2 * COOK compatible decoder
3 * Copyright (c) 2003 Sascha Sommer
4 * Copyright (c) 2005 Benjamin Larsson
6 * This file is part of Libav.
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * Cook compatible decoder. Bastardization of the G.722.1 standard.
26 * This decoder handles RealNetworks, RealAudio G2 data.
27 * Cook is identified by the codec name cook in RM files.
29 * To use this decoder, a calling application must supply the extradata
30 * bytes provided from the RM container; 8+ bytes for mono streams and
31 * 16+ for stereo streams (maybe more).
33 * Codec technicalities (all this assume a buffer length of 1024):
34 * Cook works with several different techniques to achieve its compression.
35 * In the timedomain the buffer is divided into 8 pieces and quantized. If
36 * two neighboring pieces have different quantization index a smooth
37 * quantization curve is used to get a smooth overlap between the different
39 * To get to the transformdomain Cook uses a modulated lapped transform.
40 * The transform domain has 50 subbands with 20 elements each. This
41 * means only a maximum of 50*20=1000 coefficients are used out of the 1024
45 #include "libavutil/lfg.h"
49 #include "bytestream.h"
51 #include "libavutil/audioconvert.h"
56 /* the different Cook versions */
57 #define MONO 0x1000001
58 #define STEREO 0x1000002
59 #define JOINT_STEREO 0x1000003
60 #define MC_COOK 0x2000000 // multichannel Cook, not supported
62 #define SUBBAND_SIZE 20
63 #define MAX_SUBPACKETS 5
75 int samples_per_frame;
79 int samples_per_channel;
80 int log2_numvector_size;
81 unsigned int channel_mask;
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;
133 int samples_per_channel;
136 int discarded_packets;
143 VLC envelope_quant_index[13];
144 VLC sqvh[7]; // scalar quantization
146 /* generatable tables and related variables */
147 int gain_size_factor;
148 float gain_table[23];
152 uint8_t* decoded_bytes_buffer;
153 DECLARE_ALIGNED(32, float, mono_mdct_output)[2048];
154 float decode_buffer_1[1024];
155 float decode_buffer_2[1024];
156 float decode_buffer_0[1060]; /* static allocation for joint decode */
158 const float *cplscales[5];
160 COOKSubpacket subpacket[MAX_SUBPACKETS];
163 static float pow2tab[127];
164 static float rootpow2tab[127];
166 /*************** init functions ***************/
168 /* table generator */
169 static av_cold void init_pow2table(void)
172 for (i = -63; i < 64; i++) {
173 pow2tab[63 + i] = pow(2, i);
174 rootpow2tab[63 + i] = sqrt(pow(2, i));
178 /* table generator */
179 static av_cold void init_gain_table(COOKContext *q)
182 q->gain_size_factor = q->samples_per_channel / 8;
183 for (i = 0; i < 23; i++)
184 q->gain_table[i] = pow(pow2tab[i + 52],
185 (1.0 / (double) q->gain_size_factor));
189 static av_cold int init_cook_vlc_tables(COOKContext *q)
194 for (i = 0; i < 13; i++) {
195 result |= init_vlc(&q->envelope_quant_index[i], 9, 24,
196 envelope_quant_index_huffbits[i], 1, 1,
197 envelope_quant_index_huffcodes[i], 2, 2, 0);
199 av_log(q->avctx, AV_LOG_DEBUG, "sqvh VLC init\n");
200 for (i = 0; i < 7; i++) {
201 result |= init_vlc(&q->sqvh[i], vhvlcsize_tab[i], vhsize_tab[i],
202 cvh_huffbits[i], 1, 1,
203 cvh_huffcodes[i], 2, 2, 0);
206 for (i = 0; i < q->num_subpackets; i++) {
207 if (q->subpacket[i].joint_stereo == 1) {
208 result |= init_vlc(&q->subpacket[i].channel_coupling, 6,
