c3304eae2e3dc83917713d671ec62858f11c2605
[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
56 #include "cookdata.h"
57
58 /* the different Cook versions */
59 #define MONO 0x1000001
60 #define STEREO 0x1000002
61 #define JOINT_STEREO 0x1000003
62 #define MC_COOK 0x2000000 // multichannel Cook, not supported
63
64 #define SUBBAND_SIZE 20
65 #define MAX_SUBPACKETS 5
66
67 typedef struct cook_gains {
68 int *now;
69 int *previous;
70 } cook_gains;
71
72 typedef struct COOKSubpacket {
73 int ch_idx;
74 int size;
75 int num_channels;
76 int cookversion;
77 int subbands;
78 int js_subband_start;
79 int js_vlc_bits;
80 int samples_per_channel;
81 int log2_numvector_size;
82 unsigned int channel_mask;
83 VLC channel_coupling;
84 int joint_stereo;
85 int bits_per_subpacket;
86 int bits_per_subpdiv;
87 int total_subbands;
88 int numvector_size; // 1 << log2_numvector_size;
89
90 float mono_previous_buffer1[1024];
91 float mono_previous_buffer2[1024];
92
93 cook_gains gains1;
94 cook_gains gains2;
95 int gain_1[9];
96 int gain_2[9];
97 int gain_3[9];
98 int gain_4[9];
99 } COOKSubpacket;
100
101 typedef struct cook {
102 /*
103 * The following 5 functions provide the lowlevel arithmetic on
104 * the internal audio buffers.
105 */
106 void (*scalar_dequant)(struct cook *q, int index, int quant_index,
107 int *subband_coef_index, int *subband_coef_sign,
108 float *mlt_p);
109
110 void (*decouple)(struct cook *q,
111 COOKSubpacket *p,
112 int subband,
113 float f1, float f2,
114 float *decode_buffer,
115 float *mlt_buffer1, float *mlt_buffer2);
116
117 void (*imlt_window)(struct cook *q, float *buffer1,
118 cook_gains *gains_ptr, float *previous_buffer);
119
120 void (*interpolate)(struct cook *q, float *buffer,
121 int gain_index, int gain_index_next);
122
123 void (*saturate_output)(struct cook *q, float *out);
124
125 AVCodecContext* avctx;
126 AudioDSPContext adsp;
127 BitstreamContext bc;
128 /* stream data */
129 int num_vectors;
130 int samples_per_channel;
131 /* states */
132 AVLFG random_state;
133 int discarded_packets;
134
135 /* transform data */
136 FFTContext mdct_ctx;
137 float* mlt_window;
138
139 /* VLC data */
140 VLC envelope_quant_index[13];
141 VLC sqvh[7]; // scalar quantization
142
143 /* generate tables and related variables */
144 int gain_size_factor;
145 float gain_table[23];
146
147 /* data buffers */
148
149 uint8_t* decoded_bytes_buffer;
150 DECLARE_ALIGNED(32, float, mono_mdct_output)[2048];
151 float decode_buffer_1[1024];
152 float decode_buffer_2[1024];
153 float decode_buffer_0[1060]; /* static allocation for joint decode */
154
155 const float *cplscales[5];
156 int num_subpackets;
157 COOKSubpacket subpacket[MAX_SUBPACKETS];
158 } COOKContext;
159
160 static float pow2tab[127];
161 static float rootpow2tab[127];
162
163 /*************** init functions ***************/
164
165 /* table generator */
166 static av_cold void init_pow2table(void)
167 {
168 int i;
169 for (i = -63; i < 64; i++) {
170 pow2tab[63 + i] = pow(2, i);
171 rootpow2tab[63 + i] = sqrt(pow(2, i));
172 }
173 }
174
175 /* table generator */
176 static av_cold void init_gain_table(COOKContext *q)
177 {
178 int i;
179 q->gain_size_factor = q->samples_per_channel / 8;
180 for (i = 0; i < 23; i++)
181 q->gain_table[i] = pow(pow2tab[i + 52],
182 (1.0 / (double) q->gain_size_factor));
183 }
184
185
186 static av_cold int init_cook_vlc_tables(COOKContext *q)
187 {
188 int i, result;
189
190 result = 0;
191 for (i = 0; i < 13; i++) {
192 result |= init_vlc(&q->envelope_quant_index[i], 9, 24,
193 envelope_quant_index_huffbits[i], 1, 1,
194 envelope_quant_index_huffcodes[i], 2, 2, 0);
195 }
196 av_log(q->avctx, AV_LOG_DEBUG, "sqvh VLC init\n");
197 for (i = 0; i < 7; i++) {
198 result |= init_vlc(&q->sqvh[i], vhvlcsize_tab[i], vhsize_tab[i],
199 cvh_huffbits[i], 1, 1,
200 cvh_huffcodes[i], 2, 2, 0);
201 }
202
203 for (i = 0; i < q->num_subpackets; i++) {
204 if (q->subpacket[i].joint_stereo == 1) {
205 result |= init_vlc(&q->subpacket[i].channel_coupling, 6,
206 (1 << q->subpacket[i].js_vlc_bits) - 1,
207 ccpl_huffbits[q->subpacket[i].js_vlc_bits - 2], 1, 1,
208 ccpl_huffcodes[q->subpacket[i].js_vlc_bits - 2], 2, 2, 0);
