2 * COOK compatible decoder
3 * Copyright (c) 2003 Sascha Sommer
4 * Copyright (c) 2005 Benjamin Larsson
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
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.
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).
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
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
45 #include "libavutil/lfg.h"
49 #include "bytestream.h"
51 #include "libavutil/audioconvert.h"
56 /* the different Cook versions */
57 #define MONO 0x1000001
58 #define STEREO 0x1000002
59 #define JOINT_STEREO 0x1000003
60 #define MC_COOK 0x2000000 //multichannel Cook, not supported
62 #define SUBBAND_SIZE 20
63 #define MAX_SUBPACKETS 5
75 int samples_per_frame;
79 int samples_per_channel;
80 int log2_numvector_size;
81 unsigned int channel_mask;
82 VLC ccpl; ///< channel coupling
84 int bits_per_subpacket;
87 int numvector_size; ///< 1 << log2_numvector_size;
89 float mono_previous_buffer1[1024];
90 float mono_previous_buffer2[1024];
100 typedef struct cook {
102 * The following 5 functions provide the lowlevel arithmetic on
103 * the internal audio buffers.
105 void (* scalar_dequant)(struct cook *q, int index, int quant_index,
106 int* subband_coef_index, int* subband_coef_sign,
109 void (* decouple) (struct cook *q,
113 float *decode_buffer,
114 float *mlt_buffer1, float *mlt_buffer2);
116 void (* imlt_window) (struct cook *q, float *buffer1,
117 cook_gains *gains_ptr, float *previous_buffer);
119 void (* interpolate) (struct cook *q, float* buffer,
120 int gain_index, int gain_index_next);
122 void (* saturate_output) (struct cook *q, int chan, float *out);
124 AVCodecContext* avctx;
132 int samples_per_channel;
135 int discarded_packets;
142 VLC envelope_quant_index[13];
143 VLC sqvh[7]; //scalar quantization
145 /* generatable tables and related variables */
146 int gain_size_factor;
147 float gain_table[23];
151 uint8_t* decoded_bytes_buffer;
152 DECLARE_ALIGNED(32, float, mono_mdct_output)[2048];
153 float decode_buffer_1[1024];
154 float decode_buffer_2[1024];
155 float decode_buffer_0[1060]; /* static allocation for joint decode */
157 const float *cplscales[5];
159 COOKSubpacket subpacket[MAX_SUBPACKETS];
162 static float pow2tab[127];
163 static float rootpow2tab[127];
165 /*************** init functions ***************/
167 /* table generator */
168 static av_cold void init_pow2table(void){
170 for (i=-63 ; i<64 ; i++){
171 pow2tab[63+i]= pow(2, i);
172 rootpow2tab[63+i]=sqrt(pow(2, i));
176 /* table generator */
177 static av_cold void init_gain_table(COOKContext *q) {
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));
187 static av_cold int init_cook_vlc_tables(COOKContext *q) {
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);
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);
203 for(i=0;i<q->num_subpackets;i++){
204 if (q->subpacket[i].joint_stereo==1){
205 result |= init_vlc (&q->subpacket[i].ccpl, 6, (1<<q->subpacket[i].js_vlc_bits)-1,
206 ccpl_huffbits[q->subpacket[i].js_vlc_bits-2], 1, 1,
207 ccpl_huffcodes[q->subpacket[i].js_vlc_bits-2], 2, 2, 0);
208 av_log(q->avctx,AV_LOG_DEBUG,"subpacket %i Joint-stereo VLC used.\n",i);
212 av_log(q->avctx,AV_LOG_DEBUG,"VLC tables initialized.\n");
216 static av_cold int init_cook_mlt(COOKContext *q) {
218 int mlt_size = q->samples_per_channel;
220 if ((q->mlt_window = av_malloc(mlt_size * sizeof(*q->mlt_window))) == 0)
221 return AVERROR(ENOMEM);
223 /* Initialize the MLT window: simple sine window. */
224 ff_sine_window_init(q->mlt_window, mlt_size);
225 for(j=0 ; j<mlt_size ; j++)
226 q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
228 /* Initialize the MDCT. */
229 if ((ret = ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size)+1, 1, 1.0/32768.0))) {
230 av_free(q->mlt_window);
233 av_log(q->avctx,AV_LOG_DEBUG,"MDCT initialized, order = %d.\n",
234 av_log2(mlt_size)+1);
239 static const float *maybe_reformat_buffer32 (COOKContext *q, const float *ptr, int n)
245 static av_cold void init_cplscales_table (COOKContext *q) {
248 q->cplscales[i] = maybe_reformat_buffer32 (q, cplscales[i], (1<<(i+2))-1);
251 /*************** init functions end ***********/
253 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes)+3) % 4)
254 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
257 * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
258 * Why? No idea, some checksum/error detection method maybe.
