FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
cook.c
Go to the documentation of this file.
1 /*
2  * COOK compatible decoder
3  * Copyright (c) 2003 Sascha Sommer
4  * Copyright (c) 2005 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  * Cook compatible decoder. Bastardization of the G.722.1 standard.
26  * This decoder handles RealNetworks, RealAudio G2 data.
27  * Cook is identified by the codec name cook in RM files.
28  *
29  * To use this decoder, a calling application must supply the extradata
30  * bytes provided from the RM container; 8+ bytes for mono streams and
31  * 16+ for stereo streams (maybe more).
32  *
33  * Codec technicalities (all this assume a buffer length of 1024):
34  * Cook works with several different techniques to achieve its compression.
35  * In the timedomain the buffer is divided into 8 pieces and quantized. If
36  * two neighboring pieces have different quantization index a smooth
37  * quantization curve is used to get a smooth overlap between the different
38  * pieces.
39  * To get to the transformdomain Cook uses a modulated lapped transform.
40  * The transform domain has 50 subbands with 20 elements each. This
41  * means only a maximum of 50*20=1000 coefficients are used out of the 1024
42  * available.
43  */
44 
46 #include "libavutil/lfg.h"
47 #include "avcodec.h"
48 #include "get_bits.h"
49 #include "dsputil.h"
50 #include "bytestream.h"
51 #include "fft.h"
52 #include "internal.h"
53 #include "sinewin.h"
54 
55 #include "cookdata.h"
56 
57 /* the different Cook versions */
58 #define MONO 0x1000001
59 #define STEREO 0x1000002
60 #define JOINT_STEREO 0x1000003
61 #define MC_COOK 0x2000000 // multichannel Cook, not supported
62 
63 #define SUBBAND_SIZE 20
64 #define MAX_SUBPACKETS 5
65 
66 typedef struct {
67  int *now;
68  int *previous;
69 } cook_gains;
70 
71 typedef struct {
72  int ch_idx;
73  int size;
76  int subbands;
81  unsigned int channel_mask;
87  int numvector_size; // 1 << log2_numvector_size;
88 
89  float mono_previous_buffer1[1024];
90  float mono_previous_buffer2[1024];
91 
94  int gain_1[9];
95  int gain_2[9];
96  int gain_3[9];
97  int gain_4[9];
99 
100 typedef struct cook {
101  /*
102  * The following 5 functions provide the lowlevel arithmetic on
103  * the internal audio buffers.
104  */
105  void (*scalar_dequant)(struct cook *q, int index, int quant_index,
106  int *subband_coef_index, int *subband_coef_sign,
107  float *mlt_p);
108 
109  void (*decouple)(struct cook *q,
110  COOKSubpacket *p,
111  int subband,
112  float f1, float f2,
113  float *decode_buffer,
114  float *mlt_buffer1, float *mlt_buffer2);
115 
116  void (*imlt_window)(struct cook *q, float *buffer1,
117  cook_gains *gains_ptr, float *previous_buffer);
118 
119  void (*interpolate)(struct cook *q, float *buffer,
120  int gain_index, int gain_index_next);
121 
122  void (*saturate_output)(struct cook *q, float *out);
123 
127  /* stream data */
130  /* states */
133 
134  /* transform data */
136  float* mlt_window;
137 
138  /* VLC data */
139  VLC envelope_quant_index[13];
140  VLC sqvh[7]; // scalar quantization
141 
142  /* generatable tables and related variables */
144  float gain_table[23];
145 
146  /* data buffers */
147 
149  DECLARE_ALIGNED(32, float, mono_mdct_output)[2048];
150  float decode_buffer_1[1024];
151  float decode_buffer_2[1024];
152  float decode_buffer_0[1060]; /* static allocation for joint decode */
153 
154  const float *cplscales[5];
157 } COOKContext;
158 
159 static float pow2tab[127];
160 static float rootpow2tab[127];
161 
162 /*************** init functions ***************/
163 
164 /* table generator */
165 static av_cold void init_pow2table(void)
166 {
167  int i;
168  for (i = -63; i < 64; i++) {
169  pow2tab[63 + i] = pow(2, i);
170  rootpow2tab[63 + i] = sqrt(pow(2, i));
171  }
172 }
173 
174 /* table generator */
176 {
177  int i;
179  for (i = 0; i < 23; i++)
180  q->gain_table[i] = pow(pow2tab[i + 52],
181  (1.0 / (double) q->gain_size_factor));
182 }
183 
184 
186 {
187  int i, result;
188 
189  result = 0;
190  for (i = 0; i < 13; i++) {
191  result |= init_vlc(&q->envelope_quant_index[i], 9, 24,
193  envelope_quant_index_huffcodes[i], 2, 2, 0);
194  }
195  av_log(q->avctx, AV_LOG_DEBUG, "sqvh VLC init\n");
196  for (i = 0; i < 7; i++) {
197  result |= init_vlc(&q->sqvh[i], vhvlcsize_tab[i], vhsize_tab[i],
