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"
58 /* the different Cook versions */
59 #define MONO 0x1000001
60 #define STEREO 0x1000002
61 #define JOINT_STEREO 0x1000003
62 #define MC_COOK 0x2000000 // multichannel Cook, not supported
64 #define SUBBAND_SIZE 20
65 #define MAX_SUBPACKETS 5
80 int samples_per_channel;
81 int log2_numvector_size;
82 unsigned int channel_mask;
85 int bits_per_subpacket;
88 int numvector_size; // 1 << log2_numvector_size;
90 float mono_previous_buffer1[1024];
91 float mono_previous_buffer2[1024];
101 typedef struct cook {
103 * The following 5 functions provide the lowlevel arithmetic on
104 * the internal audio buffers.
106 void (*scalar_dequant)(struct cook *q, int index, int quant_index,
107 int *subband_coef_index, int *subband_coef_sign,
110 void (*decouple)(struct cook *q,
114 float *decode_buffer,
115 float *mlt_buffer1, float *mlt_buffer2);
117 void (*imlt_window)(struct cook *q, float *buffer1,
118 cook_gains *gains_ptr, float *previous_buffer);
120 void (*interpolate)(struct cook *q, float *buffer,
121 int gain_index, int gain_index_next);
123 void (*saturate_output)(struct cook *q, float *out);
125 AVCodecContext* avctx;
130 int samples_per_channel;
133 int discarded_packets;
140 VLC envelope_quant_index[13];
141 VLC sqvh[7]; // scalar quantization
143 /* generatable tables and related variables */
144 int gain_size_factor;
145 float gain_table[23];
149 uint8_t* decoded_bytes_buffer;
150 DECLARE_ALIGNED(32, float, mono_mdct_output)[2048];
151 float decode_buffer_1[1024];
152 float decode_buffer_2[1024];
153 float decode_buffer_0[1060]; /* static allocation for joint decode */
155 const float *cplscales[5];
157 COOKSubpacket subpacket[MAX_SUBPACKETS];
160 static float pow2tab[127];
161 static float rootpow2tab[127];
163 /*************** init functions ***************/
165 /* table generator */
166 static av_cold void init_pow2table(void)
169 for (i = -63; i < 64; i++) {
170 pow2tab[63 + i] = pow(2, i);
171 rootpow2tab[63 + i] = sqrt(pow(2, i));
175 /* table generator */
176 static av_cold void init_gain_table(COOKContext *q)
179 q->gain_size_factor = q->samples_per_channel / 8;
180 for (i = 0; i < 23; i++)
181 q->gain_table[i] = pow(pow2tab[i + 52],
182 (1.0 / (double) q->gain_size_factor));
186 static av_cold int init_cook_vlc_tables(COOKContext *q)
191 for (i = 0; i < 13; i++) {
192 result |= init_vlc(&q->envelope_quant_index[i], 9, 24,
193 envelope_quant_index_huffbits[i], 1, 1,
194 envelope_quant_index_huffcodes[i], 2, 2, 0);
196 av_log(q->avctx, AV_LOG_DEBUG, "sqvh VLC init\n");
197 for (i = 0; i < 7; i++) {
198 result |= init_vlc(&q->sqvh[i], vhvlcsize_tab[i], vhsize_tab[i],
199 cvh_huffbits[i], 1, 1,
200 cvh_huffcodes[i], 2, 2, 0);
203 for (i = 0; i < q->num_subpackets; i++) {
204 if (q->subpacket[i].joint_stereo == 1) {
205 result |= init_vlc(&q->subpacket[i].channel_coupling, 6,
206 (1 << q->subpacket[i].js_vlc_bits) - 1,
207 ccpl_huffbits[q->subpacket[i].js_vlc_bits - 2], 1, 1,
208 ccpl_huffcodes[q->subpacket[i].js_vlc_bits - 2], 2, 2, 0);
209 av_log(q->avctx, AV_LOG_DEBUG, "subpacket %i Joint-stereo VLC used.\n", i);
213 av_log(q->avctx, AV_LOG_DEBUG, "VLC tables initialized.\n");
217 static av_cold int init_cook_mlt(COOKContext *q)
220 int mlt_size = q->samples_per_channel;
222 if ((q->mlt_window = av_malloc(mlt_size * sizeof(*q->mlt_window))) == 0)
223 return AVERROR(ENOMEM);
225 /* Initialize the MLT window: simple sine window. */
226 ff_sine_window_init(q->mlt_window, mlt_size);
227 for (j = 0; j < mlt_size; j++)
228 q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
230 /* Initialize the MDCT. */
231 if ((ret = ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size) + 1, 1, 1.0 / 32768.0))) {
232 av_freep(&q->mlt_window);
235 av_log(q->avctx, AV_LOG_DEBUG, "MDCT initialized, order = %d.\n",
236 av_log2(mlt_size) + 1);
241 static av_cold void init_cplscales_table(COOKContext *q)
244 for (i = 0; i < 5; i++)
245 q->cplscales[i] = cplscales[i];
248 /*************** init functions end ***********/
250 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes) + 3) % 4)
251 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
254 * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
255 * Why? No idea, some checksum/error detection method maybe.
