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/channel_layout.h"
46 #include "libavutil/lfg.h"
50 #include "bytestream.h"
57 /* the different Cook versions */
58 #define MONO 0x1000001
59 #define STEREO 0x1000002
60 #define JOINT_STEREO 0x1000003
61 #define MC_COOK 0x2000000 // multichannel Cook, not supported
63 #define SUBBAND_SIZE 20
64 #define MAX_SUBPACKETS 5
79 int samples_per_channel;
80 int log2_numvector_size;
81 unsigned int channel_mask;
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, float *out);
124 AVCodecContext* avctx;
129 int samples_per_channel;
132 int discarded_packets;
139 VLC envelope_quant_index[13];
140 VLC sqvh[7]; // scalar quantization
142 /* generatable tables and related variables */
143 int gain_size_factor;
144 float gain_table[23];
148 uint8_t* decoded_bytes_buffer;
149 DECLARE_ALIGNED(32, float, mono_mdct_output)[2048];
150 float decode_buffer_1[1024];
151 float decode_buffer_2[1024];
152 float decode_buffer_0[1060]; /* static allocation for joint decode */
154 const float *cplscales[5];
156 COOKSubpacket subpacket[MAX_SUBPACKETS];
159 static float pow2tab[127];
160 static float rootpow2tab[127];
162 /*************** init functions ***************/
164 /* table generator */
165 static av_cold void init_pow2table(void)
168 for (i = -63; i < 64; i++) {
169 pow2tab[63 + i] = pow(2, i);
170 rootpow2tab[63 + i] = sqrt(pow(2, i));
174 /* table generator */
175 static av_cold void init_gain_table(COOKContext *q)
178 q->gain_size_factor = q->samples_per_channel / 8;
179 for (i = 0; i < 23; i++)
180 q->gain_table[i] = pow(pow2tab[i + 52],
181 (1.0 / (double) q->gain_size_factor));
185 static av_cold int init_cook_vlc_tables(COOKContext *q)
190 for (i = 0; i < 13; i++) {
191 result |= init_vlc(&q->envelope_quant_index[i], 9, 24,
192 envelope_quant_index_huffbits[i], 1, 1,
193 envelope_quant_index_huffcodes[i], 2, 2, 0);
195 av_log(q->avctx, AV_LOG_DEBUG, "sqvh VLC init\n");
196 for (i = 0; i < 7; i++) {
197 result |= init_vlc(&q->sqvh[i], vhvlcsize_tab[i], vhsize_tab[i],
198 cvh_huffbits[i], 1, 1,
199 cvh_huffcodes[i], 2, 2, 0);
202 for (i = 0; i < q->num_subpackets; i++) {
203 if (q->subpacket[i].joint_stereo == 1) {
204 result |= init_vlc(&q->subpacket[i].channel_coupling, 6,
205 (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)
219 int mlt_size = q->samples_per_channel;
221 if ((q->mlt_window = av_malloc(mlt_size * sizeof(*q->mlt_window))) == 0)
222 return AVERROR(ENOMEM);
224 /* Initialize the MLT window: simple sine window. */
225 ff_sine_window_init(q->mlt_window, mlt_size);
226 for (j = 0; j < mlt_size; j++)
227 q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
229 /* Initialize the MDCT. */
230 if ((ret = ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size) + 1, 1, 1.0 / 32768.0))) {
231 av_free(q->mlt_window);
234 av_log(q->avctx, AV_LOG_DEBUG, "MDCT initialized, order = %d.\n",
235 av_log2(mlt_size) + 1);
240 static av_cold void init_cplscales_table(COOKContext *q)
243 for (i = 0; i < 5; i++)
244 q->cplscales[i] = cplscales[i];
247 /*************** init functions end ***********/
249 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes) + 3) % 4)
250 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
253 * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
254 * Why? No idea, some checksum/error detection method maybe.
256 * Out buffer size: extra bytes are needed to cope with
257 * padding/misalignment.
258 * Subpackets passed to the decoder can contain two, consecutive
259 * half-subpackets, of identical but arbitrary size.