260 * Out buffer size: extra bytes are needed to cope with
261 * padding/misalignment.
262 * Subpackets passed to the decoder can contain two, consecutive
263 * half-subpackets, of identical but arbitrary size.
264 * 1234 1234 1234 1234 extraA extraB
265 * Case 1: AAAA BBBB 0 0
266 * Case 2: AAAA ABBB BB-- 3 3
267 * Case 3: AAAA AABB BBBB 2 2
268 * Case 4: AAAA AAAB BBBB BB-- 1 5
270 * Nice way to waste CPU cycles.
272 * @param inbuffer pointer to byte array of indata
273 * @param out pointer to byte array of outdata
274 * @param bytes number of bytes
277 static inline int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
278 static const uint32_t tab[4] = {
279 AV_BE2NE32C(0x37c511f2), AV_BE2NE32C(0xf237c511),
280 AV_BE2NE32C(0x11f237c5), AV_BE2NE32C(0xc511f237),
285 uint32_t* obuf = (uint32_t*) out;
286 /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
287 * I'm too lazy though, should be something like
288 * for(i=0 ; i<bitamount/64 ; i++)
289 * (int64_t)out[i] = 0x37c511f237c511f2^av_be2ne64(int64_t)in[i]);
290 * Buffer alignment needs to be checked. */
292 off = (intptr_t)inbuffer & 3;
293 buf = (const uint32_t*) (inbuffer - off);
296 for (i = 0; i < bytes/4; i++)
297 obuf[i] = c ^ buf[i];
306 static av_cold int cook_decode_close(AVCodecContext *avctx)
309 COOKContext *q = avctx->priv_data;
310 av_log(avctx,AV_LOG_DEBUG, "Deallocating memory.\n");
312 /* Free allocated memory buffers. */
313 av_free(q->mlt_window);
314 av_free(q->decoded_bytes_buffer);
316 /* Free the transform. */
317 ff_mdct_end(&q->mdct_ctx);
319 /* Free the VLC tables. */
320 for (i=0 ; i<13 ; i++) {
321 free_vlc(&q->envelope_quant_index[i]);
323 for (i=0 ; i<7 ; i++) {
324 free_vlc(&q->sqvh[i]);
326 for (i=0 ; i<q->num_subpackets ; i++) {
327 free_vlc(&q->subpacket[i].ccpl);
330 av_log(avctx,AV_LOG_DEBUG,"Memory deallocated.\n");
336 * Fill the gain array for the timedomain quantization.
338 * @param gb pointer to the GetBitContext
339 * @param gaininfo array[9] of gain indexes
342 static void decode_gain_info(GetBitContext *gb, int *gaininfo)
346 while (get_bits1(gb)) {}
347 n = get_bits_count(gb) - 1; //amount of elements*2 to update
351 int index = get_bits(gb, 3);
352 int gain = get_bits1(gb) ? get_bits(gb, 4) - 7 : -1;
354 while (i <= index) gaininfo[i++] = gain;
356 while (i <= 8) gaininfo[i++] = 0;
360 * Create the quant index table needed for the envelope.
362 * @param q pointer to the COOKContext
363 * @param quant_index_table pointer to the array
366 static void decode_envelope(COOKContext *q, COOKSubpacket *p, int* quant_index_table) {
369 quant_index_table[0]= get_bits(&q->gb,6) - 6; //This is used later in categorize
371 for (i=1 ; i < p->total_subbands ; i++){
373 if (i >= p->js_subband_start * 2) {
374 vlc_index-=p->js_subband_start;
377 if(vlc_index < 1) vlc_index = 1;
379 if (vlc_index>13) vlc_index = 13; //the VLC tables >13 are identical to No. 13
381 j = get_vlc2(&q->gb, q->envelope_quant_index[vlc_index-1].table,
382 q->envelope_quant_index[vlc_index-1].bits,2);
383 quant_index_table[i] = quant_index_table[i-1] + j - 12; //differential encoding
388 * Calculate the category and category_index vector.