257 * Out buffer size: extra bytes are needed to cope with
258 * padding/misalignment.
259 * Subpackets passed to the decoder can contain two, consecutive
260 * half-subpackets, of identical but arbitrary size.
261 * 1234 1234 1234 1234 extraA extraB
262 * Case 1: AAAA BBBB 0 0
263 * Case 2: AAAA ABBB BB-- 3 3
264 * Case 3: AAAA AABB BBBB 2 2
265 * Case 4: AAAA AAAB BBBB BB-- 1 5
267 * Nice way to waste CPU cycles.
269 * @param inbuffer pointer to byte array of indata
270 * @param out pointer to byte array of outdata
271 * @param bytes number of bytes
273 static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
275 static const uint32_t tab[4] = {
276 AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
277 AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
282 uint32_t *obuf = (uint32_t *) out;
283 /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
284 * I'm too lazy though, should be something like
285 * for (i = 0; i < bitamount / 64; i++)
286 * (int64_t) out[i] = 0x37c511f237c511f2 ^ av_be2ne64(int64_t) in[i]);
287 * Buffer alignment needs to be checked. */
289 off = (intptr_t) inbuffer & 3;
290 buf = (const uint32_t *) (inbuffer - off);
293 for (i = 0; i < bytes / 4; i++)
294 obuf[i] = c ^ buf[i];
299 static av_cold int cook_decode_close(AVCodecContext *avctx)
302 COOKContext *q = avctx->priv_data;
303 av_log(avctx, AV_LOG_DEBUG, "Deallocating memory.\n");
305 /* Free allocated memory buffers. */
306 av_freep(&q->mlt_window);
307 av_freep(&q->decoded_bytes_buffer);
309 /* Free the transform. */
310 ff_mdct_end(&q->mdct_ctx);
312 /* Free the VLC tables. */
313 for (i = 0; i < 13; i++)
314 ff_free_vlc(&q->envelope_quant_index[i]);
315 for (i = 0; i < 7; i++)
316 ff_free_vlc(&q->sqvh[i]);
317 for (i = 0; i < q->num_subpackets; i++)
318 ff_free_vlc(&q->subpacket[i].channel_coupling);
320 av_log(avctx, AV_LOG_DEBUG, "Memory deallocated.\n");
326 * Fill the gain array for the timedomain quantization.
328 * @param gb pointer to the GetBitContext
329 * @param gaininfo array[9] of gain indexes
331 static void decode_gain_info(GetBitContext *gb, int *gaininfo)
335 n = get_unary(gb, 0, get_bits_left(gb)); // amount of elements*2 to update
339 int index = get_bits(gb, 3);
340 int gain = get_bits1(gb) ? get_bits(gb, 4) - 7 : -1;
343 gaininfo[i++] = gain;
350 * Create the quant index table needed for the envelope.
352 * @param q pointer to the COOKContext
353 * @param quant_index_table pointer to the array
355 static int decode_envelope(COOKContext *q, COOKSubpacket *p,
356 int *quant_index_table)
360 quant_index_table[0] = get_bits(&q->gb, 6) - 6; // This is used later in categorize
362 for (i = 1; i < p->total_subbands; i++) {
364 if (i >= p->js_subband_start * 2) {
365 vlc_index -= p->js_subband_start;
372 vlc_index = 13; // the VLC tables >13 are identical to No. 13
374 j = get_vlc2(&q->gb, q->envelope_quant_index[vlc_index - 1].table,
375 q->envelope_quant_index[vlc_index - 1].bits, 2);
376 quant_index_table[i] = quant_index_table[i - 1] + j - 12; // differential encoding
377 if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
378 av_log(q->avctx, AV_LOG_ERROR,
379 "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
380 quant_index_table[i], i);
381 return AVERROR_INVALIDDATA;
389 * Calculate the category and category_index vector.
