2 * COOK compatible decoder
3 * Copyright (c) 2003 Sascha Sommer
4 * Copyright (c) 2005 Benjamin Larsson
6 * This file is part of Libav.
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.
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.
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
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)
171 for (i = -63; i < 64; i++) {
172 pow2tab[63 + i] = pow(2, i);
173 rootpow2tab[63 + i] = sqrt(pow(2, i));
177 /* table generator */
178 static av_cold void init_gain_table(COOKContext *q)
181 q->gain_size_factor = q->samples_per_channel / 8;
182 for (i = 0; i < 23; i++)
183 q->gain_table[i] = pow(pow2tab[i + 52],
184 (1.0 / (double) q->gain_size_factor));
188 static av_cold int init_cook_vlc_tables(COOKContext *q)
193 for (i = 0; i < 13; i++) {
194 result |= init_vlc(&q->envelope_quant_index[i], 9, 24,
195 envelope_quant_index_huffbits[i], 1, 1,
196 envelope_quant_index_huffcodes[i], 2, 2, 0);
198 av_log(q->avctx, AV_LOG_DEBUG, "sqvh VLC init\n");
199 for (i = 0; i < 7; i++) {
200 result |= init_vlc(&q->sqvh[i], vhvlcsize_tab[i], vhsize_tab[i],
201 cvh_huffbits[i], 1, 1,
202 cvh_huffcodes[i], 2, 2, 0);
205 for (i = 0; i < q->num_subpackets; i++) {
206 if (q->subpacket[i].joint_stereo == 1) {
207 result |= init_vlc(&q->subpacket[i].ccpl, 6, (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);
214 av_log(q->avctx, AV_LOG_DEBUG, "VLC tables initialized.\n");
218 static av_cold int init_cook_mlt(COOKContext *q)
221 int mlt_size = q->samples_per_channel;
223 if ((q->mlt_window = av_malloc(mlt_size * sizeof(*q->mlt_window))) == 0)
224 return AVERROR(ENOMEM);
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);
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);
236 av_log(q->avctx, AV_LOG_DEBUG, "MDCT initialized, order = %d.\n",
237 av_log2(mlt_size) + 1);
242 static const float *maybe_reformat_buffer32(COOKContext *q, const float *ptr, int n)
248 static av_cold void init_cplscales_table(COOKContext *q)
251 for (i = 0; i < 5; i++)
252 q->cplscales[i] = maybe_reformat_buffer32(q, cplscales[i], (1 << (i + 2)) - 1);
255 /*************** init functions end ***********/
257 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes) + 3) % 4)
258 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
261 * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
262 * Why? No idea, some checksum/error detection method maybe.
264 * Out buffer size: extra bytes are needed to cope with
265 * padding/misalignment.
266 * Subpackets passed to the decoder can contain two, consecutive
267 * half-subpackets, of identical but arbitrary size.
268 * 1234 1234 1234 1234 extraA extraB
269 * Case 1: AAAA BBBB 0 0
270 * Case 2: AAAA ABBB BB-- 3 3
271 * Case 3: AAAA AABB BBBB 2 2
272 * Case 4: AAAA AAAB BBBB BB-- 1 5
274 * Nice way to waste CPU cycles.
