vlc: Add header #include when the types are used
[libav.git] / libavcodec / cook.c
1 /*
2 * COOK compatible decoder
3 * Copyright (c) 2003 Sascha Sommer
4 * Copyright (c) 2005 Benjamin Larsson
5 *
6 * This file is part of Libav.
7 *
8 * Libav 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 * Libav 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 Libav; 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
45 #include "libavutil/channel_layout.h"
46 #include "libavutil/lfg.h"
47
48 #include "audiodsp.h"
49 #include "avcodec.h"
50 #include "bitstream.h"
51 #include "bytestream.h"
52 #include "fft.h"
53 #include "internal.h"
54 #include "sinewin.h"
55 #include "vlc.h"
56
57 #include "cookdata.h"
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
68 typedef struct cook_gains {
69 int *now;
70 int *previous;
71 } cook_gains;
72
73 typedef struct COOKSubpacket {
74 int ch_idx;
75 int size;
76 int num_channels;
77 int cookversion;
78 int subbands;
79 int js_subband_start;
80 int js_vlc_bits;
81 int samples_per_channel;
82 int log2_numvector_size;
83 unsigned int channel_mask;
84 VLC channel_coupling;
85 int joint_stereo;
86 int bits_per_subpacket;
87 int bits_per_subpdiv;
88 int total_subbands;
89 int numvector_size; // 1 << log2_numvector_size;
90
91 float mono_previous_buffer1[1024];
92 float mono_previous_buffer2[1024];
93
94 cook_gains gains1;
95 cook_gains gains2;
96 int gain_1[9];
97 int gain_2[9];
98 int gain_3[9];
99 int gain_4[9];
100 } COOKSubpacket;
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,
112 COOKSubpacket *p,
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
121 void (*interpolate)(struct cook *q, float *buffer,
122 int gain_index, int gain_index_next);
123
124 void (*saturate_output)(struct cook *q, float *out);
125
126 AVCodecContext* avctx;
127 AudioDSPContext adsp;
128 BitstreamContext bc;
129 /* stream data */
130 int num_vectors;
131 int samples_per_channel;
132 /* states */
133 AVLFG random_state;
134 int discarded_packets;
135
136 /* transform data */
137 FFTContext mdct_ctx;
138 float* mlt_window;
139
140 /* VLC data */
141 VLC envelope_quant_index[13];
142 VLC sqvh[7]; // scalar quantization
143
144 /* generate tables and related variables */
145 int gain_size_factor;
146 float gain_table[23];
147
148 /* data buffers */
149
150 uint8_t* decoded_bytes_buffer;
151 DECLARE_ALIGNED(32, float, mono_mdct_output)[2048];
152 float decode_buffer_1[1024];
153 float decode_buffer_2[1024];
154 float decode_buffer_0[1060]; /* static allocation for joint decode */
155
156 const float *cplscales[5];
157 int num_subpackets;
158 COOKSubpacket subpacket[MAX_SUBPACKETS];
159 } COOKContext;
160
161 static float pow2tab[127];
162 static float rootpow2tab[127];
163
164 /*************** init functions ***************/
165
166 /* table generator */
167 static av_cold void init_pow2table(void)
168 {
169 int i;
170 for (i = -63; i < 64; i++) {
171 pow2tab[63 + i] = pow(2, i);
172 rootpow2tab[63 + i] = sqrt(pow(2, i));
173 }
174 }
175
176 /* table generator */
177 static av_cold void init_gain_table(COOKContext *q)
178 {
179 int i;
180 q->gain_size_factor = q->samples_per_channel / 8;
181 for (i = 0; i < 23; i++)
182 q->gain_table[i] = pow(pow2tab[i + 52],
183 (1.0 / (double) q->gain_size_factor));
184 }
185
186
187 static av_cold int init_cook_vlc_tables(COOKContext *q)
188 {
189 int i, result;
190
191 result = 0;
192 for (i = 0; i < 13; i++) {
193 result |= init_vlc(&q->envelope_quant_index[i], 9, 24,
194 envelope_quant_index_huffbits[i], 1, 1,
195 envelope_quant_index_huffcodes[i], 2, 2, 0);
196 }
197 av_log(q->avctx, AV_LOG_DEBUG, "sqvh VLC init\n");
198 for (i = 0; i < 7; i++) {
199 result |= init_vlc(&q->sqvh[i], vhvlcsize_tab[i], vhsize_tab[i],
200 cvh_huffbits[i], 1, 1,
201 cvh_huffcodes[i], 2, 2, 0);
202 }
203
204 for (i = 0; i < q->num_subpackets; i++) {
205 if (q->subpacket[i].joint_stereo == 1) {
206 result |= init_vlc(&q->subpacket[i].channel_coupling, 6,
207 (1 << q->subpacket[i].js_vlc_bits) - 1,
208 ccpl_huffbits[q->subpacket[i].js_vlc_bits - 2], 1, 1,
209 ccpl_huffcodes[q->subpacket[i].js_vlc_bits - 2], 2, 2, 0);
