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)
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(0x37c511f2), AV_BE2NE32C(0xf237c511),
284 AV_BE2NE32C(0x11f237c5), AV_BE2NE32C(0xc511f237),
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
392 if (quant_index_table[i] < -63 || quant_index_table[i] > 64) {
393 av_log(NULL, AV_LOG_ERROR, "quant_index_table value out of bounds\n");
394 return AVERROR_INVALIDDATA;
402 * Calculate the category and category_index vector.
404 * @param q pointer to the COOKContext
405 * @param quant_index_table pointer to the array
406 * @param category pointer to the category array
407 * @param category_index pointer to the category_index array
409 static void categorize(COOKContext *q, COOKSubpacket *p, const int *quant_index_table,
410 int *category, int *category_index)
412 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
416 int tmp_categorize_array[128 * 2];
417 int tmp_categorize_array1_idx = p->numvector_size;
418 int tmp_categorize_array2_idx = p->numvector_size;
420 bits_left = p->bits_per_subpacket - get_bits_count(&q->gb);
422 if (bits_left > q->samples_per_channel) {
423 bits_left = q->samples_per_channel +
424 ((bits_left - q->samples_per_channel) * 5) / 8;
425 //av_log(q->avctx, AV_LOG_ERROR, "bits_left = %d\n",bits_left);
428 memset(&exp_index1, 0, sizeof(exp_index1));
429 memset(&exp_index2, 0, sizeof(exp_index2));
430 memset(&tmp_categorize_array, 0, sizeof(tmp_categorize_array));
435 for (i = 32; i > 0; i = i / 2) {
438 for (j = p->total_subbands; j > 0; j--) {
439 exp_idx = av_clip((i - quant_index_table[index] + bias) / 2, 0, 7);
441 num_bits += expbits_tab[exp_idx];
443 if (num_bits >= bits_left - 32)
447 /* Calculate total number of bits. */
449 for (i = 0; i < p->total_subbands; i++) {
450 exp_idx = av_clip((bias - quant_index_table[i]) / 2, 0, 7);
451 num_bits += expbits_tab[exp_idx];
452 exp_index1[i] = exp_idx;
453 exp_index2[i] = exp_idx;
455 tmpbias1 = tmpbias2 = num_bits;
457 for (j = 1; j < p->numvector_size; j++) {
458 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
461 for (i = 0; i < p->total_subbands; i++) {
462 if (exp_index1[i] < 7) {
463 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
472 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
473 tmpbias1 -= expbits_tab[exp_index1[index]] -
474 expbits_tab[exp_index1[index] + 1];
479 for (i = 0; i < p->total_subbands; i++) {
480 if (exp_index2[i] > 0) {
481 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
490 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
491 tmpbias2 -= expbits_tab[exp_index2[index]] -
492 expbits_tab[exp_index2[index] - 1];
497 for (i = 0; i < p->total_subbands; i++)
498 category[i] = exp_index2[i];
500 for (i = 0; i < p->numvector_size - 1; i++)
501 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
506 * Expand the category vector.
508 * @param q pointer to the COOKContext
509 * @param category pointer to the category array
510 * @param category_index pointer to the category_index array
512 static inline void expand_category(COOKContext *q, int *category,
516 for (i = 0; i < q->num_vectors; i++)
517 ++category[category_index[i]];
521 * The real requantization of the mltcoefs
523 * @param q pointer to the COOKContext
525 * @param quant_index quantisation index
526 * @param subband_coef_index array of indexes to quant_centroid_tab
527 * @param subband_coef_sign signs of coefficients
528 * @param mlt_p pointer into the mlt buffer
530 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
531 int *subband_coef_index, int *subband_coef_sign,
537 for (i = 0; i < SUBBAND_SIZE; i++) {
538 if (subband_coef_index[i]) {
539 f1 = quant_centroid_tab[index][subband_coef_index[i]];
540 if (subband_coef_sign[i])
543 /* noise coding if subband_coef_index[i] == 0 */
544 f1 = dither_tab[index];
545 if (av_lfg_get(&q->random_state) < 0x80000000)
548 mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
552 * Unpack the subband_coef_index and subband_coef_sign vectors.