276 * @param inbuffer pointer to byte array of indata
277 * @param out pointer to byte array of outdata
278 * @param bytes number of bytes
280 static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
282 static const uint32_t tab[4] = {
283 AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
284 AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
289 uint32_t *obuf = (uint32_t *) out;
290 /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
291 * I'm too lazy though, should be something like
292 * for (i = 0; i < bitamount / 64; i++)
293 * (int64_t) out[i] = 0x37c511f237c511f2 ^ av_be2ne64(int64_t) in[i]);
294 * Buffer alignment needs to be checked. */
296 off = (intptr_t) inbuffer & 3;
297 buf = (const uint32_t *) (inbuffer - off);
300 for (i = 0; i < bytes / 4; i++)
301 obuf[i] = c ^ buf[i];
309 static av_cold int cook_decode_close(AVCodecContext *avctx)
312 COOKContext *q = avctx->priv_data;
313 av_log(avctx, AV_LOG_DEBUG, "Deallocating memory.\n");
315 /* Free allocated memory buffers. */
316 av_free(q->mlt_window);
317 av_free(q->decoded_bytes_buffer);
319 /* Free the transform. */
320 ff_mdct_end(&q->mdct_ctx);
322 /* Free the VLC tables. */
323 for (i = 0; i < 13; i++)
324 ff_free_vlc(&q->envelope_quant_index[i]);
325 for (i = 0; i < 7; i++)
326 ff_free_vlc(&q->sqvh[i]);
327 for (i = 0; i < q->num_subpackets; i++)
328 ff_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
341 static void decode_gain_info(GetBitContext *gb, int *gaininfo)
345 while (get_bits1(gb)) {
349 n = get_bits_count(gb) - 1; // amount of elements*2 to update
353 int index = get_bits(gb, 3);
354 int gain = get_bits1(gb) ? get_bits(gb, 4) - 7 : -1;
357 gaininfo[i++] = gain;
364 * Create the quant index table needed for the envelope.
366 * @param q pointer to the COOKContext
367 * @param quant_index_table pointer to the array
369 static int decode_envelope(COOKContext *q, COOKSubpacket *p,
370 int *quant_index_table)
374 quant_index_table[0] = get_bits(&q->gb, 6) - 6; // This is used later in categorize
376 for (i = 1; i < p->total_subbands; i++) {
378 if (i >= p->js_subband_start * 2) {
379 vlc_index -= p->js_subband_start;
386 vlc_index = 13; // the VLC tables >13 are identical to No. 13
388 j = get_vlc2(&q->gb, q->envelope_quant_index[vlc_index - 1].table,
389 q->envelope_quant_index[vlc_index - 1].bits, 2);
390 quant_index_table[i] = quant_index_table[i - 1] + j - 12; // differential encoding
391 if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
392 av_log(q->avctx, AV_LOG_ERROR,
393 "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
394 quant_index_table[i], i);
395 return AVERROR_INVALIDDATA;
403 * Calculate the category and category_index vector.
405 * @param q pointer to the COOKContext
406 * @param quant_index_table pointer to the array
407 * @param category pointer to the category array
408 * @param category_index pointer to the category_index array
410 static void categorize(COOKContext *q, COOKSubpacket *p, int *quant_index_table,
411 int *category, int *category_index)
413 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
414 int exp_index2[102] = { 0 };
415 int exp_index1[102] = { 0 };
417 int tmp_categorize_array[128 * 2] = { 0 };
418 int tmp_categorize_array1_idx = p->numvector_size;
419 int tmp_categorize_array2_idx = p->numvector_size;
421 bits_left = p->bits_per_subpacket - get_bits_count(&q->gb);
423 if (bits_left > q->samples_per_channel) {
424 bits_left = q->samples_per_channel +
425 ((bits_left - q->samples_per_channel) * 5) / 8;
426 //av_log(q->avctx, AV_LOG_ERROR, "bits_left = %d\n",bits_left);
432 for (i = 32; i > 0; i = i / 2) {
435 for (j = p->total_subbands; j > 0; j--) {
436 exp_idx = av_clip((i - quant_index_table[index] + bias) / 2, 0, 7);
438 num_bits += expbits_tab[exp_idx];
440 if (num_bits >= bits_left - 32)
444 /* Calculate total number of bits. */
446 for (i = 0; i < p->total_subbands; i++) {
447 exp_idx = av_clip((bias - quant_index_table[i]) / 2, 0, 7);
448 num_bits += expbits_tab[exp_idx];
449 exp_index1[i] = exp_idx;
450 exp_index2[i] = exp_idx;
452 tmpbias1 = tmpbias2 = num_bits;
454 for (j = 1; j < p->numvector_size; j++) {
455 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
458 for (i = 0; i < p->total_subbands; i++) {
459 if (exp_index1[i] < 7) {
460 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
469 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
470 tmpbias1 -= expbits_tab[exp_index1[index]] -
471 expbits_tab[exp_index1[index] + 1];
476 for (i = 0; i < p->total_subbands; i++) {
477 if (exp_index2[i] > 0) {
478 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
487 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
488 tmpbias2 -= expbits_tab[exp_index2[index]] -
489 expbits_tab[exp_index2[index] - 1];
494 for (i = 0; i < p->total_subbands; i++)
495 category[i] = exp_index2[i];
497 for (i = 0; i < p->numvector_size - 1; i++)
498 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
503 * Expand the category vector.