210 av_log(q->avctx, AV_LOG_DEBUG, "subpacket %i Joint-stereo VLC used.\n", i);
211 }
212 }
213
214 av_log(q->avctx, AV_LOG_DEBUG, "VLC tables initialized.\n");
215 return result;
216 }
217
218 static av_cold int init_cook_mlt(COOKContext *q)
219 {
220 int j, ret;
221 int mlt_size = q->samples_per_channel;
222
223 if ((q->mlt_window = av_malloc(mlt_size * sizeof(*q->mlt_window))) == 0)
224 return AVERROR(ENOMEM);
225
226 /* Initialize the MLT window: simple sine window. */
227 ff_sine_window_init(q->mlt_window, mlt_size);
228 for (j = 0; j < mlt_size; j++)
229 q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
230
231 /* Initialize the MDCT. */
232 if ((ret = ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size) + 1, 1, 1.0 / 32768.0))) {
233 av_free(q->mlt_window);
234 return ret;
235 }
236 av_log(q->avctx, AV_LOG_DEBUG, "MDCT initialized, order = %d.\n",
237 av_log2(mlt_size) + 1);
238
239 return 0;
240 }
241
242 static av_cold void init_cplscales_table(COOKContext *q)
243 {
244 int i;
245 for (i = 0; i < 5; i++)
246 q->cplscales[i] = cplscales[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 */
274 static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
275 {
276 static const uint32_t tab[4] = {
277 AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
278 AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
279 };
280 int i, off;
281 uint32_t c;
282 const uint32_t *buf;
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
300 static av_cold int cook_decode_close(AVCodecContext *avctx)
301 {
302 int i;
303 COOKContext *q = avctx->priv_data;
304 av_log(avctx, AV_LOG_DEBUG, "Deallocating memory.\n");
305
306 /* Free allocated memory buffers. */
307 av_free(q->mlt_window);
308 av_free(q->decoded_bytes_buffer);
309
310 /* Free the transform. */
311 ff_mdct_end(&q->mdct_ctx);
312
313 /* Free the VLC tables. */
314 for (i = 0; i < 13; i++)
315 ff_free_vlc(&q->envelope_quant_index[i]);
316 for (i = 0; i < 7; i++)
317 ff_free_vlc(&q->sqvh[i]);
318 for (i = 0; i < q->num_subpackets; i++)
319 ff_free_vlc(&q->subpacket[i].channel_coupling);
320
321 av_log(avctx, AV_LOG_DEBUG, "Memory deallocated.\n");
322
323 return 0;
324 }
325
326 /**
327 * Fill the gain array for the timedomain quantization.
328 *
329 * @param bc pointer to the BitstreamContext
330 * @param gaininfo array[9] of gain indexes
331 */
332 static void decode_gain_info(BitstreamContext *bc, int *gaininfo)
333 {
334 int i, n;
335
336 while (bitstream_read_bit(bc)) {
337 /* NOTHING */
338 }
339
340 n = bitstream_tell(bc) - 1; // amount of elements * 2 to update
341
342 i = 0;
343 while (n--) {
344 int index = bitstream_read(bc, 3);
345 int gain = bitstream_read_bit(bc) ? bitstream_read(bc, 4) - 7 : -1;
346
347 while (i <= index)
348 gaininfo[i++] = gain;
349 }
350 while (i <= 8)
351 gaininfo[i++] = 0;
352 }
353
354 /**
355 * Create the quant index table needed for the envelope.
356 *
357 * @param q pointer to the COOKContext
358 * @param quant_index_table pointer to the array
359 */
360 static int decode_envelope(COOKContext *q, COOKSubpacket *p,
361 int *quant_index_table)
362 {
363 int i, j, vlc_index;
364
365 quant_index_table[0] = bitstream_read(&q->bc, 6) - 6; // This is used later in categorize
366
367 for (i = 1; i < p->total_subbands; i++) {
368 vlc_index = i;
369 if (i >= p->js_subband_start * 2) {
370 vlc_index -= p->js_subband_start;
371 } else {
372 vlc_index /= 2;
373 if (vlc_index < 1)
374 vlc_index = 1;
375 }
376 if (vlc_index > 13)
377 vlc_index = 13; // the VLC tables >13 are identical to No. 13
378
379 j = bitstream_read_vlc(&q->bc, q->envelope_quant_index[vlc_index - 1].table,
380 q->envelope_quant_index[vlc_index - 1].bits, 2);
381 quant_index_table[i] = quant_index_table[i - 1] + j - 12; // differential encoding
382 if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
383 av_log(q->avctx, AV_LOG_ERROR,
384 "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
385 quant_index_table[i], i);
386 return AVERROR_INVALIDDATA;
387 }
388 }
389
390 return 0;
391 }
392
393 /**
394 * Calculate the category and category_index vector.