198  cvh_huffbits[i], 1, 1,
199  cvh_huffcodes[i], 2, 2, 0);
200  }
201 
202  for (i = 0; i < q->num_subpackets; i++) {
203  if (q->subpacket[i].joint_stereo == 1) {
204  result |= init_vlc(&q->subpacket[i].channel_coupling, 6,
205  (1 << q->subpacket[i].js_vlc_bits) - 1,
206  ccpl_huffbits[q->subpacket[i].js_vlc_bits - 2], 1, 1,
207  ccpl_huffcodes[q->subpacket[i].js_vlc_bits - 2], 2, 2, 0);
208  av_log(q->avctx, AV_LOG_DEBUG, "subpacket %i Joint-stereo VLC used.\n", i);
209  }
210  }
211 
212  av_log(q->avctx, AV_LOG_DEBUG, "VLC tables initialized.\n");
213  return result;
214 }
215 
217 {
218  int j, ret;
219  int mlt_size = q->samples_per_channel;
220 
221  if ((q->mlt_window = av_malloc(mlt_size * sizeof(*q->mlt_window))) == 0)
222  return AVERROR(ENOMEM);
223 
224  /* Initialize the MLT window: simple sine window. */
225  ff_sine_window_init(q->mlt_window, mlt_size);
226  for (j = 0; j < mlt_size; j++)
227  q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
228 
229  /* Initialize the MDCT. */
230  if ((ret = ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size) + 1, 1, 1.0 / 32768.0))) {
231  av_free(q->mlt_window);
232  return ret;
233  }
234  av_log(q->avctx, AV_LOG_DEBUG, "MDCT initialized, order = %d.\n",
235  av_log2(mlt_size) + 1);
236 
237  return 0;
238 }
239 
241 {
242  int i;
243  for (i = 0; i < 5; i++)
244  q->cplscales[i] = cplscales[i];
245 }
246 
247 /*************** init functions end ***********/
248 
249 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes) + 3) % 4)
250 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
251 
252 /**
253  * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
254  * Why? No idea, some checksum/error detection method maybe.
255  *
256  * Out buffer size: extra bytes are needed to cope with
257  * padding/misalignment.
258  * Subpackets passed to the decoder can contain two, consecutive
259  * half-subpackets, of identical but arbitrary size.
260  * 1234 1234 1234 1234 extraA extraB
261  * Case 1: AAAA BBBB 0 0
262  * Case 2: AAAA ABBB BB-- 3 3
263  * Case 3: AAAA AABB BBBB 2 2
264  * Case 4: AAAA AAAB BBBB BB-- 1 5
265  *
266  * Nice way to waste CPU cycles.
267  *
268  * @param inbuffer pointer to byte array of indata
269  * @param out pointer to byte array of outdata
270  * @param bytes number of bytes
271  */
272 static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
273 {
274  static const uint32_t tab[4] = {
275  AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
276  AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
277  };
278  int i, off;
279  uint32_t c;
280  const uint32_t *buf;
281  uint32_t *obuf = (uint32_t *) out;
282  /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
283  * I'm too lazy though, should be something like
284  * for (i = 0; i < bitamount / 64; i++)
285  * (int64_t) out[i] = 0x37c511f237c511f2 ^ av_be2ne64(int64_t) in[i]);
286  * Buffer alignment needs to be checked. */
287 
288  off = (intptr_t) inbuffer & 3;
289  buf = (const uint32_t *) (inbuffer - off);
290  c = tab[off];
291  bytes += 3 + off;
292  for (i = 0; i < bytes / 4; i++)
293  obuf[i] = c ^ buf[i];
294 
295  return off;
296 }
297 
299 {
300  int i;
301  COOKContext *q = avctx->priv_data;
302  av_log(avctx, AV_LOG_DEBUG, "Deallocating memory.\n");
303 
304  /* Free allocated memory buffers. */
305  av_free(q->mlt_window);
307 
308  /* Free the transform. */
309  ff_mdct_end(&q->mdct_ctx);
310 
311  /* Free the VLC tables. */
312  for (i = 0; i < 13; i++)
314  for (i = 0; i < 7; i++)
315  ff_free_vlc(&q->sqvh[i]);
316  for (i = 0; i < q->num_subpackets; i++)
318 
319  av_log(avctx, AV_LOG_DEBUG, "Memory deallocated.\n");
320 
321  return 0;
322 }
323 
324 /**
325  * Fill the gain array for the timedomain quantization.
326  *
327  * @param gb pointer to the GetBitContext
328  * @param gaininfo array[9] of gain indexes
329  */
330 static void decode_gain_info(GetBitContext *gb, int *gaininfo)
331 {
332  int i, n;
333 
334  while (get_bits1(gb)) {
335  /* NOTHING */
336  }
337 
338  n = get_bits_count(gb) - 1; // amount of elements*2 to update
339 
340  i = 0;
341  while (n--) {
342  int index = get_bits(gb, 3);
343  int gain = get_bits1(gb) ? get_bits(gb, 4) - 7 : -1;
344 
345  while (i <= index)
346  gaininfo[i++] = gain;
347  }
348  while (i <= 8)
349  gaininfo[i++] = 0;
350 }
351 
352 /**
353  * Create the quant index table needed for the envelope.
