FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
atrac1.c
Go to the documentation of this file.
1 /*
2  * Atrac 1 compatible decoder
3  * Copyright (c) 2009 Maxim Poliakovski
4  * Copyright (c) 2009 Benjamin Larsson
5  *
6  * This file is part of FFmpeg.
7  *
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.
12  *
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.
17  *
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
21  */
22 
23 /**
24  * @file
25  * Atrac 1 compatible decoder.
26  * This decoder handles raw ATRAC1 data and probably SDDS data.
27  */
28 
29 /* Many thanks to Tim Craig for all the help! */
30 
31 #include <math.h>
32 #include <stddef.h>
33 #include <stdio.h>
34 
35 #include "avcodec.h"
36 #include "get_bits.h"
37 #include "dsputil.h"
38 #include "fft.h"
39 #include "internal.h"
40 #include "sinewin.h"
41 
42 #include "atrac.h"
43 #include "atrac1data.h"
44 
45 #define AT1_MAX_BFU 52 ///< max number of block floating units in a sound unit
46 #define AT1_SU_SIZE 212 ///< number of bytes in a sound unit
47 #define AT1_SU_SAMPLES 512 ///< number of samples in a sound unit
48 #define AT1_FRAME_SIZE AT1_SU_SIZE * 2
49 #define AT1_SU_MAX_BITS AT1_SU_SIZE * 8
50 #define AT1_MAX_CHANNELS 2
51 
52 #define AT1_QMF_BANDS 3
53 #define IDX_LOW_BAND 0
54 #define IDX_MID_BAND 1
55 #define IDX_HIGH_BAND 2
56 
57 /**
58  * Sound unit struct, one unit is used per channel
59  */
60 typedef struct {
61  int log2_block_count[AT1_QMF_BANDS]; ///< log2 number of blocks in a band
62  int num_bfus; ///< number of Block Floating Units
63  float* spectrum[2];
64  DECLARE_ALIGNED(32, float, spec1)[AT1_SU_SAMPLES]; ///< mdct buffer
65  DECLARE_ALIGNED(32, float, spec2)[AT1_SU_SAMPLES]; ///< mdct buffer
66  DECLARE_ALIGNED(32, float, fst_qmf_delay)[46]; ///< delay line for the 1st stacked QMF filter
67  DECLARE_ALIGNED(32, float, snd_qmf_delay)[46]; ///< delay line for the 2nd stacked QMF filter
68  DECLARE_ALIGNED(32, float, last_qmf_delay)[256+23]; ///< delay line for the last stacked QMF filter
69 } AT1SUCtx;
70 
71 /**
72  * The atrac1 context, holds all needed parameters for decoding
73  */
74 typedef struct {
76  AT1SUCtx SUs[AT1_MAX_CHANNELS]; ///< channel sound unit
77  DECLARE_ALIGNED(32, float, spec)[AT1_SU_SAMPLES]; ///< the mdct spectrum buffer
78 
79  DECLARE_ALIGNED(32, float, low)[256];
80  DECLARE_ALIGNED(32, float, mid)[256];
81  DECLARE_ALIGNED(32, float, high)[512];
82  float* bands[3];
83  FFTContext mdct_ctx[3];
85 } AT1Ctx;
86 
87 /** size of the transform in samples in the long mode for each QMF band */
88 static const uint16_t samples_per_band[3] = {128, 128, 256};
89 static const uint8_t mdct_long_nbits[3] = {7, 7, 8};
90 
91 
92 static void at1_imdct(AT1Ctx *q, float *spec, float *out, int nbits,
93  int rev_spec)
94 {
95  FFTContext* mdct_context = &q->mdct_ctx[nbits - 5 - (nbits > 6)];
96  int transf_size = 1 << nbits;
97 
98  if (rev_spec) {
99  int i;
100  for (i = 0; i < transf_size / 2; i++)
101  FFSWAP(float, spec[i], spec[transf_size - 1 - i]);
102  }
103  mdct_context->imdct_half(mdct_context, out, spec);
104 }
105 
106 
107 static int at1_imdct_block(AT1SUCtx* su, AT1Ctx *q)
108 {
109  int band_num, band_samples, log2_block_count, nbits, num_blocks, block_size;
110  unsigned int start_pos, ref_pos = 0, pos = 0;
111 
112  for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) {
113  float *prev_buf;
114  int j;
115 
116  band_samples = samples_per_band[band_num];
117  log2_block_count = su->log2_block_count[band_num];
118 
119  /* number of mdct blocks in the current QMF band: 1 - for long mode */
120  /* 4 for short mode(low/middle bands) and 8 for short mode(high band)*/
121  num_blocks = 1 << log2_block_count;
122 
123  if (num_blocks == 1) {
124  /* mdct block size in samples: 128 (long mode, low & mid bands), */
125  /* 256 (long mode, high band) and 32 (short mode, all bands) */
