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
47
56
58
59 /* the different Cook versions */
60 #define MONO 0x1000001
61 #define STEREO 0x1000002
62 #define JOINT_STEREO 0x1000003
63 #define MC_COOK 0x2000000 // multichannel Cook, not supported
64
65 #define SUBBAND_SIZE 20
66 #define MAX_SUBPACKETS 5
67
72
90
93
101
102 typedef struct cook {
103 /*
104 * The following 5 functions provide the lowlevel arithmetic on
105 * the internal audio buffers.
106 */
107 void (*scalar_dequant)(
struct cook *q,
int index,
int quant_index,
108 int *subband_coef_index, int *subband_coef_sign,
109 float *mlt_p);
110
111 void (*decouple)(
struct cook *q,
113 int subband,
114 float f1, float f2,
115 float *decode_buffer,
116 float *mlt_buffer1, float *mlt_buffer2);
117
118 void (*imlt_window)(
struct cook *q,
float *buffer1,
119 cook_gains *gains_ptr,
float *previous_buffer);
120
122 int gain_index, int gain_index_next);
123
124 void (*saturate_output)(
struct cook *q,
float *
out);
125
129 /* stream data */
132 /* states */
135
136 /* transform data */
139
140 /* VLC data */
141 VLC envelope_quant_index[13];
142 VLC sqvh[7];
// scalar quantization
143
144 /* generatable tables and related variables */
146 float gain_table[23];
147
148 /* data buffers */
149
152 float decode_buffer_1[1024];
153 float decode_buffer_2[1024];
154 float decode_buffer_0[1060];
/* static allocation for joint decode */
155
160
163
164 /*************** init functions ***************/
165
166 /* table generator */
168 {
169 int i;
170 for (i = -63; i < 64; i++) {
173 }
174 }
175
176 /* table generator */
178 {
179 int i;
181 for (i = 0; i < 23; i++)
184 }
185
186
188 {
189 int i, result;
190
191 result = 0;
192 for (i = 0; i < 13; i++) {
196 }
198 for (i = 0; i < 7; i++) {
202 }
203
211 }
212 }
213
215 return result;
216 }
217
219 {
222
225
226 /* Initialize the MLT window: simple sine window. */
228 for (j = 0; j < mlt_size; j++)
230
231 /* Initialize the MDCT. */
235 }
238
239 return 0;
240 }
241
243 {
244 int i;
245 for (i = 0; i < 5; i++)
247 }
248
249 /*************** init functions end ***********/
250
251 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes) + 3) % 4)
252 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
253
254 /**
255 * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
256 * Why? No idea, some checksum/error detection method maybe.
257 *
258 * Out buffer size: extra bytes are needed to cope with
259 * padding/misalignment.
260 * Subpackets passed to the decoder can contain two, consecutive
261 * half-subpackets, of identical but arbitrary size.
262 * 1234 1234 1234 1234 extraA extraB
263 * Case 1: AAAA BBBB 0 0
264 * Case 2: AAAA ABBB BB-- 3 3
265 * Case 3: AAAA AABB BBBB 2 2
266 * Case 4: AAAA AAAB BBBB BB-- 1 5
267 *
268 * Nice way to waste CPU cycles.
269 *
270 * @param inbuffer pointer to byte array of indata
271 * @param out pointer to byte array of outdata
272 * @param bytes number of bytes
273 */
275 {
276 static const uint32_t
tab[4] = {
279 };
280 int i, off;
283 uint32_t *obuf = (uint32_t *) out;
284 /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
285 * I'm too lazy though, should be something like
286 * for (i = 0; i < bitamount / 64; i++)
287 * (int64_t) out[i] = 0x37c511f237c511f2 ^ av_be2ne64(int64_t) in[i]);
288 * Buffer alignment needs to be checked. */
289
290 off = (intptr_t) inbuffer & 3;
291 buf = (const uint32_t *) (inbuffer - off);
292 c = tab[off];
293 bytes += 3 + off;
294 for (i = 0; i < bytes / 4; i++)
295 obuf[i] = c ^ buf[i];
296
297 return off;
298 }
299
301 {
302 int i;
305
306 /* Free allocated memory buffers. */
309
310 /* Free the transform. */
312
313 /* Free the VLC tables. */
314 for (i = 0; i < 13; i++)
316 for (i = 0; i < 7; i++)
320
322
323 return 0;
324 }
325
326 /**
327 * Fill the gain array for the timedomain quantization.