354  *
355  * @param q pointer to the COOKContext
356  * @param quant_index_table pointer to the array
357  */
359  int *quant_index_table)
360 {
361  int i, j, vlc_index;
362 
363  quant_index_table[0] = get_bits(&q->gb, 6) - 6; // This is used later in categorize
364 
365  for (i = 1; i < p->total_subbands; i++) {
366  vlc_index = i;
367  if (i >= p->js_subband_start * 2) {
368  vlc_index -= p->js_subband_start;
369  } else {
370  vlc_index /= 2;
371  if (vlc_index < 1)
372  vlc_index = 1;
373  }
374  if (vlc_index > 13)
375  vlc_index = 13; // the VLC tables >13 are identical to No. 13
376 
377  j = get_vlc2(&q->gb, q->envelope_quant_index[vlc_index - 1].table,
378  q->envelope_quant_index[vlc_index - 1].bits, 2);
379  quant_index_table[i] = quant_index_table[i - 1] + j - 12; // differential encoding
380  if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
382  "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
383  quant_index_table[i], i);
384  return AVERROR_INVALIDDATA;
385  }
386  }
387 
388  return 0;
389 }
390 
391 /**
392  * Calculate the category and category_index vector.
393  *
394  * @param q pointer to the COOKContext
395  * @param quant_index_table pointer to the array
396  * @param category pointer to the category array
397  * @param category_index pointer to the category_index array
398  */
399 static void categorize(COOKContext *q, COOKSubpacket *p, const int *quant_index_table,
400  int *category, int *category_index)
401 {
402  int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
403  int exp_index2[102] = { 0 };
404  int exp_index1[102] = { 0 };
405 
406  int tmp_categorize_array[128 * 2] = { 0 };
407  int tmp_categorize_array1_idx = p->numvector_size;
408  int tmp_categorize_array2_idx = p->numvector_size;
409 
410  bits_left = p->bits_per_subpacket - get_bits_count(&q->gb);
411 
412  if (bits_left > q->samples_per_channel)
413  bits_left = q->samples_per_channel +
414  ((bits_left - q->samples_per_channel) * 5) / 8;
415 
416  bias = -32;
417 
418  /* Estimate bias. */
419  for (i = 32; i > 0; i = i / 2) {
420  num_bits = 0;
421  index = 0;
422  for (j = p->total_subbands; j > 0; j--) {
423  exp_idx = av_clip((i - quant_index_table[index] + bias) / 2, 0, 7);
424  index++;
425  num_bits += expbits_tab[exp_idx];
426  }
427  if (num_bits >= bits_left - 32)
428  bias += i;
429  }
430 
431  /* Calculate total number of bits. */
432  num_bits = 0;
433  for (i = 0; i < p->total_subbands; i++) {
434  exp_idx = av_clip((bias - quant_index_table[i]) / 2, 0, 7);
435  num_bits += expbits_tab[exp_idx];
436  exp_index1[i] = exp_idx;
437  exp_index2[i] = exp_idx;
438  }
439  tmpbias1 = tmpbias2 = num_bits;
440 
441  for (j = 1; j < p->numvector_size; j++) {
442  if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
443  int max = -999999;
444  index = -1;
445  for (i = 0; i < p->total_subbands; i++) {
446  if (exp_index1[i] < 7) {
447  v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
448  if (v >= max) {
449  max = v;
450  index = i;
451  }
452  }
453  }
454  if (index == -1)
455  break;
456  tmp_categorize_array[tmp_categorize_array1_idx++] = index;
457  tmpbias1 -= expbits_tab[exp_index1[index]] -
458  expbits_tab[exp_index1[index] + 1];
459  ++exp_index1[index];
460  } else { /* <--- */
461  int min = 999999;
462  index = -1;
463  for (i = 0; i < p->total_subbands; i++) {
464  if (exp_index2[i] > 0) {
465  v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
466  if (v < min) {
467  min = v;
468  index = i;
469  }
470  }
471  }
472  if (index == -1)
473  break;
474  tmp_categorize_array[--tmp_categorize_array2_idx] = index;
475  tmpbias2 -= expbits_tab[exp_index2[index]] -
476  expbits_tab[exp_index2[index] - 1];
477  --exp_index2[index];
478  }
479  }
480 
481  for (i = 0; i < p->total_subbands; i++)
482  category[i] = exp_index2[i];
483 
484  for (i = 0; i < p->numvector_size - 1; i++)
485  category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
486 }
487 
488 
489 /**
490  * Expand the category vector.