209 av_log(q->avctx, AV_LOG_DEBUG, "subpacket %i Joint-stereo VLC used.\n", i);
210 }
211 }
212
213 av_log(q->avctx, AV_LOG_DEBUG, "VLC tables initialized.\n");
214 return result;
215 }
216
217 static av_cold int init_cook_mlt(COOKContext *q)
218 {
219 int j, ret;
220 int mlt_size = q->samples_per_channel;
221
222 if ((q->mlt_window = av_malloc(mlt_size * sizeof(*q->mlt_window))) == 0)
223 return AVERROR(ENOMEM);
224
225 /* Initialize the MLT window: simple sine window. */
226 ff_sine_window_init(q->mlt_window, mlt_size);
227 for (j = 0; j < mlt_size; j++)
228 q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
229
230 /* Initialize the MDCT. */
231 if ((ret = ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size) + 1, 1, 1.0 / 32768.0))) {
232 av_free(q->mlt_window);
233 return ret;
234 }
235 av_log(q->avctx, AV_LOG_DEBUG, "MDCT initialized, order = %d.\n",
236 av_log2(mlt_size) + 1);
237
238 return 0;
239 }
240
241 static av_cold void init_cplscales_table(COOKContext *q)
242 {
243 int i;
244 for (i = 0; i < 5; i++)
245 q->cplscales[i] = cplscales[i];
246 }
247
248 /*************** init functions end ***********/
249
250 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes) + 3) % 4)
251 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
252
253 /**
254 * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
255 * Why? No idea, some checksum/error detection method maybe.
256 *
257 * Out buffer size: extra bytes are needed to cope with
258 * padding/misalignment.
259 * Subpackets passed to the decoder can contain two, consecutive
260 * half-subpackets, of identical but arbitrary size.
261 * 1234 1234 1234 1234 extraA extraB
262 * Case 1: AAAA BBBB 0 0
263 * Case 2: AAAA ABBB BB-- 3 3
264 * Case 3: AAAA AABB BBBB 2 2
265 * Case 4: AAAA AAAB BBBB BB-- 1 5
266 *
267 * Nice way to waste CPU cycles.
268 *
269 * @param inbuffer pointer to byte array of indata
270 * @param out pointer to byte array of outdata
271 * @param bytes number of bytes
272 */
273 static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
274 {
275 static const uint32_t tab[4] = {
276 AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
277 AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
278 };
279 int i, off;
280 uint32_t c;
281 const uint32_t *buf;
282 uint32_t *obuf = (uint32_t *) out;
283 /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
284 * I'm too lazy though, should be something like
285 * for (i = 0; i < bitamount / 64; i++)
286 * (int64_t) out[i] = 0x37c511f237c511f2 ^ av_be2ne64(int64_t) in[i]);
287 * Buffer alignment needs to be checked. */
288
289 off = (intptr_t) inbuffer & 3;
290 buf = (const uint32_t *) (inbuffer - off);
291 c = tab[off];
292 bytes += 3 + off;
293 for (i = 0; i < bytes / 4; i++)
294 obuf[i] = c ^ buf[i];
295
296 return off;
297 }
298
299 static av_cold int cook_decode_close(AVCodecContext *avctx)
300 {
301 int i;
302 COOKContext *q = avctx->priv_data;
303 av_log(avctx, AV_LOG_DEBUG, "Deallocating memory.\n");
304
305 /* Free allocated memory buffers. */
306 av_free(q->mlt_window);
307 av_free(q->decoded_bytes_buffer);
308
309 /* Free the transform. */
310 ff_mdct_end(&q->mdct_ctx);
311
312 /* Free the VLC tables. */
313 for (i = 0; i < 13; i++)
314 ff_free_vlc(&q->envelope_quant_index[i]);
315 for (i = 0; i < 7; i++)
316 ff_free_vlc(&q->sqvh[i]);
317 for (i = 0; i < q->num_subpackets; i++)
318 ff_free_vlc(&q->subpacket[i].channel_coupling);
319
320 av_log(avctx, AV_LOG_DEBUG, "Memory deallocated.\n");
321
322 return 0;
323 }
324
325 /**
326 * Fill the gain array for the timedomain quantization.
327 *
328 * @param bc pointer to the BitstreamContext
329 * @param gaininfo array[9] of gain indexes
330 */
331 static void decode_gain_info(BitstreamContext *bc, int *gaininfo)
332 {
333 int i, n;
334
335 while (bitstream_read_bit(bc)) {
336 /* NOTHING */
337 }
338
339 n = bitstream_tell(bc) - 1; // amount of elements * 2 to update
340
341 i = 0;
342 while (n--) {
343 int index = bitstream_read(bc, 3);
344 int gain = bitstream_read_bit(bc) ? bitstream_read(bc, 4) - 7 : -1;
345
346 while (i <= index)
347 gaininfo[i++] = gain;
348 }
349 while (i <= 8)
350 gaininfo[i++] = 0;
351 }
352
353 /**
354 * Create the quant index table needed for the envelope.
