb_channels <= 0) {
424 av_log(avctx, AV_LOG_ERROR, "invalid number of channels %d\n",
426 return AVERROR_INVALIDDATA;
427 } else if (s->nb_channels > WMAPRO_MAX_CHANNELS) {
428 avpriv_request_sample(avctx,
429 "More than %d channels", WMAPRO_MAX_CHANNELS);
430 return AVERROR_PATCHWELCOME;
433 /** init previous block len */
434 for (i = 0; i < s->nb_channels; i++)
435 s->channel[i].prev_block_len = s->samples_per_frame;
437 /** extract lfe channel position */
440 if (channel_mask & 8) {
442 for (mask = 1; mask < 16; mask <<= 1) {
443 if (channel_mask & mask)
448 INIT_VLC_STATIC(&sf_vlc, SCALEVLCBITS, HUFF_SCALE_SIZE,
449 scale_huffbits, 1, 1,
450 scale_huffcodes, 2, 2, 616);
452 INIT_VLC_STATIC(&sf_rl_vlc, VLCBITS, HUFF_SCALE_RL_SIZE,
453 scale_rl_huffbits, 1, 1,
454 scale_rl_huffcodes, 4, 4, 1406);
456 INIT_VLC_STATIC(&coef_vlc[0], VLCBITS, HUFF_COEF0_SIZE,
457 coef0_huffbits, 1, 1,
458 coef0_huffcodes, 4, 4, 2108);
460 INIT_VLC_STATIC(&coef_vlc[1], VLCBITS, HUFF_COEF1_SIZE,
461 coef1_huffbits, 1, 1,
462 coef1_huffcodes, 4, 4, 3912);
464 INIT_VLC_STATIC(&vec4_vlc, VLCBITS, HUFF_VEC4_SIZE,
466 vec4_huffcodes, 2, 2, 604);
468 INIT_VLC_STATIC(&vec2_vlc, VLCBITS, HUFF_VEC2_SIZE,
470 vec2_huffcodes, 2, 2, 562);
472 INIT_VLC_STATIC(&vec1_vlc, VLCBITS, HUFF_VEC1_SIZE,
474 vec1_huffcodes, 2, 2, 562);
476 /** calculate number of scale factor bands and their offsets
477 for every possible block size */
478 for (i = 0; i < num_possible_block_sizes; i++) {
479 int subframe_len = s->samples_per_frame >> i;
482 int rate = get_rate(avctx);
484 s->sfb_offsets[i][0] = 0;
486 for (x = 0; x < MAX_BANDS-1 && s->sfb_offsets[i][band - 1] < subframe_len; x++) {
487 int offset = (subframe_len * 2 * critical_freq[x]) / rate + 2;
489 if (offset > s->sfb_offsets[i][band - 1])
490 s->sfb_offsets[i][band++] = offset;
492 if (offset >= subframe_len)
495 s->sfb_offsets[i][band - 1] = subframe_len;
496 s->num_sfb[i] = band - 1;
497 if (s->num_sfb[i] <= 0) {
498 av_log(avctx, AV_LOG_ERROR, "num_sfb invalid\n");
499 return AVERROR_INVALIDDATA;
504 /** Scale factors can be shared between blocks of different size
505 as every block has a different scale factor band layout.
506 The matrix sf_offsets is needed to find the correct scale factor.
509 for (i = 0; i < num_possible_block_sizes; i++) {
511 for (b = 0; b < s->num_sfb[i]; b++) {
513 int offset = ((s->sfb_offsets[i][b]
514 + s->sfb_offsets[i][b + 1] - 1) << i) >> 1;
515 for (x = 0; x < num_possible_block_sizes; x++) {
517 while (s->sfb_offsets[x][v + 1] << x < offset) {
519 av_assert0(v < MAX_BANDS);
521 s->sf_offsets[i][x][b] = v;
526 /** init MDCT, FIXME: only init needed sizes */
527 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++)
528 ff_mdct_init(&s->mdct_ctx[i], WMAPRO_BLOCK_MIN_BITS+1+i, 1,
529 1.0 / (1 << (WMAPRO_BLOCK_MIN_BITS + i - 1))
530 / (1 << (s->bits_per_sample - 1)));
532 /** init MDCT windows: simple sine window */
533 for (i = 0; i < WMAPRO_BLOCK_SIZES; i++) {
534 const int win_idx = WMAPRO_BLOCK_MAX_BITS - i;
535 ff_init_ff_sine_windows(win_idx);
536 s->windows[WMAPRO_BLOCK_SIZES - i - 1] = ff_sine_windows[win_idx];
539 /** calculate subwoofer cutoff values */
540 for (i = 0; i < num_possible_block_sizes; i++) {
541 int block_size = s->samples_per_frame >> i;
542 int cutoff = (440*block_size + 3LL * (s->avctx->sample_rate >> 1) - 1)
543 / s->avctx->sample_rate;
544 s->subwoofer_cutoffs[i] = av_clip(cutoff, 4, block_size);
547 /** calculate sine values for the decorrelation matrix */
548 for (i = 0; i < 33; i++)
549 sin64[i] = sin(i*M_PI / 64.0);
551 if (avctx->debug & FF_DEBUG_BITSTREAM)
554 avctx->channel_layout = channel_mask;
560 *@brief Initialize the decoder.
561 *@param avctx codec context
562 *@return 0 on success, -1 otherwise
564 static av_cold int wmapro_decode_init(AVCodecContext *avctx)
566 WMAProDecodeCtx *s = avctx->priv_data;
568 return decode_init(s, avctx);
572 *@brief Decode the subframe length.
574 *@param offset sample offset in the frame
575 *@return decoded subframe length on success, < 0 in case of an error
577 static int decode_subframe_length(WMAProDecodeCtx *s, int offset)
579 int frame_len_shift = 0;
582 /** no need to read from the bitstream when only one length is possible */
583 if (offset == s->samples_per_frame - s->min_samples_per_subframe)
584 return s->min_samples_per_subframe;
586 if (get_bits_left(&s->gb) < 1)
587 return AVERROR_INVALIDDATA;
589 /** 1 bit indicates if the subframe is of maximum length */
590 if (s->max_subframe_len_bit) {
591 if (get_bits1(&s->gb))
592 frame_len_shift = 1 + get_bits(&s->gb, s->subframe_len_bits-1);
594 frame_len_shift = get_bits(&s->gb, s->subframe_len_bits);
596 subframe_len = s->samples_per_frame >> frame_len_shift;
598 /** sanity check the length */
599 if (subframe_len < s->min_samples_per_subframe ||
600 subframe_len > s->samples_per_frame) {
601 av_log(s->avctx, AV_LOG_ERROR, "broken frame: subframe_len %i\n",
603 return AVERROR_INVALIDDATA;
609 *@brief Decode how the data in the frame is split into subframes.