395 *
396 * @param q pointer to the COOKContext
397 * @param quant_index_table pointer to the array
398 * @param category pointer to the category array
399 * @param category_index pointer to the category_index array
400 */
401 static void categorize(COOKContext *q, COOKSubpacket *p, int *quant_index_table,
402 int *category, int *category_index)
403 {
404 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
405 int exp_index2[102] = { 0 };
406 int exp_index1[102] = { 0 };
407
408 int tmp_categorize_array[128 * 2] = { 0 };
409 int tmp_categorize_array1_idx = p->numvector_size;
410 int tmp_categorize_array2_idx = p->numvector_size;
411
412 bits_left = p->bits_per_subpacket - bitstream_tell(&q->bc);
413
414 if (bits_left > q->samples_per_channel)
415 bits_left = q->samples_per_channel +
416 ((bits_left - q->samples_per_channel) * 5) / 8;
417
418 bias = -32;
419
420 /* Estimate bias. */
421 for (i = 32; i > 0; i = i / 2) {
422 num_bits = 0;
423 index = 0;
424 for (j = p->total_subbands; j > 0; j--) {
425 exp_idx = av_clip((i - quant_index_table[index] + bias) / 2, 0, 7);
426 index++;
427 num_bits += expbits_tab[exp_idx];
428 }
429 if (num_bits >= bits_left - 32)
430 bias += i;
431 }
432
433 /* Calculate total number of bits. */
434 num_bits = 0;
435 for (i = 0; i < p->total_subbands; i++) {
436 exp_idx = av_clip((bias - quant_index_table[i]) / 2, 0, 7);
437 num_bits += expbits_tab[exp_idx];
438 exp_index1[i] = exp_idx;
439 exp_index2[i] = exp_idx;
440 }
441 tmpbias1 = tmpbias2 = num_bits;
442
443 for (j = 1; j < p->numvector_size; j++) {
444 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
445 int max = -999999;
446 index = -1;
447 for (i = 0; i < p->total_subbands; i++) {
448 if (exp_index1[i] < 7) {
449 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
450 if (v >= max) {
451 max = v;
452 index = i;
453 }
454 }
455 }
456 if (index == -1)
457 break;
458 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
459 tmpbias1 -= expbits_tab[exp_index1[index]] -
460 expbits_tab[exp_index1[index] + 1];
461 ++exp_index1[index];
462 } else { /* <--- */
463 int min = 999999;
464 index = -1;
465 for (i = 0; i < p->total_subbands; i++) {
466 if (exp_index2[i] > 0) {
467 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
468 if (v < min) {
469 min = v;
470 index = i;
471 }
472 }
473 }
474 if (index == -1)
475 break;
476 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
477 tmpbias2 -= expbits_tab[exp_index2[index]] -
478 expbits_tab[exp_index2[index] - 1];
479 --exp_index2[index];
480 }
481 }
482
483 for (i = 0; i < p->total_subbands; i++)
484 category[i] = exp_index2[i];
485
486 for (i = 0; i < p->numvector_size - 1; i++)
487 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
488 }
489
490
491 /**
492 * Expand the category vector.
493 *
494 * @param q pointer to the COOKContext
495 * @param category pointer to the category array
496 * @param category_index pointer to the category_index array
497 */
498 static inline void expand_category(COOKContext *q, int *category,
499 int *category_index)
500 {
501 int i;
502 for (i = 0; i < q->num_vectors; i++)
503 {
504 int idx = category_index[i];
505 if (++category[idx] >= FF_ARRAY_ELEMS(dither_tab))
506 --category[idx];
507 }
508 }
509
510 /**
511 * The real requantization of the mltcoefs
512 *
513 * @param q pointer to the COOKContext
514 * @param index index
515 * @param quant_index quantisation index
516 * @param subband_coef_index array of indexes to quant_centroid_tab
517 * @param subband_coef_sign signs of coefficients
518 * @param mlt_p pointer into the mlt buffer
519 */
520 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
521 int *subband_coef_index, int *subband_coef_sign,
522 float *mlt_p)
523 {
524 int i;
525 float f1;
526
527 for (i = 0; i < SUBBAND_SIZE; i++) {
528 if (subband_coef_index[i]) {
529 f1 = quant_centroid_tab[index][subband_coef_index[i]];
530 if (subband_coef_sign[i])
531 f1 = -f1;
532 } else {
533 /* noise coding if subband_coef_index[i] == 0 */
534 f1 = dither_tab[index];
535 if (av_lfg_get(&q->random_state) < 0x80000000)
536 f1 = -f1;
537 }
538 mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
539 }
540 }
541 /**
542 * Unpack the subband_coef_index and subband_coef_sign vectors.
