1 /*
2 * IMC compatible decoder
3 * Copyright (c) 2002-2004 Maxim Poliakovski
4 * Copyright (c) 2006 Benjamin Larsson
5 * Copyright (c) 2006 Konstantin Shishkov
6 *
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24 /**
25 * @file
26 * IMC - Intel Music Coder
27 * A mdct based codec using a 256 points large transform
28 * divided into 32 bands with some mix of scale factors.
29 * Only mono is supported.
30 *
31 */
32
33
34 #include <math.h>
35 #include <stddef.h>
36 #include <stdio.h>
37
48
50
51 #define IMC_BLOCK_SIZE 64
52 #define IMC_FRAME_ID 0x21
55
65
77
79
82
85
86 /** MDCT tables */
87 //@{
93 //@}
94
97
103
105
107 float weights1[31], weights2[31];
109
111
112 #define VLC_TABLES_SIZE 9512
113
115 0, 640, 1156, 1732, 2308, 2852, 3396, 3924,
117 };
118
120
122 {
123 return 3.5 * atan((freq / 7500.0) * (freq / 7500.0)) + 13.0 * atan(freq * 0.00076);
124 }
125
127 {
128 double freqmin[32], freqmid[32], freqmax[32];
129 double scale = sampling_rate / (256.0 * 2.0 * 2.0);
130 double nyquist_freq = sampling_rate * 0.5;
131 double freq, bark, prev_bark = 0,
tf,
tb;
132 int i, j;
133
134 for (i = 0; i < 32; i++) {
137
138 if (i > 0) {
139 tb = bark - prev_bark;
140 q->
weights1[i - 1] = pow(10.0, -1.0 * tb);
141 q->
weights2[i - 1] = pow(10.0, -2.7 * tb);
142 }
143 prev_bark = bark;
144
145 freqmid[i] = freq;
146
148 while (
tf < nyquist_freq) {
151 if (tb > bark + 0.5)
152 break;
153 }
155
160 if (tb <= bark - 0.5)
161 break;
162 }
164 }
165
166 for (i = 0; i < 32; i++) {
167 freq = freqmax[i];
168 for (j = 31; j > 0 && freq <= freqmid[j]; j--);
170
171 freq = freqmin[i];
172 for (j = 0; j < 32 && freq >= freqmid[j]; j++);
174 }
175 }
176
178 {
181 double r1, r2;
182
185
189 }
190
191 for (j = 0; j < avctx->
channels; j++) {
193
194 for (i = 0; i <
BANDS; i++)
196
199 }
200
201 /* Build mdct window, a simple sine window normalized with sqrt(2) */
203 for (i = 0; i <
COEFFS; i++)
205 for (i = 0; i < COEFFS / 2; i++) {
206 q->
post_cos[i] = (1.0f / 32768) * cos(i / 256.0 *
M_PI);
207 q->
post_sin[i] = (1.0f / 32768) * sin(i / 256.0 *
M_PI);
208
209 r1 = sin((i * 4.0 + 1.0) / 1024.0 *
M_PI);
210 r2 = cos((i * 4.0 + 1.0) / 1024.0 *
M_PI);
211
212 if (i & 0x1) {
214 q->
pre_coef2[i] = -(r1 - r2) * sqrt(2.0);
215 } else {
216 q->
pre_coef1[i] = -(r1 + r2) * sqrt(2.0);
218 }
219 }
220
221 /* Generate a square root table */
222
223 for (i = 0; i < 30; i++)
225
226 /* initialize the VLC tables */
227 for (i = 0; i < 4 ; i++) {
228 for (j = 0; j < 4; j++) {
230 huffman_vlc[i][j].