491  *
492  * @param q pointer to the COOKContext
493  * @param category pointer to the category array
494  * @param category_index pointer to the category_index array
495  */
496 static inline void expand_category(COOKContext *q, int *category,
497  int *category_index)
498 {
499  int i;
500  for (i = 0; i < q->num_vectors; i++)
501  {
502  int idx = category_index[i];
503  if (++category[idx] >= FF_ARRAY_ELEMS(dither_tab))
504  --category[idx];
505  }
506 }
507 
508 /**
509  * The real requantization of the mltcoefs
510  *
511  * @param q pointer to the COOKContext
512  * @param index index
513  * @param quant_index quantisation index
514  * @param subband_coef_index array of indexes to quant_centroid_tab
515  * @param subband_coef_sign signs of coefficients
516  * @param mlt_p pointer into the mlt buffer
517  */
518 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
519  int *subband_coef_index, int *subband_coef_sign,
520  float *mlt_p)
521 {
522  int i;
523  float f1;
524 
525  for (i = 0; i < SUBBAND_SIZE; i++) {
526  if (subband_coef_index[i]) {
527  f1 = quant_centroid_tab[index][subband_coef_index[i]];
528  if (subband_coef_sign[i])
529  f1 = -f1;
530  } else {
531  /* noise coding if subband_coef_index[i] == 0 */
532  f1 = dither_tab[index];
533  if (av_lfg_get(&q->random_state) < 0x80000000)
534  f1 = -f1;
535  }
536  mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
537  }
538 }
539 /**
540  * Unpack the subband_coef_index and subband_coef_sign vectors.
541  *
542  * @param q pointer to the COOKContext
543  * @param category pointer to the category array
544  * @param subband_coef_index array of indexes to quant_centroid_tab
545  * @param subband_coef_sign signs of coefficients
546  */
547 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
548  int *subband_coef_index, int *subband_coef_sign)
549 {
550  int i, j;
551  int vlc, vd, tmp, result;
552 
553  vd = vd_tab[category];
554  result = 0;
555  for (i = 0; i < vpr_tab[category]; i++) {
556  vlc = get_vlc2(&q->gb, q->sqvh[category].table, q->sqvh[category].bits, 3);
557  if (p->bits_per_subpacket < get_bits_count(&q->gb)) {
558  vlc = 0;
559  result = 1;
560  }
561  for (j = vd - 1; j >= 0; j--) {
562  tmp = (vlc * invradix_tab[category]) / 0x100000;
563  subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
564  vlc = tmp;
565  }
566  for (j = 0; j < vd; j++) {
567  if (subband_coef_index[i * vd + j]) {
568  if (get_bits_count(&q->gb) < p->bits_per_subpacket) {
569  subband_coef_sign[i * vd + j] = get_bits1(&q->gb);
570  } else {
571  result = 1;
572  subband_coef_sign[i * vd + j] = 0;
573  }
574  } else {
575  subband_coef_sign[i * vd + j] = 0;
576  }
577  }
578  }
579  return result;
580 }
581 
582 
583 /**
584  * Fill the mlt_buffer with mlt coefficients.
585  *
586  * @param q pointer to the COOKContext
587  * @param category pointer to the category array
588  * @param quant_index_table pointer to the array
589  * @param mlt_buffer pointer to mlt coefficients
590  */
591 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
592  int *quant_index_table, float *mlt_buffer)
593 {
594  /* A zero in this table means that the subband coefficient is
595  random noise coded. */
596  int subband_coef_index[SUBBAND_SIZE];
597  /* A zero in this table means that the subband coefficient is a
598  positive multiplicator. */
599  int subband_coef_sign[SUBBAND_SIZE];
600  int band, j;
601  int index = 0;
602 
603  for (band = 0; band < p->total_subbands; band++) {
604  index = category[band];
605  if (category[band] < 7) {
606  if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
607  index = 7;
608  for (j = 0; j < p->total_subbands; j++)
609  category[band + j] = 7;
610  }
611  }
612  if (index >= 7) {
613  memset(subband_coef_index, 0, sizeof(subband_coef_index));
614  memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
615  }
616  q->scalar_dequant(q, index, quant_index_table[band],
617  subband_coef_index, subband_coef_sign,
618  &mlt_buffer[band * SUBBAND_SIZE]);
619  }
620 
621  /* FIXME: should this be removed, or moved into loop above? */
623  return;
624 }
625 
626 
627 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
628 {
629  int category_index[128] = { 0 };
630  int category[128] = { 0 };
631  int quant_index_table[102];
632  int res, i;
633 
634  if ((res = decode_envelope(q, p, quant_index_table)) < 0)
635  return res;
637  categorize(q, p, quant_index_table, category, category_index);
638  expand_category(q, category, category_index);
639  for (i=0; i<p->total_subbands; i++) {
640  if (category[i] > 7)
641  return AVERROR_INVALIDDATA;
642  }
643  decode_vectors(q, p, category, quant_index_table, mlt_buffer);
644 
645  return 0;
646 }
647 
648 
649 /**
650  * the actual requantization of the timedomain samples
651  *
652  * @param q pointer to the COOKContext
653  * @param buffer pointer to the timedomain buffer
654  * @param gain_index index for the block multiplier
655  * @param gain_index_next index for the next block multiplier
656  */
657 static void interpolate_float(COOKContext *q, float *buffer,
658  int gain_index, int gain_index_next)
659 {
660  int i;
661  float fc1, fc2;
662  fc1 = pow2tab[gain_index + 63];
663 
664  if (gain_index == gain_index_next) { // static gain
665  for (i = 0; i < q->gain_size_factor; i++)
666  buffer[i] *= fc1;
667  } else { // smooth gain
668  fc2 = q->gain_table[11 + (gain_index_next - gain_index)];
669  for (i = 0; i < q->gain_size_factor; i++) {
670  buffer[i] *= fc1;
671  fc1 *= fc2;
672  }
673  }
674 }
675 
676 /**
677  * Apply transform window, overlap buffers.