209 (1 << q->subpacket[i].js_vlc_bits) - 1,
210 ccpl_huffbits[q->subpacket[i].js_vlc_bits - 2], 1, 1,
211 ccpl_huffcodes[q->subpacket[i].js_vlc_bits - 2], 2, 2, 0);
212 av_log(q->avctx, AV_LOG_DEBUG, "subpacket %i Joint-stereo VLC used.\n", i);
216 av_log(q->avctx, AV_LOG_DEBUG, "VLC tables initialized.\n");
220 static av_cold int init_cook_mlt(COOKContext *q)
223 int mlt_size = q->samples_per_channel;
225 if ((q->mlt_window = av_malloc(mlt_size * sizeof(*q->mlt_window))) == 0)
226 return AVERROR(ENOMEM);
228 /* Initialize the MLT window: simple sine window. */
229 ff_sine_window_init(q->mlt_window, mlt_size);
230 for (j = 0; j < mlt_size; j++)
231 q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
233 /* Initialize the MDCT. */
234 if ((ret = ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size) + 1, 1, 1.0 / 32768.0))) {
235 av_free(q->mlt_window);
238 av_log(q->avctx, AV_LOG_DEBUG, "MDCT initialized, order = %d.\n",
239 av_log2(mlt_size) + 1);
244 static av_cold void init_cplscales_table(COOKContext *q)
247 for (i = 0; i < 5; i++)
248 q->cplscales[i] = cplscales[i];
251 /*************** init functions end ***********/
253 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes) + 3) % 4)
254 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
257 * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
258 * Why? No idea, some checksum/error detection method maybe.
260 * Out buffer size: extra bytes are needed to cope with
261 * padding/misalignment.
262 * Subpackets passed to the decoder can contain two, consecutive
263 * half-subpackets, of identical but arbitrary size.
264 * 1234 1234 1234 1234 extraA extraB
265 * Case 1: AAAA BBBB 0 0
266 * Case 2: AAAA ABBB BB-- 3 3
267 * Case 3: AAAA AABB BBBB 2 2
268 * Case 4: AAAA AAAB BBBB BB-- 1 5
270 * Nice way to waste CPU cycles.
272 * @param inbuffer pointer to byte array of indata
273 * @param out pointer to byte array of outdata
274 * @param bytes number of bytes
276 static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
278 static const uint32_t tab[4] = {
279 AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
280 AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
285 uint32_t *obuf = (uint32_t *) out;
286 /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
287 * I'm too lazy though, should be something like
288 * for (i = 0; i < bitamount / 64; i++)
289 * (int64_t) out[i] = 0x37c511f237c511f2 ^ av_be2ne64(int64_t) in[i]);
290 * Buffer alignment needs to be checked. */
292 off = (intptr_t) inbuffer & 3;
293 buf = (const uint32_t *) (inbuffer - off);
296 for (i = 0; i < bytes / 4; i++)
297 obuf[i] = c ^ buf[i];
305 static av_cold int cook_decode_close(AVCodecContext *avctx)
308 COOKContext *q = avctx->priv_data;
309 av_log(avctx, AV_LOG_DEBUG, "Deallocating memory.\n");
311 /* Free allocated memory buffers. */
312 av_free(q->mlt_window);
313 av_free(q->decoded_bytes_buffer);
315 /* Free the transform. */
316 ff_mdct_end(&q->mdct_ctx);
318 /* Free the VLC tables. */
319 for (i = 0; i < 13; i++)
320 ff_free_vlc(&q->envelope_quant_index[i]);
321 for (i = 0; i < 7; i++)
322 ff_free_vlc(&q->sqvh[i]);
323 for (i = 0; i < q->num_subpackets; i++)
324 ff_free_vlc(&q->subpacket[i].channel_coupling);
326 av_log(avctx, AV_LOG_DEBUG, "Memory deallocated.\n");
332 * Fill the gain array for the timedomain quantization.