table_allocated = vlc_offsets[i * 4 + j + 1] - vlc_offsets[i * 4 + j];
234 }
235 }
236
239 } else {
244 }
245
249 }
255
256 return 0;
257 }
258
260 float *flcoeffs2, int *bandWidthT,
261 float *flcoeffs3, float *flcoeffs5)
262 {
266 float snr_limit = 1.e-30;
267 float accum = 0.0;
268 int i, cnt2;
269
270 for (i = 0; i <
BANDS; i++) {
271 flcoeffs5[i] = workT2[i] = 0.0;
272 if (bandWidthT[i]) {
273 workT1[i] = flcoeffs1[i] * flcoeffs1[i];
274 flcoeffs3[i] = 2.0 * flcoeffs2[i];
275 } else {
276 workT1[i] = 0.0;
277 flcoeffs3[i] = -30000.0;
278 }
279 workT3[i] = bandWidthT[i] * workT1[i] * 0.01;
280 if (workT3[i] <= snr_limit)
281 workT3[i] = 0.0;
282 }
283
284 for (i = 0; i <
BANDS; i++) {
285 for (cnt2 = i; cnt2 < q->
cyclTab[i]; cnt2++)
286 flcoeffs5[cnt2] = flcoeffs5[cnt2] + workT3[i];
287 workT2[cnt2 - 1] = workT2[cnt2 - 1] + workT3[i];
288 }
289
290 for (i = 1; i <
BANDS; i++) {
291 accum = (workT2[i - 1] + accum) * q->
weights1[i - 1];
292 flcoeffs5[i] += accum;
293 }
294
295 for (i = 0; i <
BANDS; i++)
296 workT2[i] = 0.0;
297
298 for (i = 0; i <
BANDS; i++) {
299 for (cnt2 = i - 1; cnt2 > q->
cyclTab2[i]; cnt2--)
300 flcoeffs5[cnt2] += workT3[i];
301 workT2[cnt2+1] += workT3[i];
302 }
303
304 accum = 0.0;
305
306 for (i = BANDS-2; i >= 0; i--) {
307 accum = (workT2[i+1] + accum) * q->
weights2[i];
308 flcoeffs5[i] += accum;
309 // there is missing code here, but it seems to never be triggered
310 }
311 }
312
313
315 int *levlCoeffs)
316 {
317 int i;
322
323 s = stream_format_code >> 1;
324 hufftab[0] = &huffman_vlc[
s][0];
325 hufftab[1] = &huffman_vlc[
s][1];
326 hufftab[2] = &huffman_vlc[
s][2];
327 hufftab[3] = &huffman_vlc[
s][3];
329
330 if (stream_format_code & 4)
331 start = 1;
332 if (start)
334 for (i = start; i <
BANDS; i++) {
336 hufftab[cb_sel[i]]->
bits, 2);
337 if (levlCoeffs[i] == 17)
339 }
340 }
341
343 int *levlCoeffs)
344 {
345 int i;
346
349 for (i = 1; i <
BANDS; i++)
351 }
352
354 float *flcoeffs1, float *flcoeffs2)
355 {
357 float tmp, tmp2;
358 // maybe some frequency division thingy
359
360 flcoeffs1[0] = 20000.0 /
exp2 (levlCoeffBuf[0] * 0.18945);
// 0.18945 = log2(10) * 0.05703125
361 flcoeffs2[0] =
log2f(flcoeffs1[0]);
362 tmp = flcoeffs1[0];
363 tmp2 = flcoeffs2[0];
364
365 for (i = 1; i <
BANDS; i++) {
366 level = levlCoeffBuf[i];
367 if (level == 16) {
368 flcoeffs1[i] = 1.0;
369 flcoeffs2[i] = 0.0;
370 } else {
371 if (level < 17)
372 level -= 7;
373 else if (level <= 24)
374 level -= 32;
375 else
376 level -= 16;
377
379 tmp2 += 0.83048 *
level;
// 0.83048 = log2(10) * 0.25
380 flcoeffs1[i] = tmp;
381 flcoeffs2[i] = tmp2;
382 }
383 }
384 }
385
386
388 float *old_floor, float *flcoeffs1,
389 float *flcoeffs2)
390 {
391 int i;
392 /* FIXME maybe flag_buf = noise coding and flcoeffs1 = new scale factors
393 * and flcoeffs2 old scale factors
394 * might be incomplete due to a missing table that is in the binary code
395 */
396 for (i = 0; i <
BANDS; i++) {
397 flcoeffs1[i] = 0;
398 if (levlCoeffBuf[i] < 16) {
399 flcoeffs1[i] =
imc_exp_tab2[levlCoeffBuf[i]] * old_floor[i];
400 flcoeffs2[i] = (levlCoeffBuf[i] - 7) * 0.83048 + flcoeffs2[i]; // 0.83048 = log2(10) * 0.25
401 } else {
402 flcoeffs1[i] = old_floor[i];
403 }
404 }
405 }
406
408 float *flcoeffs1, float *flcoeffs2)
409 {
411 float tmp, tmp2;
412
414 flcoeffs1[pos] = 20000.0 / pow (2, levlCoeffBuf[0] * 0.18945); // 0.18945 = log2(10) * 0.05703125
415 flcoeffs2[pos] =
log2f(flcoeffs1[0]);
416 tmp = flcoeffs1[pos];
417 tmp2 = flcoeffs2[pos];
418
419 levlCoeffBuf++;
420 for (i = 0; i <
BANDS; i++) {