543 *
544 * @param q pointer to the COOKContext
545 * @param category pointer to the category array
546 * @param subband_coef_index array of indexes to quant_centroid_tab
547 * @param subband_coef_sign signs of coefficients
548 */
549 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
550 int *subband_coef_index, int *subband_coef_sign)
551 {
552 int i, j;
553 int vlc, vd, tmp, result;
554
555 vd = vd_tab[category];
556 result = 0;
557 for (i = 0; i < vpr_tab[category]; i++) {
558 vlc = bitstream_read_vlc(&q->bc, q->sqvh[category].table, q->sqvh[category].bits, 3);
559 if (p->bits_per_subpacket < bitstream_tell(&q->bc)) {
560 vlc = 0;
561 result = 1;
562 }
563 for (j = vd - 1; j >= 0; j--) {
564 tmp = (vlc * invradix_tab[category]) / 0x100000;
565 subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
566 vlc = tmp;
567 }
568 for (j = 0; j < vd; j++) {
569 if (subband_coef_index[i * vd + j]) {
570 if (bitstream_tell(&q->bc) < p->bits_per_subpacket) {
571 subband_coef_sign[i * vd + j] = bitstream_read_bit(&q->bc);
572 } else {
573 result = 1;
574 subband_coef_sign[i * vd + j] = 0;
575 }
576 } else {
577 subband_coef_sign[i * vd + j] = 0;
578 }
579 }
580 }
581 return result;
582 }
583
584
585 /**
586 * Fill the mlt_buffer with mlt coefficients.
587 *
588 * @param q pointer to the COOKContext
589 * @param category pointer to the category array
590 * @param quant_index_table pointer to the array
591 * @param mlt_buffer pointer to mlt coefficients
592 */
593 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
594 int *quant_index_table, float *mlt_buffer)
595 {
596 /* A zero in this table means that the subband coefficient is
597 random noise coded. */
598 int subband_coef_index[SUBBAND_SIZE];
599 /* A zero in this table means that the subband coefficient is a
600 positive multiplicator. */
601 int subband_coef_sign[SUBBAND_SIZE];
602 int band, j;
603 int index = 0;
604
605 for (band = 0; band < p->total_subbands; band++) {
606 index = category[band];
607 if (category[band] < 7) {
608 if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
609 index = 7;
610 for (j = 0; j < p->total_subbands; j++)
611 category[band + j] = 7;
612 }
613 }
614 if (index >= 7) {
615 memset(subband_coef_index, 0, sizeof(subband_coef_index));
616 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
617 }
618 q->scalar_dequant(q, index, quant_index_table[band],
619 subband_coef_index, subband_coef_sign,
620 &mlt_buffer[band * SUBBAND_SIZE]);
621 }
622
623 /* FIXME: should this be removed, or moved into loop above? */
624 if (p->total_subbands * SUBBAND_SIZE >= q->samples_per_channel)
625 return;
626 }
627
628
629 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
630 {
631 int category_index[128] = { 0 };
632 int category[128] = { 0 };
633 int quant_index_table[102];
634 int res;
635
636 if ((res = decode_envelope(q, p, quant_index_table)) < 0)
637 return res;
638 q->num_vectors = bitstream_read(&q->bc, p->log2_numvector_size);
639 categorize(q, p, quant_index_table, category, category_index);
640 expand_category(q, category, category_index);
641 decode_vectors(q, p, category, quant_index_table, mlt_buffer);
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 */
655 static void interpolate_float(COOKContext *q, float *buffer,
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
663 for (i = 0; i < q->gain_size_factor; i++)
664 buffer[i] *= fc1;
665 } else { // smooth gain
666 fc2 = q->gain_table[11 + (gain_index_next - gain_index)];
667 for (i = 0; i < q->gain_size_factor; i++) {
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 */
682 static void imlt_window_float(COOKContext *q, float *inbuffer,
683 cook_gains *gains_ptr, float *previous_buffer)
684 {
685 const float fc = pow2tab[gains_ptr->previous[0] + 63];
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 */
694 for (i = 0; i < q->samples_per_channel; i++)
695 inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
696 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - 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 */
710 static void imlt_gain(COOKContext *q, float *inbuffer,
711 cook_gains *gains_ptr, float *previous_buffer)
712 {
713 float *buffer0 = q->mono_mdct_output;
714 float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
715 int i;
716
717 /* Inverse modified discrete cosine transform */
718 q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
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])
725 q->interpolate(q, &buffer1[q->gain_size_factor * i],
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,
730 q->samples_per_channel * sizeof(*previous_buffer));
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 */
740 static void decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
741 {
742 int i;
743 int vlc = bitstream_read_bit(&q->bc);
744 int start = cplband[p->js_subband_start];
745 int end = cplband[p->subbands - 1];
746 int length = end - start + 1;
747
748 if (start > end)
749 return;
750
751 if (vlc)
752 for (i = 0; i < length; i++)
753 decouple_tab[start + i] =
754 bitstream_read_vlc(&q->bc,
755 p->channel_coupling.table,
756 p->channel_coupling.bits, 2);
757 else
758 for (i = 0; i < length; i++)
759 decouple_tab[start + i] = bitstream_read(&q->bc, p->js_vlc_bits);
760 }
761
762 /*
763 * function decouples a pair of signals from a single signal via multiplication.