554 * @param q pointer to the COOKContext
555 * @param category pointer to the category array
556 * @param subband_coef_index array of indexes to quant_centroid_tab
557 * @param subband_coef_sign signs of coefficients
559 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
560 int *subband_coef_index, int *subband_coef_sign)
563 int vlc, vd, tmp, result;
565 vd = vd_tab[category];
567 for (i = 0; i < vpr_tab[category]; i++) {
568 vlc = get_vlc2(&q->gb, q->sqvh[category].table, q->sqvh[category].bits, 3);
569 if (p->bits_per_subpacket < get_bits_count(&q->gb)) {
573 for (j = vd - 1; j >= 0; j--) {
574 tmp = (vlc * invradix_tab[category]) / 0x100000;
575 subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
578 for (j = 0; j < vd; j++) {
579 if (subband_coef_index[i * vd + j]) {
580 if (get_bits_count(&q->gb) < p->bits_per_subpacket) {
581 subband_coef_sign[i * vd + j] = get_bits1(&q->gb);
584 subband_coef_sign[i * vd + j] = 0;
587 subband_coef_sign[i * vd + j] = 0;
596 * Fill the mlt_buffer with mlt coefficients.
598 * @param q pointer to the COOKContext
599 * @param category pointer to the category array
600 * @param quant_index_table pointer to the array
601 * @param mlt_buffer pointer to mlt coefficients
603 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
604 int *quant_index_table, float *mlt_buffer)
606 /* A zero in this table means that the subband coefficient is
607 random noise coded. */
608 int subband_coef_index[SUBBAND_SIZE];
609 /* A zero in this table means that the subband coefficient is a
610 positive multiplicator. */
611 int subband_coef_sign[SUBBAND_SIZE];
615 for (band = 0; band < p->total_subbands; band++) {
616 index = category[band];
617 if (category[band] < 7) {
618 if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
620 for (j = 0; j < p->total_subbands; j++)
621 category[band + j] = 7;
625 memset(subband_coef_index, 0, sizeof(subband_coef_index));
626 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
628 q->scalar_dequant(q, index, quant_index_table[band],
629 subband_coef_index, subband_coef_sign,
630 &mlt_buffer[band * SUBBAND_SIZE]);
633 /* FIXME: should this be removed, or moved into loop above? */
634 if (p->total_subbands * SUBBAND_SIZE >= q->samples_per_channel)
640 * function for decoding mono data
642 * @param q pointer to the COOKContext
643 * @param mlt_buffer pointer to mlt coefficients
645 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
647 int category_index[128];
648 int quant_index_table[102];
652 memset(&category, 0, sizeof(category));
653 memset(&category_index, 0, sizeof(category_index));
655 if ((ret = decode_envelope(q, p, quant_index_table)) < 0)
657 q->num_vectors = get_bits(&q->gb, p->log2_numvector_size);
658 categorize(q, p, quant_index_table, category, category_index);
659 expand_category(q, category, category_index);
660 for (i=0; i<p->total_subbands; i++) {
662 return AVERROR_INVALIDDATA;
664 decode_vectors(q, p, category, quant_index_table, mlt_buffer);
671 * the actual requantization of the timedomain samples
673 * @param q pointer to the COOKContext
674 * @param buffer pointer to the timedomain buffer
675 * @param gain_index index for the block multiplier
676 * @param gain_index_next index for the next block multiplier
678 static void interpolate_float(COOKContext *q, float *buffer,
679 int gain_index, int gain_index_next)
683 fc1 = pow2tab[gain_index + 63];
685 if (gain_index == gain_index_next) { // static gain
686 for (i = 0; i < q->gain_size_factor; i++)
688 } else { // smooth gain
689 fc2 = q->gain_table[11 + (gain_index_next - gain_index)];
690 for (i = 0; i < q->gain_size_factor; i++) {
698 * Apply transform window, overlap buffers.
700 * @param q pointer to the COOKContext
701 * @param inbuffer pointer to the mltcoefficients
702 * @param gains_ptr current and previous gains
703 * @param previous_buffer pointer to the previous buffer to be used for overlapping
705 static void imlt_window_float(COOKContext *q, float *inbuffer,
706 cook_gains *gains_ptr, float *previous_buffer)
708 const float fc = pow2tab[gains_ptr->previous[0] + 63];
710 /* The weird thing here, is that the two halves of the time domain
711 * buffer are swapped. Also, the newest data, that we save away for
712 * next frame, has the wrong sign. Hence the subtraction below.
713 * Almost sounds like a complex conjugate/reverse data/FFT effect.
716 /* Apply window and overlap */
717 for (i = 0; i < q->samples_per_channel; i++)
718 inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
719 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
723 * The modulated lapped transform, this takes transform coefficients
724 * and transforms them into timedomain samples.