505 * @param q pointer to the COOKContext
506 * @param category pointer to the category array
507 * @param category_index pointer to the category_index array
509 static inline void expand_category(COOKContext *q, int *category,
513 for (i = 0; i < q->num_vectors; i++)
515 int idx = category_index[i];
516 if (++category[idx] >= FF_ARRAY_ELEMS(dither_tab))
522 * The real requantization of the mltcoefs
524 * @param q pointer to the COOKContext
526 * @param quant_index quantisation index
527 * @param subband_coef_index array of indexes to quant_centroid_tab
528 * @param subband_coef_sign signs of coefficients
529 * @param mlt_p pointer into the mlt buffer
531 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
532 int *subband_coef_index, int *subband_coef_sign,
538 for (i = 0; i < SUBBAND_SIZE; i++) {
539 if (subband_coef_index[i]) {
540 f1 = quant_centroid_tab[index][subband_coef_index[i]];
541 if (subband_coef_sign[i])
544 /* noise coding if subband_coef_index[i] == 0 */
545 f1 = dither_tab[index];
546 if (av_lfg_get(&q->random_state) < 0x80000000)
549 mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
553 * Unpack the subband_coef_index and subband_coef_sign vectors.
555 * @param q pointer to the COOKContext
556 * @param category pointer to the category array
557 * @param subband_coef_index array of indexes to quant_centroid_tab
558 * @param subband_coef_sign signs of coefficients
560 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
561 int *subband_coef_index, int *subband_coef_sign)
564 int vlc, vd, tmp, result;
566 vd = vd_tab[category];
568 for (i = 0; i < vpr_tab[category]; i++) {
569 vlc = get_vlc2(&q->gb, q->sqvh[category].table, q->sqvh[category].bits, 3);
570 if (p->bits_per_subpacket < get_bits_count(&q->gb)) {
574 for (j = vd - 1; j >= 0; j--) {
575 tmp = (vlc * invradix_tab[category]) / 0x100000;
576 subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
579 for (j = 0; j < vd; j++) {
580 if (subband_coef_index[i * vd + j]) {
581 if (get_bits_count(&q->gb) < p->bits_per_subpacket) {
582 subband_coef_sign[i * vd + j] = get_bits1(&q->gb);
585 subband_coef_sign[i * vd + j] = 0;
588 subband_coef_sign[i * vd + j] = 0;
597 * Fill the mlt_buffer with mlt coefficients.