260 * 1234 1234 1234 1234 extraA extraB
261 * Case 1: AAAA BBBB 0 0
262 * Case 2: AAAA ABBB BB-- 3 3
263 * Case 3: AAAA AABB BBBB 2 2
264 * Case 4: AAAA AAAB BBBB BB-- 1 5
266 * Nice way to waste CPU cycles.
268 * @param inbuffer pointer to byte array of indata
269 * @param out pointer to byte array of outdata
270 * @param bytes number of bytes
272 static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
274 static const uint32_t tab[4] = {
275 AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
276 AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
281 uint32_t *obuf = (uint32_t *) out;
282 /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
283 * I'm too lazy though, should be something like
284 * for (i = 0; i < bitamount / 64; i++)
285 * (int64_t) out[i] = 0x37c511f237c511f2 ^ av_be2ne64(int64_t) in[i]);
286 * Buffer alignment needs to be checked. */
288 off = (intptr_t) inbuffer & 3;
289 buf = (const uint32_t *) (inbuffer - off);
292 for (i = 0; i < bytes / 4; i++)
293 obuf[i] = c ^ buf[i];
298 static av_cold int cook_decode_close(AVCodecContext *avctx)
301 COOKContext *q = avctx->priv_data;
302 av_log(avctx, AV_LOG_DEBUG, "Deallocating memory.\n");
304 /* Free allocated memory buffers. */
305 av_free(q->mlt_window);
306 av_free(q->decoded_bytes_buffer);
308 /* Free the transform. */
309 ff_mdct_end(&q->mdct_ctx);
311 /* Free the VLC tables. */
312 for (i = 0; i < 13; i++)
313 ff_free_vlc(&q->envelope_quant_index[i]);
314 for (i = 0; i < 7; i++)
315 ff_free_vlc(&q->sqvh[i]);
316 for (i = 0; i < q->num_subpackets; i++)
317 ff_free_vlc(&q->subpacket[i].channel_coupling);
319 av_log(avctx, AV_LOG_DEBUG, "Memory deallocated.\n");
325 * Fill the gain array for the timedomain quantization.
327 * @param gb pointer to the GetBitContext
328 * @param gaininfo array[9] of gain indexes
330 static void decode_gain_info(GetBitContext *gb, int *gaininfo)
334 while (get_bits1(gb)) {
338 n = get_bits_count(gb) - 1; // amount of elements*2 to update
342 int index = get_bits(gb, 3);
343 int gain = get_bits1(gb) ? get_bits(gb, 4) - 7 : -1;
346 gaininfo[i++] = gain;
353 * Create the quant index table needed for the envelope.
355 * @param q pointer to the COOKContext
356 * @param quant_index_table pointer to the array
358 static int decode_envelope(COOKContext *q, COOKSubpacket *p,
359 int *quant_index_table)
363 quant_index_table[0] = get_bits(&q->gb, 6) - 6; // This is used later in categorize
365 for (i = 1; i < p->total_subbands; i++) {
367 if (i >= p->js_subband_start * 2) {
368 vlc_index -= p->js_subband_start;
375 vlc_index = 13; // the VLC tables >13 are identical to No. 13
377 j = get_vlc2(&q->gb, q->envelope_quant_index[vlc_index - 1].table,
378 q->envelope_quant_index[vlc_index - 1].bits, 2);
379 quant_index_table[i] = quant_index_table[i - 1] + j - 12; // differential encoding
380 if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
381 av_log(q->avctx, AV_LOG_ERROR,
382 "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
383 quant_index_table[i], i);
384 return AVERROR_INVALIDDATA;
392 * Calculate the category and category_index vector.