391 * @param q pointer to the COOKContext
392 * @param quant_index_table pointer to the array
393 * @param category pointer to the category array
394 * @param category_index pointer to the category_index array
396 static void categorize(COOKContext *q, COOKSubpacket *p, const int *quant_index_table,
397 int *category, int *category_index)
399 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
400 int exp_index2[102] = { 0 };
401 int exp_index1[102] = { 0 };
403 int tmp_categorize_array[128 * 2] = { 0 };
404 int tmp_categorize_array1_idx = p->numvector_size;
405 int tmp_categorize_array2_idx = p->numvector_size;
407 bits_left = p->bits_per_subpacket - get_bits_count(&q->gb);
409 if (bits_left > q->samples_per_channel)
410 bits_left = q->samples_per_channel +
411 ((bits_left - q->samples_per_channel) * 5) / 8;
416 for (i = 32; i > 0; i = i / 2) {
419 for (j = p->total_subbands; j > 0; j--) {
420 exp_idx = av_clip((i - quant_index_table[index] + bias) / 2, 0, 7);
422 num_bits += expbits_tab[exp_idx];
424 if (num_bits >= bits_left - 32)
428 /* Calculate total number of bits. */
430 for (i = 0; i < p->total_subbands; i++) {
431 exp_idx = av_clip((bias - quant_index_table[i]) / 2, 0, 7);
432 num_bits += expbits_tab[exp_idx];
433 exp_index1[i] = exp_idx;
434 exp_index2[i] = exp_idx;
436 tmpbias1 = tmpbias2 = num_bits;
438 for (j = 1; j < p->numvector_size; j++) {
439 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
442 for (i = 0; i < p->total_subbands; i++) {
443 if (exp_index1[i] < 7) {
444 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
453 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
454 tmpbias1 -= expbits_tab[exp_index1[index]] -
455 expbits_tab[exp_index1[index] + 1];
460 for (i = 0; i < p->total_subbands; i++) {
461 if (exp_index2[i] > 0) {
462 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
471 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
472 tmpbias2 -= expbits_tab[exp_index2[index]] -
473 expbits_tab[exp_index2[index] - 1];
478 for (i = 0; i < p->total_subbands; i++)
479 category[i] = exp_index2[i];
481 for (i = 0; i < p->numvector_size - 1; i++)
482 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
487 * Expand the category vector.
489 * @param q pointer to the COOKContext
490 * @param category pointer to the category array
491 * @param category_index pointer to the category_index array
493 static inline void expand_category(COOKContext *q, int *category,
497 for (i = 0; i < q->num_vectors; i++)
499 int idx = category_index[i];
500 if (++category[idx] >= FF_ARRAY_ELEMS(dither_tab))
506 * The real requantization of the mltcoefs
508 * @param q pointer to the COOKContext
510 * @param quant_index quantisation index
511 * @param subband_coef_index array of indexes to quant_centroid_tab
512 * @param subband_coef_sign signs of coefficients
513 * @param mlt_p pointer into the mlt buffer
515 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
516 int *subband_coef_index, int *subband_coef_sign,
522 for (i = 0; i < SUBBAND_SIZE; i++) {
523 if (subband_coef_index[i]) {
524 f1 = quant_centroid_tab[index][subband_coef_index[i]];
525 if (subband_coef_sign[i])
528 /* noise coding if subband_coef_index[i] == 0 */
529 f1 = dither_tab[index];
530 if (av_lfg_get(&q->random_state) < 0x80000000)
533 mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
537 * Unpack the subband_coef_index and subband_coef_sign vectors.
539 * @param q pointer to the COOKContext
540 * @param category pointer to the category array
541 * @param subband_coef_index array of indexes to quant_centroid_tab
542 * @param subband_coef_sign signs of coefficients
544 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
545 int *subband_coef_index, int *subband_coef_sign)
548 int vlc, vd, tmp, result;
550 vd = vd_tab[category];
552 for (i = 0; i < vpr_tab[category]; i++) {
553 vlc = get_vlc2(&q->gb, q->sqvh[category].table, q->sqvh[category].bits, 3);
554 if (p->bits_per_subpacket < get_bits_count(&q->gb)) {
558 for (j = vd - 1; j >= 0; j--) {
559 tmp = (vlc * invradix_tab[category]) / 0x100000;
560 subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
563 for (j = 0; j < vd; j++) {
564 if (subband_coef_index[i * vd + j]) {
565 if (get_bits_count(&q->gb) < p->bits_per_subpacket) {
566 subband_coef_sign[i * vd + j] = get_bits1(&q->gb);
569 subband_coef_sign[i * vd + j] = 0;
572 subband_coef_sign[i * vd + j] = 0;
581 * Fill the mlt_buffer with mlt coefficients.