610 * Every WMA frame contains the encoded data for a fixed number of
611 * samples per channel. The data for every channel might be split
612 * into several subframes. This function will reconstruct the list of
613 * subframes for every channel.
615 * If the subframes are not evenly split, the algorithm estimates the
616 * channels with the lowest number of total samples.
617 * Afterwards, for each of these channels a bit is read from the
618 * bitstream that indicates if the channel contains a subframe with the
619 * next subframe size that is going to be read from the bitstream or not.
620 * If a channel contains such a subframe, the subframe size gets added to
621 * the channel's subframe list.
622 * The algorithm repeats these steps until the frame is properly divided
623 * between the individual channels.
626 *@return 0 on success, < 0 in case of an error
628 static int decode_tilehdr(WMAProDecodeCtx *s)
630 uint16_t num_samples[WMAPRO_MAX_CHANNELS] = { 0 };/**< sum of samples for all currently known subframes of a channel */
631 uint8_t contains_subframe[WMAPRO_MAX_CHANNELS]; /**< flag indicating if a channel contains the current subframe */
632 int channels_for_cur_subframe = s->nb_channels; /**< number of channels that contain the current subframe */
633 int fixed_channel_layout = 0; /**< flag indicating that all channels use the same subframe offsets and sizes */
634 int min_channel_len = 0; /**< smallest sum of samples (channels with this length will be processed first) */
637 /* Should never consume more than 3073 bits (256 iterations for the
638 * while loop when always the minimum amount of 128 samples is subtracted
639 * from missing samples in the 8 channel case).
640 * 1 + BLOCK_MAX_SIZE * MAX_CHANNELS / BLOCK_MIN_SIZE * (MAX_CHANNELS + 4)
643 /** reset tiling information */
644 for (c = 0; c < s->nb_channels; c++)
645 s->channel[c].num_subframes = 0;
647 if (s->max_num_subframes == 1 || get_bits1(&s->gb))
648 fixed_channel_layout = 1;
650 /** loop until the frame data is split between the subframes */
654 /** check which channels contain the subframe */
655 for (c = 0; c < s->nb_channels; c++) {
656 if (num_samples[c] == min_channel_len) {
657 if (fixed_channel_layout || channels_for_cur_subframe == 1 ||
658 (min_channel_len == s->samples_per_frame - s->min_samples_per_subframe))
659 contains_subframe[c] = 1;
661 contains_subframe[c] = get_bits1(&s->gb);
663 contains_subframe[c] = 0;
666 /** get subframe length, subframe_len == 0 is not allowed */
667 if ((subframe_len = decode_subframe_length(s, min_channel_len)) <= 0)
668 return AVERROR_INVALIDDATA;
670 /** add subframes to the individual channels and find new min_channel_len */
671 min_channel_len += subframe_len;
672 for (c = 0; c < s->nb_channels; c++) {
673 WMAProChannelCtx* chan = &s->channel[c];
675 if (contains_subframe[c]) {
676 if (chan->num_subframes >= MAX_SUBFRAMES) {
677 av_log(s->avctx, AV_LOG_ERROR,
678 "broken frame: num subframes > 31\n");
679 return AVERROR_INVALIDDATA;
681 chan->subframe_len[chan->num_subframes] = subframe_len;
682 num_samples[c] += subframe_len;
683 ++chan->num_subframes;
684 if (num_samples[c] > s->samples_per_frame) {
685 av_log(s->avctx, AV_LOG_ERROR, "broken frame: "
686 "channel len > samples_per_frame\n");
687 return AVERROR_INVALIDDATA;
689 } else if (num_samples[c] <= min_channel_len) {
690 if (num_samples[c] < min_channel_len) {
691 channels_for_cur_subframe = 0;
692 min_channel_len = num_samples[c];
694 ++channels_for_cur_subframe;
697 } while (min_channel_len < s->samples_per_frame);
699 for (c = 0; c < s->nb_channels; c++) {
702 for (i = 0; i < s->channel[c].num_subframes; i++) {
703 ff_dlog(s->avctx, "frame[%i] channel[%i] subframe[%i]"
704 " len %i\n", s->frame_num, c, i,
705 s->channel[c].subframe_len[i]);
706 s->channel[c].subframe_offset[i] = offset;
707 offset += s->channel[c].subframe_len[i];
715 *@brief Calculate a decorrelation matrix from the bitstream parameters.
716 *@param s codec context
717 *@param chgroup channel group for which the matrix needs to be calculated
719 static void decode_decorrelation_matrix(WMAProDecodeCtx *s,
720 WMAProChannelGrp *chgroup)
724 int8_t rotation_offset[WMAPRO_MAX_CHANNELS * WMAPRO_MAX_CHANNELS];
725 memset(chgroup->decorrelation_matrix, 0, s->nb_channels *
726 s->nb_channels * sizeof(*chgroup->decorrelation_matrix));
728 for (i = 0; i < chgroup->num_channels * (chgroup->num_channels - 1) >> 1; i++)
729 rotation_offset[i] = get_bits(&s->gb, 6);
731 for (i = 0; i < chgroup->num_channels; i++)
732 chgroup->decorrelation_matrix[chgroup->num_channels * i + i] =
733 get_bits1(&s->gb) ? 1.0 : -1.0;
735 for (i = 1; i < chgroup->num_channels; i++) {
737 for (x = 0; x < i; x++) {
739 for (y = 0; y < i + 1; y++) {
740 float v1 = chgroup->decorrelation_matrix[x * chgroup->num_channels + y];
741 float v2 = chgroup->decorrelation_matrix[i * chgroup->num_channels + y];
742 int n = rotation_offset[offset + x];
748 cosv = sin64[32 - n];
750 sinv = sin64[64 - n];
751 cosv = -sin64[n - 32];
754 chgroup->decorrelation_matrix[y + x * chgroup->num_channels] =
755 (v1 * sinv) - (v2 * cosv);
756 chgroup->decorrelation_matrix[y + i * chgroup->num_channels] =
757 (v1 * cosv) + (v2 * sinv);
765 *@brief Decode channel transformation parameters
766 *@param s codec context
767 *@return >= 0 in case of success, < 0 in case of bitstream errors
769 static int decode_channel_transform(WMAProDecodeCtx* s)
772 /* should never consume more than 1921 bits for the 8 channel case
773 * 1 + MAX_CHANNELS * (MAX_CHANNELS + 2 + 3 * MAX_CHANNELS * MAX_CHANNELS