126  block_size = band_samples >> log2_block_count;
127 
128  /* calc transform size in bits according to the block_size_mode */
129  nbits = mdct_long_nbits[band_num] - log2_block_count;
130 
131  if (nbits != 5 && nbits != 7 && nbits != 8)
132  return AVERROR_INVALIDDATA;
133  } else {
134  block_size = 32;
135  nbits = 5;
136  }
137 
138  start_pos = 0;
139  prev_buf = &su->spectrum[1][ref_pos + band_samples - 16];
140  for (j=0; j < num_blocks; j++) {
141  at1_imdct(q, &q->spec[pos], &su->spectrum[0][ref_pos + start_pos], nbits, band_num);
142 
143  /* overlap and window */
144  q->dsp.vector_fmul_window(&q->bands[band_num][start_pos], prev_buf,
145  &su->spectrum[0][ref_pos + start_pos], ff_sine_32, 16);
146 
147  prev_buf = &su->spectrum[0][ref_pos+start_pos + 16];
148  start_pos += block_size;
149  pos += block_size;
150  }
151 
152  if (num_blocks == 1)
153  memcpy(q->bands[band_num] + 32, &su->spectrum[0][ref_pos + 16], 240 * sizeof(float));
154 
155  ref_pos += band_samples;
156  }
157 
158  /* Swap buffers so the mdct overlap works */
159  FFSWAP(float*, su->spectrum[0], su->spectrum[1]);
160 
161  return 0;
162 }
163 
164 /**
165  * Parse the block size mode byte
166  */
167 
168 static int at1_parse_bsm(GetBitContext* gb, int log2_block_cnt[AT1_QMF_BANDS])
169 {
170  int log2_block_count_tmp, i;
171 
172  for (i = 0; i < 2; i++) {
173  /* low and mid band */
174  log2_block_count_tmp = get_bits(gb, 2);
175  if (log2_block_count_tmp & 1)
176  return AVERROR_INVALIDDATA;
177  log2_block_cnt[i] = 2 - log2_block_count_tmp;
178  }
179 
180  /* high band */
181  log2_block_count_tmp = get_bits(gb, 2);
182  if (log2_block_count_tmp != 0 && log2_block_count_tmp != 3)
183  return AVERROR_INVALIDDATA;
184  log2_block_cnt[IDX_HIGH_BAND] = 3 - log2_block_count_tmp;
185 
186  skip_bits(gb, 2);
187  return 0;
188 }
189 
190 
192  float spec[AT1_SU_SAMPLES])
193 {
194  int bits_used, band_num, bfu_num, i;
195  uint8_t idwls[AT1_MAX_BFU]; ///< the word length indexes for each BFU
196  uint8_t idsfs[AT1_MAX_BFU]; ///< the scalefactor indexes for each BFU
197 
198  /* parse the info byte (2nd byte) telling how much BFUs were coded */
199  su->num_bfus = bfu_amount_tab1[get_bits(gb, 3)];
200 
201  /* calc number of consumed bits:
202  num_BFUs * (idwl(4bits) + idsf(6bits)) + log2_block_count(8bits) + info_byte(8bits)
203  + info_byte_copy(8bits) + log2_block_count_copy(8bits) */
204  bits_used = su->num_bfus * 10 + 32 +
205  bfu_amount_tab2[get_bits(gb, 2)] +
206  (bfu_amount_tab3[get_bits(gb, 3)] << 1);
207 
208  /* get word length index (idwl) for each BFU */
209  for (i = 0; i < su->num_bfus; i++)
210  idwls[i] = get_bits(gb, 4);
211 
212  /* get scalefactor index (idsf) for each BFU */
213  for (i = 0; i < su->num_bfus; i++)
214  idsfs[i] = get_bits(gb, 6);
215 
216  /* zero idwl/idsf for empty BFUs */
217  for (i = su->num_bfus; i < AT1_MAX_BFU; i++)
218  idwls[i] = idsfs[i] = 0;
219 
220  /* read in the spectral data and reconstruct MDCT spectrum of this channel */
221  for (band_num = 0; band_num < AT1_QMF_BANDS; band_num++) {
222  for (bfu_num = bfu_bands_t[band_num]; bfu_num < bfu_bands_t[band_num+1]; bfu_num++) {
223  int pos;
224 
225  int num_specs = specs_per_bfu[bfu_num];
226  int word_len = !!idwls[bfu_num] + idwls[bfu_num];
227  float scale_factor = ff_atrac_sf_table[idsfs[bfu_num]];
228  bits_used += word_len * num_specs; /* add number of bits consumed by current BFU */
229 
230  /* check for bitstream overflow */
231  if (bits_used > AT1_SU_MAX_BITS)
232  return AVERROR_INVALIDDATA;
233 
234  /* get the position of the 1st spec according to the block size mode */
235  pos = su->log2_block_count[band_num] ? bfu_start_short[bfu_num] : bfu_start_long[bfu_num];
236 
237  if (word_len) {
238  float max_quant = 1.0 / (float)((1 << (word_len - 1)) - 1);
239 
240  for (i = 0; i < num_specs; i++) {
241  /* read in a quantized spec and convert it to