355 *
356 * @param q pointer to the COOKContext
357 * @param quant_index_table pointer to the array
358 */
359 static int decode_envelope(COOKContext *q, COOKSubpacket *p,
360 int *quant_index_table)
361 {
362 int i, j, vlc_index;
363
364 quant_index_table[0] = bitstream_read(&q->bc, 6) - 6; // This is used later in categorize
365
366 for (i = 1; i < p->total_subbands; i++) {
367 vlc_index = i;
368 if (i >= p->js_subband_start * 2) {
369 vlc_index -= p->js_subband_start;
370 } else {
371 vlc_index /= 2;
372 if (vlc_index < 1)
373 vlc_index = 1;
374 }
375 if (vlc_index > 13)
376 vlc_index = 13; // the VLC tables >13 are identical to No. 13
377
378 j = bitstream_read_vlc(&q->bc, q->envelope_quant_index[vlc_index - 1].table,
379 q->envelope_quant_index[vlc_index - 1].bits, 2);
380 quant_index_table[i] = quant_index_table[i - 1] + j - 12; // differential encoding
381 if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
382 av_log(q->avctx, AV_LOG_ERROR,
383 "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
384 quant_index_table[i], i);
385 return AVERROR_INVALIDDATA;
386 }
387 }
388
389 return 0;
390 }
391
392 /**
393 * Calculate the category and category_index vector.
394 *
395 * @param q pointer to the COOKContext
396 * @param quant_index_table pointer to the array
397 * @param category pointer to the category array
398 * @param category_index pointer to the category_index array
399 */
400 static void categorize(COOKContext *q, COOKSubpacket *p, int *quant_index_table,
401 int *category, int *category_index)
402 {
403 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
404 int exp_index2[102] = { 0 };
405 int exp_index1[102] = { 0 };
406
407 int tmp_categorize_array[128 * 2] = { 0 };
408 int tmp_categorize_array1_idx = p->numvector_size;
409 int tmp_categorize_array2_idx = p->numvector_size;
410
411 bits_left = p->bits_per_subpacket - bitstream_tell(&q->bc);
412
413 if (bits_left > q->samples_per_channel)
414 bits_left = q->samples_per_channel +
415 ((bits_left - q->samples_per_channel) * 5) / 8;
416
417 bias = -32;
418
419 /* Estimate bias. */
420 for (i = 32; i > 0; i = i / 2) {
421 num_bits = 0;
422 index = 0;
423 for (j = p->total_subbands; j > 0; j--) {
424 exp_idx = av_clip((i - quant_index_table[index] + bias) / 2, 0, 7);
425 index++;
426 num_bits += expbits_tab[exp_idx];
427 }
428 if (num_bits >= bits_left - 32)
429 bias += i;
430 }
431
432 /* Calculate total number of bits. */
433 num_bits = 0;
434 for (i = 0; i < p->total_subbands; i++) {
435 exp_idx = av_clip((bias - quant_index_table[i]) / 2, 0, 7);
436 num_bits += expbits_tab[exp_idx];
437 exp_index1[i] = exp_idx;
438 exp_index2[i] = exp_idx;
439 }
440 tmpbias1 = tmpbias2 = num_bits;
441
442 for (j = 1; j < p->numvector_size; j++) {
443 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
444 int max = -999999;
445 index = -1;
446 for (i = 0; i < p->total_subbands; i++) {
447 if (exp_index1[i] < 7) {
448 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
449 if (v >= max) {
450 max = v;
451 index = i;
452 }
453 }
454 }
455 if (index == -1)
456 break;
457 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
458 tmpbias1 -= expbits_tab[exp_index1[index]] -
459 expbits_tab[exp_index1[index] + 1];
460 ++exp_index1[index];
461 } else { /* <--- */
462 int min = 999999;
463 index = -1;
464 for (i = 0; i < p->total_subbands; i++) {
465 if (exp_index2[i] > 0) {
466 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
467 if (v < min) {
468 min = v;
469 index = i;
470 }
471 }
472 }
473 if (index == -1)
474 break;
475 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
476 tmpbias2 -= expbits_tab[exp_index2[index]] -
477 expbits_tab[exp_index2[index] - 1];
478 --exp_index2[index];
479 }
480 }
481
482 for (i = 0; i < p->total_subbands; i++)
483 category[i] = exp_index2[i];
484
485 for (i = 0; i < p->numvector_size - 1; i++)
486 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
487 }
488
489
490 /**
491 * Expand the category vector.