421 if (i == pos)
422 continue;
423 level = *levlCoeffBuf++;
424 flcoeffs1[i] = tmp *
powf(10.0, -level * 0.4375);
//todo tab
425 flcoeffs2[i] = tmp2 - 1.4533435415 *
level;
// 1.4533435415 = log2(10) * 0.4375
426 }
427 }
428
429 /**
430 * Perform bit allocation depending on bits available
431 */
433 int stream_format_code, int freebits, int flag)
434 {
435 int i, j;
436 const float limit = -1.e20;
437 float highest = 0.0;
438 int indx;
441 float summa = 0.0;
442 int iacc = 0;
443 int summer = 0;
444 int rres, cwlen;
445 float lowest = 1.e10;
446 int low_indx = 0;
447 float workT[32];
448 int flg;
449 int found_indx = 0;
450
451 for (i = 0; i <
BANDS; i++)
453
454 for (i = 0; i < BANDS - 1; i++) {
458 }
460 }
462
463 highest = highest * 0.25;
464
465 for (i = 0; i <
BANDS; i++) {
466 indx = -1;
468 indx = 0;
469
471 indx = 1;
472
474 indx = 2;
475
476 if (indx == -1)
478
480 }
481
482 if (stream_format_code & 0x2) {
487 }
488
489 for (i = (stream_format_code & 0x2) ? 4 : 0; i < BANDS - 1; i++) {
492 }
493
494 if (!iacc)
496
498 summa = (summa * 0.5 - freebits) / iacc;
499
500
501 for (i = 0; i < BANDS / 2; i++) {
502 rres = summer - freebits;
503 if ((rres >= -8) && (rres <= 8))
504 break;
505
506 summer = 0;
507 iacc = 0;
508
509 for (j = (stream_format_code & 0x2) ? 4 : 0; j <
BANDS; j++) {
510 cwlen = av_clipf(((chctx->
flcoeffs4[j] * 0.5) - summa + 0.5), 0, 6);
511
514
515 if (cwlen > 0)
517 }
518
520 t2 = 1;
521 if (freebits < summer)
522 t2 = -1;
523 if (i == 0)
525 if (flg != t2)
526 t1++;
527
528 summa = (float)(summer - freebits) / ((t1 + 1) * iacc) + summa;
529 }
530
531 for (i = (stream_format_code & 0x2) ? 4 : 0; i <
BANDS; i++) {
534 }
535
536 if (freebits > summer) {
537 for (i = 0; i <
BANDS; i++) {
538 workT[i] = (chctx->
bitsBandT[i] == 6) ? -1.e20
540 }
541
542 highest = 0.0;
543
544 do {
545 if (highest <= -1.e20)
546 break;
547
548 found_indx = 0;
549 highest = -1.e20;
550
551 for (i = 0; i <
BANDS; i++) {
552 if (workT[i] > highest) {
553 highest = workT[i];
554 found_indx = i;
555 }
556 }
557
558 if (highest > -1.e20) {
559 workT[found_indx] -= 2.0;
561 workT[found_indx] = -1.e20;
562
563 for (j =
band_tab[found_indx]; j < band_tab[found_indx + 1] && (freebits > summer); j++) {
565 summer++;
566 }
567 }
568 } while (freebits > summer);
569 }
570 if (freebits < summer) {
571 for (i = 0; i <
BANDS; i++) {
573 : 1.e20;
574 }
575 if (stream_format_code & 0x2) {
576 workT[0] = 1.e20;
577 workT[1] = 1.e20;
578 workT[2] = 1.e20;
579 workT[3] = 1.e20;
580 }
581 while (freebits < summer) {
582 lowest = 1.e10;
583 low_indx = 0;
584 for (i = 0; i <
BANDS; i++) {
585 if (workT[i] < lowest) {
586 lowest = workT[i];
587 low_indx = i;
588 }
589 }
590 // if (lowest >= 1.e10)
591 // break;
592 workT[low_indx] = lowest + 2.0;
593
595 workT[low_indx] = 1.e20;
596
597 for (j =
band_tab[low_indx]; j <
band_tab[low_indx+1] && (freebits < summer); j++) {
600 summer--;
601 }
602 }
603 }
604 }
605 return 0;
606 }
607
609 {
610 int i, j;
611
614 for (i = 0; i <
BANDS; i++) {
616 continue;
617
620
621 for (j = band_tab[i]; j < band_tab[i + 1]; j++) {
625 }
626 } else {
633 } else {
639 } else {
645 } else { // 101
647 }
648 }
649 }
650 }
651
652 if (j < band_tab[i + 1]) {
656 }
657 }
658 }
659 }
660
661 /**
662 * Increase highest' band coefficient sizes as some bits won't be used
663 */
665 int summer)
666 {
667 float workT[32];
668 int corrected = 0;
669 int i, j;
670 float highest = 0;
671 int found_indx = 0;
672
673 for (i = 0; i <
BANDS; i++) {
674 workT[i] = (chctx->
bitsBandT[i] == 6) ? -1.e20