599 * @param q pointer to the COOKContext
600 * @param category pointer to the category array
601 * @param quant_index_table pointer to the array
602 * @param mlt_buffer pointer to mlt coefficients
604 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
605 int *quant_index_table, float *mlt_buffer)
607 /* A zero in this table means that the subband coefficient is
608 random noise coded. */
609 int subband_coef_index[SUBBAND_SIZE];
610 /* A zero in this table means that the subband coefficient is a
611 positive multiplicator. */
612 int subband_coef_sign[SUBBAND_SIZE];
616 for (band = 0; band < p->total_subbands; band++) {
617 index = category[band];
618 if (category[band] < 7) {
619 if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
621 for (j = 0; j < p->total_subbands; j++)
622 category[band + j] = 7;
626 memset(subband_coef_index, 0, sizeof(subband_coef_index));
627 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
629 q->scalar_dequant(q, index, quant_index_table[band],
630 subband_coef_index, subband_coef_sign,
631 &mlt_buffer[band * SUBBAND_SIZE]);
634 /* FIXME: should this be removed, or moved into loop above? */
635 if (p->total_subbands * SUBBAND_SIZE >= q->samples_per_channel)
641 * function for decoding mono data
643 * @param q pointer to the COOKContext
644 * @param mlt_buffer pointer to mlt coefficients
646 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
648 int category_index[128] = { 0 };
649 int category[128] = { 0 };
650 int quant_index_table[102];
653 if ((res = decode_envelope(q, p, quant_index_table)) < 0)
655 q->num_vectors = get_bits(&q->gb, p->log2_numvector_size);
656 categorize(q, p, quant_index_table, category, category_index);
657 expand_category(q, category, category_index);
658 decode_vectors(q, p, category, quant_index_table, mlt_buffer);
665 * the actual requantization of the timedomain samples
667 * @param q pointer to the COOKContext
668 * @param buffer pointer to the timedomain buffer
669 * @param gain_index index for the block multiplier
670 * @param gain_index_next index for the next block multiplier
672 static void interpolate_float(COOKContext *q, float *buffer,
673 int gain_index, int gain_index_next)
677 fc1 = pow2tab[gain_index + 63];
679 if (gain_index == gain_index_next) { // static gain
680 for (i = 0; i < q->gain_size_factor; i++)
682 } else { // smooth gain
683 fc2 = q->gain_table[11 + (gain_index_next - gain_index)];
684 for (i = 0; i < q->gain_size_factor; i++) {
692 * Apply transform window, overlap buffers.
694 * @param q pointer to the COOKContext
695 * @param inbuffer pointer to the mltcoefficients
696 * @param gains_ptr current and previous gains
697 * @param previous_buffer pointer to the previous buffer to be used for overlapping
699 static void imlt_window_float(COOKContext *q, float *inbuffer,
700 cook_gains *gains_ptr, float *previous_buffer)
702 const float fc = pow2tab[gains_ptr->previous[0] + 63];
704 /* The weird thing here, is that the two halves of the time domain
705 * buffer are swapped. Also, the newest data, that we save away for
706 * next frame, has the wrong sign. Hence the subtraction below.
707 * Almost sounds like a complex conjugate/reverse data/FFT effect.
710 /* Apply window and overlap */
711 for (i = 0; i < q->samples_per_channel; i++)
712 inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
713 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
717 * The modulated lapped transform, this takes transform coefficients
718 * and transforms them into timedomain samples.
719 * Apply transform window, overlap buffers, apply gain profile
720 * and buffer management.
722 * @param q pointer to the COOKContext
723 * @param inbuffer pointer to the mltcoefficients
724 * @param gains_ptr current and previous gains
725 * @param previous_buffer pointer to the previous buffer to be used for overlapping
727 static void imlt_gain(COOKContext *q, float *inbuffer,
728 cook_gains *gains_ptr, float *previous_buffer)
730 float *buffer0 = q->mono_mdct_output;
731 float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
734 /* Inverse modified discrete cosine transform */
735 q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
737 q->imlt_window(q, buffer1, gains_ptr, previous_buffer);
739 /* Apply gain profile */
740 for (i = 0; i < 8; i++)
741 if (gains_ptr->now[i] || gains_ptr->now[i + 1])
742 q->interpolate(q, &buffer1[q->gain_size_factor * i],
743 gains_ptr->now[i], gains_ptr->now[i + 1]);
745 /* Save away the current to be previous block. */
746 memcpy(previous_buffer, buffer0,
747 q->samples_per_channel * sizeof(*previous_buffer));
752 * function for getting the jointstereo coupling information
754 * @param q pointer to the COOKContext
755 * @param decouple_tab decoupling array
758 static void decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
761 int vlc = get_bits1(&q->gb);
762 int start = cplband[p->js_subband_start];
763 int end = cplband[p->subbands - 1];
764 int length = end - start + 1;
770 for (i = 0; i < length; i++)
771 decouple_tab[start + i] = get_vlc2(&q->gb, p->ccpl.table, p->ccpl.bits, 2);
773 for (i = 0; i < length; i++)
774 decouple_tab[start + i] = get_bits(&q->gb, p->js_vlc_bits);
778 * function decouples a pair of signals from a single signal via multiplication.