492 *
493 * @param q pointer to the COOKContext
494 * @param category pointer to the category array
495 * @param category_index pointer to the category_index array
496 */
497 static inline void expand_category(COOKContext *q, int *category,
498 int *category_index)
499 {
500 int i;
501 for (i = 0; i < q->num_vectors; i++)
502 {
503 int idx = category_index[i];
504 if (++category[idx] >= FF_ARRAY_ELEMS(dither_tab))
505 --category[idx];
506 }
507 }
508
509 /**
510 * The real requantization of the mltcoefs
511 *
512 * @param q pointer to the COOKContext
513 * @param index index
514 * @param quant_index quantisation index
515 * @param subband_coef_index array of indexes to quant_centroid_tab
516 * @param subband_coef_sign signs of coefficients
517 * @param mlt_p pointer into the mlt buffer
518 */
519 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
520 int *subband_coef_index, int *subband_coef_sign,
521 float *mlt_p)
522 {
523 int i;
524 float f1;
525
526 for (i = 0; i < SUBBAND_SIZE; i++) {
527 if (subband_coef_index[i]) {
528 f1 = quant_centroid_tab[index][subband_coef_index[i]];
529 if (subband_coef_sign[i])
530 f1 = -f1;
531 } else {
532 /* noise coding if subband_coef_index[i] == 0 */
533 f1 = dither_tab[index];
534 if (av_lfg_get(&q->random_state) < 0x80000000)
535 f1 = -f1;
536 }
537 mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
538 }
539 }
540 /**
541 * Unpack the subband_coef_index and subband_coef_sign vectors.
542 *
543 * @param q pointer to the COOKContext
544 * @param category pointer to the category array
545 * @param subband_coef_index array of indexes to quant_centroid_tab
546 * @param subband_coef_sign signs of coefficients
547 */
548 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
549 int *subband_coef_index, int *subband_coef_sign)
550 {
551 int i, j;
552 int vlc, vd, tmp, result;
553
554 vd = vd_tab[category];
555 result = 0;
556 for (i = 0; i < vpr_tab[category]; i++) {
557 vlc = bitstream_read_vlc(&q->bc, q->sqvh[category].table, q->sqvh[category].bits, 3);
558 if (p->bits_per_subpacket < bitstream_tell(&q->bc)) {
559 vlc = 0;
560 result = 1;
561 }
562 for (j = vd - 1; j >= 0; j--) {
563 tmp = (vlc * invradix_tab[category]) / 0x100000;
564 subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
565 vlc = tmp;
566 }
567 for (j = 0; j < vd; j++) {
568 if (subband_coef_index[i * vd + j]) {
569 if (bitstream_tell(&q->bc) < p->bits_per_subpacket) {
570 subband_coef_sign[i * vd + j] = bitstream_read_bit(&q->bc);
571 } else {
572 result = 1;
573 subband_coef_sign[i * vd + j] = 0;
574 }
575 } else {
576 subband_coef_sign[i * vd + j] = 0;
577 }
578 }
579 }
580 return result;
581 }
582
583
584 /**
585 * Fill the mlt_buffer with mlt coefficients.
586 *
587 * @param q pointer to the COOKContext
588 * @param category pointer to the category array
589 * @param quant_index_table pointer to the array
590 * @param mlt_buffer pointer to mlt coefficients
591 */
592 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
593 int *quant_index_table, float *mlt_buffer)
594 {
595 /* A zero in this table means that the subband coefficient is
596 random noise coded. */
597 int subband_coef_index[SUBBAND_SIZE];
598 /* A zero in this table means that the subband coefficient is a
599 positive multiplicator. */
600 int subband_coef_sign[SUBBAND_SIZE];
601 int band, j;
602 int index = 0;
603
604 for (band = 0; band < p->total_subbands; band++) {
605 index = category[band];
606 if (category[band] < 7) {
607 if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
608 index = 7;
609 for (j = 0; j < p->total_subbands; j++)
610 category[band + j] = 7;
611 }
612 }
613 if (index >= 7) {
614 memset(subband_coef_index, 0, sizeof(subband_coef_index));
615 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
616 }
617 q->scalar_dequant(q, index, quant_index_table[band],
618 subband_coef_index, subband_coef_sign,
619 &mlt_buffer[band * SUBBAND_SIZE]);
620 }
621
622 /* FIXME: should this be removed, or moved into loop above? */
623 if (p->total_subbands * SUBBAND_SIZE >= q->samples_per_channel)
624 return;
625 }
626
627
628 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
629 {
630 int category_index[128] = { 0 };
631 int category[128] = { 0 };
632 int quant_index_table[102];
633 int res;
634
635 if ((res = decode_envelope(q, p, quant_index_table)) < 0)
636 return res;
637 q->num_vectors = bitstream_read(&q->bc, p->log2_numvector_size);
638 categorize(q, p, quant_index_table, category, category_index);
639 expand_category(q, category, category_index);
640 decode_vectors(q, p, category, quant_index_table, mlt_buffer);
641
642 return 0;
643 }
644
645
646 /**
647 * the actual requantization of the timedomain samples
648 *
649 * @param q pointer to the COOKContext
650 * @param buffer pointer to the timedomain buffer
651 * @param gain_index index for the block multiplier
652 * @param gain_index_next index for the next block multiplier
653 */
654 static void interpolate_float(COOKContext *q, float *buffer,
655 int gain_index, int gain_index_next)
656 {
657 int i;
658 float fc1, fc2;
659 fc1 = pow2tab[gain_index + 63];
660
661 if (gain_index == gain_index_next) { // static gain
662 for (i = 0; i < q->gain_size_factor; i++)
663 buffer[i] *= fc1;
664 } else { // smooth gain
665 fc2 = q->gain_table[11 + (gain_index_next - gain_index)];
666 for (i = 0; i < q->gain_size_factor; i++) {
667 buffer[i] *= fc1;
668 fc1 *= fc2;
669 }
670 }
671 }
672
673 /**
674 * Apply transform window, overlap buffers.