394 * @param q pointer to the COOKContext
395 * @param quant_index_table pointer to the array
396 * @param category pointer to the category array
397 * @param category_index pointer to the category_index array
399 static void categorize(COOKContext *q, COOKSubpacket *p, const int *quant_index_table,
400 int *category, int *category_index)
402 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
403 int exp_index2[102] = { 0 };
404 int exp_index1[102] = { 0 };
406 int tmp_categorize_array[128 * 2] = { 0 };
407 int tmp_categorize_array1_idx = p->numvector_size;
408 int tmp_categorize_array2_idx = p->numvector_size;
410 bits_left = p->bits_per_subpacket - get_bits_count(&q->gb);
412 if (bits_left > q->samples_per_channel)
413 bits_left = q->samples_per_channel +
414 ((bits_left - q->samples_per_channel) * 5) / 8;
419 for (i = 32; i > 0; i = i / 2) {
422 for (j = p->total_subbands; j > 0; j--) {
423 exp_idx = av_clip((i - quant_index_table[index] + bias) / 2, 0, 7);
425 num_bits += expbits_tab[exp_idx];
427 if (num_bits >= bits_left - 32)
431 /* Calculate total number of bits. */
433 for (i = 0; i < p->total_subbands; i++) {
434 exp_idx = av_clip((bias - quant_index_table[i]) / 2, 0, 7);
435 num_bits += expbits_tab[exp_idx];
436 exp_index1[i] = exp_idx;
437 exp_index2[i] = exp_idx;
439 tmpbias1 = tmpbias2 = num_bits;
441 for (j = 1; j < p->numvector_size; j++) {
442 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
445 for (i = 0; i < p->total_subbands; i++) {
446 if (exp_index1[i] < 7) {
447 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
456 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
457 tmpbias1 -= expbits_tab[exp_index1[index]] -
458 expbits_tab[exp_index1[index] + 1];
463 for (i = 0; i < p->total_subbands; i++) {
464 if (exp_index2[i] > 0) {
465 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
474 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
475 tmpbias2 -= expbits_tab[exp_index2[index]] -
476 expbits_tab[exp_index2[index] - 1];
481 for (i = 0; i < p->total_subbands; i++)
482 category[i] = exp_index2[i];
484 for (i = 0; i < p->numvector_size - 1; i++)
485 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
490 * Expand the category vector.
492 * @param q pointer to the COOKContext
493 * @param category pointer to the category array
494 * @param category_index pointer to the category_index array
496 static inline void expand_category(COOKContext *q, int *category,
500 for (i = 0; i < q->num_vectors; i++)
502 int idx = category_index[i];
503 if (++category[idx] >= FF_ARRAY_ELEMS(dither_tab))
509 * The real requantization of the mltcoefs
511 * @param q pointer to the COOKContext
513 * @param quant_index quantisation index
514 * @param subband_coef_index array of indexes to quant_centroid_tab
515 * @param subband_coef_sign signs of coefficients
516 * @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,
525 for (i = 0; i < SUBBAND_SIZE; i++) {
526 if (subband_coef_index[i]) {
527 f1 = quant_centroid_tab[index][subband_coef_index[i]];
528 if (subband_coef_sign[i])
531 /* noise coding if subband_coef_index[i] == 0 */
532 f1 = dither_tab[index];
533 if (av_lfg_get(&q->random_state) < 0x80000000)
536 mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
540 * Unpack the subband_coef_index and subband_coef_sign vectors.
542 * @param q pointer to the COOKContext
543 * @param category pointer to the category array
544 * @param subband_coef_index array of indexes to quant_centroid_tab
545 * @param subband_coef_sign signs of coefficients
547 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
548 int *subband_coef_index, int *subband_coef_sign)
551 int vlc, vd, tmp, result;
553 vd = vd_tab[category];
555 for (i = 0; i < vpr_tab[category]; i++) {
556 vlc = get_vlc2(&q->gb, q->sqvh[category].table, q->sqvh[category].bits, 3);
557 if (p->bits_per_subpacket < get_bits_count(&q->gb)) {
561 for (j = vd - 1; j >= 0; j--) {
562 tmp = (vlc * invradix_tab[category]) / 0x100000;
563 subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
566 for (j = 0; j < vd; j++) {
567 if (subband_coef_index[i * vd + j]) {
568 if (get_bits_count(&q->gb) < p->bits_per_subpacket) {
569 subband_coef_sign[i * vd + j] = get_bits1(&q->gb);
572 subband_coef_sign[i * vd + j] = 0;
575 subband_coef_sign[i * vd + j] = 0;
584 * Fill the mlt_buffer with mlt coefficients.