764 *
765 * @param q pointer to the COOKContext
766 * @param subband index of the current subband
767 * @param f1 multiplier for channel 1 extraction
768 * @param f2 multiplier for channel 2 extraction
769 * @param decode_buffer input buffer
770 * @param mlt_buffer1 pointer to left channel mlt coefficients
771 * @param mlt_buffer2 pointer to right channel mlt coefficients
772 */
773 static void decouple_float(COOKContext *q,
774 COOKSubpacket *p,
775 int subband,
776 float f1, float f2,
777 float *decode_buffer,
778 float *mlt_buffer1, float *mlt_buffer2)
779 {
780 int j, tmp_idx;
781 for (j = 0; j < SUBBAND_SIZE; j++) {
782 tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
783 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
784 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
785 }
786 }
787
788 /**
789 * function for decoding joint stereo data
790 *
791 * @param q pointer to the COOKContext
792 * @param mlt_buffer1 pointer to left channel mlt coefficients
793 * @param mlt_buffer2 pointer to right channel mlt coefficients
794 */
795 static int joint_decode(COOKContext *q, COOKSubpacket *p,
796 float *mlt_buffer_left, float *mlt_buffer_right)
797 {
798 int i, j, res;
799 int decouple_tab[SUBBAND_SIZE] = { 0 };
800 float *decode_buffer = q->decode_buffer_0;
801 int idx, cpl_tmp;
802 float f1, f2;
803 const float *cplscale;
804
805 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
806
807 /* Make sure the buffers are zeroed out. */
808 memset(mlt_buffer_left, 0, 1024 * sizeof(*mlt_buffer_left));
809 memset(mlt_buffer_right, 0, 1024 * sizeof(*mlt_buffer_right));
810 decouple_info(q, p, decouple_tab);
811 if ((res = mono_decode(q, p, decode_buffer)) < 0)
812 return res;
813
814 /* The two channels are stored interleaved in decode_buffer. */
815 for (i = 0; i < p->js_subband_start; i++) {
816 for (j = 0; j < SUBBAND_SIZE; j++) {
817 mlt_buffer_left[i * 20 + j] = decode_buffer[i * 40 + j];
818 mlt_buffer_right[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
819 }
820 }
821
822 /* When we reach js_subband_start (the higher frequencies)
823 the coefficients are stored in a coupling scheme. */
824 idx = (1 << p->js_vlc_bits) - 1;
825 for (i = p->js_subband_start; i < p->subbands; i++) {
826 cpl_tmp = cplband[i];
827 idx -= decouple_tab[cpl_tmp];
828 cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
829 f1 = cplscale[decouple_tab[cpl_tmp] + 1];
830 f2 = cplscale[idx];
831 q->decouple(q, p, i, f1, f2, decode_buffer,
832 mlt_buffer_left, mlt_buffer_right);
833 idx = (1 << p->js_vlc_bits) - 1;
834 }
835
836 return 0;
837 }
838
839 /**
840 * First part of subpacket decoding:
841 * decode raw stream bytes and read gain info.
842 *
843 * @param q pointer to the COOKContext
844 * @param inbuffer pointer to raw stream data
845 * @param gains_ptr array of current/prev gain pointers
846 */
847 static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
848 const uint8_t *inbuffer,
849 cook_gains *gains_ptr)
850 {
851 int offset;
852
853 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
854 p->bits_per_subpacket / 8);
855 bitstream_init(&q->bc, q->decoded_bytes_buffer + offset,
856 p->bits_per_subpacket);
857 decode_gain_info(&q->bc, gains_ptr->now);
858
859 /* Swap current and previous gains */
860 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
861 }
862
863 /**
864 * Saturate the output signal and interleave.