390 * @param q pointer to the COOKContext
391 * @param quant_index_table pointer to the array
392 * @param category pointer to the category array
393 * @param category_index pointer to the category_index array
396 static void categorize(COOKContext *q, COOKSubpacket *p, int* quant_index_table,
397 int* category, int* category_index){
398 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
402 int tmp_categorize_array[128*2];
403 int tmp_categorize_array1_idx=p->numvector_size;
404 int tmp_categorize_array2_idx=p->numvector_size;
406 bits_left = p->bits_per_subpacket - get_bits_count(&q->gb);
408 if(bits_left > q->samples_per_channel) {
409 bits_left = q->samples_per_channel +
410 ((bits_left - q->samples_per_channel)*5)/8;
411 //av_log(q->avctx, AV_LOG_ERROR, "bits_left = %d\n",bits_left);
414 memset(&exp_index1, 0, sizeof(exp_index1));
415 memset(&exp_index2, 0, sizeof(exp_index2));
416 memset(&tmp_categorize_array, 0, sizeof(tmp_categorize_array));
421 for (i=32 ; i>0 ; i=i/2){
424 for (j=p->total_subbands ; j>0 ; j--){
425 exp_idx = av_clip((i - quant_index_table[index] + bias) / 2, 0, 7);
427 num_bits+=expbits_tab[exp_idx];
429 if(num_bits >= bits_left - 32){
434 /* Calculate total number of bits. */
436 for (i=0 ; i<p->total_subbands ; i++) {
437 exp_idx = av_clip((bias - quant_index_table[i]) / 2, 0, 7);
438 num_bits += expbits_tab[exp_idx];
439 exp_index1[i] = exp_idx;
440 exp_index2[i] = exp_idx;
442 tmpbias1 = tmpbias2 = num_bits;
444 for (j = 1 ; j < p->numvector_size ; j++) {
445 if (tmpbias1 + tmpbias2 > 2*bits_left) { /* ---> */
448 for (i=0 ; i<p->total_subbands ; i++){
449 if (exp_index1[i] < 7) {
450 v = (-2*exp_index1[i]) - quant_index_table[i] + bias;
458 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
459 tmpbias1 -= expbits_tab[exp_index1[index]] -
460 expbits_tab[exp_index1[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;
474 if(index == -1)break;
475 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
476 tmpbias2 -= expbits_tab[exp_index2[index]] -
477 expbits_tab[exp_index2[index]-1];
482 for(i=0 ; i<p->total_subbands ; i++)
483 category[i] = exp_index2[i];
485 for(i=0 ; i<p->numvector_size-1 ; i++)
486 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
492 * Expand the category vector.
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
499 static inline void expand_category(COOKContext *q, int* category,
500 int* category_index){
502 for(i=0 ; i<q->num_vectors ; i++){
503 ++category[category_index[i]];
508 * The real requantization of the mltcoefs
510 * @param q pointer to the COOKContext
512 * @param quant_index quantisation index
513 * @param subband_coef_index array of indexes to quant_centroid_tab
514 * @param subband_coef_sign signs of coefficients
515 * @param mlt_p pointer into the mlt buffer
518 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
519 int* subband_coef_index, int* subband_coef_sign,
524 for(i=0 ; i<SUBBAND_SIZE ; i++) {
525 if (subband_coef_index[i]) {
526 f1 = quant_centroid_tab[index][subband_coef_index[i]];
527 if (subband_coef_sign[i]) f1 = -f1;
529 /* noise coding if subband_coef_index[i] == 0 */
530 f1 = dither_tab[index];
531 if (av_lfg_get(&q->random_state) < 0x80000000) f1 = -f1;
533 mlt_p[i] = f1 * rootpow2tab[quant_index+63];
537 * Unpack the subband_coef_index and subband_coef_sign vectors.
539 * @param q pointer to the COOKContext
540 * @param category pointer to the category array
541 * @param subband_coef_index array of indexes to quant_centroid_tab
542 * @param subband_coef_sign signs of coefficients
545 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category, int* subband_coef_index,
546 int* subband_coef_sign) {
548 int vlc, vd ,tmp, result;
550 vd = vd_tab[category];
552 for(i=0 ; i<vpr_tab[category] ; i++){
553 vlc = get_vlc2(&q->gb, q->sqvh[category].table, q->sqvh[category].bits, 3);
554 if (p->bits_per_subpacket < get_bits_count(&q->gb)){
558 for(j=vd-1 ; j>=0 ; j--){
559 tmp = (vlc * invradix_tab[category])/0x100000;
560 subband_coef_index[vd*i+j] = vlc - tmp * (kmax_tab[category]+1);
563 for(j=0 ; j<vd ; j++){
564 if (subband_coef_index[i*vd + j]) {
565 if(get_bits_count(&q->gb) < p->bits_per_subpacket){
566 subband_coef_sign[i*vd+j] = get_bits1(&q->gb);
569 subband_coef_sign[i*vd+j]=0;
572 subband_coef_sign[i*vd+j]=0;
581 * Fill the mlt_buffer with mlt coefficients.