774 * + MAX_CHANNELS + MAX_BANDS + 1)
777 /** in the one channel case channel transforms are pointless */
779 if (s->nb_channels > 1) {
780 int remaining_channels = s->channels_for_cur_subframe;
782 if (get_bits1(&s->gb)) {
783 avpriv_request_sample(s->avctx,
784 "Channel transform bit");
785 return AVERROR_PATCHWELCOME;
788 for (s->num_chgroups = 0; remaining_channels &&
789 s->num_chgroups < s->channels_for_cur_subframe; s->num_chgroups++) {
790 WMAProChannelGrp* chgroup = &s->chgroup[s->num_chgroups];
791 float** channel_data = chgroup->channel_data;
792 chgroup->num_channels = 0;
793 chgroup->transform = 0;
795 /** decode channel mask */
796 if (remaining_channels > 2) {
797 for (i = 0; i < s->channels_for_cur_subframe; i++) {
798 int channel_idx = s->channel_indexes_for_cur_subframe[i];
799 if (!s->channel[channel_idx].grouped
800 && get_bits1(&s->gb)) {
801 ++chgroup->num_channels;
802 s->channel[channel_idx].grouped = 1;
803 *channel_data++ = s->channel[channel_idx].coeffs;
807 chgroup->num_channels = remaining_channels;
808 for (i = 0; i < s->channels_for_cur_subframe; i++) {
809 int channel_idx = s->channel_indexes_for_cur_subframe[i];
810 if (!s->channel[channel_idx].grouped)
811 *channel_data++ = s->channel[channel_idx].coeffs;
812 s->channel[channel_idx].grouped = 1;
816 /** decode transform type */
817 if (chgroup->num_channels == 2) {
818 if (get_bits1(&s->gb)) {
819 if (get_bits1(&s->gb)) {
820 avpriv_request_sample(s->avctx,
821 "Unknown channel transform type");
822 return AVERROR_PATCHWELCOME;
825 chgroup->transform = 1;
826 if (s->nb_channels == 2) {
827 chgroup->decorrelation_matrix[0] = 1.0;
828 chgroup->decorrelation_matrix[1] = -1.0;
829 chgroup->decorrelation_matrix[2] = 1.0;
830 chgroup->decorrelation_matrix[3] = 1.0;
833 chgroup->decorrelation_matrix[0] = 0.70703125;
834 chgroup->decorrelation_matrix[1] = -0.70703125;
835 chgroup->decorrelation_matrix[2] = 0.70703125;
836 chgroup->decorrelation_matrix[3] = 0.70703125;
839 } else if (chgroup->num_channels > 2) {
840 if (get_bits1(&s->gb)) {
841 chgroup->transform = 1;
842 if (get_bits1(&s->gb)) {
843 decode_decorrelation_matrix(s, chgroup);
845 /** FIXME: more than 6 coupled channels not supported */
846 if (chgroup->num_channels > 6) {
847 avpriv_request_sample(s->avctx,
848 "Coupled channels > 6");
850 memcpy(chgroup->decorrelation_matrix,
851 default_decorrelation[chgroup->num_channels],
852 chgroup->num_channels * chgroup->num_channels *
853 sizeof(*chgroup->decorrelation_matrix));
859 /** decode transform on / off */
860 if (chgroup->transform) {
861 if (!get_bits1(&s->gb)) {
863 /** transform can be enabled for individual bands */
864 for (i = 0; i < s->num_bands; i++) {
865 chgroup->transform_band[i] = get_bits1(&s->gb);
868 memset(chgroup->transform_band, 1, s->num_bands);
871 remaining_channels -= chgroup->num_channels;
878 *@brief Extract the coefficients from the bitstream.
879 *@param s codec context
880 *@param c current channel number
881 *@return 0 on success, < 0 in case of bitstream errors
883 static int decode_coeffs(WMAProDecodeCtx *s, int c)
885 /* Integers 0..15 as single-precision floats. The table saves a
886 costly int to float conversion, and storing the values as
887 integers allows fast sign-flipping. */
888 static const uint32_t fval_tab[16] = {
889 0x00000000, 0x3f800000, 0x40000000, 0x40400000,
890 0x40800000, 0x40a00000, 0x40c00000, 0x40e00000,
891 0x41000000, 0x41100000, 0x41200000, 0x41300000,
892 0x41400000, 0x41500000, 0x41600000, 0x41700000,
896 WMAProChannelCtx* ci = &s->channel[c];
903 ff_dlog(s->avctx, "decode coefficients for channel %i\n", c);
905 vlctable = get_bits1(&s->gb);
906 vlc = &coef_vlc[vlctable];
916 /** decode vector coefficients (consumes up to 167 bits per iteration for
917 4 vector coded large values) */
918 while ((s->transmit_num_vec_coeffs || !rl_mode) &&
919 (cur_coeff + 3 < ci->num_vec_coeffs)) {
924 idx = get_vlc2(&s->gb, vec4_vlc.table, VLCBITS, VEC4MAXDEPTH);
926 if (idx == HUFF_VEC4_SIZE - 1) {
927 for (i = 0; i < 4; i += 2) {
928 idx = get_vlc2(&s->gb, vec2_vlc.table, VLCBITS, VEC2MAXDEPTH);
929 if (idx == HUFF_VEC2_SIZE - 1) {
931 v0 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
932 if (v0 == HUFF_VEC1_SIZE - 1)
933 v0 += ff_wma_get_large_val(&s->gb);
934 v1 = get_vlc2(&s->gb, vec1_vlc.table, VLCBITS, VEC1MAXDEPTH);
935 if (v1 == HUFF_VEC1_SIZE - 1)
936 v1 += ff_wma_get_large_val(&s->gb);
937 vals[i ] = av_float2int(v0);
938 vals[i+1] = av_float2int(v1);
940 vals[i] = fval_tab[symbol_to_vec2[idx] >> 4 ];
941 vals[i+1] = fval_tab[symbol_to_vec2[idx] & 0xF];
945 vals[0] = fval_tab[ symbol_to_vec4[idx] >> 12 ];
946 vals[1] = fval_tab[(symbol_to_vec4[idx] >> 8) & 0xF];
947 vals[2] = fval_tab[(symbol_to_vec4[idx] >> 4) & 0xF];
948 vals[3] = fval_tab[ symbol_to_vec4[idx] & 0xF];
952 for (i = 0; i < 4; i++) {
954 uint32_t sign = get_bits1(&s->gb) - 1;
955 AV_WN32A(&ci->coeffs[cur_coeff], vals[i] ^ sign << 31);
958 ci->coeffs[cur_coeff] = 0;
959 /** switch to run level mode when subframe_len / 128 zeros
960 were found in a row */
961 rl_mode |= (++num_zeros > s->subframe_len >> 8);
967 /** decode run level coded coefficients */
968 if (cur_coeff < s->subframe_len) {
969 memset(&ci->coeffs[cur_coeff], 0,
970 sizeof(*ci->coeffs) * (s->subframe_len - cur_coeff));
971 if (ff_wma_run_level_decode(s->avctx, &s->gb, vlc,
972 level, run, 1, ci->coeffs,
973 cur_coeff, s->subframe_len,
974 s->subframe_len, s->esc_len, 0))
975 return AVERROR_INVALIDDATA;
982 *@brief Extract scale factors from the bitstream.