583 * @param q pointer to the COOKContext
584 * @param category pointer to the category array
585 * @param quant_index_table pointer to the array
586 * @param mlt_buffer pointer to mlt coefficients
588 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
589 int *quant_index_table, float *mlt_buffer)
591 /* A zero in this table means that the subband coefficient is
592 random noise coded. */
593 int subband_coef_index[SUBBAND_SIZE];
594 /* A zero in this table means that the subband coefficient is a
595 positive multiplicator. */
596 int subband_coef_sign[SUBBAND_SIZE];
600 for (band = 0; band < p->total_subbands; band++) {
601 index = category[band];
602 if (category[band] < 7) {
603 if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
605 for (j = 0; j < p->total_subbands; j++)
606 category[band + j] = 7;
610 memset(subband_coef_index, 0, sizeof(subband_coef_index));
611 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
613 q->scalar_dequant(q, index, quant_index_table[band],
614 subband_coef_index, subband_coef_sign,
615 &mlt_buffer[band * SUBBAND_SIZE]);
618 /* FIXME: should this be removed, or moved into loop above? */
619 if (p->total_subbands * SUBBAND_SIZE >= q->samples_per_channel)
624 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
626 int category_index[128] = { 0 };
627 int category[128] = { 0 };
628 int quant_index_table[102];
631 if ((res = decode_envelope(q, p, quant_index_table)) < 0)
633 q->num_vectors = get_bits(&q->gb, p->log2_numvector_size);
634 categorize(q, p, quant_index_table, category, category_index);
635 expand_category(q, category, category_index);
636 for (i=0; i<p->total_subbands; i++) {
638 return AVERROR_INVALIDDATA;
640 decode_vectors(q, p, category, quant_index_table, mlt_buffer);
647 * the actual requantization of the timedomain samples
649 * @param q pointer to the COOKContext
650 * @param buffer pointer to the timedomain buffer
651 * @param gain_index index for the block multiplier
652 * @param gain_index_next index for the next block multiplier
654 static void interpolate_float(COOKContext *q, float *buffer,
655 int gain_index, int gain_index_next)
659 fc1 = pow2tab[gain_index + 63];
661 if (gain_index == gain_index_next) { // static gain
662 for (i = 0; i < q->gain_size_factor; i++)
664 } else { // smooth gain
665 fc2 = q->gain_table[11 + (gain_index_next - gain_index)];
666 for (i = 0; i < q->gain_size_factor; i++) {
674 * Apply transform window, overlap buffers.
676 * @param q pointer to the COOKContext
677 * @param inbuffer pointer to the mltcoefficients
678 * @param gains_ptr current and previous gains
679 * @param previous_buffer pointer to the previous buffer to be used for overlapping
681 static void imlt_window_float(COOKContext *q, float *inbuffer,
682 cook_gains *gains_ptr, float *previous_buffer)
684 const float fc = pow2tab[gains_ptr->previous[0] + 63];
686 /* The weird thing here, is that the two halves of the time domain
687 * buffer are swapped. Also, the newest data, that we save away for
688 * next frame, has the wrong sign. Hence the subtraction below.
689 * Almost sounds like a complex conjugate/reverse data/FFT effect.
692 /* Apply window and overlap */
693 for (i = 0; i < q->samples_per_channel; i++)
694 inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
695 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
699 * The modulated lapped transform, this takes transform coefficients
700 * and transforms them into timedomain samples.
701 * Apply transform window, overlap buffers, apply gain profile
702 * and buffer management.