334 * @param gb pointer to the GetBitContext
335 * @param gaininfo array[9] of gain indexes
337 static void decode_gain_info(GetBitContext *gb, int *gaininfo)
341 while (get_bits1(gb)) {
345 n = get_bits_count(gb) - 1; // amount of elements*2 to update
349 int index = get_bits(gb, 3);
350 int gain = get_bits1(gb) ? get_bits(gb, 4) - 7 : -1;
353 gaininfo[i++] = gain;
360 * Create the quant index table needed for the envelope.
362 * @param q pointer to the COOKContext
363 * @param quant_index_table pointer to the array
365 static int decode_envelope(COOKContext *q, COOKSubpacket *p,
366 int *quant_index_table)
370 quant_index_table[0] = get_bits(&q->gb, 6) - 6; // This is used later in categorize
372 for (i = 1; i < p->total_subbands; i++) {
374 if (i >= p->js_subband_start * 2) {
375 vlc_index -= p->js_subband_start;
382 vlc_index = 13; // the VLC tables >13 are identical to No. 13
384 j = get_vlc2(&q->gb, q->envelope_quant_index[vlc_index - 1].table,
385 q->envelope_quant_index[vlc_index - 1].bits, 2);
386 quant_index_table[i] = quant_index_table[i - 1] + j - 12; // differential encoding
387 if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
388 av_log(q->avctx, AV_LOG_ERROR,
389 "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
390 quant_index_table[i], i);
391 return AVERROR_INVALIDDATA;
399 * Calculate the category and category_index vector.
401 * @param q pointer to the COOKContext
402 * @param quant_index_table pointer to the array
403 * @param category pointer to the category array
404 * @param category_index pointer to the category_index array
406 static void categorize(COOKContext *q, COOKSubpacket *p, int *quant_index_table,
407 int *category, int *category_index)
409 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
410 int exp_index2[102] = { 0 };
411 int exp_index1[102] = { 0 };
413 int tmp_categorize_array[128 * 2] = { 0 };
414 int tmp_categorize_array1_idx = p->numvector_size;
415 int tmp_categorize_array2_idx = p->numvector_size;
417 bits_left = p->bits_per_subpacket - get_bits_count(&q->gb);
419 if (bits_left > q->samples_per_channel)
420 bits_left = q->samples_per_channel +
421 ((bits_left - q->samples_per_channel) * 5) / 8;
426 for (i = 32; i > 0; i = i / 2) {
429 for (j = p->total_subbands; j > 0; j--) {
430 exp_idx = av_clip((i - quant_index_table[index] + bias) / 2, 0, 7);
432 num_bits += expbits_tab[exp_idx];
434 if (num_bits >= bits_left - 32)
438 /* Calculate total number of bits. */
440 for (i = 0; i < p->total_subbands; i++) {
441 exp_idx = av_clip((bias - quant_index_table[i]) / 2, 0, 7);
442 num_bits += expbits_tab[exp_idx];
443 exp_index1[i] = exp_idx;
444 exp_index2[i] = exp_idx;
446 tmpbias1 = tmpbias2 = num_bits;
448 for (j = 1; j < p->numvector_size; j++) {
449 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
452 for (i = 0; i < p->total_subbands; i++) {
453 if (exp_index1[i] < 7) {
454 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
463 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
464 tmpbias1 -= expbits_tab[exp_index1[index]] -
465 expbits_tab[exp_index1[index] + 1];
470 for (i = 0; i < p->total_subbands; i++) {
471 if (exp_index2[i] > 0) {
472 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
481 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
482 tmpbias2 -= expbits_tab[exp_index2[index]] -
483 expbits_tab[exp_index2[index] - 1];
488 for (i = 0; i < p->total_subbands; i++)
489 category[i] = exp_index2[i];
491 for (i = 0; i < p->numvector_size - 1; i++)
492 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
497 * Expand the category vector.