983 *@param s codec context
984 *@return 0 on success, < 0 in case of bitstream errors
986 static int decode_scale_factors(WMAProDecodeCtx* s)
990 /** should never consume more than 5344 bits
991 * MAX_CHANNELS * (1 + MAX_BANDS * 23)
994 for (i = 0; i < s->channels_for_cur_subframe; i++) {
995 int c = s->channel_indexes_for_cur_subframe[i];
998 s->channel[c].scale_factors = s->channel[c].saved_scale_factors[!s->channel[c].scale_factor_idx];
999 sf_end = s->channel[c].scale_factors + s->num_bands;
1001 /** resample scale factors for the new block size
1002 * as the scale factors might need to be resampled several times
1003 * before some new values are transmitted, a backup of the last
1004 * transmitted scale factors is kept in saved_scale_factors
1006 if (s->channel[c].reuse_sf) {
1007 const int8_t* sf_offsets = s->sf_offsets[s->table_idx][s->channel[c].table_idx];
1009 for (b = 0; b < s->num_bands; b++)
1010 s->channel[c].scale_factors[b] =
1011 s->channel[c].saved_scale_factors[s->channel[c].scale_factor_idx][*sf_offsets++];
1014 if (!s->channel[c].cur_subframe || get_bits1(&s->gb)) {
1016 if (!s->channel[c].reuse_sf) {
1018 /** decode DPCM coded scale factors */
1019 s->channel[c].scale_factor_step = get_bits(&s->gb, 2) + 1;
1020 val = 45 / s->channel[c].scale_factor_step;
1021 for (sf = s->channel[c].scale_factors; sf < sf_end; sf++) {
1022 val += get_vlc2(&s->gb, sf_vlc.table, SCALEVLCBITS, SCALEMAXDEPTH) - 60;
1027 /** run level decode differences to the resampled factors */
1028 for (i = 0; i < s->num_bands; i++) {
1034 idx = get_vlc2(&s->gb, sf_rl_vlc.table, VLCBITS, SCALERLMAXDEPTH);
1037 uint32_t code = get_bits(&s->gb, 14);
1039 sign = (code & 1) - 1;
1040 skip = (code & 0x3f) >> 1;
1041 } else if (idx == 1) {
1044 skip = scale_rl_run[idx];
1045 val = scale_rl_level[idx];
1046 sign = get_bits1(&s->gb)-1;
1050 if (i >= s->num_bands) {
1051 av_log(s->avctx, AV_LOG_ERROR,
1052 "invalid scale factor coding\n");
1053 return AVERROR_INVALIDDATA;
1055 s->channel[c].scale_factors[i] += (val ^ sign) - sign;
1059 s->channel[c].scale_factor_idx = !s->channel[c].scale_factor_idx;
1060 s->channel[c].table_idx = s->table_idx;
1061 s->channel[c].reuse_sf = 1;
1064 /** calculate new scale factor maximum */
1065 s->channel[c].max_scale_factor = s->channel[c].scale_factors[0];
1066 for (sf = s->channel[c].scale_factors + 1; sf < sf_end; sf++) {
1067 s->channel[c].max_scale_factor =
1068 FFMAX(s->channel[c].max_scale_factor, *sf);
1076 *@brief Reconstruct the individual channel data.
1077 *@param s codec context
1079 static void inverse_channel_transform(WMAProDecodeCtx *s)
1083 for (i = 0; i < s->num_chgroups; i++) {
1084 if (s->chgroup[i].transform) {
1085 float data[WMAPRO_MAX_CHANNELS];
1086 const int num_channels = s->chgroup[i].num_channels;
1087 float** ch_data = s->chgroup[i].channel_data;
1088 float** ch_end = ch_data + num_channels;
1089 const int8_t* tb = s->chgroup[i].transform_band;
1092 /** multichannel decorrelation */
1093 for (sfb = s->cur_sfb_offsets;
1094 sfb < s->cur_sfb_offsets + s->num_bands; sfb++) {
1097 /** multiply values with the decorrelation_matrix */
1098 for (y = sfb[0]; y < FFMIN(sfb[1], s->subframe_len); y++) {
1099 const float* mat = s->chgroup[i].decorrelation_matrix;
1100 const float* data_end = data + num_channels;
1101 float* data_ptr = data;
1104 for (ch = ch_data; ch < ch_end; ch++)
1105 *data_ptr++ = (*ch)[y];
1107 for (ch = ch_data; ch < ch_end; ch++) {
1110 while (data_ptr < data_end)
1111 sum += *data_ptr++ * *mat++;
1116 } else if (s->nb_channels == 2) {
1117 int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
1118 s->fdsp->vector_fmul_scalar(ch_data[0] + sfb[0],
1119 ch_data[0] + sfb[0],
1121 s->fdsp->vector_fmul_scalar(ch_data[1] + sfb[0],
1122 ch_data[1] + sfb[0],
1131 *@brief Apply sine window and reconstruct the output buffer.
1132 *@param s codec context
1134 static void wmapro_window(WMAProDecodeCtx *s)
1137 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1138 int c = s->channel_indexes_for_cur_subframe[i];
1139 const float* window;
1140 int winlen = s->channel[c].prev_block_len;
1141 float* start = s->channel[c].coeffs - (winlen >> 1);
1143 if (s->subframe_len < winlen) {
1144 start += (winlen - s->subframe_len) >> 1;
1145 winlen = s->subframe_len;
1148 window = s->windows[av_log2(winlen) - WMAPRO_BLOCK_MIN_BITS];
1152 s->fdsp->vector_fmul_window(start, start, start + winlen,
1155 s->channel[c].prev_block_len = s->subframe_len;
1160 *@brief Decode a single subframe (block).