704 * @param q pointer to the COOKContext
705 * @param inbuffer pointer to the mltcoefficients
706 * @param gains_ptr current and previous gains
707 * @param previous_buffer pointer to the previous buffer to be used for overlapping
709 static void imlt_gain(COOKContext *q, float *inbuffer,
710 cook_gains *gains_ptr, float *previous_buffer)
712 float *buffer0 = q->mono_mdct_output;
713 float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
716 /* Inverse modified discrete cosine transform */
717 q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
719 q->imlt_window(q, buffer1, gains_ptr, previous_buffer);
721 /* Apply gain profile */
722 for (i = 0; i < 8; i++)
723 if (gains_ptr->now[i] || gains_ptr->now[i + 1])
724 q->interpolate(q, &buffer1[q->gain_size_factor * i],
725 gains_ptr->now[i], gains_ptr->now[i + 1]);
727 /* Save away the current to be previous block. */
728 memcpy(previous_buffer, buffer0,
729 q->samples_per_channel * sizeof(*previous_buffer));
734 * function for getting the jointstereo coupling information
736 * @param q pointer to the COOKContext
737 * @param decouple_tab decoupling array
739 static int decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
742 int vlc = get_bits1(&q->gb);
743 int start = cplband[p->js_subband_start];
744 int end = cplband[p->subbands - 1];
745 int length = end - start + 1;
751 for (i = 0; i < length; i++)
752 decouple_tab[start + i] = get_vlc2(&q->gb,
753 p->channel_coupling.table,
754 p->channel_coupling.bits, 2);
756 for (i = 0; i < length; i++) {
757 int v = get_bits(&q->gb, p->js_vlc_bits);
758 if (v == (1<<p->js_vlc_bits)-1) {
759 av_log(q->avctx, AV_LOG_ERROR, "decouple value too large\n");
760 return AVERROR_INVALIDDATA;
762 decouple_tab[start + i] = v;
768 * function decouples a pair of signals from a single signal via multiplication.
770 * @param q pointer to the COOKContext
771 * @param subband index of the current subband
772 * @param f1 multiplier for channel 1 extraction
773 * @param f2 multiplier for channel 2 extraction
774 * @param decode_buffer input buffer
775 * @param mlt_buffer1 pointer to left channel mlt coefficients
776 * @param mlt_buffer2 pointer to right channel mlt coefficients
778 static void decouple_float(COOKContext *q,
782 float *decode_buffer,
783 float *mlt_buffer1, float *mlt_buffer2)
786 for (j = 0; j < SUBBAND_SIZE; j++) {
787 tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
788 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
789 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
794 * function for decoding joint stereo data
796 * @param q pointer to the COOKContext
797 * @param mlt_buffer1 pointer to left channel mlt coefficients
798 * @param mlt_buffer2 pointer to right channel mlt coefficients
800 static int joint_decode(COOKContext *q, COOKSubpacket *p,
801 float *mlt_buffer_left, float *mlt_buffer_right)
804 int decouple_tab[SUBBAND_SIZE] = { 0 };
805 float *decode_buffer = q->decode_buffer_0;
808 const float *cplscale;
810 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
812 /* Make sure the buffers are zeroed out. */
813 memset(mlt_buffer_left, 0, 1024 * sizeof(*mlt_buffer_left));
814 memset(mlt_buffer_right, 0, 1024 * sizeof(*mlt_buffer_right));
815 if ((res = decouple_info(q, p, decouple_tab)) < 0)
817 if ((res = mono_decode(q, p, decode_buffer)) < 0)
819 /* The two channels are stored interleaved in decode_buffer. */
820 for (i = 0; i < p->js_subband_start; i++) {
821 for (j = 0; j < SUBBAND_SIZE; j++) {
822 mlt_buffer_left[i * 20 + j] = decode_buffer[i * 40 + j];
823 mlt_buffer_right[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
827 /* When we reach js_subband_start (the higher frequencies)
828 the coefficients are stored in a coupling scheme. */
829 idx = (1 << p->js_vlc_bits) - 1;
830 for (i = p->js_subband_start; i < p->subbands; i++) {
831 cpl_tmp = cplband[i];
832 idx -= decouple_tab[cpl_tmp];
833 cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
834 f1 = cplscale[decouple_tab[cpl_tmp] + 1];
836 q->decouple(q, p, i, f1, f2, decode_buffer,
837 mlt_buffer_left, mlt_buffer_right);
838 idx = (1 << p->js_vlc_bits) - 1;
845 * First part of subpacket decoding:
846 * decode raw stream bytes and read gain info.