675 *
676 * @param q pointer to the COOKContext
677 * @param inbuffer pointer to the mltcoefficients
678 * @param gains_ptr current and previous gains
679 * @param previous_buffer pointer to the previous buffer to be used for overlapping
680 */
681 static void imlt_window_float(COOKContext *q, float *inbuffer,
682 cook_gains *gains_ptr, float *previous_buffer)
683 {
684 const float fc = pow2tab[gains_ptr->previous[0] + 63];
685 int i;
686 /* The weird thing here, is that the two halves of the time domain
687 * buffer are swapped. Also, the newest data, that we save away for
688 * next frame, has the wrong sign. Hence the subtraction below.
689 * Almost sounds like a complex conjugate/reverse data/FFT effect.
690 */
691
692 /* Apply window and overlap */
693 for (i = 0; i < q->samples_per_channel; i++)
694 inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
695 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
696 }
697
698 /**
699 * The modulated lapped transform, this takes transform coefficients
700 * and transforms them into timedomain samples.
701 * Apply transform window, overlap buffers, apply gain profile
702 * and buffer management.
703 *
704 * @param q pointer to the COOKContext
705 * @param inbuffer pointer to the mltcoefficients
706 * @param gains_ptr current and previous gains
707 * @param previous_buffer pointer to the previous buffer to be used for overlapping
708 */
709 static void imlt_gain(COOKContext *q, float *inbuffer,
710 cook_gains *gains_ptr, float *previous_buffer)
711 {
712 float *buffer0 = q->mono_mdct_output;
713 float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
714 int i;
715
716 /* Inverse modified discrete cosine transform */
717 q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
718
719 q->imlt_window(q, buffer1, gains_ptr, previous_buffer);
720
721 /* Apply gain profile */
722 for (i = 0; i < 8; i++)
723 if (gains_ptr->now[i] || gains_ptr->now[i + 1])
724 q->interpolate(q, &buffer1[q->gain_size_factor * i],
725 gains_ptr->now[i], gains_ptr->now[i + 1]);
726
727 /* Save away the current to be previous block. */
728 memcpy(previous_buffer, buffer0,
729 q->samples_per_channel * sizeof(*previous_buffer));
730 }
731
732
733 /**
734 * function for getting the jointstereo coupling information
735 *
736 * @param q pointer to the COOKContext
737 * @param decouple_tab decoupling array
738 */
739 static void decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
740 {
741 int i;
742 int vlc = bitstream_read_bit(&q->bc);
743 int start = cplband[p->js_subband_start];
744 int end = cplband[p->subbands - 1];
745 int length = end - start + 1;
746
747 if (start > end)
748 return;
749
750 if (vlc)
751 for (i = 0; i < length; i++)
752 decouple_tab[start + i] =
753 bitstream_read_vlc(&q->bc,
754 p->channel_coupling.table,
755 p->channel_coupling.bits, 2);
756 else
757 for (i = 0; i < length; i++)
758 decouple_tab[start + i] = bitstream_read(&q->bc, p->js_vlc_bits);
759 }
760
761 /*
762 * function decouples a pair of signals from a single signal via multiplication.
763 *
764 * @param q pointer to the COOKContext
765 * @param subband index of the current subband
766 * @param f1 multiplier for channel 1 extraction
767 * @param f2 multiplier for channel 2 extraction
768 * @param decode_buffer input buffer
769 * @param mlt_buffer1 pointer to left channel mlt coefficients
770 * @param mlt_buffer2 pointer to right channel mlt coefficients
771 */
772 static void decouple_float(COOKContext *q,
773 COOKSubpacket *p,
774 int subband,
775 float f1, float f2,
776 float *decode_buffer,
777 float *mlt_buffer1, float *mlt_buffer2)
778 {
779 int j, tmp_idx;
780 for (j = 0; j < SUBBAND_SIZE; j++) {
781 tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
782 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
783 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
784 }
785 }
786
787 /**
788 * function for decoding joint stereo data
789 *
790 * @param q pointer to the COOKContext
791 * @param mlt_buffer1 pointer to left channel mlt coefficients
792 * @param mlt_buffer2 pointer to right channel mlt coefficients
793 */
794 static int joint_decode(COOKContext *q, COOKSubpacket *p,
795 float *mlt_buffer_left, float *mlt_buffer_right)
796 {
797 int i, j, res;
798 int decouple_tab[SUBBAND_SIZE] = { 0 };
799 float *decode_buffer = q->decode_buffer_0;
800 int idx, cpl_tmp;
801 float f1, f2;
802 const float *cplscale;
803
804 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
805
806 /* Make sure the buffers are zeroed out. */
807 memset(mlt_buffer_left, 0, 1024 * sizeof(*mlt_buffer_left));
808 memset(mlt_buffer_right, 0, 1024 * sizeof(*mlt_buffer_right));
809 decouple_info(q, p, decouple_tab);
810 if ((res = mono_decode(q, p, decode_buffer)) < 0)
811 return res;
812
813 /* The two channels are stored interleaved in decode_buffer. */
814 for (i = 0; i < p->js_subband_start; i++) {
815 for (j = 0; j < SUBBAND_SIZE; j++) {
816 mlt_buffer_left[i * 20 + j] = decode_buffer[i * 40 + j];
817 mlt_buffer_right[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
818 }
819 }
820
821 /* When we reach js_subband_start (the higher frequencies)
822 the coefficients are stored in a coupling scheme. */
823 idx = (1 << p->js_vlc_bits) - 1;
824 for (i = p->js_subband_start; i < p->subbands; i++) {
825 cpl_tmp = cplband[i];
826 idx -= decouple_tab[cpl_tmp];
827 cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
828 f1 = cplscale[decouple_tab[cpl_tmp] + 1];
829 f2 = cplscale[idx];
830 q->decouple(q, p, i, f1, f2, decode_buffer,
831 mlt_buffer_left, mlt_buffer_right);
832 idx = (1 << p->js_vlc_bits) - 1;
833 }
834
835 return 0;
836 }
837
838 /**
839 * First part of subpacket decoding:
840 * decode raw stream bytes and read gain info.