583 * @param q pointer to the COOKContext
584 * @param category pointer to the category array
585 * @param quant_index_table pointer to the array
586 * @param mlt_buffer pointer to mlt coefficients
590 static void decode_vectors(COOKContext* q, COOKSubpacket* p, int* category,
591 int *quant_index_table, float* mlt_buffer){
592 /* A zero in this table means that the subband coefficient is
593 random noise coded. */
594 int subband_coef_index[SUBBAND_SIZE];
595 /* A zero in this table means that the subband coefficient is a
596 positive multiplicator. */
597 int subband_coef_sign[SUBBAND_SIZE];
601 for(band=0 ; band<p->total_subbands ; band++){
602 index = category[band];
603 if(category[band] < 7){
604 if(unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)){
606 for(j=0 ; j<p->total_subbands ; j++) category[band+j]=7;
610 memset(subband_coef_index, 0, sizeof(subband_coef_index));
611 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
613 q->scalar_dequant(q, index, quant_index_table[band],
614 subband_coef_index, subband_coef_sign,
615 &mlt_buffer[band * SUBBAND_SIZE]);
618 if(p->total_subbands*SUBBAND_SIZE >= q->samples_per_channel){
620 } /* FIXME: should this be removed, or moved into loop above? */
625 * function for decoding mono data
627 * @param q pointer to the COOKContext
628 * @param mlt_buffer pointer to mlt coefficients
631 static void mono_decode(COOKContext *q, COOKSubpacket *p, float* mlt_buffer) {
633 int category_index[128];
634 int quant_index_table[102];
637 memset(&category, 0, sizeof(category));
638 memset(&category_index, 0, sizeof(category_index));
640 decode_envelope(q, p, quant_index_table);
641 q->num_vectors = get_bits(&q->gb,p->log2_numvector_size);
642 categorize(q, p, quant_index_table, category, category_index);
643 expand_category(q, category, category_index);
644 decode_vectors(q, p, category, quant_index_table, mlt_buffer);
649 * the actual requantization of the timedomain samples
651 * @param q pointer to the COOKContext
652 * @param buffer pointer to the timedomain buffer
653 * @param gain_index index for the block multiplier
654 * @param gain_index_next index for the next block multiplier
657 static void interpolate_float(COOKContext *q, float* buffer,
658 int gain_index, int gain_index_next){
661 fc1 = pow2tab[gain_index+63];
663 if(gain_index == gain_index_next){ //static gain
664 for(i=0 ; i<q->gain_size_factor ; i++){
667 } else { //smooth gain
668 fc2 = q->gain_table[11 + (gain_index_next-gain_index)];
669 for(i=0 ; i<q->gain_size_factor ; i++){
677 * Apply transform window, overlap buffers.
679 * @param q pointer to the COOKContext
680 * @param inbuffer pointer to the mltcoefficients
681 * @param gains_ptr current and previous gains
682 * @param previous_buffer pointer to the previous buffer to be used for overlapping
685 static void imlt_window_float (COOKContext *q, float *inbuffer,
686 cook_gains *gains_ptr, float *previous_buffer)
688 const float fc = pow2tab[gains_ptr->previous[0] + 63];
690 /* The weird thing here, is that the two halves of the time domain
691 * buffer are swapped. Also, the newest data, that we save away for
692 * next frame, has the wrong sign. Hence the subtraction below.
693 * Almost sounds like a complex conjugate/reverse data/FFT effect.
696 /* Apply window and overlap */
697 for(i = 0; i < q->samples_per_channel; i++){
698 inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
699 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
704 * The modulated lapped transform, this takes transform coefficients
705 * and transforms them into timedomain samples.
706 * Apply transform window, overlap buffers, apply gain profile
707 * and buffer management.
709 * @param q pointer to the COOKContext
710 * @param inbuffer pointer to the mltcoefficients
711 * @param gains_ptr current and previous gains
712 * @param previous_buffer pointer to the previous buffer to be used for overlapping
715 static void imlt_gain(COOKContext *q, float *inbuffer,
716 cook_gains *gains_ptr, float* previous_buffer)
718 float *buffer0 = q->mono_mdct_output;
719 float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
722 /* Inverse modified discrete cosine transform */
723 q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
725 q->imlt_window (q, buffer1, gains_ptr, previous_buffer);
727 /* Apply gain profile */
728 for (i = 0; i < 8; i++) {
729 if (gains_ptr->now[i] || gains_ptr->now[i + 1])
730 q->interpolate(q, &buffer1[q->gain_size_factor * i],
731 gains_ptr->now[i], gains_ptr->now[i + 1]);
734 /* Save away the current to be previous block. */
735 memcpy(previous_buffer, buffer0,
736 q->samples_per_channel * sizeof(*previous_buffer));
741 * function for getting the jointstereo coupling information
743 * @param q pointer to the COOKContext
744 * @param decouple_tab decoupling array
747 static void decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
750 int vlc = get_bits1(&q->gb);
751 int start = cplband[p->js_subband_start];
752 int end = cplband[p->subbands-1];
753 int length = end - start + 1;
759 for (i = 0; i < length; i++)
760 decouple_tab[start + i] = get_vlc2(&q->gb, p->ccpl.table, p->ccpl.bits, 2);
762 for (i = 0; i < length; i++)
763 decouple_tab[start + i] = get_bits(&q->gb, p->js_vlc_bits);
768 * function decouples a pair of signals from a single signal via multiplication.