725 * Apply transform window, overlap buffers, apply gain profile
726 * and buffer management.
728 * @param q pointer to the COOKContext
729 * @param inbuffer pointer to the mltcoefficients
730 * @param gains_ptr current and previous gains
731 * @param previous_buffer pointer to the previous buffer to be used for overlapping
733 static void imlt_gain(COOKContext *q, float *inbuffer,
734 cook_gains *gains_ptr, float *previous_buffer)
736 float *buffer0 = q->mono_mdct_output;
737 float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
740 /* Inverse modified discrete cosine transform */
741 q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
743 q->imlt_window(q, buffer1, gains_ptr, previous_buffer);
745 /* Apply gain profile */
746 for (i = 0; i < 8; i++)
747 if (gains_ptr->now[i] || gains_ptr->now[i + 1])
748 q->interpolate(q, &buffer1[q->gain_size_factor * i],
749 gains_ptr->now[i], gains_ptr->now[i + 1]);
751 /* Save away the current to be previous block. */
752 memcpy(previous_buffer, buffer0,
753 q->samples_per_channel * sizeof(*previous_buffer));
758 * function for getting the jointstereo coupling information
760 * @param q pointer to the COOKContext
761 * @param decouple_tab decoupling array
764 static void decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
767 int vlc = get_bits1(&q->gb);
768 int start = cplband[p->js_subband_start];
769 int end = cplband[p->subbands - 1];
770 int length = end - start + 1;
776 for (i = 0; i < length; i++)
777 decouple_tab[start + i] = get_vlc2(&q->gb, p->ccpl.table, p->ccpl.bits, 2);
779 for (i = 0; i < length; i++)
780 decouple_tab[start + i] = get_bits(&q->gb, p->js_vlc_bits);
784 * function decouples a pair of signals from a single signal via multiplication.
786 * @param q pointer to the COOKContext
787 * @param subband index of the current subband
788 * @param f1 multiplier for channel 1 extraction
789 * @param f2 multiplier for channel 2 extraction
790 * @param decode_buffer input buffer
791 * @param mlt_buffer1 pointer to left channel mlt coefficients
792 * @param mlt_buffer2 pointer to right channel mlt coefficients
794 static void decouple_float(COOKContext *q,
798 float *decode_buffer,
799 float *mlt_buffer1, float *mlt_buffer2)
802 for (j = 0; j < SUBBAND_SIZE; j++) {
803 tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
804 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
805 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
810 * function for decoding joint stereo data
812 * @param q pointer to the COOKContext
813 * @param mlt_buffer1 pointer to left channel mlt coefficients
814 * @param mlt_buffer2 pointer to right channel mlt coefficients
816 static int joint_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer1,
820 int decouple_tab[SUBBAND_SIZE];
821 float *decode_buffer = q->decode_buffer_0;
824 const float *cplscale;
826 memset(decouple_tab, 0, sizeof(decouple_tab));
827 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
829 /* Make sure the buffers are zeroed out. */
830 memset(mlt_buffer1, 0, 1024 * sizeof(*mlt_buffer1));
831 memset(mlt_buffer2, 0, 1024 * sizeof(*mlt_buffer2));
832 decouple_info(q, p, decouple_tab);
833 if ((ret = mono_decode(q, p, decode_buffer)) < 0)
835 /* The two channels are stored interleaved in decode_buffer. */
836 for (i = 0; i < p->js_subband_start; i++) {
837 for (j = 0; j < SUBBAND_SIZE; j++) {
838 mlt_buffer1[i * 20 + j] = decode_buffer[i * 40 + j];
839 mlt_buffer2[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
843 /* When we reach js_subband_start (the higher frequencies)
844 the coefficients are stored in a coupling scheme. */
845 idx = (1 << p->js_vlc_bits) - 1;
846 for (i = p->js_subband_start; i < p->subbands; i++) {
847 cpl_tmp = cplband[i];
848 idx -= decouple_tab[cpl_tmp];
849 cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
850 f1 = cplscale[decouple_tab[cpl_tmp]];
851 f2 = cplscale[idx - 1];
852 q->decouple(q, p, i, f1, f2, decode_buffer, mlt_buffer1, mlt_buffer2);
853 idx = (1 << p->js_vlc_bits) - 1;
859 * First part of subpacket decoding:
860 * decode raw stream bytes and read gain info.