499 * @param q pointer to the COOKContext
500 * @param category pointer to the category array
501 * @param category_index pointer to the category_index array
503 static inline void expand_category(COOKContext *q, int *category,
507 for (i = 0; i < q->num_vectors; i++)
509 int idx = category_index[i];
510 if (++category[idx] >= FF_ARRAY_ELEMS(dither_tab))
516 * The real requantization of the mltcoefs
518 * @param q pointer to the COOKContext
520 * @param quant_index quantisation index
521 * @param subband_coef_index array of indexes to quant_centroid_tab
522 * @param subband_coef_sign signs of coefficients
523 * @param mlt_p pointer into the mlt buffer
525 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
526 int *subband_coef_index, int *subband_coef_sign,
532 for (i = 0; i < SUBBAND_SIZE; i++) {
533 if (subband_coef_index[i]) {
534 f1 = quant_centroid_tab[index][subband_coef_index[i]];
535 if (subband_coef_sign[i])
538 /* noise coding if subband_coef_index[i] == 0 */
539 f1 = dither_tab[index];
540 if (av_lfg_get(&q->random_state) < 0x80000000)
543 mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
547 * Unpack the subband_coef_index and subband_coef_sign vectors.
549 * @param q pointer to the COOKContext
550 * @param category pointer to the category array
551 * @param subband_coef_index array of indexes to quant_centroid_tab
552 * @param subband_coef_sign signs of coefficients
554 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
555 int *subband_coef_index, int *subband_coef_sign)
558 int vlc, vd, tmp, result;
560 vd = vd_tab[category];
562 for (i = 0; i < vpr_tab[category]; i++) {
563 vlc = get_vlc2(&q->gb, q->sqvh[category].table, q->sqvh[category].bits, 3);
564 if (p->bits_per_subpacket < get_bits_count(&q->gb)) {
568 for (j = vd - 1; j >= 0; j--) {
569 tmp = (vlc * invradix_tab[category]) / 0x100000;
570 subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
573 for (j = 0; j < vd; j++) {
574 if (subband_coef_index[i * vd + j]) {
575 if (get_bits_count(&q->gb) < p->bits_per_subpacket) {
576 subband_coef_sign[i * vd + j] = get_bits1(&q->gb);
579 subband_coef_sign[i * vd + j] = 0;
582 subband_coef_sign[i * vd + j] = 0;
591 * Fill the mlt_buffer with mlt coefficients.
593 * @param q pointer to the COOKContext
594 * @param category pointer to the category array
595 * @param quant_index_table pointer to the array
596 * @param mlt_buffer pointer to mlt coefficients
598 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
599 int *quant_index_table, float *mlt_buffer)
601 /* A zero in this table means that the subband coefficient is
602 random noise coded. */
603 int subband_coef_index[SUBBAND_SIZE];
604 /* A zero in this table means that the subband coefficient is a
605 positive multiplicator. */
606 int subband_coef_sign[SUBBAND_SIZE];
610 for (band = 0; band < p->total_subbands; band++) {
611 index = category[band];
612 if (category[band] < 7) {
613 if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
615 for (j = 0; j < p->total_subbands; j++)
616 category[band + j] = 7;
620 memset(subband_coef_index, 0, sizeof(subband_coef_index));
621 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
623 q->scalar_dequant(q, index, quant_index_table[band],
624 subband_coef_index, subband_coef_sign,
625 &mlt_buffer[band * SUBBAND_SIZE]);
628 /* FIXME: should this be removed, or moved into loop above? */
629 if (p->total_subbands * SUBBAND_SIZE >= q->samples_per_channel)
635 * function for decoding mono data
637 * @param q pointer to the COOKContext
638 * @param mlt_buffer pointer to mlt coefficients
640 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
642 int category_index[128] = { 0 };
643 int category[128] = { 0 };
644 int quant_index_table[102];
647 if ((res = decode_envelope(q, p, quant_index_table)) < 0)
649 q->num_vectors = get_bits(&q->gb, p->log2_numvector_size);
650 categorize(q, p, quant_index_table, category, category_index);
651 expand_category(q, category, category_index);
652 decode_vectors(q, p, category, quant_index_table, mlt_buffer);
659 * the actual requantization of the timedomain samples
661 * @param q pointer to the COOKContext
662 * @param buffer pointer to the timedomain buffer
663 * @param gain_index index for the block multiplier
664 * @param gain_index_next index for the next block multiplier
666 static void interpolate_float(COOKContext *q, float *buffer,
667 int gain_index, int gain_index_next)
671 fc1 = pow2tab[gain_index + 63];
673 if (gain_index == gain_index_next) { // static gain
674 for (i = 0; i < q->gain_size_factor; i++)
676 } else { // smooth gain
677 fc2 = q->gain_table[11 + (gain_index_next - gain_index)];
678 for (i = 0; i < q->gain_size_factor; i++) {
686 * Apply transform window, overlap buffers.