770 * @param q pointer to the COOKContext
771 * @param subband index of the current subband
772 * @param f1 multiplier for channel 1 extraction
773 * @param f2 multiplier for channel 2 extraction
774 * @param decode_buffer input buffer
775 * @param mlt_buffer1 pointer to left channel mlt coefficients
776 * @param mlt_buffer2 pointer to right channel mlt coefficients
778 static void decouple_float (COOKContext *q,
782 float *decode_buffer,
783 float *mlt_buffer1, float *mlt_buffer2)
786 for (j=0 ; j<SUBBAND_SIZE ; j++) {
787 tmp_idx = ((p->js_subband_start + subband)*SUBBAND_SIZE)+j;
788 mlt_buffer1[SUBBAND_SIZE*subband + j] = f1 * decode_buffer[tmp_idx];
789 mlt_buffer2[SUBBAND_SIZE*subband + j] = f2 * decode_buffer[tmp_idx];
794 * function for decoding joint stereo data
796 * @param q pointer to the COOKContext
797 * @param mlt_buffer1 pointer to left channel mlt coefficients
798 * @param mlt_buffer2 pointer to right channel mlt coefficients
801 static void joint_decode(COOKContext *q, COOKSubpacket *p, float* mlt_buffer1,
802 float* mlt_buffer2) {
804 int decouple_tab[SUBBAND_SIZE];
805 float *decode_buffer = q->decode_buffer_0;
808 const float* cplscale;
810 memset(decouple_tab, 0, sizeof(decouple_tab));
811 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
813 /* Make sure the buffers are zeroed out. */
814 memset(mlt_buffer1, 0, 1024 * sizeof(*mlt_buffer1));
815 memset(mlt_buffer2, 0, 1024 * sizeof(*mlt_buffer2));
816 decouple_info(q, p, decouple_tab);
817 mono_decode(q, p, decode_buffer);
819 /* The two channels are stored interleaved in decode_buffer. */
820 for (i=0 ; i<p->js_subband_start ; i++) {
821 for (j=0 ; j<SUBBAND_SIZE ; j++) {
822 mlt_buffer1[i*20+j] = decode_buffer[i*40+j];
823 mlt_buffer2[i*20+j] = decode_buffer[i*40+20+j];
827 /* When we reach js_subband_start (the higher frequencies)
828 the coefficients are stored in a coupling scheme. */
829 idx = (1 << p->js_vlc_bits) - 1;
830 for (i=p->js_subband_start ; i<p->subbands ; i++) {
831 cpl_tmp = cplband[i];
832 idx -=decouple_tab[cpl_tmp];
833 cplscale = q->cplscales[p->js_vlc_bits-2]; //choose decoupler table
834 f1 = cplscale[decouple_tab[cpl_tmp]];
835 f2 = cplscale[idx-1];
836 q->decouple (q, p, i, f1, f2, decode_buffer, mlt_buffer1, mlt_buffer2);
837 idx = (1 << p->js_vlc_bits) - 1;
842 * First part of subpacket decoding:
843 * decode raw stream bytes and read gain info.
845 * @param q pointer to the COOKContext
846 * @param inbuffer pointer to raw stream data
847 * @param gains_ptr array of current/prev gain pointers
851 decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p, const uint8_t *inbuffer,
852 cook_gains *gains_ptr)
856 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
857 p->bits_per_subpacket/8);
858 init_get_bits(&q->gb, q->decoded_bytes_buffer + offset,
859 p->bits_per_subpacket);
860 decode_gain_info(&q->gb, gains_ptr->now);
862 /* Swap current and previous gains */
863 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
867 * Saturate the output signal and interleave.
869 * @param q pointer to the COOKContext
870 * @param chan channel to saturate
871 * @param out pointer to the output vector
873 static void saturate_output_float(COOKContext *q, int chan, float *out)
876 float *output = q->mono_mdct_output + q->samples_per_channel;
877 for (j = 0; j < q->samples_per_channel; j++) {
878 out[chan + q->nb_channels * j] = av_clipf(output[j], -1.0, 1.0);
883 * Final part of subpacket decoding:
884 * Apply modulated lapped transform, gain compensation,
885 * clip and convert to integer.