678  *
679  * @param q pointer to the COOKContext
680  * @param inbuffer pointer to the mltcoefficients
681  * @param gains_ptr current and previous gains
682  * @param previous_buffer pointer to the previous buffer to be used for overlapping
683  */
684 static void imlt_window_float(COOKContext *q, float *inbuffer,
685  cook_gains *gains_ptr, float *previous_buffer)
686 {
687  const float fc = pow2tab[gains_ptr->previous[0] + 63];
688  int i;
689  /* The weird thing here, is that the two halves of the time domain
690  * buffer are swapped. Also, the newest data, that we save away for
691  * next frame, has the wrong sign. Hence the subtraction below.
692  * Almost sounds like a complex conjugate/reverse data/FFT effect.
693  */
694 
695  /* Apply window and overlap */
696  for (i = 0; i < q->samples_per_channel; i++)
697  inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
698  previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
699 }
700 
701 /**
702  * The modulated lapped transform, this takes transform coefficients
703  * and transforms them into timedomain samples.
704  * Apply transform window, overlap buffers, apply gain profile
705  * and buffer management.
706  *
707  * @param q pointer to the COOKContext
708  * @param inbuffer pointer to the mltcoefficients
709  * @param gains_ptr current and previous gains
710  * @param previous_buffer pointer to the previous buffer to be used for overlapping
711  */
712 static void imlt_gain(COOKContext *q, float *inbuffer,
713  cook_gains *gains_ptr, float *previous_buffer)
714 {
715  float *buffer0 = q->mono_mdct_output;
716  float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
717  int i;
718 
719  /* Inverse modified discrete cosine transform */
720  q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
721 
722  q->imlt_window(q, buffer1, gains_ptr, previous_buffer);
723 
724  /* Apply gain profile */
725  for (i = 0; i < 8; i++)
726  if (gains_ptr->now[i] || gains_ptr->now[i + 1])
727  q->interpolate(q, &buffer1[q->gain_size_factor * i],
728  gains_ptr->now[i], gains_ptr->now[i + 1]);
729 
730  /* Save away the current to be previous block. */
731  memcpy(previous_buffer, buffer0,
732  q->samples_per_channel * sizeof(*previous_buffer));
733 }
734 
735 
736 /**
737  * function for getting the jointstereo coupling information
738  *
739  * @param q pointer to the COOKContext
740  * @param decouple_tab decoupling array
741  */
742 static int decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
743 {
744  int i;
745  int vlc = get_bits1(&q->gb);
746  int start = cplband[p->js_subband_start];
747  int end = cplband[p->subbands - 1];
748  int length = end - start + 1;
749 
750  if (start > end)
751  return 0;
752 
753  if (vlc)
754  for (i = 0; i < length; i++)
755  decouple_tab[start + i] = get_vlc2(&q->gb,
757  p->channel_coupling.bits, 2);
758  else
759  for (i = 0; i < length; i++) {
760  int v = get_bits(&q->gb, p->js_vlc_bits);
761  if (v == (1<<p->js_vlc_bits)-1) {
762  av_log(q->avctx, AV_LOG_ERROR, "decouple value too large\n");
763  return AVERROR_INVALIDDATA;
764  }
765  decouple_tab[start + i] = v;
766  }
767  return 0;
768 }
769 
770 /**
771  * function decouples a pair of signals from a single signal via multiplication.
772  *
773  * @param q pointer to the COOKContext
774  * @param subband index of the current subband
775  * @param f1 multiplier for channel 1 extraction
776  * @param f2 multiplier for channel 2 extraction
777  * @param decode_buffer input buffer
778  * @param mlt_buffer1 pointer to left channel mlt coefficients
779  * @param mlt_buffer2 pointer to right channel mlt coefficients
780  */
782  COOKSubpacket *p,
783  int subband,
784  float f1, float f2,
785  float *decode_buffer,
786  float *mlt_buffer1, float *mlt_buffer2)
787 {
788  int j, tmp_idx;
789  for (j = 0; j < SUBBAND_SIZE; j++) {
790  tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
791  mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
792  mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
793  }
794 }
795 
796 /**
797  * function for decoding joint stereo data
798  *
799  * @param q pointer to the COOKContext
800  * @param mlt_buffer1 pointer to left channel mlt coefficients
801  * @param mlt_buffer2 pointer to right channel mlt coefficients
802  */
804  float *mlt_buffer_left, float *mlt_buffer_right)
805 {
806  int i, j, res;
807  int decouple_tab[SUBBAND_SIZE] = { 0 };
808  float *decode_buffer = q->decode_buffer_0;
809  int idx, cpl_tmp;
810  float f1, f2;
811  const float *cplscale;
812 
813  memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
814 
815  /* Make sure the buffers are zeroed out. */
816  memset(mlt_buffer_left, 0, 1024 * sizeof(*mlt_buffer_left));
817  memset(mlt_buffer_right, 0, 1024 * sizeof(*mlt_buffer_right));
818  if ((res = decouple_info(q, p, decouple_tab)) < 0)
819  return res;
820  if ((res = mono_decode(q, p, decode_buffer)) < 0)
821  return res;
822  /* The two channels are stored interleaved in decode_buffer. */
823  for (i = 0; i < p->js_subband_start; i++) {
824  for (j = 0; j < SUBBAND_SIZE; j++) {
825  mlt_buffer_left[i * 20 + j] = decode_buffer[i * 40 + j];
826  mlt_buffer_right[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
827  }
828  }
829 
830  /* When we reach js_subband_start (the higher frequencies)
831  the coefficients are stored in a coupling scheme. */
832  idx = (1 << p->js_vlc_bits) - 1;
833  for (i = p->js_subband_start; i < p->subbands; i++) {
834  cpl_tmp = cplband[i];
835  idx -= decouple_tab[cpl_tmp];
836  cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
837  f1 = cplscale[decouple_tab[cpl_tmp] + 1];
838  f2 = cplscale[idx];
839  q->decouple(q, p, i, f1, f2, decode_buffer,
840  mlt_buffer_left, mlt_buffer_right);
841  idx = (1 << p->js_vlc_bits) - 1;
842  }
843 
844  return 0;
845 }
846 
847 /**
848  * First part of subpacket decoding:
849  * decode raw stream bytes and read gain info.