780 * @param q pointer to the COOKContext
781 * @param subband index of the current subband
782 * @param f1 multiplier for channel 1 extraction
783 * @param f2 multiplier for channel 2 extraction
784 * @param decode_buffer input buffer
785 * @param mlt_buffer1 pointer to left channel mlt coefficients
786 * @param mlt_buffer2 pointer to right channel mlt coefficients
788 static void decouple_float(COOKContext *q,
792 float *decode_buffer,
793 float *mlt_buffer1, float *mlt_buffer2)
796 for (j = 0; j < SUBBAND_SIZE; j++) {
797 tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
798 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
799 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
804 * function for decoding joint stereo data
806 * @param q pointer to the COOKContext
807 * @param mlt_buffer1 pointer to left channel mlt coefficients
808 * @param mlt_buffer2 pointer to right channel mlt coefficients
810 static int joint_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer1,
814 int decouple_tab[SUBBAND_SIZE] = { 0 };
815 float *decode_buffer = q->decode_buffer_0;
818 const float *cplscale;
820 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
822 /* Make sure the buffers are zeroed out. */
823 memset(mlt_buffer1, 0, 1024 * sizeof(*mlt_buffer1));
824 memset(mlt_buffer2, 0, 1024 * sizeof(*mlt_buffer2));
825 decouple_info(q, p, decouple_tab);
826 if ((res = mono_decode(q, p, decode_buffer)) < 0)
829 /* The two channels are stored interleaved in decode_buffer. */
830 for (i = 0; i < p->js_subband_start; i++) {
831 for (j = 0; j < SUBBAND_SIZE; j++) {
832 mlt_buffer1[i * 20 + j] = decode_buffer[i * 40 + j];
833 mlt_buffer2[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
837 /* When we reach js_subband_start (the higher frequencies)
838 the coefficients are stored in a coupling scheme. */
839 idx = (1 << p->js_vlc_bits) - 1;
840 for (i = p->js_subband_start; i < p->subbands; i++) {
841 cpl_tmp = cplband[i];
842 idx -= decouple_tab[cpl_tmp];
843 cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
844 f1 = cplscale[decouple_tab[cpl_tmp] + 1];
846 q->decouple(q, p, i, f1, f2, decode_buffer, mlt_buffer1, mlt_buffer2);
847 idx = (1 << p->js_vlc_bits) - 1;
854 * First part of subpacket decoding:
855 * decode raw stream bytes and read gain info.
857 * @param q pointer to the COOKContext
858 * @param inbuffer pointer to raw stream data
859 * @param gains_ptr array of current/prev gain pointers
861 static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
862 const uint8_t *inbuffer,
863 cook_gains *gains_ptr)
867 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
868 p->bits_per_subpacket / 8);
869 init_get_bits(&q->gb, q->decoded_bytes_buffer + offset,
870 p->bits_per_subpacket);
871 decode_gain_info(&q->gb, gains_ptr->now);
873 /* Swap current and previous gains */
874 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
878 * Saturate the output signal and interleave.
880 * @param q pointer to the COOKContext
881 * @param chan channel to saturate
882 * @param out pointer to the output vector
884 static void saturate_output_float(COOKContext *q, int chan, float *out)
887 float *output = q->mono_mdct_output + q->samples_per_channel;
888 for (j = 0; j < q->samples_per_channel; j++) {
889 out[chan + q->nb_channels * j] = av_clipf(output[j], -1.0, 1.0);
894 * Final part of subpacket decoding:
895 * Apply modulated lapped transform, gain compensation,
896 * clip and convert to integer.
898 * @param q pointer to the COOKContext
899 * @param decode_buffer pointer to the mlt coefficients
900 * @param gains_ptr array of current/prev gain pointers
901 * @param previous_buffer pointer to the previous buffer to be used for overlapping
902 * @param out pointer to the output buffer
903 * @param chan 0: left or single channel, 1: right channel
905 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
906 cook_gains *gains_ptr, float *previous_buffer,
907 float *out, int chan)
909 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
911 q->saturate_output(q, chan, out);
916 * Cook subpacket decoding. This function returns one decoded subpacket,
917 * usually 1024 samples per channel.