688 * @param q pointer to the COOKContext
689 * @param inbuffer pointer to the mltcoefficients
690 * @param gains_ptr current and previous gains
691 * @param previous_buffer pointer to the previous buffer to be used for overlapping
693 static void imlt_window_float(COOKContext *q, float *inbuffer,
694 cook_gains *gains_ptr, float *previous_buffer)
696 const float fc = pow2tab[gains_ptr->previous[0] + 63];
698 /* The weird thing here, is that the two halves of the time domain
699 * buffer are swapped. Also, the newest data, that we save away for
700 * next frame, has the wrong sign. Hence the subtraction below.
701 * Almost sounds like a complex conjugate/reverse data/FFT effect.
704 /* Apply window and overlap */
705 for (i = 0; i < q->samples_per_channel; i++)
706 inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
707 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
711 * The modulated lapped transform, this takes transform coefficients
712 * and transforms them into timedomain samples.
713 * Apply transform window, overlap buffers, apply gain profile
714 * and buffer management.
716 * @param q pointer to the COOKContext
717 * @param inbuffer pointer to the mltcoefficients
718 * @param gains_ptr current and previous gains
719 * @param previous_buffer pointer to the previous buffer to be used for overlapping
721 static void imlt_gain(COOKContext *q, float *inbuffer,
722 cook_gains *gains_ptr, float *previous_buffer)
724 float *buffer0 = q->mono_mdct_output;
725 float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
728 /* Inverse modified discrete cosine transform */
729 q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
731 q->imlt_window(q, buffer1, gains_ptr, previous_buffer);
733 /* Apply gain profile */
734 for (i = 0; i < 8; i++)
735 if (gains_ptr->now[i] || gains_ptr->now[i + 1])
736 q->interpolate(q, &buffer1[q->gain_size_factor * i],
737 gains_ptr->now[i], gains_ptr->now[i + 1]);
739 /* Save away the current to be previous block. */
740 memcpy(previous_buffer, buffer0,
741 q->samples_per_channel * sizeof(*previous_buffer));
746 * function for getting the jointstereo coupling information
748 * @param q pointer to the COOKContext
749 * @param decouple_tab decoupling array
752 static void decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
755 int vlc = get_bits1(&q->gb);
756 int start = cplband[p->js_subband_start];
757 int end = cplband[p->subbands - 1];
758 int length = end - start + 1;
764 for (i = 0; i < length; i++)
765 decouple_tab[start + i] = get_vlc2(&q->gb,
766 p->channel_coupling.table,
767 p->channel_coupling.bits, 2);
769 for (i = 0; i < length; i++)
770 decouple_tab[start + i] = get_bits(&q->gb, p->js_vlc_bits);
774 * function decouples a pair of signals from a single signal via multiplication.