887 * @param q pointer to the COOKContext
888 * @param decode_buffer pointer to the mlt coefficients
889 * @param gains_ptr array of current/prev gain pointers
890 * @param previous_buffer pointer to the previous buffer to be used for overlapping
891 * @param out pointer to the output buffer
892 * @param chan 0: left or single channel, 1: right channel
896 mlt_compensate_output(COOKContext *q, float *decode_buffer,
897 cook_gains *gains_ptr, float *previous_buffer,
898 float *out, int chan)
900 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
902 q->saturate_output(q, chan, out);
907 * Cook subpacket decoding. This function returns one decoded subpacket,
908 * usually 1024 samples per channel.
910 * @param q pointer to the COOKContext
911 * @param inbuffer pointer to the inbuffer
912 * @param outbuffer pointer to the outbuffer
914 static void decode_subpacket(COOKContext *q, COOKSubpacket *p,
915 const uint8_t *inbuffer, float *outbuffer)
917 int sub_packet_size = p->size;
919 // for (i=0 ; i<sub_packet_size ; i++) {
920 // av_log(q->avctx, AV_LOG_ERROR, "%02x", inbuffer[i]);
922 // av_log(q->avctx, AV_LOG_ERROR, "\n");
923 memset(q->decode_buffer_1,0,sizeof(q->decode_buffer_1));
924 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
926 if (p->joint_stereo) {
927 joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2);
929 mono_decode(q, p, q->decode_buffer_1);
931 if (p->num_channels == 2) {
932 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size/2, &p->gains2);
933 mono_decode(q, p, q->decode_buffer_2);
937 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
938 p->mono_previous_buffer1, outbuffer, p->ch_idx);
940 if (p->num_channels == 2) {
941 if (p->joint_stereo) {
942 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
943 p->mono_previous_buffer2, outbuffer, p->ch_idx + 1);
945 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
946 p->mono_previous_buffer2, outbuffer, p->ch_idx + 1);
954 * Cook frame decoding
956 * @param avctx pointer to the AVCodecContext
959 static int cook_decode_frame(AVCodecContext *avctx, void *data,
960 int *got_frame_ptr, AVPacket *avpkt)
962 const uint8_t *buf = avpkt->data;
963 int buf_size = avpkt->size;
964 COOKContext *q = avctx->priv_data;
965 float *samples = NULL;
970 if (buf_size < avctx->block_align)
973 /* get output buffer */
974 if (q->discarded_packets >= 2) {
975 q->frame.nb_samples = q->samples_per_channel;
976 if ((ret = avctx->get_buffer(avctx, &q->frame)) < 0) {
977 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
980 samples = (float *)q->frame.data[0];
983 /* estimate subpacket sizes */
984 q->subpacket[0].size = avctx->block_align;
986 for(i=1;i<q->num_subpackets;i++){
987 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
988 q->subpacket[0].size -= q->subpacket[i].size + 1;
989 if (q->subpacket[0].size < 0) {
990 av_log(avctx,AV_LOG_DEBUG,"frame subpacket size total > avctx->block_align!\n");
991 return AVERROR_INVALIDDATA;
995 /* decode supbackets */
996 for(i=0;i<q->num_subpackets;i++){
997 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size*8)>>q->subpacket[i].bits_per_subpdiv;
998 q->subpacket[i].ch_idx = chidx;
999 av_log(avctx,AV_LOG_DEBUG,"subpacket[%i] size %i js %i %i block_align %i\n",i,q->subpacket[i].size,q->subpacket[i].joint_stereo,offset,avctx->block_align);
1000 decode_subpacket(q, &q->subpacket[i], buf + offset, samples);
1001 offset += q->subpacket[i].size;
1002 chidx += q->subpacket[i].num_channels;
1003 av_log(avctx,AV_LOG_DEBUG,"subpacket[%i] %i %i\n",i,q->subpacket[i].size * 8,get_bits_count(&q->gb));
1006 /* Discard the first two frames: no valid audio. */
1007 if (q->discarded_packets < 2) {
1008 q->discarded_packets++;
1010 return avctx->block_align;
1014 *(AVFrame *)data = q->frame;
1016 return avctx->block_align;
1020 static void dump_cook_context(COOKContext *q)
1023 #define PRINT(a,b) av_log(q->avctx,AV_LOG_ERROR," %s = %d\n", a, b);
1024 av_log(q->avctx,AV_LOG_ERROR,"COOKextradata\n");