919 * @param q pointer to the COOKContext
920 * @param inbuffer pointer to the inbuffer
921 * @param outbuffer pointer to the outbuffer
923 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
924 const uint8_t *inbuffer, float *outbuffer)
926 int sub_packet_size = p->size;
929 // for (i = 0; i < sub_packet_size ; i++)
930 // av_log(q->avctx, AV_LOG_ERROR, "%02x", inbuffer[i]);
931 // av_log(q->avctx, AV_LOG_ERROR, "\n");
932 memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
933 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
935 if (p->joint_stereo) {
936 if ((res = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
939 if ((res = mono_decode(q, p, q->decode_buffer_1)) < 0)
942 if (p->num_channels == 2) {
943 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
944 if ((res = mono_decode(q, p, q->decode_buffer_2)) < 0)
949 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
950 p->mono_previous_buffer1, outbuffer, p->ch_idx);
952 if (p->num_channels == 2)
954 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
955 p->mono_previous_buffer2, outbuffer, p->ch_idx + 1);
957 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
958 p->mono_previous_buffer2, outbuffer, p->ch_idx + 1);
965 * Cook frame decoding
967 * @param avctx pointer to the AVCodecContext
969 static int cook_decode_frame(AVCodecContext *avctx, void *data,
970 int *got_frame_ptr, AVPacket *avpkt)
972 const uint8_t *buf = avpkt->data;
973 int buf_size = avpkt->size;
974 COOKContext *q = avctx->priv_data;
975 float *samples = NULL;
980 if (buf_size < avctx->block_align)
983 /* get output buffer */
984 if (q->discarded_packets >= 2) {
985 q->frame.nb_samples = q->samples_per_channel;
986 if ((ret = avctx->get_buffer(avctx, &q->frame)) < 0) {
987 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
990 samples = (float *) q->frame.data[0];
993 /* estimate subpacket sizes */
994 q->subpacket[0].size = avctx->block_align;
996 for (i = 1; i < q->num_subpackets; i++) {
997 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
998 q->subpacket[0].size -= q->subpacket[i].size + 1;
999 if (q->subpacket[0].size < 0) {
1000 av_log(avctx, AV_LOG_DEBUG,
1001 "frame subpacket size total > avctx->block_align!\n");
1002 return AVERROR_INVALIDDATA;
1006 /* decode supbackets */
1007 for (i = 0; i < q->num_subpackets; i++) {
1008 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
1009 q->subpacket[i].bits_per_subpdiv;
1010 q->subpacket[i].ch_idx = chidx;
1011 av_log(avctx, AV_LOG_DEBUG,
1012 "subpacket[%i] size %i js %i %i block_align %i\n",
1013 i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
1014 avctx->block_align);
1016 if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
1018 offset += q->subpacket[i].size;
1019 chidx += q->subpacket[i].num_channels;
1020 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
1021 i, q->subpacket[i].size * 8, get_bits_count(&q->gb));
1024 /* Discard the first two frames: no valid audio. */
1025 if (q->discarded_packets < 2) {
1026 q->discarded_packets++;
1028 return avctx->block_align;
1032 *(AVFrame *) data = q->frame;
1034 return avctx->block_align;
1038 static void dump_cook_context(COOKContext *q)
1041 #define PRINT(a, b) av_log(q->avctx, AV_LOG_ERROR, " %s = %d\n", a, b);
1042 av_log(q->avctx, AV_LOG_ERROR, "COOKextradata\n");
1043 av_log(q->avctx, AV_LOG_ERROR, "cookversion=%x\n", q->subpacket[0].cookversion);
1044 if (q->subpacket[0].cookversion > STEREO) {
1045 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1046 PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1048 av_log(q->avctx, AV_LOG_ERROR, "COOKContext\n");
1049 PRINT("nb_channels", q->nb_channels);
1050 PRINT("bit_rate", q->bit_rate);
1051 PRINT("sample_rate", q->sample_rate);
1052 PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1053 PRINT("samples_per_frame", q->subpacket[0].samples_per_frame);