676 }
677
678 while (corrected < summer) {
679 if (highest <= -1.e20)
680 break;
681
682 highest = -1.e20;
683
684 for (i = 0; i <
BANDS; i++) {
685 if (workT[i] > highest) {
686 highest = workT[i];
687 found_indx = i;
688 }
689 }
690
691 if (highest > -1.e20) {
692 workT[found_indx] -= 2.0;
693 if (++(chctx->
bitsBandT[found_indx]) == 6)
694 workT[found_indx] = -1.e20;
695
696 for (j =
band_tab[found_indx]; j <
band_tab[found_indx+1] && (corrected < summer); j++) {
699 corrected++;
700 }
701 }
702 }
703 }
704 }
705
707 {
708 int i;
712
713 /* prerotation */
714 for (i = 0; i <
COEFFS / 2; i++) {
719 }
720
721 /* FFT */
724
725 /* postrotation, window and reorder */
726 for (i = 0; i <
COEFFS / 2; i++) {
733 dst1 += 2;
734 dst2 -= 2;
736 }
737 }
738
740 int stream_format_code)
741 {
742 int i, j;
743 int middle_value, cw_len, max_size;
744 const float *quantizer;
745
746 for (i = 0; i <
BANDS; i++) {
750
751 if (cw_len <= 0 || chctx->skipFlags[j])
752 continue;
753
754 max_size = 1 << cw_len;
755 middle_value = max_size >> 1;
756
759
760 if (cw_len >= 4) {
764 else
766 }else{
770 else
772 }
773 }
774 }
775 return 0;
776 }
777
778
780 {
781 int i, j, cw_len, cw;
782
783 for (i = 0; i <
BANDS; i++) {
785 continue;
789 cw = 0;
790
792 av_dlog(NULL,
"Band %i coeff %i cw_len %i\n", i, j, cw_len);
794 }
795
798
800 }
801 }
802 }
803 return 0;
804 }
805
807 {
808 int i, j;
810
811 for (i = 0; i <
BANDS; i++) {
817 if ((((band_tab[i + 1] - band_tab[i]) * 1.5) > chctx->
sumLenArr[i]) && (chctx->
sumLenArr[i] > 0))
819 }
820
822
823 for (i = 0; i <
BANDS; i++) {
825 /* band has flag set and at least one coded coefficient */
829 }
830 }
831
832 /* calculate bits left, bits needed and adjust bit allocation */
833 bits = summer = 0;
834
835 for (i = 0; i <
BANDS; i++) {
841 }
842 }
845 }
846 }
848 }
849
851 {
852 int stream_format_code;
853 int imc_hdr, i, j,
ret;
854 int flag;
856 int counter, bitscount;
858
859
860 /* Check the frame header */
862 if (imc_hdr & 0x18) {
866 }
868
869 if (stream_format_code & 0x04)
871
873 for (i = 0; i <
BANDS; i++)
878 }
879
881 if (stream_format_code & 0x1)
884 else if (stream_format_code & 0x1)
886 else
888
889 if (stream_format_code & 0x4)
892 else
895
896 for(i=0; i<
BANDS; i++) {
900 }
901 }
902
904
905 counter = 0;
906 if (stream_format_code & 0x1) {
907 for (i = 0; i <
BANDS; i++) {
912 }
913 } else {
914 for (i = 0; i <
BANDS; i++) {
917 counter++;
918 } else
920 }
921
923 for (i = 0; i < BANDS - 1; i++)
926
930 }
931
932 bitscount = 0;
933 /* first 4 bands will be assigned 5 bits per coefficient */
934 if (stream_format_code & 0x2) {
935 bitscount += 15;
936
941 for (i = 1; i < 4; i++) {
942 if (stream_format_code & 0x1)
943 bits = 5;
944 else
950 }
951 }
952 }
955 if (!(stream_format_code & 0x2))
956 bitscount += 16;
957 }
958
961 flag)) < 0) {
965 }
966
967 if (stream_format_code & 0x1) {
968 for (i = 0; i <
BANDS; i++)
970 } else {
972 }
973
974 for (i = 0; i <
BANDS; i++) {
976
980 }
981
983
988 }
989
994 }
995
997
999
1000 return 0;
1001 }
1002
1004 int *got_frame_ptr,
AVPacket *avpkt)
1005 {
1008 int buf_size = avpkt->
size;
1010
1012
1014
1015 if (buf_size < IMC_BLOCK_SIZE * avctx->channels) {
1018 }
1019
1020 /* get output buffer */
1024
1025 for (i = 0; i < avctx->
channels; i++) {
1027
1029
1031
1033
1036 }
1037
1041 }
1042
1043 *got_frame_ptr = 1;
1044
1046 }
1047
1048
1050 {
1052
1054
1055 return 0;
1056 }
1057
1059 {
1061
1064 }
1065
1066 #if CONFIG_IMC_DECODER
1080 };
1081 #endif
1082 #if CONFIG_IAC_DECODER
1096 };
1097 #endif