776 * @param q pointer to the COOKContext
777 * @param subband index of the current subband
778 * @param f1 multiplier for channel 1 extraction
779 * @param f2 multiplier for channel 2 extraction
780 * @param decode_buffer input buffer
781 * @param mlt_buffer1 pointer to left channel mlt coefficients
782 * @param mlt_buffer2 pointer to right channel mlt coefficients
784 static void decouple_float(COOKContext *q,
788 float *decode_buffer,
789 float *mlt_buffer1, float *mlt_buffer2)
792 for (j = 0; j < SUBBAND_SIZE; j++) {
793 tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
794 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
795 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
800 * function for decoding joint stereo data
802 * @param q pointer to the COOKContext
803 * @param mlt_buffer1 pointer to left channel mlt coefficients
804 * @param mlt_buffer2 pointer to right channel mlt coefficients
806 static int joint_decode(COOKContext *q, COOKSubpacket *p,
807 float *mlt_buffer_left, float *mlt_buffer_right)
810 int decouple_tab[SUBBAND_SIZE] = { 0 };
811 float *decode_buffer = q->decode_buffer_0;
814 const float *cplscale;
816 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
818 /* Make sure the buffers are zeroed out. */
819 memset(mlt_buffer_left, 0, 1024 * sizeof(*mlt_buffer_left));
820 memset(mlt_buffer_right, 0, 1024 * sizeof(*mlt_buffer_right));
821 decouple_info(q, p, decouple_tab);
822 if ((res = mono_decode(q, p, decode_buffer)) < 0)
825 /* The two channels are stored interleaved in decode_buffer. */
826 for (i = 0; i < p->js_subband_start; i++) {
827 for (j = 0; j < SUBBAND_SIZE; j++) {
828 mlt_buffer_left[i * 20 + j] = decode_buffer[i * 40 + j];
829 mlt_buffer_right[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
833 /* When we reach js_subband_start (the higher frequencies)
834 the coefficients are stored in a coupling scheme. */
835 idx = (1 << p->js_vlc_bits) - 1;
836 for (i = p->js_subband_start; i < p->subbands; i++) {
837 cpl_tmp = cplband[i];
838 idx -= decouple_tab[cpl_tmp];
839 cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
840 f1 = cplscale[decouple_tab[cpl_tmp] + 1];
842 q->decouple(q, p, i, f1, f2, decode_buffer,
843 mlt_buffer_left, mlt_buffer_right);
844 idx = (1 << p->js_vlc_bits) - 1;
851 * First part of subpacket decoding:
852 * decode raw stream bytes and read gain info.
854 * @param q pointer to the COOKContext
855 * @param inbuffer pointer to raw stream data
856 * @param gains_ptr array of current/prev gain pointers
858 static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
859 const uint8_t *inbuffer,
860 cook_gains *gains_ptr)
864 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
865 p->bits_per_subpacket / 8);
866 init_get_bits(&q->gb, q->decoded_bytes_buffer + offset,
867 p->bits_per_subpacket);
868 decode_gain_info(&q->gb, gains_ptr->now);
870 /* Swap current and previous gains */
871 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
875 * Saturate the output signal and interleave.
877 * @param q pointer to the COOKContext
878 * @param out pointer to the output vector
880 static void saturate_output_float(COOKContext *q, float *out)
882 q->dsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
883 -1.0f, 1.0f, FFALIGN(q->samples_per_channel, 8));
888 * Final part of subpacket decoding:
889 * Apply modulated lapped transform, gain compensation,
890 * clip and convert to integer.
892 * @param q pointer to the COOKContext
893 * @param decode_buffer pointer to the mlt coefficients
894 * @param gains_ptr array of current/prev gain pointers
895 * @param previous_buffer pointer to the previous buffer to be used for overlapping
896 * @param out pointer to the output buffer
898 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
899 cook_gains *gains_ptr, float *previous_buffer,
902 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
904 q->saturate_output(q, out);
909 * Cook subpacket decoding. This function returns one decoded subpacket,
910 * usually 1024 samples per channel.