328 *
329 * @param gb pointer to the GetBitContext
330 * @param gaininfo array[9] of gain indexes
331 */
333 {
335
337
338 i = 0;
339 while (n--) {
342
343 while (i <= index)
344 gaininfo[i++] = gain;
345 }
346 while (i <= 8)
347 gaininfo[i++] = 0;
348 }
349
350 /**
351 * Create the quant index table needed for the envelope.
352 *
353 * @param q pointer to the COOKContext
354 * @param quant_index_table pointer to the array
355 */
357 int *quant_index_table)
358 {
359 int i, j, vlc_index;
360
361 quant_index_table[0] =
get_bits(&q->
gb, 6) - 6;
// This is used later in categorize
362
364 vlc_index = i;
367 } else {
368 vlc_index /= 2;
369 if (vlc_index < 1)
370 vlc_index = 1;
371 }
372 if (vlc_index > 13)
373 vlc_index = 13; // the VLC tables >13 are identical to No. 13
374
377 quant_index_table[i] = quant_index_table[i - 1] + j - 12; // differential encoding
378 if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
380 "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
381 quant_index_table[i], i);
383 }
384 }
385
386 return 0;
387 }
388
389 /**
390 * Calculate the category and category_index vector.
391 *
392 * @param q pointer to the COOKContext
393 * @param quant_index_table pointer to the array
394 * @param category pointer to the category array
395 * @param category_index pointer to the category_index array
396 */
399 {
400 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits,
index,
v, i, j;
401 int exp_index2[102] = { 0 };
402 int exp_index1[102] = { 0 };
403
404 int tmp_categorize_array[128 * 2] = { 0 };
407
409
413
414 bias = -32;
415
416 /* Estimate bias. */
417 for (i = 32; i > 0; i = i / 2) {
418 num_bits = 0;
419 index = 0;
421 exp_idx = av_clip_uintp2((i - quant_index_table[index] + bias) / 2, 3);
422 index++;
424 }
425 if (num_bits >= bits_left - 32)
426 bias += i;
427 }
428
429 /* Calculate total number of bits. */
430 num_bits = 0;
432 exp_idx = av_clip_uintp2((bias - quant_index_table[i]) / 2, 3);
434 exp_index1[i] = exp_idx;
435 exp_index2[i] = exp_idx;
436 }
437 tmpbias1 = tmpbias2 = num_bits;
438
440 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
441 int max = -999999;
442 index = -1;
444 if (exp_index1[i] < 7) {
445 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
446 if (v >= max) {
448 index = i;
449 }
450 }
451 }
452 if (index == -1)
453 break;
454 tmp_categorize_array[tmp_categorize_array1_idx++] =
index;
458 } else { /* <--- */
460 index = -1;
462 if (exp_index2[i] > 0) {
463 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
464 if (v < min) {
466 index = i;
467 }
468 }
469 }
470 if (index == -1)
471 break;
472 tmp_categorize_array[--tmp_categorize_array2_idx] =
index;
476 }
477 }
478
480 category[i] = exp_index2[i];
481
483 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
484 }
485
486
487 /**
488 * Expand the category vector.
489 *
490 * @param q pointer to the COOKContext
491 * @param category pointer to the category array
492 * @param category_index pointer to the category_index array
493 */
495 int *category_index)
496 {
497 int i;
499 {
500 int idx = category_index[i];
502 --category[idx];
503 }
504 }
505
506 /**
507 * The real requantization of the mltcoefs
508 *
509 * @param q pointer to the COOKContext
510 * @param index index
511 * @param quant_index quantisation index
512 * @param subband_coef_index array of indexes to quant_centroid_tab
513 * @param subband_coef_sign signs of coefficients
514 * @param mlt_p pointer into the mlt buffer
515 */
517 int *subband_coef_index, int *subband_coef_sign,
518 float *mlt_p)
519 {
520 int i;
521 float f1;
522
524 if (subband_coef_index[i]) {
526 if (subband_coef_sign[i])
527 f1 = -f1;
528 } else {
529 /* noise coding if subband_coef_index[i] == 0 */
532 f1 = -f1;
533 }
535 }
536 }
537 /**
538 * Unpack the subband_coef_index and subband_coef_sign vectors.