586 * @param q pointer to the COOKContext
587 * @param category pointer to the category array
588 * @param quant_index_table pointer to the array
589 * @param mlt_buffer pointer to mlt coefficients
591 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
592 int *quant_index_table, float *mlt_buffer)
594 /* A zero in this table means that the subband coefficient is
595 random noise coded. */
596 int subband_coef_index[SUBBAND_SIZE];
597 /* A zero in this table means that the subband coefficient is a
598 positive multiplicator. */
599 int subband_coef_sign[SUBBAND_SIZE];
603 for (band = 0; band < p->total_subbands; band++) {
604 index = category[band];
605 if (category[band] < 7) {
606 if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
608 for (j = 0; j < p->total_subbands; j++)
609 category[band + j] = 7;
613 memset(subband_coef_index, 0, sizeof(subband_coef_index));
614 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
616 q->scalar_dequant(q, index, quant_index_table[band],
617 subband_coef_index, subband_coef_sign,
618 &mlt_buffer[band * SUBBAND_SIZE]);
621 /* FIXME: should this be removed, or moved into loop above? */
622 if (p->total_subbands * SUBBAND_SIZE >= q->samples_per_channel)
627 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
629 int category_index[128] = { 0 };
630 int category[128] = { 0 };
631 int quant_index_table[102];
634 if ((res = decode_envelope(q, p, quant_index_table)) < 0)
636 q->num_vectors = get_bits(&q->gb, p->log2_numvector_size);
637 categorize(q, p, quant_index_table, category, category_index);
638 expand_category(q, category, category_index);
639 for (i=0; i<p->total_subbands; i++) {
641 return AVERROR_INVALIDDATA;
643 decode_vectors(q, p, category, quant_index_table, mlt_buffer);
650 * the actual requantization of the timedomain samples
652 * @param q pointer to the COOKContext
653 * @param buffer pointer to the timedomain buffer
654 * @param gain_index index for the block multiplier
655 * @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)
662 fc1 = pow2tab[gain_index + 63];
664 if (gain_index == gain_index_next) { // static gain
665 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
684 static void imlt_window_float(COOKContext *q, float *inbuffer,
685 cook_gains *gains_ptr, float *previous_buffer)
687 const float fc = pow2tab[gains_ptr->previous[0] + 63];
689 /* The weird thing here, is that the two halves of the time domain
690 * buffer are swapped. Also, the newest data, that we save away for
691 * next frame, has the wrong sign. Hence the subtraction below.
692 * Almost sounds like a complex conjugate/reverse data/FFT effect.
695 /* Apply window and overlap */
696 for (i = 0; i < q->samples_per_channel; i++)
697 inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
698 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
702 * The modulated lapped transform, this takes transform coefficients
703 * and transforms them into timedomain samples.
704 * Apply transform window, overlap buffers, apply gain profile
705 * and buffer management.
707 * @param q pointer to the COOKContext
708 * @param inbuffer pointer to the mltcoefficients
709 * @param gains_ptr current and previous gains
710 * @param previous_buffer pointer to the previous buffer to be used for overlapping
712 static void imlt_gain(COOKContext *q, float *inbuffer,
713 cook_gains *gains_ptr, float *previous_buffer)
715 float *buffer0 = q->mono_mdct_output;
716 float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
719 /* Inverse modified discrete cosine transform */
720 q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
722 q->imlt_window(q, buffer1, gains_ptr, previous_buffer);
724 /* Apply gain profile */
725 for (i = 0; i < 8; i++)
726 if (gains_ptr->now[i] || gains_ptr->now[i + 1])
727 q->interpolate(q, &buffer1[q->gain_size_factor * i],
728 gains_ptr->now[i], gains_ptr->now[i + 1]);
730 /* Save away the current to be previous block. */
731 memcpy(previous_buffer, buffer0,
732 q->samples_per_channel * sizeof(*previous_buffer));
737 * function for getting the jointstereo coupling information
739 * @param q pointer to the COOKContext
740 * @param decouple_tab decoupling array
742 static int decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
745 int vlc = get_bits1(&q->gb);
746 int start = cplband[p->js_subband_start];
747 int end = cplband[p->subbands - 1];
748 int length = end - start + 1;
754 for (i = 0; i < length; i++)
755 decouple_tab[start + i] = get_vlc2(&q->gb,
756 p->channel_coupling.table,
757 p->channel_coupling.bits, 2);
759 for (i = 0; i < length; i++) {
760 int v = get_bits(&q->gb, p->js_vlc_bits);
761 if (v == (1<<p->js_vlc_bits)-1) {
762 av_log(q->avctx, AV_LOG_ERROR, "decouple value too large\n");
763 return AVERROR_INVALIDDATA;
765 decouple_tab[start + i] = v;
771 * function decouples a pair of signals from a single signal via multiplication.