841 *
842 * @param q pointer to the COOKContext
843 * @param inbuffer pointer to raw stream data
844 * @param gains_ptr array of current/prev gain pointers
845 */
846 static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
847 const uint8_t *inbuffer,
848 cook_gains *gains_ptr)
849 {
850 int offset;
851
852 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
853 p->bits_per_subpacket / 8);
854 bitstream_init(&q->bc, q->decoded_bytes_buffer + offset,
855 p->bits_per_subpacket);
856 decode_gain_info(&q->bc, gains_ptr->now);
857
858 /* Swap current and previous gains */
859 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
860 }
861
862 /**
863 * Saturate the output signal and interleave.
864 *
865 * @param q pointer to the COOKContext
866 * @param out pointer to the output vector
867 */
868 static void saturate_output_float(COOKContext *q, float *out)
869 {
870 q->adsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
871 FFALIGN(q->samples_per_channel, 8), -1.0f, 1.0f);
872 }
873
874
875 /**
876 * Final part of subpacket decoding:
877 * Apply modulated lapped transform, gain compensation,
878 * clip and convert to integer.
879 *
880 * @param q pointer to the COOKContext
881 * @param decode_buffer pointer to the mlt coefficients
882 * @param gains_ptr array of current/prev gain pointers
883 * @param previous_buffer pointer to the previous buffer to be used for overlapping
884 * @param out pointer to the output buffer
885 */
886 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
887 cook_gains *gains_ptr, float *previous_buffer,
888 float *out)
889 {
890 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
891 if (out)
892 q->saturate_output(q, out);
893 }
894
895
896 /**
897 * Cook subpacket decoding. This function returns one decoded subpacket,
898 * usually 1024 samples per channel.
899 *
900 * @param q pointer to the COOKContext
901 * @param inbuffer pointer to the inbuffer
902 * @param outbuffer pointer to the outbuffer
903 */
904 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
905 const uint8_t *inbuffer, float **outbuffer)
906 {
907 int sub_packet_size = p->size;
908 int res;
909
910 memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
911 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
912
913 if (p->joint_stereo) {
914 if ((res = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
915 return res;
916 } else {
917 if ((res = mono_decode(q, p, q->decode_buffer_1)) < 0)
918 return res;
919
920 if (p->num_channels == 2) {
921 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
922 if ((res = mono_decode(q, p, q->decode_buffer_2)) < 0)
923 return res;
924 }
925 }
926
927 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
928 p->mono_previous_buffer1,
929 outbuffer ? outbuffer[p->ch_idx] : NULL);
930
931 if (p->num_channels == 2)
932 if (p->joint_stereo)
933 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
934 p->mono_previous_buffer2,
935 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
936 else
937 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
938 p->mono_previous_buffer2,
939 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
940
941 return 0;
942 }
943
944
945 static int cook_decode_frame(AVCodecContext *avctx, void *data,
946 int *got_frame_ptr, AVPacket *avpkt)
947 {
948 AVFrame *frame = data;
949 const uint8_t *buf = avpkt->data;
950 int buf_size = avpkt->size;
951 COOKContext *q = avctx->priv_data;
952 float **samples = NULL;
953 int i, ret;
954 int offset = 0;
955 int chidx = 0;
956
957 if (buf_size < avctx->block_align)
958 return buf_size;
959
960 /* get output buffer */
961 if (q->discarded_packets >= 2) {
962 frame->nb_samples = q->samples_per_channel;
963 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
964 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
965 return ret;
966 }
967 samples = (float **)frame->extended_data;
968 }
969
970 /* estimate subpacket sizes */
971 q->subpacket[0].size = avctx->block_align;
972
973 for (i = 1; i < q->num_subpackets; i++) {
974 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
975 q->subpacket[0].size -= q->subpacket[i].size + 1;
976 if (q->subpacket[0].size < 0) {
977 av_log(avctx, AV_LOG_DEBUG,
978 "frame subpacket size total > avctx->block_align!\n");
979 return AVERROR_INVALIDDATA;
980 }
981 }
982
983 /* decode supbackets */
984 for (i = 0; i < q->num_subpackets; i++) {
985 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
986 q->subpacket[i].bits_per_subpdiv;
987 q->subpacket[i].ch_idx = chidx;
988 av_log(avctx, AV_LOG_DEBUG,
989 "subpacket[%i] size %i js %i %i block_align %i\n",
990 i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