850  *
851  * @param q pointer to the COOKContext
852  * @param inbuffer pointer to raw stream data
853  * @param gains_ptr array of current/prev gain pointers
854  */
856  const uint8_t *inbuffer,
857  cook_gains *gains_ptr)
858 {
859  int offset;
860 
861  offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
862  p->bits_per_subpacket / 8);
863  init_get_bits(&q->gb, q->decoded_bytes_buffer + offset,
864  p->bits_per_subpacket);
865  decode_gain_info(&q->gb, gains_ptr->now);
866 
867  /* Swap current and previous gains */
868  FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
869 }
870 
871 /**
872  * Saturate the output signal and interleave.
873  *
874  * @param q pointer to the COOKContext
875  * @param out pointer to the output vector
876  */
877 static void saturate_output_float(COOKContext *q, float *out)
878 {
880  -1.0f, 1.0f, FFALIGN(q->samples_per_channel, 8));
881 }
882 
883 
884 /**
885  * Final part of subpacket decoding:
886  * Apply modulated lapped transform, gain compensation,
887  * clip and convert to integer.
888  *
889  * @param q pointer to the COOKContext
890  * @param decode_buffer pointer to the mlt coefficients
891  * @param gains_ptr array of current/prev gain pointers
892  * @param previous_buffer pointer to the previous buffer to be used for overlapping
893  * @param out pointer to the output buffer
894  */
895 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
896  cook_gains *gains_ptr, float *previous_buffer,
897  float *out)
898 {
899  imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
900  if (out)
901  q->saturate_output(q, out);
902 }
903 
904 
905 /**
906  * Cook subpacket decoding. This function returns one decoded subpacket,
907  * usually 1024 samples per channel.
908  *
909  * @param q pointer to the COOKContext
910  * @param inbuffer pointer to the inbuffer
911  * @param outbuffer pointer to the outbuffer
912  */
914  const uint8_t *inbuffer, float **outbuffer)
915 {
916  int sub_packet_size = p->size;
917  int res;
918 
919  memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
920  decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
921 
922  if (p->joint_stereo) {
923  if ((res = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
924  return res;
925  } else {
926  if ((res = mono_decode(q, p, q->decode_buffer_1)) < 0)
927  return res;
928 
929  if (p->num_channels == 2) {
930  decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
931  if ((res = mono_decode(q, p, q->decode_buffer_2)) < 0)
932  return res;
933  }
934  }
935 
938  outbuffer ? outbuffer[p->ch_idx] : NULL);
939 
940  if (p->num_channels == 2) {
941  if (p->joint_stereo)
944  outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
945  else
948  outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
949  }
950 
951  return 0;
952 }
953 
954 
955 static int cook_decode_frame(AVCodecContext *avctx, void *data,
956  int *got_frame_ptr, AVPacket *avpkt)
957 {
958  AVFrame *frame = data;
959  const uint8_t *buf = avpkt->data;
960  int buf_size = avpkt->size;
961  COOKContext *q = avctx->priv_data;
962  float **samples = NULL;
963  int i, ret;
964  int offset = 0;
965  int chidx = 0;
966 
967  if (buf_size < avctx->block_align)
968  return buf_size;
969 
970  /* get output buffer */
971  if (q->discarded_packets >= 2) {
972  frame->nb_samples = q->samples_per_channel;
973  if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
974  return ret;
975  samples = (float **)frame->extended_data;
976  }
977 
978  /* estimate subpacket sizes */
979  q->subpacket[0].size = avctx->block_align;
980 
981  for (i = 1; i < q->num_subpackets; i++) {
982  q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
983  q->subpacket[0].size -= q->subpacket[i].size + 1;
984  if (q->subpacket[0].size < 0) {
985  av_log(avctx, AV_LOG_DEBUG,
986  "frame subpacket size total > avctx->block_align!\n");
987  return AVERROR_INVALIDDATA;
988  }
989  }
990 
991  /* decode supbackets */
992  for (i = 0; i < q->num_subpackets; i++) {
993  q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
995  q->subpacket[i].ch_idx = chidx;
996  av_log(avctx, AV_LOG_DEBUG,
997  "subpacket[%i] size %i js %i %i block_align %i\n",
998  i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
999  avctx->block_align);
1000 
1001  if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
1002  return ret;
1003  offset += q->subpacket[i].size;