862 * @param q pointer to the COOKContext
863 * @param inbuffer pointer to raw stream data
864 * @param gains_ptr array of current/prev gain pointers
866 static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
867 const uint8_t *inbuffer,
868 cook_gains *gains_ptr)
872 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
873 p->bits_per_subpacket / 8);
874 init_get_bits(&q->gb, q->decoded_bytes_buffer + offset,
875 p->bits_per_subpacket);
876 decode_gain_info(&q->gb, gains_ptr->now);
878 /* Swap current and previous gains */
879 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
883 * Saturate the output signal and interleave.
885 * @param q pointer to the COOKContext
886 * @param chan channel to saturate
887 * @param out pointer to the output vector
889 static void saturate_output_float(COOKContext *q, int chan, float *out)
892 float *output = q->mono_mdct_output + q->samples_per_channel;
893 for (j = 0; j < q->samples_per_channel; j++) {
894 out[chan + q->nb_channels * j] = av_clipf(output[j], -1.0, 1.0);
899 * Final part of subpacket decoding:
900 * Apply modulated lapped transform, gain compensation,
901 * clip and convert to integer.
903 * @param q pointer to the COOKContext
904 * @param decode_buffer pointer to the mlt coefficients
905 * @param gains_ptr array of current/prev gain pointers
906 * @param previous_buffer pointer to the previous buffer to be used for overlapping
907 * @param out pointer to the output buffer
908 * @param chan 0: left or single channel, 1: right channel
910 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
911 cook_gains *gains_ptr, float *previous_buffer,
912 float *out, int chan)
914 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
916 q->saturate_output(q, chan, out);
921 * Cook subpacket decoding. This function returns one decoded subpacket,
922 * usually 1024 samples per channel.
924 * @param q pointer to the COOKContext
925 * @param inbuffer pointer to the inbuffer
926 * @param outbuffer pointer to the outbuffer
928 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
929 const uint8_t *inbuffer, float *outbuffer)
931 int sub_packet_size = p->size;
934 // for (i = 0; i < sub_packet_size ; i++)
935 // av_log(q->avctx, AV_LOG_ERROR, "%02x", inbuffer[i]);
936 // av_log(q->avctx, AV_LOG_ERROR, "\n");
937 memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
938 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
940 if (p->joint_stereo) {
941 if ((ret = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
944 if ((ret = mono_decode(q, p, q->decode_buffer_1)) < 0)
947 if (p->num_channels == 2) {
948 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
949 if ((ret = mono_decode(q, p, q->decode_buffer_2)) < 0)
954 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
955 p->mono_previous_buffer1, outbuffer, p->ch_idx);
957 if (p->num_channels == 2)
959 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
960 p->mono_previous_buffer2, outbuffer, p->ch_idx + 1);
962 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
963 p->mono_previous_buffer2, outbuffer, p->ch_idx + 1);
968 * Cook frame decoding
970 * @param avctx pointer to the AVCodecContext
972 static int cook_decode_frame(AVCodecContext *avctx, void *data,
973 int *got_frame_ptr, AVPacket *avpkt)
975 const uint8_t *buf = avpkt->data;
976 int buf_size = avpkt->size;
977 COOKContext *q = avctx->priv_data;
978 float *samples = NULL;
983 if (buf_size < avctx->block_align)
986 /* get output buffer */
987 if (q->discarded_packets >= 2) {
988 q->frame.nb_samples = q->samples_per_channel;
989 if ((ret = avctx->get_buffer(avctx, &q->frame)) < 0) {
990 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
993 samples = (float *) q->frame.data[0];
996 /* estimate subpacket sizes */
997 q->subpacket[0].size = avctx->block_align;
999 for (i = 1; i < q->num_subpackets; i++) {
1000 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
1001 q->subpacket[0].size -= q->subpacket[i].size + 1;
1002 if (q->subpacket[0].size < 0) {
1003 av_log(avctx, AV_LOG_DEBUG,
1004 "frame subpacket size total > avctx->block_align!\n");
1005 return AVERROR_INVALIDDATA;
1009 /* decode supbackets */
1010 for (i = 0; i < q->num_subpackets; i++) {
1011 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
1012 q->subpacket[i].bits_per_subpdiv;
1013 q->subpacket[i].ch_idx = chidx;
1014 av_log(avctx, AV_LOG_DEBUG,
1015 "subpacket[%i] size %i js %i %i block_align %i\n",