991 avctx->block_align);
992
993 if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
994 return ret;
995 offset += q->subpacket[i].size;
996 chidx += q->subpacket[i].num_channels;
997 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
998 i, q->subpacket[i].size * 8, bitstream_tell(&q->bc));
999 }
1000
1001 /* Discard the first two frames: no valid audio. */
1002 if (q->discarded_packets < 2) {
1003 q->discarded_packets++;
1004 *got_frame_ptr = 0;
1005 return avctx->block_align;
1006 }
1007
1008 *got_frame_ptr = 1;
1009
1010 return avctx->block_align;
1011 }
1012
1013 #ifdef DEBUG
1014 static void dump_cook_context(COOKContext *q)
1015 {
1016 //int i=0;
1017 #define PRINT(a, b) ff_dlog(q->avctx, " %s = %d\n", a, b);
1018 ff_dlog(q->avctx, "COOKextradata\n");
1019 ff_dlog(q->avctx, "cookversion=%x\n", q->subpacket[0].cookversion);
1020 if (q->subpacket[0].cookversion > STEREO) {
1021 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1022 PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1023 }
1024 ff_dlog(q->avctx, "COOKContext\n");
1025 PRINT("nb_channels", q->avctx->channels);
1026 PRINT("bit_rate", q->avctx->bit_rate);
1027 PRINT("sample_rate", q->avctx->sample_rate);
1028 PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1029 PRINT("subbands", q->subpacket[0].subbands);
1030 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1031 PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1032 PRINT("numvector_size", q->subpacket[0].numvector_size);
1033 PRINT("total_subbands", q->subpacket[0].total_subbands);
1034 }
1035 #endif
1036
1037 /**
1038 * Cook initialization
1039 *
1040 * @param avctx pointer to the AVCodecContext
1041 */
1042 static av_cold int cook_decode_init(AVCodecContext *avctx)
1043 {
1044 COOKContext *q = avctx->priv_data;
1045 GetByteContext gb;
1046 int s = 0;
1047 unsigned int channel_mask = 0;
1048 int samples_per_frame;
1049 int ret;
1050 q->avctx = avctx;
1051
1052 /* Take care of the codec specific extradata. */
1053 if (avctx->extradata_size < 8) {
1054 av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1055 return AVERROR_INVALIDDATA;
1056 }
1057 av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1058
1059 bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
1060
1061 /* Take data from the AVCodecContext (RM container). */
1062 if (!avctx->channels) {
1063 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1064 return AVERROR_INVALIDDATA;
1065 }
1066
1067 /* Initialize RNG. */
1068 av_lfg_init(&q->random_state, 0);
1069
1070 ff_audiodsp_init(&q->adsp);
1071
1072 while (bytestream2_get_bytes_left(&gb)) {
1073 /* 8 for mono, 16 for stereo, ? for multichannel
1074 Swap to right endianness so we don't need to care later on. */
1075 q->subpacket[s].cookversion = bytestream2_get_be32(&gb);
1076 samples_per_frame = bytestream2_get_be16(&gb);
1077 q->subpacket[s].subbands = bytestream2_get_be16(&gb);
1078 bytestream2_get_be32(&gb); // Unknown unused
1079 q->subpacket[s].js_subband_start = bytestream2_get_be16(&gb);
1080 q->subpacket[s].js_vlc_bits = bytestream2_get_be16(&gb);
1081
1082 /* Initialize extradata related variables. */
1083 q->subpacket[s].samples_per_channel = samples_per_frame / avctx->channels;
1084 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1085
1086 /* Initialize default data states. */
1087 q->subpacket[s].log2_numvector_size = 5;
1088 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1089 q->subpacket[s].num_channels = 1;
1090
1091 /* Initialize version-dependent variables */
1092
1093 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1094 q->subpacket[s].cookversion);
1095 q->subpacket[s].joint_stereo = 0;
1096 switch (q->subpacket[s].cookversion) {
1097 case MONO:
1098 if (avctx->channels != 1) {
1099 avpriv_request_sample(avctx, "Container channels != 1");
1100 return AVERROR_PATCHWELCOME;
1101 }
1102 av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1103 break;
1104 case STEREO:
1105 if (avctx->channels != 1) {
1106 q->subpacket[s].bits_per_subpdiv = 1;
1107 q->subpacket[s].num_channels = 2;
1108 }
1109 av_log(avctx, AV_LOG_DEBUG, "STEREO\n");
1110 break;
1111 case JOINT_STEREO:
1112 if (avctx->channels != 2) {
1113 avpriv_request_sample(avctx, "Container channels != 2");
1114 return AVERROR_PATCHWELCOME;
1115 }
1116 av_log(avctx, AV_LOG_DEBUG, "JOINT_STEREO\n");
1117 if (avctx->extradata_size >= 16) {
1118 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1119 q->subpacket[s].js_subband_start;
1120 q->subpacket[s].joint_stereo = 1;
1121 q->subpacket[s].num_channels = 2;
1122 }
1123 if (q->subpacket[s].samples_per_channel > 256) {