1004  chidx += q->subpacket[i].num_channels;
1005  av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
1006  i, q->subpacket[i].size * 8, get_bits_count(&q->gb));
1007  }
1008 
1009  /* Discard the first two frames: no valid audio. */
1010  if (q->discarded_packets < 2) {
1011  q->discarded_packets++;
1012  *got_frame_ptr = 0;
1013  return avctx->block_align;
1014  }
1015 
1016  *got_frame_ptr = 1;
1017 
1018  return avctx->block_align;
1019 }
1020 
1021 #ifdef DEBUG
1022 static void dump_cook_context(COOKContext *q)
1023 {
1024  //int i=0;
1025 #define PRINT(a, b) av_dlog(q->avctx, " %s = %d\n", a, b);
1026  av_dlog(q->avctx, "COOKextradata\n");
1027  av_dlog(q->avctx, "cookversion=%x\n", q->subpacket[0].cookversion);
1028  if (q->subpacket[0].cookversion > STEREO) {
1029  PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1030  PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1031  }
1032  av_dlog(q->avctx, "COOKContext\n");
1033  PRINT("nb_channels", q->avctx->channels);
1034  PRINT("bit_rate", q->avctx->bit_rate);
1035  PRINT("sample_rate", q->avctx->sample_rate);
1036  PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1037  PRINT("subbands", q->subpacket[0].subbands);
1038  PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1039  PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1040  PRINT("numvector_size", q->subpacket[0].numvector_size);
1041  PRINT("total_subbands", q->subpacket[0].total_subbands);
1042 }
1043 #endif
1044 
1045 /**
1046  * Cook initialization
1047  *
1048  * @param avctx pointer to the AVCodecContext
1049  */
1051 {
1052  COOKContext *q = avctx->priv_data;
1053  const uint8_t *edata_ptr = avctx->extradata;
1054  const uint8_t *edata_ptr_end = edata_ptr + avctx->extradata_size;
1055  int extradata_size = avctx->extradata_size;
1056  int s = 0;
1057  unsigned int channel_mask = 0;
1058  int samples_per_frame = 0;
1059  int ret;
1060  q->avctx = avctx;
1061 
1062  /* Take care of the codec specific extradata. */
1063  if (extradata_size <= 0) {
1064  av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1065  return AVERROR_INVALIDDATA;
1066  }
1067  av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1068 
1069  /* Take data from the AVCodecContext (RM container). */
1070  if (!avctx->channels) {
1071  av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1072  return AVERROR_INVALIDDATA;
1073  }
1074 
1075  /* Initialize RNG. */
1076  av_lfg_init(&q->random_state, 0);
1077 
1078  ff_dsputil_init(&q->dsp, avctx);
1079 
1080  while (edata_ptr < edata_ptr_end) {
1081  /* 8 for mono, 16 for stereo, ? for multichannel
1082  Swap to right endianness so we don't need to care later on. */
1083  if (extradata_size >= 8) {
1084  q->subpacket[s].cookversion = bytestream_get_be32(&edata_ptr);
1085  samples_per_frame = bytestream_get_be16(&edata_ptr);
1086  q->subpacket[s].subbands = bytestream_get_be16(&edata_ptr);
1087  extradata_size -= 8;
1088  }
1089  if (extradata_size >= 8) {
1090  bytestream_get_be32(&edata_ptr); // Unknown unused
1091  q->subpacket[s].js_subband_start = bytestream_get_be16(&edata_ptr);
1092  if (q->subpacket[s].js_subband_start >= 51) {
1093  av_log(avctx, AV_LOG_ERROR, "js_subband_start %d is too large\n", q->subpacket[s].js_subband_start);
1094  return AVERROR_INVALIDDATA;
1095  }
1096 
1097  q->subpacket[s].js_vlc_bits = bytestream_get_be16(&edata_ptr);
1098  extradata_size -= 8;
1099  }
1100 
1101  /* Initialize extradata related variables. */
1102  q->subpacket[s].samples_per_channel = samples_per_frame / avctx->channels;
1103  q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1104 
1105  /* Initialize default data states. */
1108  q->subpacket[s].num_channels = 1;
1109 
1110  /* Initialize version-dependent variables */
1111 
1112  av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1113  q->subpacket[s].cookversion);
1114  q->subpacket[s].joint_stereo = 0;
1115  switch (q->subpacket[s].cookversion) {
1116  case MONO:
1117  if (avctx->channels != 1) {
1118  avpriv_request_sample(avctx, "Container channels != 1");
1119  return AVERROR_PATCHWELCOME;
1120  }
1121  av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1122  break;
1123  case STEREO:
1124  if (avctx->channels != 1) {
1125  q->subpacket[s].bits_per_subpdiv = 1;
1126  q->subpacket[s].num_channels = 2;
1127  }
1128  av_log(avctx, AV_LOG_DEBUG, "STEREO\n");
1129  break;
1130  case JOINT_STEREO:
1131  if (avctx->channels != 2) {