1025 av_log(q->avctx,AV_LOG_ERROR,"cookversion=%x\n",q->subpacket[0].cookversion);
1026 if (q->subpacket[0].cookversion > STEREO) {
1027 PRINT("js_subband_start",q->subpacket[0].js_subband_start);
1028 PRINT("js_vlc_bits",q->subpacket[0].js_vlc_bits);
1030 av_log(q->avctx,AV_LOG_ERROR,"COOKContext\n");
1031 PRINT("nb_channels",q->nb_channels);
1032 PRINT("bit_rate",q->bit_rate);
1033 PRINT("sample_rate",q->sample_rate);
1034 PRINT("samples_per_channel",q->subpacket[0].samples_per_channel);
1035 PRINT("samples_per_frame",q->subpacket[0].samples_per_frame);
1036 PRINT("subbands",q->subpacket[0].subbands);
1037 PRINT("js_subband_start",q->subpacket[0].js_subband_start);
1038 PRINT("log2_numvector_size",q->subpacket[0].log2_numvector_size);
1039 PRINT("numvector_size",q->subpacket[0].numvector_size);
1040 PRINT("total_subbands",q->subpacket[0].total_subbands);
1044 static av_cold int cook_count_channels(unsigned int mask){
1047 for(i = 0;i<32;i++){
1055 * Cook initialization
1057 * @param avctx pointer to the AVCodecContext
1060 static av_cold int cook_decode_init(AVCodecContext *avctx)
1062 COOKContext *q = avctx->priv_data;
1063 const uint8_t *edata_ptr = avctx->extradata;
1064 const uint8_t *edata_ptr_end = edata_ptr + avctx->extradata_size;
1065 int extradata_size = avctx->extradata_size;
1067 unsigned int channel_mask = 0;
1071 /* Take care of the codec specific extradata. */
1072 if (extradata_size <= 0) {
1073 av_log(avctx,AV_LOG_ERROR,"Necessary extradata missing!\n");
1074 return AVERROR_INVALIDDATA;
1076 av_log(avctx,AV_LOG_DEBUG,"codecdata_length=%d\n",avctx->extradata_size);
1078 /* Take data from the AVCodecContext (RM container). */
1079 q->sample_rate = avctx->sample_rate;
1080 q->nb_channels = avctx->channels;
1081 q->bit_rate = avctx->bit_rate;
1083 /* Initialize RNG. */
1084 av_lfg_init(&q->random_state, 0);
1086 while(edata_ptr < edata_ptr_end){
1087 /* 8 for mono, 16 for stereo, ? for multichannel
1088 Swap to right endianness so we don't need to care later on. */
1089 if (extradata_size >= 8){
1090 q->subpacket[s].cookversion = bytestream_get_be32(&edata_ptr);
1091 q->subpacket[s].samples_per_frame = bytestream_get_be16(&edata_ptr);
1092 q->subpacket[s].subbands = bytestream_get_be16(&edata_ptr);
1093 extradata_size -= 8;
1095 if (extradata_size >= 8) {
1096 bytestream_get_be32(&edata_ptr); //Unknown unused
1097 q->subpacket[s].js_subband_start = bytestream_get_be16(&edata_ptr);
1098 q->subpacket[s].js_vlc_bits = bytestream_get_be16(&edata_ptr);
1099 extradata_size -= 8;
1102 /* Initialize extradata related variables. */
1103 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame / q->nb_channels;
1104 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1106 /* Initialize default data states. */
1107 q->subpacket[s].log2_numvector_size = 5;
1108 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1109 q->subpacket[s].num_channels = 1;
1111 /* Initialize version-dependent variables */
1113 av_log(avctx,AV_LOG_DEBUG,"subpacket[%i].cookversion=%x\n",s,q->subpacket[s].cookversion);
1114 q->subpacket[s].joint_stereo = 0;
1115 switch (q->subpacket[s].cookversion) {
1117 if (q->nb_channels != 1) {
1118 av_log_ask_for_sample(avctx, "Container channels != 1.\n");
1119 return AVERROR_PATCHWELCOME;
1121 av_log(avctx,AV_LOG_DEBUG,"MONO\n");
1124 if (q->nb_channels != 1) {
1125 q->subpacket[s].bits_per_subpdiv = 1;
1126 q->subpacket[s].num_channels = 2;
1128 av_log(avctx,AV_LOG_DEBUG,"STEREO\n");
1131 if (q->nb_channels != 2) {
1132 av_log_ask_for_sample(avctx, "Container channels != 2.\n");
1133 return AVERROR_PATCHWELCOME;
1135 av_log(avctx,AV_LOG_DEBUG,"JOINT_STEREO\n");
1136 if (avctx->extradata_size >= 16){
1137 q->subpacket[s].total_subbands = q->subpacket[s].subbands + q->subpacket[s].js_subband_start;
1138 q->subpacket[s].joint_stereo = 1;
1139 q->subpacket[s].num_channels = 2;
1141 if (q->subpacket[s].samples_per_channel > 256) {
1142 q->subpacket[s].log2_numvector_size = 6;
1144 if (q->subpacket[s].samples_per_channel > 512) {
1145 q->subpacket[s].log2_numvector_size = 7;