1016 i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
1017 avctx->block_align);
1019 if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
1021 offset += q->subpacket[i].size;
1022 chidx += q->subpacket[i].num_channels;
1023 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
1024 i, q->subpacket[i].size * 8, get_bits_count(&q->gb));
1027 /* Discard the first two frames: no valid audio. */
1028 if (q->discarded_packets < 2) {
1029 q->discarded_packets++;
1031 return avctx->block_align;
1035 *(AVFrame *) data = q->frame;
1037 return avctx->block_align;
1041 static void dump_cook_context(COOKContext *q)
1044 #define PRINT(a, b) av_log(q->avctx, AV_LOG_ERROR, " %s = %d\n", a, b);
1045 av_log(q->avctx, AV_LOG_ERROR, "COOKextradata\n");
1046 av_log(q->avctx, AV_LOG_ERROR, "cookversion=%x\n", q->subpacket[0].cookversion);
1047 if (q->subpacket[0].cookversion > STEREO) {
1048 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1049 PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1051 av_log(q->avctx, AV_LOG_ERROR, "COOKContext\n");
1052 PRINT("nb_channels", q->nb_channels);
1053 PRINT("bit_rate", q->bit_rate);
1054 PRINT("sample_rate", q->sample_rate);
1055 PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1056 PRINT("samples_per_frame", q->subpacket[0].samples_per_frame);
1057 PRINT("subbands", q->subpacket[0].subbands);
1058 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1059 PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1060 PRINT("numvector_size", q->subpacket[0].numvector_size);
1061 PRINT("total_subbands", q->subpacket[0].total_subbands);
1065 static av_cold int cook_count_channels(unsigned int mask)
1069 for (i = 0; i < 32; i++)
1070 if (mask & (1 << i))
1076 * Cook initialization
1078 * @param avctx pointer to the AVCodecContext
1080 static av_cold int cook_decode_init(AVCodecContext *avctx)
1082 COOKContext *q = avctx->priv_data;
1083 const uint8_t *edata_ptr = avctx->extradata;
1084 const uint8_t *edata_ptr_end = edata_ptr + avctx->extradata_size;
1085 int extradata_size = avctx->extradata_size;
1087 unsigned int channel_mask = 0;
1091 /* Take care of the codec specific extradata. */
1092 if (extradata_size <= 0) {
1093 av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1094 return AVERROR_INVALIDDATA;
1096 av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1098 /* Take data from the AVCodecContext (RM container). */
1099 q->sample_rate = avctx->sample_rate;
1100 q->nb_channels = avctx->channels;
1101 q->bit_rate = avctx->bit_rate;
1102 if (!q->nb_channels) {
1103 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1104 return AVERROR_INVALIDDATA;
1107 /* Initialize RNG. */
1108 av_lfg_init(&q->random_state, 0);
1110 while (edata_ptr < edata_ptr_end) {
1111 /* 8 for mono, 16 for stereo, ? for multichannel
1112 Swap to right endianness so we don't need to care later on. */
1113 if (extradata_size >= 8) {
1114 q->subpacket[s].cookversion = bytestream_get_be32(&edata_ptr);
1115 q->subpacket[s].samples_per_frame = bytestream_get_be16(&edata_ptr);
1116 q->subpacket[s].subbands = bytestream_get_be16(&edata_ptr);
1117 extradata_size -= 8;
1119 if (extradata_size >= 8) {
1120 bytestream_get_be32(&edata_ptr); // Unknown unused
1121 q->subpacket[s].js_subband_start = bytestream_get_be16(&edata_ptr);
1122 q->subpacket[s].js_vlc_bits = bytestream_get_be16(&edata_ptr);
1123 extradata_size -= 8;
1126 /* Initialize extradata related variables. */
1127 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame / q->nb_channels;
1128 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1130 /* Initialize default data states. */
1131 q->subpacket[s].log2_numvector_size = 5;
1132 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1133 q->subpacket[s].num_channels = 1;
1135 /* Initialize version-dependent variables */
1137 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1138 q->subpacket[s].cookversion);
1139 q->subpacket[s].joint_stereo = 0;
1140 switch (q->subpacket[s].cookversion) {
1142 if (q->nb_channels != 1) {
1143 av_log_ask_for_sample(avctx, "Container channels != 1.\n");
1144 return AVERROR_PATCHWELCOME;
1146 av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1149 if (q->nb_channels != 1) {
1150 q->subpacket[s].bits_per_subpdiv = 1;
1151 q->subpacket[s].num_channels = 2;
1153 av_log(avctx, AV_LOG_DEBUG, "STEREO\n");
1156 if (q->nb_channels != 2) {
1157 av_log_ask_for_sample(avctx, "Container channels != 2.\n");