1161 *@param s codec context
1162 *@return 0 on success, < 0 when decoding failed
1164 static int decode_subframe(WMAProDecodeCtx *s)
1166 int offset = s->samples_per_frame;
1167 int subframe_len = s->samples_per_frame;
1169 int total_samples = s->samples_per_frame * s->nb_channels;
1170 int transmit_coeffs = 0;
1171 int cur_subwoofer_cutoff;
1173 s->subframe_offset = get_bits_count(&s->gb);
1175 /** reset channel context and find the next block offset and size
1176 == the next block of the channel with the smallest number of
1179 for (i = 0; i < s->nb_channels; i++) {
1180 s->channel[i].grouped = 0;
1181 if (offset > s->channel[i].decoded_samples) {
1182 offset = s->channel[i].decoded_samples;
1184 s->channel[i].subframe_len[s->channel[i].cur_subframe];
1189 "processing subframe with offset %i len %i\n", offset, subframe_len);
1191 /** get a list of all channels that contain the estimated block */
1192 s->channels_for_cur_subframe = 0;
1193 for (i = 0; i < s->nb_channels; i++) {
1194 const int cur_subframe = s->channel[i].cur_subframe;
1195 /** subtract already processed samples */
1196 total_samples -= s->channel[i].decoded_samples;
1198 /** and count if there are multiple subframes that match our profile */
1199 if (offset == s->channel[i].decoded_samples &&
1200 subframe_len == s->channel[i].subframe_len[cur_subframe]) {
1201 total_samples -= s->channel[i].subframe_len[cur_subframe];
1202 s->channel[i].decoded_samples +=
1203 s->channel[i].subframe_len[cur_subframe];
1204 s->channel_indexes_for_cur_subframe[s->channels_for_cur_subframe] = i;
1205 ++s->channels_for_cur_subframe;
1209 /** check if the frame will be complete after processing the
1212 s->parsed_all_subframes = 1;
1215 ff_dlog(s->avctx, "subframe is part of %i channels\n",
1216 s->channels_for_cur_subframe);
1218 /** calculate number of scale factor bands and their offsets */
1219 s->table_idx = av_log2(s->samples_per_frame/subframe_len);
1220 s->num_bands = s->num_sfb[s->table_idx];
1221 s->cur_sfb_offsets = s->sfb_offsets[s->table_idx];
1222 cur_subwoofer_cutoff = s->subwoofer_cutoffs[s->table_idx];
1224 /** configure the decoder for the current subframe */
1225 offset += s->samples_per_frame >> 1;
1227 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1228 int c = s->channel_indexes_for_cur_subframe[i];
1230 s->channel[c].coeffs = &s->channel[c].out[offset];
1233 s->subframe_len = subframe_len;
1234 s->esc_len = av_log2(s->subframe_len - 1) + 1;
1236 /** skip extended header if any */
1237 if (get_bits1(&s->gb)) {
1239 if (!(num_fill_bits = get_bits(&s->gb, 2))) {
1240 int len = get_bits(&s->gb, 4);
1241 num_fill_bits = get_bitsz(&s->gb, len) + 1;
1244 if (num_fill_bits >= 0) {
1245 if (get_bits_count(&s->gb) + num_fill_bits > s->num_saved_bits) {
1246 av_log(s->avctx, AV_LOG_ERROR, "invalid number of fill bits\n");
1247 return AVERROR_INVALIDDATA;
1250 skip_bits_long(&s->gb, num_fill_bits);
1254 /** no idea for what the following bit is used */
1255 if (get_bits1(&s->gb)) {
1256 avpriv_request_sample(s->avctx, "Reserved bit");
1257 return AVERROR_PATCHWELCOME;
1261 if (decode_channel_transform(s) < 0)
1262 return AVERROR_INVALIDDATA;
1265 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1266 int c = s->channel_indexes_for_cur_subframe[i];
1267 if ((s->channel[c].transmit_coefs = get_bits1(&s->gb)))
1268 transmit_coeffs = 1;
1271 av_assert0(s->subframe_len <= WMAPRO_BLOCK_MAX_SIZE);
1272 if (transmit_coeffs) {
1274 int quant_step = 90 * s->bits_per_sample >> 4;
1276 /** decode number of vector coded coefficients */
1277 if ((s->transmit_num_vec_coeffs = get_bits1(&s->gb))) {
1278 int num_bits = av_log2((s->subframe_len + 3)/4) + 1;
1279 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1280 int c = s->channel_indexes_for_cur_subframe[i];
1281 int num_vec_coeffs = get_bits(&s->gb, num_bits) << 2;
1282 if (num_vec_coeffs > s->subframe_len) {
1283 av_log(s->avctx, AV_LOG_ERROR, "num_vec_coeffs %d is too large\n", num_vec_coeffs);
1284 return AVERROR_INVALIDDATA;
1286 av_assert0(num_vec_coeffs + offset <= FF_ARRAY_ELEMS(s->channel[c].out));
1287 s->channel[c].num_vec_coeffs = num_vec_coeffs;
1290 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1291 int c = s->channel_indexes_for_cur_subframe[i];
1292 s->channel[c].num_vec_coeffs = s->subframe_len;
1295 /** decode quantization step */
1296 step = get_sbits(&s->gb, 6);
1298 if (step == -32 || step == 31) {
1299 const int sign = (step == 31) - 1;
1301 while (get_bits_count(&s->gb) + 5 < s->num_saved_bits &&
1302 (step = get_bits(&s->gb, 5)) == 31) {
1305 quant_step += ((quant + step) ^ sign) - sign;
1307 if (quant_step < 0) {
1308 av_log(s->avctx, AV_LOG_DEBUG, "negative quant step\n");
1311 /** decode quantization step modifiers for every channel */
1313 if (s->channels_for_cur_subframe == 1) {
1314 s->channel[s->channel_indexes_for_cur_subframe[0]].quant_step = quant_step;
1316 int modifier_len = get_bits(&s->gb, 3);
1317 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1318 int c = s->channel_indexes_for_cur_subframe[i];
1319 s->channel[c].quant_step = quant_step;
1320 if (get_bits1(&s->gb)) {
1322 s->channel[c].quant_step += get_bits(&s->gb, modifier_len) + 1;
1324 ++s->channel[c].quant_step;
1329 /** decode scale factors */
1330 if (decode_scale_factors(s) < 0)
1331 return AVERROR_INVALIDDATA;
1334 ff_dlog(s->avctx, "BITSTREAM: subframe header length was %i\n",
1335 get_bits_count(&s->gb) - s->subframe_offset);
1337 /** parse coefficients */
1338 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1339 int c = s->channel_indexes_for_cur_subframe[i];
1340 if (s->channel[c].transmit_coefs &&
1341 get_bits_count(&s->gb) < s->num_saved_bits) {
1342 decode_coeffs(s, c);
1344 memset(s->channel[c].coeffs, 0,
1345 sizeof(*s->channel[c].coeffs) * subframe_len);
1348 ff_dlog(s->avctx, "BITSTREAM: subframe length was %i\n",
1349 get_bits_count(&s->gb) - s->subframe_offset);
1351 if (transmit_coeffs) {
1352 FFTContext *mdct = &s->mdct_ctx[av_log2(subframe_len) - WMAPRO_BLOCK_MIN_BITS];
1353 /** reconstruct the per channel data */
1354 inverse_channel_transform(s);
1355 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1356 int c = s->channel_indexes_for_cur_subframe[i];
1357 const int* sf = s->channel[c].scale_factors;
1360 if (c == s->lfe_channel)
1361 memset(&s->tmp[cur_subwoofer_cutoff], 0, sizeof(*s->tmp) *
1362 (subframe_len - cur_subwoofer_cutoff));
1364 /** inverse quantization and rescaling */
1365 for (b = 0; b < s->num_bands; b++) {
1366 const int end = FFMIN(s->cur_sfb_offsets[b+1], s->subframe_len);
1367 const int exp = s->channel[c].quant_step -
1368 (s->channel[c].max_scale_factor - *sf++) *
1369 s->channel[c].scale_factor_step;
1370 const float quant = ff_exp10(exp / 20.0);
1371 int start = s->cur_sfb_offsets[b];
1372 s->fdsp->vector_fmul_scalar(s->tmp + start,
1373 s->channel[c].coeffs + start,
1374 quant, end - start);
1377 /** apply imdct (imdct_half == DCTIV with reverse) */
1378 mdct->imdct_half(mdct, s->channel[c].coeffs, s->tmp);
1382 /** window and overlapp-add */
1385 /** handled one subframe */
1386 for (i = 0; i < s->channels_for_cur_subframe; i++) {
1387 int c = s->channel_indexes_for_cur_subframe[i];
1388 if (s->channel[c].cur_subframe >= s->channel[c].num_subframes) {
1389 av_log(s->avctx, AV_LOG_ERROR, "broken subframe\n");
1390 return AVERROR_INVALIDDATA;
1392 ++s->channel[c].cur_subframe;
1399 *@brief Decode one WMA frame.