1149 av_log(avctx,AV_LOG_DEBUG,"MULTI_CHANNEL\n");
1150 if(extradata_size >= 4)
1151 channel_mask |= q->subpacket[s].channel_mask = bytestream_get_be32(&edata_ptr);
1153 if(cook_count_channels(q->subpacket[s].channel_mask) > 1){
1154 q->subpacket[s].total_subbands = q->subpacket[s].subbands + q->subpacket[s].js_subband_start;
1155 q->subpacket[s].joint_stereo = 1;
1156 q->subpacket[s].num_channels = 2;
1157 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame >> 1;
1159 if (q->subpacket[s].samples_per_channel > 256) {
1160 q->subpacket[s].log2_numvector_size = 6;
1162 if (q->subpacket[s].samples_per_channel > 512) {
1163 q->subpacket[s].log2_numvector_size = 7;
1166 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame;
1170 av_log_ask_for_sample(avctx, "Unknown Cook version.\n");
1171 return AVERROR_PATCHWELCOME;
1174 if(s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1175 av_log(avctx,AV_LOG_ERROR,"different number of samples per channel!\n");
1176 return AVERROR_INVALIDDATA;
1178 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1181 /* Initialize variable relations */
1182 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1184 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1185 if (q->subpacket[s].total_subbands > 53) {
1186 av_log_ask_for_sample(avctx, "total_subbands > 53\n");
1187 return AVERROR_PATCHWELCOME;
1190 if ((q->subpacket[s].js_vlc_bits > 6) || (q->subpacket[s].js_vlc_bits < 2*q->subpacket[s].joint_stereo)) {
1191 av_log(avctx,AV_LOG_ERROR,"js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1192 q->subpacket[s].js_vlc_bits, 2*q->subpacket[s].joint_stereo);
1193 return AVERROR_INVALIDDATA;
1196 if (q->subpacket[s].subbands > 50) {
1197 av_log_ask_for_sample(avctx, "subbands > 50\n");
1198 return AVERROR_PATCHWELCOME;
1200 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1201 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1202 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1203 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1205 q->num_subpackets++;
1207 if (s > MAX_SUBPACKETS) {
1208 av_log_ask_for_sample(avctx, "Too many subpackets > 5\n");
1209 return AVERROR_PATCHWELCOME;
1212 /* Generate tables */
1215 init_cplscales_table(q);
1217 if ((ret = init_cook_vlc_tables(q)))
1221 if(avctx->block_align >= UINT_MAX/2)
1222 return AVERROR(EINVAL);
1224 /* Pad the databuffer with:
1225 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1226 FF_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1227 q->decoded_bytes_buffer =
1228 av_mallocz(avctx->block_align
1229 + DECODE_BYTES_PAD1(avctx->block_align)
1230 + FF_INPUT_BUFFER_PADDING_SIZE);
1231 if (q->decoded_bytes_buffer == NULL)
1232 return AVERROR(ENOMEM);
1234 /* Initialize transform. */
1235 if ((ret = init_cook_mlt(q)))
1238 /* Initialize COOK signal arithmetic handling */
1240 q->scalar_dequant = scalar_dequant_float;
1241 q->decouple = decouple_float;
1242 q->imlt_window = imlt_window_float;
1243 q->interpolate = interpolate_float;
1244 q->saturate_output = saturate_output_float;
1247 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1248 if ((q->samples_per_channel == 256) || (q->samples_per_channel == 512) || (q->samples_per_channel == 1024)) {
1250 av_log_ask_for_sample(avctx,
1251 "unknown amount of samples_per_channel = %d\n",
1252 q->samples_per_channel);
1253 return AVERROR_PATCHWELCOME;
1256 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
1258 avctx->channel_layout = channel_mask;
1260 avctx->channel_layout = (avctx->channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1262 avcodec_get_frame_defaults(&q->frame);
1263 avctx->coded_frame = &q->frame;
1266 dump_cook_context(q);
1272 AVCodec ff_cook_decoder =
1275 .type = AVMEDIA_TYPE_AUDIO,
1276 .id = CODEC_ID_COOK,
1277 .priv_data_size = sizeof(COOKContext),
1278 .init = cook_decode_init,
1279 .close = cook_decode_close,
1280 .decode = cook_decode_frame,
1281 .capabilities = CODEC_CAP_DR1,
1282 .long_name = NULL_IF_CONFIG_SMALL("COOK"),