848 * @param q pointer to the COOKContext
849 * @param inbuffer pointer to raw stream data
850 * @param gains_ptr array of current/prev gain pointers
852 static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
853 const uint8_t *inbuffer,
854 cook_gains *gains_ptr)
858 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
859 p->bits_per_subpacket / 8);
860 init_get_bits(&q->gb, q->decoded_bytes_buffer + offset,
861 p->bits_per_subpacket);
862 decode_gain_info(&q->gb, gains_ptr->now);
864 /* Swap current and previous gains */
865 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
869 * Saturate the output signal and interleave.
871 * @param q pointer to the COOKContext
872 * @param out pointer to the output vector
874 static void saturate_output_float(COOKContext *q, float *out)
876 q->dsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
877 -1.0f, 1.0f, FFALIGN(q->samples_per_channel, 8));
882 * Final part of subpacket decoding:
883 * Apply modulated lapped transform, gain compensation,
884 * clip and convert to integer.
886 * @param q pointer to the COOKContext
887 * @param decode_buffer pointer to the mlt coefficients
888 * @param gains_ptr array of current/prev gain pointers
889 * @param previous_buffer pointer to the previous buffer to be used for overlapping
890 * @param out pointer to the output buffer
892 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
893 cook_gains *gains_ptr, float *previous_buffer,
896 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
898 q->saturate_output(q, out);
903 * Cook subpacket decoding. This function returns one decoded subpacket,
904 * usually 1024 samples per channel.
906 * @param q pointer to the COOKContext
907 * @param inbuffer pointer to the inbuffer
908 * @param outbuffer pointer to the outbuffer
910 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
911 const uint8_t *inbuffer, float **outbuffer)
913 int sub_packet_size = p->size;
916 memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
917 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
919 if (p->joint_stereo) {
920 if ((res = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
923 if ((res = mono_decode(q, p, q->decode_buffer_1)) < 0)
926 if (p->num_channels == 2) {
927 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
928 if ((res = mono_decode(q, p, q->decode_buffer_2)) < 0)
933 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
934 p->mono_previous_buffer1,
935 outbuffer ? outbuffer[p->ch_idx] : NULL);
937 if (p->num_channels == 2) {
939 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
940 p->mono_previous_buffer2,
941 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
943 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
944 p->mono_previous_buffer2,
945 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
952 static int cook_decode_frame(AVCodecContext *avctx, void *data,
953 int *got_frame_ptr, AVPacket *avpkt)
955 AVFrame *frame = data;
956 const uint8_t *buf = avpkt->data;
957 int buf_size = avpkt->size;
958 COOKContext *q = avctx->priv_data;
959 float **samples = NULL;
964 if (buf_size < avctx->block_align)
967 /* get output buffer */
968 if (q->discarded_packets >= 2) {
969 frame->nb_samples = q->samples_per_channel;
970 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
972 samples = (float **)frame->extended_data;
975 /* estimate subpacket sizes */
976 q->subpacket[0].size = avctx->block_align;
978 for (i = 1; i < q->num_subpackets; i++) {
979 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
980 q->subpacket[0].size -= q->subpacket[i].size + 1;
981 if (q->subpacket[0].size < 0) {
982 av_log(avctx, AV_LOG_DEBUG,
983 "frame subpacket size total > avctx->block_align!\n");
984 return AVERROR_INVALIDDATA;
988 /* decode supbackets */
989 for (i = 0; i < q->num_subpackets; i++) {
990 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
991 q->subpacket[i].bits_per_subpdiv;
992 q->subpacket[i].ch_idx = chidx;
993 av_log(avctx, AV_LOG_DEBUG,
994 "subpacket[%i] size %i js %i %i block_align %i\n",
995 i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
998 if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
1000 offset += q->subpacket[i].size;
1001 chidx += q->subpacket[i].num_channels;
1002 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
1003 i, q->subpacket[i].size * 8, get_bits_count(&q->gb));
1006 /* Discard the first two frames: no valid audio. */