539 *
540 * @param q pointer to the COOKContext
541 * @param category pointer to the category array
542 * @param subband_coef_index array of indexes to quant_centroid_tab
543 * @param subband_coef_sign signs of coefficients
544 */
546 int *subband_coef_index, int *subband_coef_sign)
547 {
548 int i, j;
549 int vlc, vd, tmp, result;
550
552 result = 0;
556 vlc = 0;
557 result = 1;
558 }
559 for (j = vd - 1; j >= 0; j--) {
562 vlc = tmp;
563 }
564 for (j = 0; j < vd; j++) {
565 if (subband_coef_index[i * vd + j]) {
568 } else {
569 result = 1;
570 subband_coef_sign[i * vd + j] = 0;
571 }
572 } else {
573 subband_coef_sign[i * vd + j] = 0;
574 }
575 }
576 }
577 return result;
578 }
579
580
581 /**
582 * Fill the mlt_buffer with mlt coefficients.
583 *
584 * @param q pointer to the COOKContext
585 * @param category pointer to the category array
586 * @param quant_index_table pointer to the array
587 * @param mlt_buffer pointer to mlt coefficients
588 */
590 int *quant_index_table, float *mlt_buffer)
591 {
592 /* A zero in this table means that the subband coefficient is
593 random noise coded. */
595 /* A zero in this table means that the subband coefficient is a
596 positive multiplicator. */
600
602 index = category[
band];
603 if (category[band] < 7) {
604 if (
unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
605 index = 7;
607 category[band + j] = 7;
608 }
609 }
610 if (index >= 7) {
611 memset(subband_coef_index, 0, sizeof(subband_coef_index));
612 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
613 }
615 subband_coef_index, subband_coef_sign,
617 }
618
619 /* FIXME: should this be removed, or moved into loop above? */
621 return;
622 }
623
624
626 {
627 int category_index[128] = { 0 };
629 int quant_index_table[102];
630 int res, i;
631
633 return res;
635 categorize(q, p, quant_index_table, category, category_index);
638 if (category[i] > 7)
640 }
642
643 return 0;
644 }
645
646
647 /**
648 * the actual requantization of the timedomain samples
649 *
650 * @param q pointer to the COOKContext
651 * @param buffer pointer to the timedomain buffer
652 * @param gain_index index for the block multiplier
653 * @param gain_index_next index for the next block multiplier
654 */
656 int gain_index, int gain_index_next)
657 {
658 int i;
659 float fc1, fc2;
660 fc1 =
pow2tab[gain_index + 63];
661
662 if (gain_index == gain_index_next) { // static gain
664 buffer[i] *= fc1;
665 } else { // smooth gain
666 fc2 = q->
gain_table[11 + (gain_index_next - gain_index)];
668 buffer[i] *= fc1;
669 fc1 *= fc2;
670 }
671 }
672 }
673
674 /**
675 * Apply transform window, overlap buffers.
676 *
677 * @param q pointer to the COOKContext
678 * @param inbuffer pointer to the mltcoefficients
679 * @param gains_ptr current and previous gains
680 * @param previous_buffer pointer to the previous buffer to be used for overlapping
681 */
683 cook_gains *gains_ptr,
float *previous_buffer)
684 {
686 int i;
687 /* The weird thing here, is that the two halves of the time domain
688 * buffer are swapped. Also, the newest data, that we save away for
689 * next frame, has the wrong sign. Hence the subtraction below.
690 * Almost sounds like a complex conjugate/reverse data/FFT effect.
691 */
692
693 /* Apply window and overlap */
695 inbuffer[i] = inbuffer[i] * fc * q->
mlt_window[i] -
697 }
698
699 /**
700 * The modulated lapped transform, this takes transform coefficients
701 * and transforms them into timedomain samples.
702 * Apply transform window, overlap buffers, apply gain profile
703 * and buffer management.