865 *
866 * @param q pointer to the COOKContext
867 * @param out pointer to the output vector
868 */
869 static void saturate_output_float(COOKContext *q, float *out)
870 {
871 q->adsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
872 FFALIGN(q->samples_per_channel, 8), -1.0f, 1.0f);
873 }
874
875
876 /**
877 * Final part of subpacket decoding:
878 * Apply modulated lapped transform, gain compensation,
879 * clip and convert to integer.
880 *
881 * @param q pointer to the COOKContext
882 * @param decode_buffer pointer to the mlt coefficients
883 * @param gains_ptr array of current/prev gain pointers
884 * @param previous_buffer pointer to the previous buffer to be used for overlapping
885 * @param out pointer to the output buffer
886 */
887 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
888 cook_gains *gains_ptr, float *previous_buffer,
889 float *out)
890 {
891 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
892 if (out)
893 q->saturate_output(q, out);
894 }
895
896
897 /**
898 * Cook subpacket decoding. This function returns one decoded subpacket,
899 * usually 1024 samples per channel.
900 *
901 * @param q pointer to the COOKContext
902 * @param inbuffer pointer to the inbuffer
903 * @param outbuffer pointer to the outbuffer
904 */
905 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
906 const uint8_t *inbuffer, float **outbuffer)
907 {
908 int sub_packet_size = p->size;
909 int res;
910
911 memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
912 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
913
914 if (p->joint_stereo) {
915 if ((res = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
916 return res;
917 } else {
918 if ((res = mono_decode(q, p, q->decode_buffer_1)) < 0)
919 return res;
920
921 if (p->num_channels == 2) {
922 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
923 if ((res = mono_decode(q, p, q->decode_buffer_2)) < 0)
924 return res;
925 }
926 }
927
928 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
929 p->mono_previous_buffer1,
930 outbuffer ? outbuffer[p->ch_idx] : NULL);
931
932 if (p->num_channels == 2)
933 if (p->joint_stereo)
934 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
935 p->mono_previous_buffer2,
936 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
937 else
938 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
939 p->mono_previous_buffer2,
940 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
941
942 return 0;
943 }
944
945
946 static int cook_decode_frame(AVCodecContext *avctx, void *data,
947 int *got_frame_ptr, AVPacket *avpkt)
948 {
949 AVFrame *frame = data;
950 const uint8_t *buf = avpkt->data;
951 int buf_size = avpkt->size;
952 COOKContext *q = avctx->priv_data;
953 float **samples = NULL;
954 int i, ret;
955 int offset = 0;
956 int chidx = 0;
957
958 if (buf_size < avctx->block_align)
959 return buf_size;
960
961 /* get output buffer */
962 if (q->discarded_packets >= 2) {
963 frame->nb_samples = q->samples_per_channel;
964 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
965 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
966 return ret;
967 }
968 samples = (float **)frame->extended_data;
969 }
970
971 /* estimate subpacket sizes */
972 q->subpacket[0].size = avctx->block_align;
973
974 for (i = 1; i < q->num_subpackets; i++) {
975 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
976 q->subpacket[0].size -= q->subpacket[i].size + 1;
977 if (q->subpacket[0].size < 0) {
978 av_log(avctx, AV_LOG_DEBUG,
979 "frame subpacket size total > avctx->block_align!\n");
980 return AVERROR_INVALIDDATA;
981 }
982 }
983
984 /* decode supbackets */
985 for (i = 0; i < q->num_subpackets; i++) {
986 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
987 q->subpacket[i].bits_per_subpdiv;
988 q->subpacket[i].ch_idx = chidx;
989 av_log(avctx, AV_LOG_DEBUG,
990 "subpacket[%i] size %i js %i %i block_align %i\n",
991 i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
992 avctx->block_align);
993
994 if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
995 return ret;
996 offset += q->subpacket[i].size;
997 chidx += q->subpacket[i].num_channels;
998 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
999 i, q->subpacket[i].size * 8, bitstream_tell(&q->bc));
1000 }
1001
1002 /* Discard the first two frames: no valid audio. */
1003 if (q->discarded_packets < 2) {
1004 q->discarded_packets++;