1158 return AVERROR_PATCHWELCOME;
1160 av_log(avctx, AV_LOG_DEBUG, "JOINT_STEREO\n");
1161 if (avctx->extradata_size >= 16) {
1162 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1163 q->subpacket[s].js_subband_start;
1164 q->subpacket[s].joint_stereo = 1;
1165 q->subpacket[s].num_channels = 2;
1167 if (q->subpacket[s].samples_per_channel > 256) {
1168 q->subpacket[s].log2_numvector_size = 6;
1170 if (q->subpacket[s].samples_per_channel > 512) {
1171 q->subpacket[s].log2_numvector_size = 7;
1175 av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1176 if (extradata_size >= 4)
1177 channel_mask |= q->subpacket[s].channel_mask = bytestream_get_be32(&edata_ptr);
1179 if (cook_count_channels(q->subpacket[s].channel_mask) > 1) {
1180 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1181 q->subpacket[s].js_subband_start;
1182 q->subpacket[s].joint_stereo = 1;
1183 q->subpacket[s].num_channels = 2;
1184 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame >> 1;
1186 if (q->subpacket[s].samples_per_channel > 256) {
1187 q->subpacket[s].log2_numvector_size = 6;
1189 if (q->subpacket[s].samples_per_channel > 512) {
1190 q->subpacket[s].log2_numvector_size = 7;
1193 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame;
1197 av_log_ask_for_sample(avctx, "Unknown Cook version.\n");
1198 return AVERROR_PATCHWELCOME;
1201 if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1202 av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1203 return AVERROR_INVALIDDATA;
1205 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1208 /* Initialize variable relations */
1209 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1211 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1212 if (q->subpacket[s].total_subbands > 53) {
1213 av_log_ask_for_sample(avctx, "total_subbands > 53\n");
1214 return AVERROR_PATCHWELCOME;
1217 if ((q->subpacket[s].js_vlc_bits > 6) ||
1218 (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1219 av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1220 q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1221 return AVERROR_INVALIDDATA;
1224 if (q->subpacket[s].subbands > 50) {
1225 av_log_ask_for_sample(avctx, "subbands > 50\n");
1226 return AVERROR_PATCHWELCOME;
1228 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1229 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1230 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1231 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1233 q->num_subpackets++;
1235 if (s > MAX_SUBPACKETS) {
1236 av_log_ask_for_sample(avctx, "Too many subpackets > 5\n");
1237 return AVERROR_PATCHWELCOME;
1240 /* Generate tables */
1243 init_cplscales_table(q);
1245 if ((ret = init_cook_vlc_tables(q)))
1249 if (avctx->block_align >= UINT_MAX / 2)
1250 return AVERROR(EINVAL);
1252 /* Pad the databuffer with:
1253 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1254 FF_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1255 q->decoded_bytes_buffer =
1256 av_mallocz(avctx->block_align
1257 + DECODE_BYTES_PAD1(avctx->block_align)
1258 + FF_INPUT_BUFFER_PADDING_SIZE);
1259 if (q->decoded_bytes_buffer == NULL)
1260 return AVERROR(ENOMEM);
1262 /* Initialize transform. */
1263 if ((ret = init_cook_mlt(q)))
1266 /* Initialize COOK signal arithmetic handling */
1268 q->scalar_dequant = scalar_dequant_float;
1269 q->decouple = decouple_float;
1270 q->imlt_window = imlt_window_float;
1271 q->interpolate = interpolate_float;
1272 q->saturate_output = saturate_output_float;
1275 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1276 if ((q->samples_per_channel == 256) || (q->samples_per_channel == 512)
1277 || (q->samples_per_channel == 1024)) {
1279 av_log_ask_for_sample(avctx,
1280 "unknown amount of samples_per_channel = %d\n",
1281 q->samples_per_channel);
1282 return AVERROR_PATCHWELCOME;
1285 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
1287 avctx->channel_layout = channel_mask;
1289 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1291 avcodec_get_frame_defaults(&q->frame);
1292 avctx->coded_frame = &q->frame;
1295 dump_cook_context(q);
1300 AVCodec ff_cook_decoder = {
1302 .type = AVMEDIA_TYPE_AUDIO,
1303 .id = CODEC_ID_COOK,
1304 .priv_data_size = sizeof(COOKContext),
1305 .init = cook_decode_init,
1306 .close = cook_decode_close,
1307 .decode = cook_decode_frame,
1308 .capabilities = CODEC_CAP_DR1,
1309 .long_name = NULL_IF_CONFIG_SMALL("COOK"),