704 *
705 * @param q pointer to the COOKContext
706 * @param inbuffer pointer to the mltcoefficients
707 * @param gains_ptr current and previous gains
708 * @param previous_buffer pointer to the previous buffer to be used for overlapping
709 */
711 cook_gains *gains_ptr,
float *previous_buffer)
712 {
715 int i;
716
717 /* Inverse modified discrete cosine transform */
719
720 q->
imlt_window(q, buffer1, gains_ptr, previous_buffer);
721
722 /* Apply gain profile */
723 for (i = 0; i < 8; i++)
724 if (gains_ptr->
now[i] || gains_ptr->
now[i + 1])
726 gains_ptr->
now[i], gains_ptr->
now[i + 1]);
727
728 /* Save away the current to be previous block. */
729 memcpy(previous_buffer, buffer0,
731 }
732
733
734 /**
735 * function for getting the jointstereo coupling information
736 *
737 * @param q pointer to the COOKContext
738 * @param decouple_tab decoupling array
739 */
741 {
742 int i;
746 int length = end - start + 1;
747
748 if (start > end)
749 return 0;
750
751 if (vlc)
752 for (i = 0; i <
length; i++)
756 else
757 for (i = 0; i <
length; i++) {
762 }
763 decouple_tab[start + i] =
v;
764 }
765 return 0;
766 }
767
768 /**
769 * function decouples a pair of signals from a single signal via multiplication.
770 *
771 * @param q pointer to the COOKContext
772 * @param subband index of the current subband
773 * @param f1 multiplier for channel 1 extraction
774 * @param f2 multiplier for channel 2 extraction
775 * @param decode_buffer input buffer
776 * @param mlt_buffer1 pointer to left channel mlt coefficients
777 * @param mlt_buffer2 pointer to right channel mlt coefficients
778 */
781 int subband,
782 float f1, float f2,
783 float *decode_buffer,
784 float *mlt_buffer1, float *mlt_buffer2)
785 {
786 int j, tmp_idx;
789 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
790 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
791 }
792 }
793
794 /**
795 * function for decoding joint stereo data
796 *
797 * @param q pointer to the COOKContext
798 * @param mlt_buffer1 pointer to left channel mlt coefficients
799 * @param mlt_buffer2 pointer to right channel mlt coefficients
800 */
802 float *mlt_buffer_left, float *mlt_buffer_right)
803 {
804 int i, j, res;
807 int idx, cpl_tmp;
808 float f1, f2;
809 const float *cplscale;
810
812
813 /* Make sure the buffers are zeroed out. */
814 memset(mlt_buffer_left, 0, 1024 * sizeof(*mlt_buffer_left));
815 memset(mlt_buffer_right, 0, 1024 * sizeof(*mlt_buffer_right));
817 return res;
819 return res;
820 /* The two channels are stored interleaved in decode_buffer. */
823 mlt_buffer_left[i * 20 + j] = decode_buffer[i * 40 + j];
824 mlt_buffer_right[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
825 }
826 }
827
828 /* When we reach js_subband_start (the higher frequencies)
829 the coefficients are stored in a coupling scheme. */
833 idx -= decouple_tab[cpl_tmp];
835 f1 = cplscale[decouple_tab[cpl_tmp] + 1];
836 f2 = cplscale[idx];
837 q->
decouple(q, p, i, f1, f2, decode_buffer,
838 mlt_buffer_left, mlt_buffer_right);
840 }
841
842 return 0;
843 }
844
845 /**
846 * First part of subpacket decoding:
847 * decode raw stream bytes and read gain info.
848 *
849 * @param q pointer to the COOKContext
850 * @param inbuffer pointer to raw stream data
851 * @param gains_ptr array of current/prev gain pointers
852 */
856 {
858
864
865 /* Swap current and previous gains */
867 }
868
869 /**
870 * Saturate the output signal and interleave.
871 *
872 * @param q pointer to the COOKContext
873 * @param out pointer to the output vector
874 */
876 {
879 }
880
881
882 /**
883 * Final part of subpacket decoding:
884 * Apply modulated lapped transform, gain compensation,
885 * clip and convert to integer.