242  * signed int and then inverse quantization
243  */
244  spec[pos+i] = get_sbits(gb, word_len) * scale_factor * max_quant;
245  }
246  } else { /* word_len = 0 -> empty BFU, zero all specs in the empty BFU */
247  memset(&spec[pos], 0, num_specs * sizeof(float));
248  }
249  }
250  }
251 
252  return 0;
253 }
254 
255 
256 static void at1_subband_synthesis(AT1Ctx *q, AT1SUCtx* su, float *pOut)
257 {
258  float temp[256];
259  float iqmf_temp[512 + 46];
260 
261  /* combine low and middle bands */
262  ff_atrac_iqmf(q->bands[0], q->bands[1], 128, temp, su->fst_qmf_delay, iqmf_temp);
263 
264  /* delay the signal of the high band by 23 samples */
265  memcpy( su->last_qmf_delay, &su->last_qmf_delay[256], sizeof(float) * 23);
266  memcpy(&su->last_qmf_delay[23], q->bands[2], sizeof(float) * 256);
267 
268  /* combine (low + middle) and high bands */
269  ff_atrac_iqmf(temp, su->last_qmf_delay, 256, pOut, su->snd_qmf_delay, iqmf_temp);
270 }
271 
272 
273 static int atrac1_decode_frame(AVCodecContext *avctx, void *data,
274  int *got_frame_ptr, AVPacket *avpkt)
275 {
276  const uint8_t *buf = avpkt->data;
277  int buf_size = avpkt->size;
278  AT1Ctx *q = avctx->priv_data;
279  int ch, ret;
280  GetBitContext gb;
281 
282 
283  if (buf_size < 212 * avctx->channels) {
284  av_log(avctx, AV_LOG_ERROR, "Not enough data to decode!\n");
285  return AVERROR_INVALIDDATA;
286  }
287 
288  /* get output buffer */
290  if ((ret = ff_get_buffer(avctx, &q->frame)) < 0) {
291  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
292  return ret;
293  }
294 
295  for (ch = 0; ch < avctx->channels; ch++) {
296  AT1SUCtx* su = &q->SUs[ch];
297 
298  init_get_bits(&gb, &buf[212 * ch], 212 * 8);
299 
300  /* parse block_size_mode, 1st byte */
301  ret = at1_parse_bsm(&gb, su->log2_block_count);
302  if (ret < 0)
303  return ret;
304 
305  ret = at1_unpack_dequant(&gb, su, q->spec);
306  if (ret < 0)
307  return ret;
308 
309  ret = at1_imdct_block(su, q);
310  if (ret < 0)
311  return ret;
312  at1_subband_synthesis(q, su, (float *)q->frame.extended_data[ch]);
313  }
314 
315  *got_frame_ptr = 1;
316  *(AVFrame *)data = q->frame;
317 
318  return avctx->block_align;
319 }
320 
321 
323 {
324  AT1Ctx *q = avctx->priv_data;
325 
326  ff_mdct_end(&q->mdct_ctx[0]);
327  ff_mdct_end(&q->mdct_ctx[1]);
328  ff_mdct_end(&q->mdct_ctx[2]);
329 
330  return 0;
331 }
332 
333 
335 {
336  AT1Ctx *q = avctx->priv_data;
337  int ret;
338 
340 
341  if (avctx->channels < 1 || avctx->channels > AT1_MAX_CHANNELS) {
342  av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %d\n",
343  avctx->channels);
344  return AVERROR(EINVAL);
345  }
346 
347  if (avctx->block_align <= 0) {
348  av_log_ask_for_sample(avctx, "unsupported block align\n");
349  return AVERROR_PATCHWELCOME;
350  }
351 
352  /* Init the mdct transforms */
353  if ((ret = ff_mdct_init(&q->mdct_ctx[0], 6, 1, -1.0/ (1 << 15))) ||
354  (ret = ff_mdct_init(&q->mdct_ctx[1], 8, 1, -1.0/ (1 << 15))) ||
355  (ret = ff_mdct_init(&q->mdct_ctx[2], 9, 1, -1.0/ (1 << 15)))) {
356  av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
357  atrac1_decode_end(avctx);
358  return ret;
359  }
360 
362 
364 
365  ff_dsputil_init(&q->dsp, avctx);
366 
367  q->bands[0] = q->low;
368  q->bands[1] = q->mid;
369  q->bands[2] = q->high;
370 
371  /* Prepare the mdct overlap buffers */
372  q->SUs[0].spectrum[0] = q->SUs[0].spec1;
373  q->SUs[0].spectrum[1] = q->SUs[0].spec2;
374  q->SUs[1].spectrum[0] = q->SUs[1].spec1;
375  q->SUs[1].spectrum[1] = q->SUs[1].spec2;
376 
378  avctx->coded_frame = &q->frame;
379 
380  return 0;
381 }
382 
383 
385  .name = "atrac1",
386  .type = AVMEDIA_TYPE_AUDIO,
387  .id = AV_CODEC_ID_ATRAC1,
388  .priv_data_size = sizeof(AT1Ctx),
392  .capabilities = CODEC_CAP_DR1,
393  .long_name = NULL_IF_CONFIG_SMALL("Atrac 1 (Adaptive TRansform Acoustic Coding)"),
394  .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
396 };