773 * @param q pointer to the COOKContext
774 * @param subband index of the current subband
775 * @param f1 multiplier for channel 1 extraction
776 * @param f2 multiplier for channel 2 extraction
777 * @param decode_buffer input buffer
778 * @param mlt_buffer1 pointer to left channel mlt coefficients
779 * @param mlt_buffer2 pointer to right channel mlt coefficients
781 static void decouple_float(COOKContext *q,
785 float *decode_buffer,
786 float *mlt_buffer1, float *mlt_buffer2)
789 for (j = 0; j < SUBBAND_SIZE; j++) {
790 tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
791 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
792 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
797 * function for decoding joint stereo data
799 * @param q pointer to the COOKContext
800 * @param mlt_buffer1 pointer to left channel mlt coefficients
801 * @param mlt_buffer2 pointer to right channel mlt coefficients
803 static int joint_decode(COOKContext *q, COOKSubpacket *p,
804 float *mlt_buffer_left, float *mlt_buffer_right)
807 int decouple_tab[SUBBAND_SIZE] = { 0 };
808 float *decode_buffer = q->decode_buffer_0;
811 const float *cplscale;
813 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
815 /* Make sure the buffers are zeroed out. */
816 memset(mlt_buffer_left, 0, 1024 * sizeof(*mlt_buffer_left));
817 memset(mlt_buffer_right, 0, 1024 * sizeof(*mlt_buffer_right));
818 if ((res = decouple_info(q, p, decouple_tab)) < 0)
820 if ((res = mono_decode(q, p, decode_buffer)) < 0)
822 /* The two channels are stored interleaved in decode_buffer. */
823 for (i = 0; i < p->js_subband_start; i++) {
824 for (j = 0; j < SUBBAND_SIZE; j++) {
825 mlt_buffer_left[i * 20 + j] = decode_buffer[i * 40 + j];
826 mlt_buffer_right[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
830 /* When we reach js_subband_start (the higher frequencies)
831 the coefficients are stored in a coupling scheme. */
832 idx = (1 << p->js_vlc_bits) - 1;
833 for (i = p->js_subband_start; i < p->subbands; i++) {
834 cpl_tmp = cplband[i];
835 idx -= decouple_tab[cpl_tmp];
836 cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
837 f1 = cplscale[decouple_tab[cpl_tmp] + 1];
839 q->decouple(q, p, i, f1, f2, decode_buffer,
840 mlt_buffer_left, mlt_buffer_right);
841 idx = (1 << p->js_vlc_bits) - 1;
848 * First part of subpacket decoding:
849 * decode raw stream bytes and read gain info.
851 * @param q pointer to the COOKContext
852 * @param inbuffer pointer to raw stream data
853 * @param gains_ptr array of current/prev gain pointers
855 static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
856 const uint8_t *inbuffer,
857 cook_gains *gains_ptr)
861 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
862 p->bits_per_subpacket / 8);
863 init_get_bits(&q->gb, q->decoded_bytes_buffer + offset,
864 p->bits_per_subpacket);
865 decode_gain_info(&q->gb, gains_ptr->now);
867 /* Swap current and previous gains */
868 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
872 * Saturate the output signal and interleave.