1400 *@param s codec context
1401 *@return 0 if the trailer bit indicates that this is the last frame,
1402 * 1 if there are additional frames
1404 static int decode_frame(WMAProDecodeCtx *s, AVFrame *frame, int *got_frame_ptr)
1406 GetBitContext* gb = &s->gb;
1407 int more_frames = 0;
1411 /** get frame length */
1413 len = get_bits(gb, s->log2_frame_size);
1415 ff_dlog(s->avctx, "decoding frame with length %x\n", len);
1417 /** decode tile information */
1418 if (decode_tilehdr(s)) {
1423 /** read postproc transform */
1424 if (s->nb_channels > 1 && get_bits1(gb)) {
1425 if (get_bits1(gb)) {
1426 for (i = 0; i < s->nb_channels * s->nb_channels; i++)
1431 /** read drc info */
1432 if (s->dynamic_range_compression) {
1433 s->drc_gain = get_bits(gb, 8);
1434 ff_dlog(s->avctx, "drc_gain %i\n", s->drc_gain);
1437 /** no idea what these are for, might be the number of samples
1438 that need to be skipped at the beginning or end of a stream */
1439 if (get_bits1(gb)) {
1442 /** usually true for the first frame */
1443 if (get_bits1(gb)) {
1444 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1445 ff_dlog(s->avctx, "start skip: %i\n", skip);
1448 /** sometimes true for the last frame */
1449 if (get_bits1(gb)) {
1450 skip = get_bits(gb, av_log2(s->samples_per_frame * 2));
1451 ff_dlog(s->avctx, "end skip: %i\n", skip);
1456 ff_dlog(s->avctx, "BITSTREAM: frame header length was %i\n",
1457 get_bits_count(gb) - s->frame_offset);
1459 /** reset subframe states */
1460 s->parsed_all_subframes = 0;
1461 for (i = 0; i < s->nb_channels; i++) {
1462 s->channel[i].decoded_samples = 0;
1463 s->channel[i].cur_subframe = 0;
1464 s->channel[i].reuse_sf = 0;
1467 /** decode all subframes */
1468 while (!s->parsed_all_subframes) {
1469 if (decode_subframe(s) < 0) {
1475 /** copy samples to the output buffer */
1476 for (i = 0; i < s->nb_channels; i++)
1477 memcpy(frame->extended_data[i], s->channel[i].out,
1478 s->samples_per_frame * sizeof(*s->channel[i].out));
1480 for (i = 0; i < s->nb_channels; i++) {
1481 /** reuse second half of the IMDCT output for the next frame */
1482 memcpy(&s->channel[i].out[0],
1483 &s->channel[i].out[s->samples_per_frame],
1484 s->samples_per_frame * sizeof(*s->channel[i].out) >> 1);
1487 if (s->skip_frame) {
1490 av_frame_unref(frame);
1495 if (s->len_prefix) {
1496 if (len != (get_bits_count(gb) - s->frame_offset) + 2) {
1497 /** FIXME: not sure if this is always an error */
1498 av_log(s->avctx, AV_LOG_ERROR,
1499 "frame[%"PRIu32"] would have to skip %i bits\n",
1501 len - (get_bits_count(gb) - s->frame_offset) - 1);
1506 /** skip the rest of the frame data */
1507 skip_bits_long(gb, len - (get_bits_count(gb) - s->frame_offset) - 1);
1509 while (get_bits_count(gb) < s->num_saved_bits && get_bits1(gb) == 0) {
1513 /** decode trailer bit */
1514 more_frames = get_bits1(gb);
1521 *@brief Calculate remaining input buffer length.
1522 *@param s codec context
1523 *@param gb bitstream reader context
1524 *@return remaining size in bits
1526 static int remaining_bits(WMAProDecodeCtx *s, GetBitContext *gb)
1528 return s->buf_bit_size - get_bits_count(gb);
1532 *@brief Fill the bit reservoir with a (partial) frame.