1132  avpriv_request_sample(avctx, "Container channels != 2");
1133  return AVERROR_PATCHWELCOME;
1134  }
1135  av_log(avctx, AV_LOG_DEBUG, "JOINT_STEREO\n");
1136  if (avctx->extradata_size >= 16) {
1139  q->subpacket[s].joint_stereo = 1;
1140  q->subpacket[s].num_channels = 2;
1141  }
1142  if (q->subpacket[s].samples_per_channel > 256) {
1144  }
1145  if (q->subpacket[s].samples_per_channel > 512) {
1147  }
1148  break;
1149  case MC_COOK:
1150  av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1151  if (extradata_size >= 4)
1152  channel_mask |= q->subpacket[s].channel_mask = bytestream_get_be32(&edata_ptr);
1153 
1157  q->subpacket[s].joint_stereo = 1;
1158  q->subpacket[s].num_channels = 2;
1159  q->subpacket[s].samples_per_channel = samples_per_frame >> 1;
1160 
1161  if (q->subpacket[s].samples_per_channel > 256) {
1163  }
1164  if (q->subpacket[s].samples_per_channel > 512) {
1166  }
1167  } else
1168  q->subpacket[s].samples_per_channel = samples_per_frame;
1169 
1170  break;
1171  default:
1172  avpriv_request_sample(avctx, "Cook version %d",
1173  q->subpacket[s].cookversion);
1174  return AVERROR_PATCHWELCOME;
1175  }
1176 
1177  if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1178  av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1179  return AVERROR_INVALIDDATA;
1180  } else
1182 
1183 
1184  /* Initialize variable relations */
1186 
1187  /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1188  if (q->subpacket[s].total_subbands > 53) {
1189  avpriv_request_sample(avctx, "total_subbands > 53");
1190  return AVERROR_PATCHWELCOME;
1191  }
1192 
1193  if ((q->subpacket[s].js_vlc_bits > 6) ||
1194  (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1195  av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1196  q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1197  return AVERROR_INVALIDDATA;
1198  }
1199 
1200  if (q->subpacket[s].subbands > 50) {
1201  avpriv_request_sample(avctx, "subbands > 50");
1202  return AVERROR_PATCHWELCOME;
1203  }
1204  if (q->subpacket[s].subbands == 0) {
1205  avpriv_request_sample(avctx, "subbands = 0");
1206  return AVERROR_PATCHWELCOME;
1207  }
1208  q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1210  q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1212 
1213  if (q->num_subpackets + q->subpacket[s].num_channels > q->avctx->channels) {
1214  av_log(avctx, AV_LOG_ERROR, "Too many subpackets %d for channels %d\n", q->num_subpackets, q->avctx->channels);
1215  return AVERROR_INVALIDDATA;
1216  }
1217 
1218  q->num_subpackets++;
1219  s++;
1220  if (s > MAX_SUBPACKETS) {
1221  avpriv_request_sample(avctx, "subpackets > %d", MAX_SUBPACKETS);
1222  return AVERROR_PATCHWELCOME;
1223  }
1224  }
1225  /* Generate tables */
1226  init_pow2table();
1227  init_gain_table(q);
1229 
1230  if ((ret = init_cook_vlc_tables(q)))
1231  return ret;
1232 
1233 
1234  if (avctx->block_align >= UINT_MAX / 2)
1235  return AVERROR(EINVAL);
1236 
1237  /* Pad the databuffer with:
1238  DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1239  FF_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1241  av_mallocz(avctx->block_align
1242  + DECODE_BYTES_PAD1(avctx->block_align)
1244  if (q->decoded_bytes_buffer == NULL)
1245  return AVERROR(ENOMEM);
1246 
1247  /* Initialize transform. */
1248  if ((ret = init_cook_mlt(q)))
1249  return ret;
1250 
1251  /* Initialize COOK signal arithmetic handling */
1252  if (1) {
1254  q->decouple = decouple_float;
1258  }
1259 
1260  /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1261  if (q->samples_per_channel != 256 && q->samples_per_channel != 512 &&
1262  q->samples_per_channel != 1024) {
1263  avpriv_request_sample(avctx, "samples_per_channel = %d",
1264  q->samples_per_channel);
1265  return AVERROR_PATCHWELCOME;
1266  }
1267 
1268  avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1269  if (channel_mask)
1270  avctx->channel_layout = channel_mask;
1271  else
1272  avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1273 
1274 #ifdef DEBUG
1275  dump_cook_context(q);
1276 #endif
1277  return 0;
1278 }
1279 
1281  .name = "cook",
1282  .type = AVMEDIA_TYPE_AUDIO,
1283  .id = AV_CODEC_ID_COOK,
1284  .priv_data_size = sizeof(COOKContext),
1288  .capabilities = CODEC_CAP_DR1,
1289  .long_name = NULL_IF_CONFIG_SMALL("Cook / Cooker / Gecko (RealAudio G2)"),
1290  .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1292 };