1007 if (q->discarded_packets < 2) {
1008 q->discarded_packets++;
1010 return avctx->block_align;
1015 return avctx->block_align;
1019 static void dump_cook_context(COOKContext *q)
1022 #define PRINT(a, b) av_dlog(q->avctx, " %s = %d\n", a, b);
1023 av_dlog(q->avctx, "COOKextradata\n");
1024 av_dlog(q->avctx, "cookversion=%x\n", q->subpacket[0].cookversion);
1025 if (q->subpacket[0].cookversion > STEREO) {
1026 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1027 PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1029 av_dlog(q->avctx, "COOKContext\n");
1030 PRINT("nb_channels", q->avctx->channels);
1031 PRINT("bit_rate", q->avctx->bit_rate);
1032 PRINT("sample_rate", q->avctx->sample_rate);
1033 PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1034 PRINT("subbands", q->subpacket[0].subbands);
1035 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1036 PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1037 PRINT("numvector_size", q->subpacket[0].numvector_size);
1038 PRINT("total_subbands", q->subpacket[0].total_subbands);
1043 * Cook initialization
1045 * @param avctx pointer to the AVCodecContext
1047 static av_cold int cook_decode_init(AVCodecContext *avctx)
1049 COOKContext *q = avctx->priv_data;
1050 const uint8_t *edata_ptr = avctx->extradata;
1051 const uint8_t *edata_ptr_end = edata_ptr + avctx->extradata_size;
1052 int extradata_size = avctx->extradata_size;
1054 unsigned int channel_mask = 0;
1055 int samples_per_frame = 0;
1059 /* Take care of the codec specific extradata. */
1060 if (extradata_size <= 0) {
1061 av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1062 return AVERROR_INVALIDDATA;
1064 av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1066 /* Take data from the AVCodecContext (RM container). */
1067 if (!avctx->channels) {
1068 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1069 return AVERROR_INVALIDDATA;
1072 /* Initialize RNG. */
1073 av_lfg_init(&q->random_state, 0);
1075 ff_dsputil_init(&q->dsp, avctx);
1077 while (edata_ptr < edata_ptr_end) {
1078 /* 8 for mono, 16 for stereo, ? for multichannel
1079 Swap to right endianness so we don't need to care later on. */
1080 if (extradata_size >= 8) {
1081 q->subpacket[s].cookversion = bytestream_get_be32(&edata_ptr);
1082 samples_per_frame = bytestream_get_be16(&edata_ptr);
1083 q->subpacket[s].subbands = bytestream_get_be16(&edata_ptr);
1084 extradata_size -= 8;
1086 if (extradata_size >= 8) {
1087 bytestream_get_be32(&edata_ptr); // Unknown unused
1088 q->subpacket[s].js_subband_start = bytestream_get_be16(&edata_ptr);
1089 if (q->subpacket[s].js_subband_start >= 51) {
1090 av_log(avctx, AV_LOG_ERROR, "js_subband_start %d is too large\n", q->subpacket[s].js_subband_start);
1091 return AVERROR_INVALIDDATA;
1094 q->subpacket[s].js_vlc_bits = bytestream_get_be16(&edata_ptr);
1095 extradata_size -= 8;
1098 /* Initialize extradata related variables. */
1099 q->subpacket[s].samples_per_channel = samples_per_frame / avctx->channels;
1100 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1102 /* Initialize default data states. */
1103 q->subpacket[s].log2_numvector_size = 5;
1104 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1105 q->subpacket[s].num_channels = 1;
1107 /* Initialize version-dependent variables */
1109 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1110 q->subpacket[s].cookversion);
1111 q->subpacket[s].joint_stereo = 0;
1112 switch (q->subpacket[s].cookversion) {
1114 if (avctx->channels != 1) {
1115 avpriv_request_sample(avctx, "Container channels != 1");
1116 return AVERROR_PATCHWELCOME;
1118 av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1121 if (avctx->channels != 1) {
1122 q->subpacket[s].bits_per_subpdiv = 1;
1123 q->subpacket[s].num_channels = 2;
1125 av_log(avctx, AV_LOG_DEBUG, "STEREO\n");
1128 if (avctx->channels != 2) {
1129 avpriv_request_sample(avctx, "Container channels != 2");
1130 return AVERROR_PATCHWELCOME;
1132 av_log(avctx, AV_LOG_DEBUG, "JOINT_STEREO\n");
1133 if (avctx->extradata_size >= 16) {
1134 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1135 q->subpacket[s].js_subband_start;
1136 q->subpacket[s].joint_stereo = 1;
1137 q->subpacket[s].num_channels = 2;
1139 if (q->subpacket[s].samples_per_channel > 256) {
1140 q->subpacket[s].log2_numvector_size = 6;
1142 if (q->subpacket[s].samples_per_channel > 512) {