874 * @param q pointer to the COOKContext
875 * @param out pointer to the output vector
877 static void saturate_output_float(COOKContext *q, float *out)
879 q->dsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
880 -1.0f, 1.0f, FFALIGN(q->samples_per_channel, 8));
885 * Final part of subpacket decoding:
886 * Apply modulated lapped transform, gain compensation,
887 * clip and convert to integer.
889 * @param q pointer to the COOKContext
890 * @param decode_buffer pointer to the mlt coefficients
891 * @param gains_ptr array of current/prev gain pointers
892 * @param previous_buffer pointer to the previous buffer to be used for overlapping
893 * @param out pointer to the output buffer
895 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
896 cook_gains *gains_ptr, float *previous_buffer,
899 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
901 q->saturate_output(q, out);
906 * Cook subpacket decoding. This function returns one decoded subpacket,
907 * usually 1024 samples per channel.
909 * @param q pointer to the COOKContext
910 * @param inbuffer pointer to the inbuffer
911 * @param outbuffer pointer to the outbuffer
913 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
914 const uint8_t *inbuffer, float **outbuffer)
916 int sub_packet_size = p->size;
919 memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
920 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
922 if (p->joint_stereo) {
923 if ((res = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
926 if ((res = mono_decode(q, p, q->decode_buffer_1)) < 0)
929 if (p->num_channels == 2) {
930 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
931 if ((res = mono_decode(q, p, q->decode_buffer_2)) < 0)
936 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
937 p->mono_previous_buffer1,
938 outbuffer ? outbuffer[p->ch_idx] : NULL);
940 if (p->num_channels == 2) {
942 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
943 p->mono_previous_buffer2,
944 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
946 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
947 p->mono_previous_buffer2,
948 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
955 static int cook_decode_frame(AVCodecContext *avctx, void *data,
956 int *got_frame_ptr, AVPacket *avpkt)
958 AVFrame *frame = data;
959 const uint8_t *buf = avpkt->data;
960 int buf_size = avpkt->size;
961 COOKContext *q = avctx->priv_data;
962 float **samples = NULL;
967 if (buf_size < avctx->block_align)
970 /* get output buffer */
971 if (q->discarded_packets >= 2) {
972 frame->nb_samples = q->samples_per_channel;
973 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
975 samples = (float **)frame->extended_data;
978 /* estimate subpacket sizes */
979 q->subpacket[0].size = avctx->block_align;
981 for (i = 1; i < q->num_subpackets; i++) {
982 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
983 q->subpacket[0].size -= q->subpacket[i].size + 1;
984 if (q->subpacket[0].size < 0) {
985 av_log(avctx, AV_LOG_DEBUG,
986 "frame subpacket size total > avctx->block_align!\n");
987 return AVERROR_INVALIDDATA;
991 /* decode supbackets */
992 for (i = 0; i < q->num_subpackets; i++) {
993 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
994 q->subpacket[i].bits_per_subpdiv;
995 q->subpacket[i].ch_idx = chidx;
996 av_log(avctx, AV_LOG_DEBUG,
997 "subpacket[%i] size %i js %i %i block_align %i\n",
998 i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
1001 if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
1003 offset += q->subpacket[i].size;
1004 chidx += q->subpacket[i].num_channels;
1005 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
1006 i, q->subpacket[i].size * 8, get_bits_count(&q->gb));
1009 /* Discard the first two frames: no valid audio. */
1010 if (q->discarded_packets < 2) {
1011 q->discarded_packets++;
1013 return avctx->block_align;
1018 return avctx->block_align;
1022 static void dump_cook_context(COOKContext *q)
1025 #define PRINT(a, b) av_dlog(q->avctx, " %s = %d\n", a, b);
1026 av_dlog(q->avctx, "COOKextradata\n");