1533 *@param s codec context
1534 *@param gb bitstream reader context
1535 *@param len length of the partial frame
1536 *@param append decides whether to reset the buffer or not
1538 static void save_bits(WMAProDecodeCtx *s, GetBitContext* gb, int len,
1543 /** when the frame data does not need to be concatenated, the input buffer
1544 is reset and additional bits from the previous frame are copied
1545 and skipped later so that a fast byte copy is possible */
1548 s->frame_offset = get_bits_count(gb) & 7;
1549 s->num_saved_bits = s->frame_offset;
1550 init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
1553 buflen = (put_bits_count(&s->pb) + len + 8) >> 3;
1555 if (len <= 0 || buflen > MAX_FRAMESIZE) {
1556 avpriv_request_sample(s->avctx, "Too small input buffer");
1561 av_assert0(len <= put_bits_left(&s->pb));
1563 s->num_saved_bits += len;
1565 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3),
1568 int align = 8 - (get_bits_count(gb) & 7);
1569 align = FFMIN(align, len);
1570 put_bits(&s->pb, align, get_bits(gb, align));
1572 avpriv_copy_bits(&s->pb, gb->buffer + (get_bits_count(gb) >> 3), len);
1574 skip_bits_long(gb, len);
1577 PutBitContext tmp = s->pb;
1578 flush_put_bits(&tmp);
1581 init_get_bits(&s->gb, s->frame_data, s->num_saved_bits);
1582 skip_bits(&s->gb, s->frame_offset);
1585 static int decode_packet(AVCodecContext *avctx, WMAProDecodeCtx *s,
1586 void *data, int *got_frame_ptr, AVPacket *avpkt)
1588 GetBitContext* gb = &s->pgb;
1589 const uint8_t* buf = avpkt->data;
1590 int buf_size = avpkt->size;
1591 int num_bits_prev_frame;
1592 int packet_sequence_number;
1596 if (s->packet_done || s->packet_loss) {
1599 /** sanity check for the buffer length */
1600 if (avctx->codec_id == AV_CODEC_ID_WMAPRO && buf_size < avctx->block_align) {
1601 av_log(avctx, AV_LOG_ERROR, "Input packet too small (%d < %d)\n",
1602 buf_size, avctx->block_align);
1603 return AVERROR_INVALIDDATA;
1606 if (avctx->codec_id == AV_CODEC_ID_WMAPRO) {
1607 s->next_packet_start = buf_size - avctx->block_align;
1608 buf_size = avctx->block_align;
1610 s->next_packet_start = buf_size - FFMIN(buf_size, avctx->block_align);
1611 buf_size = FFMIN(buf_size, avctx->block_align);
1613 s->buf_bit_size = buf_size << 3;
1615 /** parse packet header */
1616 init_get_bits(gb, buf, s->buf_bit_size);
1617 if (avctx->codec_id != AV_CODEC_ID_XMA2) {
1618 packet_sequence_number = get_bits(gb, 4);
1621 int num_frames = get_bits(gb, 6);
1622 ff_dlog(avctx, "packet[%d]: number of frames %d\n", avctx->frame_number, num_frames);
1623 packet_sequence_number = 0;
1626 /** get number of bits that need to be added to the previous frame */
1627 num_bits_prev_frame = get_bits(gb, s->log2_frame_size);
1628 if (avctx->codec_id != AV_CODEC_ID_WMAPRO) {
1630 s->skip_packets = get_bits(gb, 8);
1631 ff_dlog(avctx, "packet[%d]: skip packets %d\n", avctx->frame_number, s->skip_packets);
1634 ff_dlog(avctx, "packet[%d]: nbpf %x\n", avctx->frame_number,
1635 num_bits_prev_frame);
1637 /** check for packet loss */
1638 if (avctx->codec_id == AV_CODEC_ID_WMAPRO && !s->packet_loss &&
1639 ((s->packet_sequence_number + 1) & 0xF) != packet_sequence_number) {
1641 av_log(avctx, AV_LOG_ERROR,
1642 "Packet loss detected! seq %"PRIx8" vs %x\n",
1643 s->packet_sequence_number, packet_sequence_number);
1645 s->packet_sequence_number = packet_sequence_number;
1647 if (num_bits_prev_frame > 0) {
1648 int remaining_packet_bits = s->buf_bit_size - get_bits_count(gb);
1649 if (num_bits_prev_frame >= remaining_packet_bits) {
1650 num_bits_prev_frame = remaining_packet_bits;
1654 /** append the previous frame data to the remaining data from the
1655 previous packet to create a full frame */
1656 save_bits(s, gb, num_bits_prev_frame, 1);
1657 ff_dlog(avctx, "accumulated %x bits of frame data\n",
1658 s->num_saved_bits - s->frame_offset);
1660 /** decode the cross packet frame if it is valid */
1661 if (!s->packet_loss)
1662 decode_frame(s, data, got_frame_ptr);
1663 } else if (s->num_saved_bits - s->frame_offset) {
1664 ff_dlog(avctx, "ignoring %x previously saved bits\n",
1665 s->num_saved_bits - s->frame_offset);
1668 if (s->packet_loss) {
1669 /** reset number of saved bits so that the decoder
1670 does not start to decode incomplete frames in the
1671 s->len_prefix == 0 case */
1672 s->num_saved_bits = 0;
1677 s->buf_bit_size = (avpkt->size - s->next_packet_start) << 3;
1678 init_get_bits(gb, avpkt->data, s->buf_bit_size);
1679 skip_bits(gb, s->packet_offset);
1680 if (s->len_prefix && remaining_bits(s, gb) > s->log2_frame_size &&
1681 (frame_size = show_bits(gb, s->log2_frame_size)) &&
1682 frame_size <= remaining_bits(s, gb)) {
1683 save_bits(s, gb, frame_size, 0);
1684 if (!s->packet_loss)
1685 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1686 } else if (!s->len_prefix
1687 && s->num_saved_bits > get_bits_count(&s->gb)) {
1688 /** when the frames do not have a length prefix, we don't know
1689 the compressed length of the individual frames
1690 however, we know what part of a new packet belongs to the
1692 therefore we save the incoming packet first, then we append
1693 the "previous frame" data from the next packet so that
1694 we get a buffer that only contains full frames */
1695 s->packet_done = !decode_frame(s, data, got_frame_ptr);
1701 if (remaining_bits(s, gb) < 0) {
1702 av_log(avctx, AV_LOG_ERROR, "Overread %d\n", -remaining_bits(s, gb));
1706 if (s->packet_done && !s->packet_loss &&
1707 remaining_bits(s, gb) > 0) {
1708 /** save the rest of the data so that it can be decoded
1709 with the next packet */
1710 save_bits(s, gb, remaining_bits(s, gb), 0);
1713 s->packet_offset = get_bits_count(gb) & 7;
1715 return AVERROR_INVALIDDATA;
1717 return get_bits_count(gb) >> 3;
1721 *@brief Decode a single WMA packet.