886 *
887 * @param q pointer to the COOKContext
888 * @param decode_buffer pointer to the mlt coefficients
889 * @param gains_ptr array of current/prev gain pointers
890 * @param previous_buffer pointer to the previous buffer to be used for overlapping
891 * @param out pointer to the output buffer
892 */
894 cook_gains *gains_ptr,
float *previous_buffer,
896 {
897 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
898 if (out)
900 }
901
902
903 /**
904 * Cook subpacket decoding. This function returns one decoded subpacket,
905 * usually 1024 samples per channel.
906 *
907 * @param q pointer to the COOKContext
908 * @param inbuffer pointer to the inbuffer
909 * @param outbuffer pointer to the outbuffer
910 */
912 const uint8_t *inbuffer,
float **outbuffer)
913 {
914 int sub_packet_size = p->
size;
915 int res;
916
919
922 return res;
923 } else {
925 return res;
926
930 return res;
931 }
932 }
933
937
942 outbuffer ? outbuffer[p->
ch_idx + 1] : NULL);
943 else
946 outbuffer ? outbuffer[p->
ch_idx + 1] : NULL);
947 }
948
949 return 0;
950 }
951
952
954 int *got_frame_ptr,
AVPacket *avpkt)
955 {
958 int buf_size = avpkt->
size;
960 float **samples =
NULL;
963 int chidx = 0;
964
965 if (buf_size < avctx->block_align)
966 return buf_size;
967
968 /* get output buffer */
974 }
975
976 /* estimate subpacket sizes */
978
984 "frame subpacket size total > avctx->block_align!\n");
986 }
987 }
988
989 /* decode supbackets */
995 "subpacket[%i] size %i js %i %i block_align %i\n",
998
1000 return ret;
1005 }
1006
1007 /* Discard the first two frames: no valid audio. */
1010 *got_frame_ptr = 0;
1012 }
1013
1014 *got_frame_ptr = 1;
1015
1017 }
1018
1019 #ifdef DEBUG
1021 {
1022 //int i=0;
1023 #define PRINT(a, b) av_dlog(q->avctx, " %s = %d\n", a, b);
1029 }
1040 }
1041 #endif
1042
1043 /**
1044 * Cook initialization
1045 *
1046 * @param avctx pointer to the AVCodecContext
1047 */
1049 {
1055 unsigned int channel_mask = 0;
1056 int samples_per_frame = 0;
1059
1060 /* Take care of the codec specific extradata. */
1061 if (extradata_size < 8) {
1064 }
1066
1067 /* Take data from the AVCodecContext (RM container). */
1071 }
1072
1073 /* Initialize RNG. */
1075
1077
1078 while (edata_ptr < edata_ptr_end) {
1079 /* 8 for mono, 16 for stereo, ? for multichannel
1080 Swap to right endianness so we don't need to care later on. */
1081 if (extradata_size >= 8) {
1083 samples_per_frame = bytestream_get_be16(&edata_ptr);
1085 extradata_size -= 8;
1086 }
1087 if (extradata_size >= 8) {
1088 bytestream_get_be32(&edata_ptr); // Unknown unused
1093 }
1094
1096 extradata_size -= 8;
1097 }
1098
1099 /* Initialize extradata related variables. */
1102
1103 /* Initialize default data states. */
1107
1108 /* Initialize version-dependent variables */
1109
1118 }
1120 break;
1125 }
1127 break;
1132 }
1139 }
1142 }
1145 }
1146 break;
1149 if (extradata_size >= 4)
1151
1158
1161 }
1164 }
1165 } else
1167
1168 break;
1169 default:
1173 }
1174
1178 } else
1180
1181
1182 /* Initialize variable relations */
1184
1185 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1189 }
1190
1196 }
1197
1201 }
1205 }
1210
1214 }
1215
1217 s++;
1221 }
1222 }
1223 /* Generate tables */
1227
1230
1231
1234
1235 /* Pad the databuffer with:
1236 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1237 FF_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1244
1245 /* Initialize transform. */
1248
1249 /* Initialize COOK signal arithmetic handling */
1250 if (1) {
1256 }
1257
1258 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1264 }
1265
1267 if (channel_mask)
1269 else
1271
1272 #ifdef DEBUG
1273 dump_cook_context(q);
1274 #endif
1275 return 0;
1276 }
1277
1290 };