1027 av_dlog(q->avctx, "cookversion=%x\n", q->subpacket[0].cookversion);
1028 if (q->subpacket[0].cookversion > STEREO) {
1029 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1030 PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1032 av_dlog(q->avctx, "COOKContext\n");
1033 PRINT("nb_channels", q->avctx->channels);
1034 PRINT("bit_rate", q->avctx->bit_rate);
1035 PRINT("sample_rate", q->avctx->sample_rate);
1036 PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1037 PRINT("subbands", q->subpacket[0].subbands);
1038 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1039 PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1040 PRINT("numvector_size", q->subpacket[0].numvector_size);
1041 PRINT("total_subbands", q->subpacket[0].total_subbands);
1046 * Cook initialization
1048 * @param avctx pointer to the AVCodecContext
1050 static av_cold int cook_decode_init(AVCodecContext *avctx)
1052 COOKContext *q = avctx->priv_data;
1053 const uint8_t *edata_ptr = avctx->extradata;
1054 const uint8_t *edata_ptr_end = edata_ptr + avctx->extradata_size;
1055 int extradata_size = avctx->extradata_size;
1057 unsigned int channel_mask = 0;
1058 int samples_per_frame = 0;
1062 /* Take care of the codec specific extradata. */
1063 if (extradata_size <= 0) {
1064 av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1065 return AVERROR_INVALIDDATA;
1067 av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1069 /* Take data from the AVCodecContext (RM container). */
1070 if (!avctx->channels) {
1071 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1072 return AVERROR_INVALIDDATA;
1075 /* Initialize RNG. */
1076 av_lfg_init(&q->random_state, 0);
1078 ff_dsputil_init(&q->dsp, avctx);
1080 while (edata_ptr < edata_ptr_end) {
1081 /* 8 for mono, 16 for stereo, ? for multichannel
1082 Swap to right endianness so we don't need to care later on. */
1083 if (extradata_size >= 8) {
1084 q->subpacket[s].cookversion = bytestream_get_be32(&edata_ptr);
1085 samples_per_frame = bytestream_get_be16(&edata_ptr);
1086 q->subpacket[s].subbands = bytestream_get_be16(&edata_ptr);
1087 extradata_size -= 8;
1089 if (extradata_size >= 8) {
1090 bytestream_get_be32(&edata_ptr); // Unknown unused
1091 q->subpacket[s].js_subband_start = bytestream_get_be16(&edata_ptr);
1092 if (q->subpacket[s].js_subband_start >= 51) {
1093 av_log(avctx, AV_LOG_ERROR, "js_subband_start %d is too large\n", q->subpacket[s].js_subband_start);
1094 return AVERROR_INVALIDDATA;
1097 q->subpacket[s].js_vlc_bits = bytestream_get_be16(&edata_ptr);
1098 extradata_size -= 8;
1101 /* Initialize extradata related variables. */
1102 q->subpacket[s].samples_per_channel = samples_per_frame / avctx->channels;
1103 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1105 /* Initialize default data states. */
1106 q->subpacket[s].log2_numvector_size = 5;
1107 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1108 q->subpacket[s].num_channels = 1;
1110 /* Initialize version-dependent variables */
1112 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1113 q->subpacket[s].cookversion);
1114 q->subpacket[s].joint_stereo = 0;
1115 switch (q->subpacket[s].cookversion) {
1117 if (avctx->channels != 1) {
1118 avpriv_request_sample(avctx, "Container channels != 1");
1119 return AVERROR_PATCHWELCOME;
1121 av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1124 if (avctx->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 (avctx->channels != 2) {
1132 avpriv_request_sample(avctx, "Container channels != 2");
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 +
1138 q->subpacket[s].js_subband_start;
1139 q->subpacket[s].joint_stereo = 1;
1140 q->subpacket[s].num_channels = 2;
1142 if (q->subpacket[s].samples_per_channel > 256) {
1143 q->subpacket[s].log2_numvector_size = 6;
1145 if (q->subpacket[s].samples_per_channel > 512) {
1146 q->subpacket[s].log2_numvector_size = 7;
1150 av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1151 if (extradata_size >= 4)
1152 channel_mask |= q->subpacket[s].channel_mask = bytestream_get_be32(&edata_ptr);
1154 if (av_get_channel_layout_nb_channels(q->subpacket[s].channel_mask) > 1) {