1005 *got_frame_ptr = 0;
1006 return avctx->block_align;
1007 }
1008
1009 *got_frame_ptr = 1;
1010
1011 return avctx->block_align;
1012 }
1013
1014 #ifdef DEBUG
1015 static void dump_cook_context(COOKContext *q)
1016 {
1017 //int i=0;
1018 #define PRINT(a, b) ff_dlog(q->avctx, " %s = %d\n", a, b);
1019 ff_dlog(q->avctx, "COOKextradata\n");
1020 ff_dlog(q->avctx, "cookversion=%x\n", q->subpacket[0].cookversion);
1021 if (q->subpacket[0].cookversion > STEREO) {
1022 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1023 PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1024 }
1025 ff_dlog(q->avctx, "COOKContext\n");
1026 PRINT("nb_channels", q->avctx->channels);
1027 PRINT("bit_rate", q->avctx->bit_rate);
1028 PRINT("sample_rate", q->avctx->sample_rate);
1029 PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1030 PRINT("subbands", q->subpacket[0].subbands);
1031 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1032 PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1033 PRINT("numvector_size", q->subpacket[0].numvector_size);
1034 PRINT("total_subbands", q->subpacket[0].total_subbands);
1035 }
1036 #endif
1037
1038 /**
1039 * Cook initialization
1040 *
1041 * @param avctx pointer to the AVCodecContext
1042 */
1043 static av_cold int cook_decode_init(AVCodecContext *avctx)
1044 {
1045 COOKContext *q = avctx->priv_data;
1046 GetByteContext gb;
1047 int s = 0;
1048 unsigned int channel_mask = 0;
1049 int samples_per_frame;
1050 int ret;
1051 q->avctx = avctx;
1052
1053 /* Take care of the codec specific extradata. */
1054 if (avctx->extradata_size < 8) {
1055 av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1056 return AVERROR_INVALIDDATA;
1057 }
1058 av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1059
1060 bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
1061
1062 /* Take data from the AVCodecContext (RM container). */
1063 if (!avctx->channels) {
1064 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1065 return AVERROR_INVALIDDATA;
1066 }
1067
1068 /* Initialize RNG. */
1069 av_lfg_init(&q->random_state, 0);
1070
1071 ff_audiodsp_init(&q->adsp);
1072
1073 while (bytestream2_get_bytes_left(&gb)) {
1074 /* 8 for mono, 16 for stereo, ? for multichannel
1075 Swap to right endianness so we don't need to care later on. */
1076 q->subpacket[s].cookversion = bytestream2_get_be32(&gb);
1077 samples_per_frame = bytestream2_get_be16(&gb);
1078 q->subpacket[s].subbands = bytestream2_get_be16(&gb);
1079 bytestream2_get_be32(&gb); // Unknown unused
1080 q->subpacket[s].js_subband_start = bytestream2_get_be16(&gb);
1081 q->subpacket[s].js_vlc_bits = bytestream2_get_be16(&gb);
1082
1083 /* Initialize extradata related variables. */
1084 q->subpacket[s].samples_per_channel = samples_per_frame / avctx->channels;
1085 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1086
1087 /* Initialize default data states. */
1088 q->subpacket[s].log2_numvector_size = 5;
1089 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1090 q->subpacket[s].num_channels = 1;
1091
1092 /* Initialize version-dependent variables */
1093
1094 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1095 q->subpacket[s].cookversion);
1096 q->subpacket[s].joint_stereo = 0;
1097 switch (q->subpacket[s].cookversion) {
1098 case MONO:
1099 if (avctx->channels != 1) {
1100 avpriv_request_sample(avctx, "Container channels != 1");
1101 return AVERROR_PATCHWELCOME;
1102 }
1103 av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1104 break;
1105 case STEREO:
1106 if (avctx->channels != 1) {
1107 q->subpacket[s].bits_per_subpdiv = 1;
1108 q->subpacket[s].num_channels = 2;
1109 }
1110 av_log(avctx, AV_LOG_DEBUG, "STEREO\n");
1111 break;
1112 case JOINT_STEREO:
1113 if (avctx->channels != 2) {
1114 avpriv_request_sample(avctx, "Container channels != 2");
1115 return AVERROR_PATCHWELCOME;
1116 }
1117 av_log(avctx, AV_LOG_DEBUG, "JOINT_STEREO\n");
1118 if (avctx->extradata_size >= 16) {
1119 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1120 q->subpacket[s].js_subband_start;
1121 q->subpacket[s].joint_stereo = 1;
1122 q->subpacket[s].num_channels = 2;
1123 }
1124 if (q->subpacket[s].samples_per_channel > 256) {
1125 q->subpacket[s].log2_numvector_size = 6;
1126 }
1127 if (q->subpacket[s].samples_per_channel > 512) {
1128 q->subpacket[s].log2_numvector_size = 7;
1129 }
1130 break;
1131 case MC_COOK:
1132 av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1133 channel_mask |= q->subpacket[s].channel_mask = bytestream2_get_be32(&gb);
1134
1135 if (av_get_channel_layout_nb_channels(q->subpacket[s].channel_mask) > 1) {
1136 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1137 q->subpacket[s].js_subband_start;
1138 q->subpacket[s].joint_stereo = 1;
1139 q->subpacket[s].num_channels = 2;
1140 q->subpacket[s].samples_per_channel = samples_per_frame >> 1;
1141
1142 if (q->subpacket[s].samples_per_channel > 256) {
1143 q->subpacket[s].log2_numvector_size = 6;
1144 }
1145 if (q->subpacket[s].samples_per_channel > 512) {
1146 q->subpacket[s].log2_numvector_size = 7;
1147 }
1148 } else
1149 q->subpacket[s].samples_per_channel = samples_per_frame;
1150
1151 break;
1152 default:
1153 avpriv_request_sample(avctx, "Cook version %d",
1154 q->subpacket[s].cookversion);
1155 return AVERROR_PATCHWELCOME;
1156 }
1157
1158 if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1159 av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1160 return AVERROR_INVALIDDATA;
1161 } else
1162 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1163
1164
1165 /* Initialize variable relations */
1166 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1167
1168 /* Try to catch some obviously faulty streams, otherwise it might be exploitable */
1169 if (q->subpacket[s].total_subbands > 53) {
1170 avpriv_request_sample(avctx, "total_subbands > 53");
1171 return AVERROR_PATCHWELCOME;
1172 }
1173
1174 if ((q->subpacket[s].js_vlc_bits > 6) ||
1175 (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1176 av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1177 q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1178 return AVERROR_INVALIDDATA;
1179 }
1180
1181 if (q->subpacket[s].subbands > 50) {
1182 avpriv_request_sample(avctx, "subbands > 50");
1183 return AVERROR_PATCHWELCOME;
1184 }
1185 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1186 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1187 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1188 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1189
1190 q->num_subpackets++;
1191 s++;
1192 if (s > MAX_SUBPACKETS) {
1193 avpriv_request_sample(avctx, "subpackets > %d", MAX_SUBPACKETS);
1194 return AVERROR_PATCHWELCOME;
1195 }
1196 }
1197 /* Generate tables */
1198 init_pow2table();
1199 init_gain_table(q);
1200 init_cplscales_table(q);
1201
1202 if ((ret = init_cook_vlc_tables(q)))
1203 return ret;
1204
1205
1206 if (avctx->block_align >= UINT_MAX / 2)
1207 return AVERROR(EINVAL);
1208
1209 /* Pad the databuffer with:
1210 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1211 AV_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1212 q->decoded_bytes_buffer =
1213 av_mallocz(avctx->block_align
1214 + DECODE_BYTES_PAD1(avctx->block_align)
1215 + AV_INPUT_BUFFER_PADDING_SIZE);
1216 if (!q->decoded_bytes_buffer)
1217 return AVERROR(ENOMEM);
1218
1219 /* Initialize transform. */
1220 if ((ret = init_cook_mlt(q)))
1221 return ret;
1222
1223 /* Initialize COOK signal arithmetic handling */
1224 if (1) {
1225 q->scalar_dequant = scalar_dequant_float;
1226 q->decouple = decouple_float;
1227 q->imlt_window = imlt_window_float;
1228 q->interpolate = interpolate_float;
1229 q->saturate_output = saturate_output_float;
1230 }
1231
1232 /* Try to catch some obviously faulty streams, otherwise it might be exploitable */
1233 if (q->samples_per_channel != 256 && q->samples_per_channel != 512 &&
1234 q->samples_per_channel != 1024) {
1235 avpriv_request_sample(avctx, "samples_per_channel = %d",
1236 q->samples_per_channel);
1237 return AVERROR_PATCHWELCOME;
1238 }
1239
1240 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1241 if (channel_mask)
1242 avctx->channel_layout = channel_mask;
1243 else
1244 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1245
1246 #ifdef DEBUG
1247 dump_cook_context(q);
1248 #endif
1249 return 0;
1250 }
1251
1252 AVCodec ff_cook_decoder = {
1253 .name = "cook",
1254 .long_name = NULL_IF_CONFIG_SMALL("Cook / Cooker / Gecko (RealAudio G2)"),
1255 .type = AVMEDIA_TYPE_AUDIO,
1256 .id = AV_CODEC_ID_COOK,
1257 .priv_data_size = sizeof(COOKContext),
1258 .init = cook_decode_init,
1259 .close = cook_decode_close,
1260 .decode = cook_decode_frame,
1261 .capabilities = AV_CODEC_CAP_DR1,
1262 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1263 AV_SAMPLE_FMT_NONE },
1264 };