1722 *@param avctx codec context
1723 *@param data the output buffer
1724 *@param avpkt input packet
1725 *@return number of bytes that were read from the input buffer
1727 static int wmapro_decode_packet(AVCodecContext *avctx, void *data,
1728 int *got_frame_ptr, AVPacket *avpkt)
1730 WMAProDecodeCtx *s = avctx->priv_data;
1731 AVFrame *frame = data;
1734 /* get output buffer */
1735 frame->nb_samples = s->samples_per_frame;
1736 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1741 return decode_packet(avctx, s, data, got_frame_ptr, avpkt);
1744 static int xma_decode_packet(AVCodecContext *avctx, void *data,
1745 int *got_frame_ptr, AVPacket *avpkt)
1747 XMADecodeCtx *s = avctx->priv_data;
1748 int got_stream_frame_ptr = 0;
1749 AVFrame *frame = data;
1750 int i, ret, offset = INT_MAX;
1752 ret = decode_packet(avctx, &s->xma[s->current_stream], s->frames[s->current_stream],
1753 &got_stream_frame_ptr, avpkt);
1755 if (got_stream_frame_ptr) {
1756 memcpy(&s->samples[s->current_stream * 2 + 0][s->offset[s->current_stream] * 512],
1757 s->frames[s->current_stream]->extended_data[0], 512 * 4);
1758 if (avctx->channels > 1)
1759 memcpy(&s->samples[s->current_stream * 2 + 1][s->offset[s->current_stream] * 512],
1760 s->frames[s->current_stream]->extended_data[1], 512 * 4);
1761 s->offset[s->current_stream]++;
1764 if (s->xma[s->current_stream].packet_done ||
1765 s->xma[s->current_stream].packet_loss) {
1768 if (s->xma[s->current_stream].skip_packets == 0) {
1770 } else if (s->xma[0].skip_packets == 0 && avctx->channels >= 2) {
1771 s->current_stream = 0;
1772 } else if (s->xma[1].skip_packets == 0 && avctx->channels >= 4) {
1773 s->current_stream = 1;
1774 } else if (s->xma[2].skip_packets == 0 && avctx->channels >= 6) {
1775 s->current_stream = 2;
1776 } else if (s->xma[3].skip_packets == 0 && avctx->channels == 8) {
1777 s->current_stream = 3;
1781 min[0] = s->xma[0].skip_packets;
1784 for (i = 1; i < avctx->channels / 2; i++) {
1785 if (s->xma[i].skip_packets < min[0]) {
1787 min[0] = s->xma[i].skip_packets;
1791 s->current_stream = min[1];
1794 for (i = 0; i < avctx->channels / 2; i++) {
1795 s->xma[i].skip_packets = FFMAX(0, s->xma[i].skip_packets - 1);
1798 for (i = 0; i < (avctx->channels + 1) / 2; i++) {
1799 offset = FFMIN(offset, s->offset[i]);
1803 frame->nb_samples = 512 * offset;
1804 if ((bret = ff_get_buffer(avctx, frame, 0)) < 0)
1807 for (i = 0; i < (avctx->channels + 1) / 2; i++) {
1808 memcpy(frame->extended_data[i * 2 + 0], s->samples[i * 2 + 0], frame->nb_samples * 4);
1809 if (avctx->channels > 1)
1810 memcpy(frame->extended_data[i * 2 + 1], s->samples[i * 2 + 1], frame->nb_samples * 4);
1811 s->offset[i] -= offset;
1813 memmove(s->samples[i * 2 + 0], s->samples[i * 2 + 0] + frame->nb_samples, s->offset[i] * 4 * 512);
1814 if (avctx->channels > 1)
1815 memmove(s->samples[i * 2 + 1], s->samples[i * 2 + 1] + frame->nb_samples, s->offset[i] * 4 * 512);
1826 static av_cold int xma_decode_init(AVCodecContext *avctx)
1828 XMADecodeCtx *s = avctx->priv_data;
1831 if (avctx->channels <= 0 || avctx->channels > 8)
1832 return AVERROR_INVALIDDATA;
1834 for (i = 0; i < (avctx->channels + 1) / 2; i++) {
1835 ret = decode_init(&s->xma[i], avctx);
1838 s->frames[i] = av_frame_alloc();
1840 return AVERROR(ENOMEM);
1841 s->frames[i]->nb_samples = 512;
1842 if ((ret = ff_get_buffer(avctx, s->frames[i], 0)) < 0) {
1843 return AVERROR(ENOMEM);
1851 static av_cold int xma_decode_end(AVCodecContext *avctx)
1853 XMADecodeCtx *s = avctx->priv_data;
1856 for (i = 0; i < avctx->channels / 2; i++) {
1857 decode_end(&s->xma[i]);
1858 av_frame_free(&s->frames[i]);
1864 static void flush(WMAProDecodeCtx *s)
1867 /** reset output buffer as a part of it is used during the windowing of a
1869 for (i = 0; i < s->nb_channels; i++)
1870 memset(s->channel[i].out, 0, s->samples_per_frame *
1871 sizeof(*s->channel[i].out));
1873 s->skip_packets = 0;
1878 *@brief Clear decoder buffers (for seeking).
1879 *@param avctx codec context
1881 static void wmapro_flush(AVCodecContext *avctx)
1883 WMAProDecodeCtx *s = avctx->priv_data;
1888 static void xma_flush(AVCodecContext *avctx)
1890 XMADecodeCtx *s = avctx->priv_data;
1892 for (i = 0; i < (avctx->channels + 1) / 2; i++)
1895 memset(s->offset, 0, sizeof(s->offset));
1896 s->current_stream = 0;
1901 *@brief wmapro decoder
1903 AVCodec ff_wmapro_decoder = {
1905 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 9 Professional"),
1906 .type = AVMEDIA_TYPE_AUDIO,
1907 .id = AV_CODEC_ID_WMAPRO,
1908 .priv_data_size = sizeof(WMAProDecodeCtx),
1909 .init = wmapro_decode_init,
1910 .close = wmapro_decode_end,
1911 .decode = wmapro_decode_packet,
1912 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1913 .flush = wmapro_flush,
1914 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1915 AV_SAMPLE_FMT_NONE },
1918 AVCodec ff_xma1_decoder = {
1920 .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 1"),
1921 .type = AVMEDIA_TYPE_AUDIO,
1922 .id = AV_CODEC_ID_XMA1,
1923 .priv_data_size = sizeof(XMADecodeCtx),
1924 .init = xma_decode_init,
1925 .close = xma_decode_end,
1926 .decode = xma_decode_packet,
1927 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1928 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1929 AV_SAMPLE_FMT_NONE },
1932 AVCodec ff_xma2_decoder = {
1934 .long_name = NULL_IF_CONFIG_SMALL("Xbox Media Audio 2"),
1935 .type = AVMEDIA_TYPE_AUDIO,
1936 .id = AV_CODEC_ID_XMA2,
1937 .priv_data_size = sizeof(XMADecodeCtx),
1938 .init = xma_decode_init,
1939 .close = xma_decode_end,
1940 .decode = xma_decode_packet,
1942 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1943 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1944 AV_SAMPLE_FMT_NONE },