1054 PRINT("subbands", q->subpacket[0].subbands);
1055 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1056 PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1057 PRINT("numvector_size", q->subpacket[0].numvector_size);
1058 PRINT("total_subbands", q->subpacket[0].total_subbands);
1062 static av_cold int cook_count_channels(unsigned int mask)
1066 for (i = 0; i < 32; i++)
1067 if (mask & (1 << i))
1073 * Cook initialization
1075 * @param avctx pointer to the AVCodecContext
1077 static av_cold int cook_decode_init(AVCodecContext *avctx)
1079 COOKContext *q = avctx->priv_data;
1080 const uint8_t *edata_ptr = avctx->extradata;
1081 const uint8_t *edata_ptr_end = edata_ptr + avctx->extradata_size;
1082 int extradata_size = avctx->extradata_size;
1084 unsigned int channel_mask = 0;
1088 /* Take care of the codec specific extradata. */
1089 if (extradata_size <= 0) {
1090 av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1091 return AVERROR_INVALIDDATA;
1093 av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1095 /* Take data from the AVCodecContext (RM container). */
1096 q->sample_rate = avctx->sample_rate;
1097 q->nb_channels = avctx->channels;
1098 q->bit_rate = avctx->bit_rate;
1099 if (!q->nb_channels) {
1100 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1101 return AVERROR_INVALIDDATA;
1104 /* Initialize RNG. */
1105 av_lfg_init(&q->random_state, 0);
1107 while (edata_ptr < edata_ptr_end) {
1108 /* 8 for mono, 16 for stereo, ? for multichannel
1109 Swap to right endianness so we don't need to care later on. */
1110 if (extradata_size >= 8) {
1111 q->subpacket[s].cookversion = bytestream_get_be32(&edata_ptr);
1112 q->subpacket[s].samples_per_frame = bytestream_get_be16(&edata_ptr);
1113 q->subpacket[s].subbands = bytestream_get_be16(&edata_ptr);
1114 extradata_size -= 8;
1116 if (extradata_size >= 8) {
1117 bytestream_get_be32(&edata_ptr); // Unknown unused
1118 q->subpacket[s].js_subband_start = bytestream_get_be16(&edata_ptr);
1119 q->subpacket[s].js_vlc_bits = bytestream_get_be16(&edata_ptr);
1120 extradata_size -= 8;
1123 /* Initialize extradata related variables. */
1124 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame / q->nb_channels;
1125 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1127 /* Initialize default data states. */
1128 q->subpacket[s].log2_numvector_size = 5;
1129 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1130 q->subpacket[s].num_channels = 1;
1132 /* Initialize version-dependent variables */
1134 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1135 q->subpacket[s].cookversion);
1136 q->subpacket[s].joint_stereo = 0;
1137 switch (q->subpacket[s].cookversion) {
1139 if (q->nb_channels != 1) {
1140 av_log_ask_for_sample(avctx, "Container channels != 1.\n");
1141 return AVERROR_PATCHWELCOME;
1143 av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1146 if (q->nb_channels != 1) {
1147 q->subpacket[s].bits_per_subpdiv = 1;
1148 q->subpacket[s].num_channels = 2;
1150 av_log(avctx, AV_LOG_DEBUG, "STEREO\n");
1153 if (q->nb_channels != 2) {
1154 av_log_ask_for_sample(avctx, "Container channels != 2.\n");
1155 return AVERROR_PATCHWELCOME;
1157 av_log(avctx, AV_LOG_DEBUG, "JOINT_STEREO\n");
1158 if (avctx->extradata_size >= 16) {
1159 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1160 q->subpacket[s].js_subband_start;
1161 q->subpacket[s].joint_stereo = 1;
1162 q->subpacket[s].num_channels = 2;
1164 if (q->subpacket[s].samples_per_channel > 256) {
1165 q->subpacket[s].log2_numvector_size = 6;
1167 if (q->subpacket[s].samples_per_channel > 512) {
1168 q->subpacket[s].log2_numvector_size = 7;
1172 av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1173 if (extradata_size >= 4)