1124 q->subpacket[s].log2_numvector_size = 6;
1125 }
1126 if (q->subpacket[s].samples_per_channel > 512) {
1127 q->subpacket[s].log2_numvector_size = 7;
1128 }
1129 break;
1130 case MC_COOK:
1131 av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1132 channel_mask |= q->subpacket[s].channel_mask = bytestream2_get_be32(&gb);
1133
1134 if (av_get_channel_layout_nb_channels(q->subpacket[s].channel_mask) > 1) {
1135 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1136 q->subpacket[s].js_subband_start;
1137 q->subpacket[s].joint_stereo = 1;
1138 q->subpacket[s].num_channels = 2;
1139 q->subpacket[s].samples_per_channel = samples_per_frame >> 1;
1140
1141 if (q->subpacket[s].samples_per_channel > 256) {
1142 q->subpacket[s].log2_numvector_size = 6;
1143 }
1144 if (q->subpacket[s].samples_per_channel > 512) {
1145 q->subpacket[s].log2_numvector_size = 7;
1146 }
1147 } else
1148 q->subpacket[s].samples_per_channel = samples_per_frame;
1149
1150 break;
1151 default:
1152 avpriv_request_sample(avctx, "Cook version %d",
1153 q->subpacket[s].cookversion);
1154 return AVERROR_PATCHWELCOME;
1155 }
1156
1157 if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1158 av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1159 return AVERROR_INVALIDDATA;
1160 } else
1161 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1162
1163
1164 /* Initialize variable relations */
1165 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1166
1167 /* Try to catch some obviously faulty streams, otherwise it might be exploitable */
1168 if (q->subpacket[s].total_subbands > 53) {
1169 avpriv_request_sample(avctx, "total_subbands > 53");
1170 return AVERROR_PATCHWELCOME;
1171 }
1172
1173 if ((q->subpacket[s].js_vlc_bits > 6) ||
1174 (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1175 av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1176 q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1177 return AVERROR_INVALIDDATA;
1178 }
1179
1180 if (q->subpacket[s].subbands > 50) {
1181 avpriv_request_sample(avctx, "subbands > 50");
1182 return AVERROR_PATCHWELCOME;
1183 }
1184 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1185 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1186 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1187 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1188
1189 q->num_subpackets++;
1190 s++;
1191 if (s > MAX_SUBPACKETS) {
1192 avpriv_request_sample(avctx, "subpackets > %d", MAX_SUBPACKETS);
1193 return AVERROR_PATCHWELCOME;
1194 }
1195 }
1196 /* Generate tables */
1197 init_pow2table();
1198 init_gain_table(q);
1199 init_cplscales_table(q);
1200
1201 if ((ret = init_cook_vlc_tables(q)))
1202 return ret;
1203
1204
1205 if (avctx->block_align >= UINT_MAX / 2)
1206 return AVERROR(EINVAL);
1207
1208 /* Pad the databuffer with:
1209 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1210 AV_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1211 q->decoded_bytes_buffer =
1212 av_mallocz(avctx->block_align
1213 + DECODE_BYTES_PAD1(avctx->block_align)
1214 + AV_INPUT_BUFFER_PADDING_SIZE);
1215 if (!q->decoded_bytes_buffer)
1216 return AVERROR(ENOMEM);
1217
1218 /* Initialize transform. */
1219 if ((ret = init_cook_mlt(q)))
1220 return ret;
1221
1222 /* Initialize COOK signal arithmetic handling */
1223 if (1) {
1224 q->scalar_dequant = scalar_dequant_float;
1225 q->decouple = decouple_float;
1226 q->imlt_window = imlt_window_float;
1227 q->interpolate = interpolate_float;
1228 q->saturate_output = saturate_output_float;
1229 }
1230
1231 /* Try to catch some obviously faulty streams, otherwise it might be exploitable */
1232 if (q->samples_per_channel != 256 && q->samples_per_channel != 512 &&
1233 q->samples_per_channel != 1024) {
1234 avpriv_request_sample(avctx, "samples_per_channel = %d",
1235 q->samples_per_channel);
1236 return AVERROR_PATCHWELCOME;
1237 }
1238
1239 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1240 if (channel_mask)
1241 avctx->channel_layout = channel_mask;
1242 else
1243 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1244
1245 #ifdef DEBUG
1246 dump_cook_context(q);
1247 #endif
1248 return 0;
1249 }
1250
1251 AVCodec ff_cook_decoder = {
1252 .name = "cook",
1253 .long_name = NULL_IF_CONFIG_SMALL("Cook / Cooker / Gecko (RealAudio G2)"),
1254 .type = AVMEDIA_TYPE_AUDIO,
1255 .id = AV_CODEC_ID_COOK,
1256 .priv_data_size = sizeof(COOKContext),
1257 .init = cook_decode_init,
1258 .close = cook_decode_close,
1259 .decode = cook_decode_frame,
1260 .capabilities = AV_CODEC_CAP_DR1,
1261 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1262 AV_SAMPLE_FMT_NONE },
1263 };