1155 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1156 q->subpacket[s].js_subband_start;
1157 q->subpacket[s].joint_stereo = 1;
1158 q->subpacket[s].num_channels = 2;
1159 q->subpacket[s].samples_per_channel = samples_per_frame >> 1;
1161 if (q->subpacket[s].samples_per_channel > 256) {
1162 q->subpacket[s].log2_numvector_size = 6;
1164 if (q->subpacket[s].samples_per_channel > 512) {
1165 q->subpacket[s].log2_numvector_size = 7;
1168 q->subpacket[s].samples_per_channel = samples_per_frame;
1172 avpriv_request_sample(avctx, "Cook version %d",
1173 q->subpacket[s].cookversion);
1174 return AVERROR_PATCHWELCOME;
1177 if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1178 av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1179 return AVERROR_INVALIDDATA;
1181 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1184 /* Initialize variable relations */
1185 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1187 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1188 if (q->subpacket[s].total_subbands > 53) {
1189 avpriv_request_sample(avctx, "total_subbands > 53");
1190 return AVERROR_PATCHWELCOME;
1193 if ((q->subpacket[s].js_vlc_bits > 6) ||
1194 (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1195 av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1196 q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1197 return AVERROR_INVALIDDATA;
1200 if (q->subpacket[s].subbands > 50) {
1201 avpriv_request_sample(avctx, "subbands > 50");
1202 return AVERROR_PATCHWELCOME;
1204 if (q->subpacket[s].subbands == 0) {
1205 avpriv_request_sample(avctx, "subbands = 0");
1206 return AVERROR_PATCHWELCOME;
1208 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1209 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1210 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1211 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1213 if (q->num_subpackets + q->subpacket[s].num_channels > q->avctx->channels) {
1214 av_log(avctx, AV_LOG_ERROR, "Too many subpackets %d for channels %d\n", q->num_subpackets, q->avctx->channels);
1215 return AVERROR_INVALIDDATA;
1218 q->num_subpackets++;
1220 if (s > MAX_SUBPACKETS) {
1221 avpriv_request_sample(avctx, "subpackets > %d", MAX_SUBPACKETS);
1222 return AVERROR_PATCHWELCOME;
1225 /* Generate tables */
1228 init_cplscales_table(q);
1230 if ((ret = init_cook_vlc_tables(q)))
1234 if (avctx->block_align >= UINT_MAX / 2)
1235 return AVERROR(EINVAL);
1237 /* Pad the databuffer with:
1238 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1239 FF_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1240 q->decoded_bytes_buffer =
1241 av_mallocz(avctx->block_align
1242 + DECODE_BYTES_PAD1(avctx->block_align)
1243 + FF_INPUT_BUFFER_PADDING_SIZE);
1244 if (q->decoded_bytes_buffer == NULL)
1245 return AVERROR(ENOMEM);
1247 /* Initialize transform. */
1248 if ((ret = init_cook_mlt(q)))
1251 /* Initialize COOK signal arithmetic handling */
1253 q->scalar_dequant = scalar_dequant_float;
1254 q->decouple = decouple_float;
1255 q->imlt_window = imlt_window_float;
1256 q->interpolate = interpolate_float;
1257 q->saturate_output = saturate_output_float;
1260 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1261 if (q->samples_per_channel != 256 && q->samples_per_channel != 512 &&
1262 q->samples_per_channel != 1024) {
1263 avpriv_request_sample(avctx, "samples_per_channel = %d",
1264 q->samples_per_channel);
1265 return AVERROR_PATCHWELCOME;
1268 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1270 avctx->channel_layout = channel_mask;
1272 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1275 dump_cook_context(q);
1280 AVCodec ff_cook_decoder = {
1282 .type = AVMEDIA_TYPE_AUDIO,
1283 .id = AV_CODEC_ID_COOK,
1284 .priv_data_size = sizeof(COOKContext),
1285 .init = cook_decode_init,
1286 .close = cook_decode_close,
1287 .decode = cook_decode_frame,
1288 .capabilities = CODEC_CAP_DR1,
1289 .long_name = NULL_IF_CONFIG_SMALL("Cook / Cooker / Gecko (RealAudio G2)"),
1290 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1291 AV_SAMPLE_FMT_NONE },