1174 channel_mask |= q->subpacket[s].channel_mask = bytestream_get_be32(&edata_ptr);
1176 if (cook_count_channels(q->subpacket[s].channel_mask) > 1) {
1177 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1178 q->subpacket[s].js_subband_start;
1179 q->subpacket[s].joint_stereo = 1;
1180 q->subpacket[s].num_channels = 2;
1181 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame >> 1;
1183 if (q->subpacket[s].samples_per_channel > 256) {
1184 q->subpacket[s].log2_numvector_size = 6;
1186 if (q->subpacket[s].samples_per_channel > 512) {
1187 q->subpacket[s].log2_numvector_size = 7;
1190 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame;
1194 av_log_ask_for_sample(avctx, "Unknown Cook version.\n");
1195 return AVERROR_PATCHWELCOME;
1198 if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1199 av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1200 return AVERROR_INVALIDDATA;
1202 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1205 /* Initialize variable relations */
1206 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1208 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1209 if (q->subpacket[s].total_subbands > 53) {
1210 av_log_ask_for_sample(avctx, "total_subbands > 53\n");
1211 return AVERROR_PATCHWELCOME;
1214 if ((q->subpacket[s].js_vlc_bits > 6) ||
1215 (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1216 av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1217 q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1218 return AVERROR_INVALIDDATA;
1221 if (q->subpacket[s].subbands > 50) {
1222 av_log_ask_for_sample(avctx, "subbands > 50\n");
1223 return AVERROR_PATCHWELCOME;
1225 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1226 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1227 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1228 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1230 q->num_subpackets++;
1232 if (s > MAX_SUBPACKETS) {
1233 av_log_ask_for_sample(avctx, "Too many subpackets > 5\n");
1234 return AVERROR_PATCHWELCOME;
1237 /* Generate tables */
1240 init_cplscales_table(q);
1242 if ((ret = init_cook_vlc_tables(q)))
1246 if (avctx->block_align >= UINT_MAX / 2)
1247 return AVERROR(EINVAL);
1249 /* Pad the databuffer with:
1250 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1251 FF_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1252 q->decoded_bytes_buffer =
1253 av_mallocz(avctx->block_align
1254 + DECODE_BYTES_PAD1(avctx->block_align)
1255 + FF_INPUT_BUFFER_PADDING_SIZE);
1256 if (q->decoded_bytes_buffer == NULL)
1257 return AVERROR(ENOMEM);
1259 /* Initialize transform. */
1260 if ((ret = init_cook_mlt(q)))
1263 /* Initialize COOK signal arithmetic handling */
1265 q->scalar_dequant = scalar_dequant_float;
1266 q->decouple = decouple_float;
1267 q->imlt_window = imlt_window_float;
1268 q->interpolate = interpolate_float;
1269 q->saturate_output = saturate_output_float;
1272 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1273 if ((q->samples_per_channel == 256) || (q->samples_per_channel == 512)
1274 || (q->samples_per_channel == 1024)) {
1276 av_log_ask_for_sample(avctx,
1277 "unknown amount of samples_per_channel = %d\n",
1278 q->samples_per_channel);
1279 return AVERROR_PATCHWELCOME;
1282 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
1284 avctx->channel_layout = channel_mask;
1286 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1288 avcodec_get_frame_defaults(&q->frame);
1289 avctx->coded_frame = &q->frame;
1292 dump_cook_context(q);
1297 AVCodec ff_cook_decoder = {
1299 .type = AVMEDIA_TYPE_AUDIO,
1300 .id = CODEC_ID_COOK,
1301 .priv_data_size = sizeof(COOKContext),
1302 .init = cook_decode_init,
1303 .close = cook_decode_close,
1304 .decode = cook_decode_frame,
1305 .capabilities = CODEC_CAP_DR1,
1306 .long_name = NULL_IF_CONFIG_SMALL("COOK"),