1143 q->subpacket[s].log2_numvector_size = 7;
1147 av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1148 if (extradata_size >= 4)
1149 channel_mask |= q->subpacket[s].channel_mask = bytestream_get_be32(&edata_ptr);
1151 if (av_get_channel_layout_nb_channels(q->subpacket[s].channel_mask) > 1) {
1152 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1153 q->subpacket[s].js_subband_start;
1154 q->subpacket[s].joint_stereo = 1;
1155 q->subpacket[s].num_channels = 2;
1156 q->subpacket[s].samples_per_channel = samples_per_frame >> 1;
1158 if (q->subpacket[s].samples_per_channel > 256) {
1159 q->subpacket[s].log2_numvector_size = 6;
1161 if (q->subpacket[s].samples_per_channel > 512) {
1162 q->subpacket[s].log2_numvector_size = 7;
1165 q->subpacket[s].samples_per_channel = samples_per_frame;
1169 avpriv_request_sample(avctx, "Cook version %d",
1170 q->subpacket[s].cookversion);
1171 return AVERROR_PATCHWELCOME;
1174 if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1175 av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1176 return AVERROR_INVALIDDATA;
1178 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1181 /* Initialize variable relations */
1182 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1184 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1185 if (q->subpacket[s].total_subbands > 53) {
1186 avpriv_request_sample(avctx, "total_subbands > 53");
1187 return AVERROR_PATCHWELCOME;
1190 if ((q->subpacket[s].js_vlc_bits > 6) ||
1191 (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1192 av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1193 q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1194 return AVERROR_INVALIDDATA;
1197 if (q->subpacket[s].subbands > 50) {
1198 avpriv_request_sample(avctx, "subbands > 50");
1199 return AVERROR_PATCHWELCOME;
1201 if (q->subpacket[s].subbands == 0) {
1202 avpriv_request_sample(avctx, "subbands = 0");
1203 return AVERROR_PATCHWELCOME;
1205 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1206 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1207 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1208 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1210 if (q->num_subpackets + q->subpacket[s].num_channels > q->avctx->channels) {
1211 av_log(avctx, AV_LOG_ERROR, "Too many subpackets %d for channels %d\n", q->num_subpackets, q->avctx->channels);
1212 return AVERROR_INVALIDDATA;
1215 q->num_subpackets++;
1217 if (s > MAX_SUBPACKETS) {
1218 avpriv_request_sample(avctx, "subpackets > %d", MAX_SUBPACKETS);
1219 return AVERROR_PATCHWELCOME;
1222 /* Generate tables */
1225 init_cplscales_table(q);
1227 if ((ret = init_cook_vlc_tables(q)))
1231 if (avctx->block_align >= UINT_MAX / 2)
1232 return AVERROR(EINVAL);
1234 /* Pad the databuffer with:
1235 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1236 FF_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1237 q->decoded_bytes_buffer =
1238 av_mallocz(avctx->block_align
1239 + DECODE_BYTES_PAD1(avctx->block_align)
1240 + FF_INPUT_BUFFER_PADDING_SIZE);
1241 if (q->decoded_bytes_buffer == NULL)
1242 return AVERROR(ENOMEM);
1244 /* Initialize transform. */
1245 if ((ret = init_cook_mlt(q)))
1248 /* Initialize COOK signal arithmetic handling */
1250 q->scalar_dequant = scalar_dequant_float;
1251 q->decouple = decouple_float;
1252 q->imlt_window = imlt_window_float;
1253 q->interpolate = interpolate_float;
1254 q->saturate_output = saturate_output_float;
1257 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1258 if (q->samples_per_channel != 256 && q->samples_per_channel != 512 &&
1259 q->samples_per_channel != 1024) {
1260 avpriv_request_sample(avctx, "samples_per_channel = %d",
1261 q->samples_per_channel);
1262 return AVERROR_PATCHWELCOME;
1265 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1267 avctx->channel_layout = channel_mask;
1269 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1272 dump_cook_context(q);
1277 AVCodec ff_cook_decoder = {
1279 .long_name = NULL_IF_CONFIG_SMALL("Cook / Cooker / Gecko (RealAudio G2)"),
1280 .type = AVMEDIA_TYPE_AUDIO,
1281 .id = AV_CODEC_ID_COOK,
1282 .priv_data_size = sizeof(COOKContext),
1283 .init = cook_decode_init,
1284 .close = cook_decode_close,
1285 .decode = cook_decode_frame,
1286 .capabilities = CODEC_CAP_DR1,
1287 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1288 AV_SAMPLE_FMT_NONE },