912 * @param q pointer to the COOKContext
913 * @param inbuffer pointer to the inbuffer
914 * @param outbuffer pointer to the outbuffer
916 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
917 const uint8_t *inbuffer, float **outbuffer)
919 int sub_packet_size = p->size;
922 memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
923 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
925 if (p->joint_stereo) {
926 if ((res = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
929 if ((res = mono_decode(q, p, q->decode_buffer_1)) < 0)
932 if (p->num_channels == 2) {
933 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
934 if ((res = mono_decode(q, p, q->decode_buffer_2)) < 0)
939 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
940 p->mono_previous_buffer1,
941 outbuffer ? outbuffer[p->ch_idx] : NULL);
943 if (p->num_channels == 2)
945 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
946 p->mono_previous_buffer2,
947 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
949 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
950 p->mono_previous_buffer2,
951 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
958 * Cook frame decoding
960 * @param avctx pointer to the AVCodecContext
962 static int cook_decode_frame(AVCodecContext *avctx, void *data,
963 int *got_frame_ptr, AVPacket *avpkt)
965 const uint8_t *buf = avpkt->data;
966 int buf_size = avpkt->size;
967 COOKContext *q = avctx->priv_data;
968 float **samples = NULL;
973 if (buf_size < avctx->block_align)
976 /* get output buffer */
977 if (q->discarded_packets >= 2) {
978 q->frame.nb_samples = q->samples_per_channel;
979 if ((ret = avctx->get_buffer(avctx, &q->frame)) < 0) {
980 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
983 samples = (float **)q->frame.extended_data;
986 /* estimate subpacket sizes */
987 q->subpacket[0].size = avctx->block_align;
989 for (i = 1; i < q->num_subpackets; i++) {
990 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
991 q->subpacket[0].size -= q->subpacket[i].size + 1;
992 if (q->subpacket[0].size < 0) {
993 av_log(avctx, AV_LOG_DEBUG,
994 "frame subpacket size total > avctx->block_align!\n");
995 return AVERROR_INVALIDDATA;
999 /* decode supbackets */
1000 for (i = 0; i < q->num_subpackets; i++) {
1001 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
1002 q->subpacket[i].bits_per_subpdiv;
1003 q->subpacket[i].ch_idx = chidx;
1004 av_log(avctx, AV_LOG_DEBUG,
1005 "subpacket[%i] size %i js %i %i block_align %i\n",
1006 i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
1007 avctx->block_align);
1009 if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
1011 offset += q->subpacket[i].size;
1012 chidx += q->subpacket[i].num_channels;
1013 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
1014 i, q->subpacket[i].size * 8, get_bits_count(&q->gb));
1017 /* Discard the first two frames: no valid audio. */
1018 if (q->discarded_packets < 2) {
1019 q->discarded_packets++;
1021 return avctx->block_align;
1025 *(AVFrame *) data = q->frame;
1027 return avctx->block_align;
1031 static void dump_cook_context(COOKContext *q)
1034 #define PRINT(a, b) av_log(q->avctx, AV_LOG_ERROR, " %s = %d\n", a, b);
1035 av_log(q->avctx, AV_LOG_ERROR, "COOKextradata\n");
1036 av_log(q->avctx, AV_LOG_ERROR, "cookversion=%x\n", q->subpacket[0].cookversion);
1037 if (q->subpacket[0].cookversion > STEREO) {
1038 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1039 PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1041 av_log(q->avctx, AV_LOG_ERROR, "COOKContext\n");
1042 PRINT("nb_channels", q->nb_channels);
1043 PRINT("bit_rate", q->bit_rate);
1044 PRINT("sample_rate", q->sample_rate);
1045 PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1046 PRINT("samples_per_frame", q->subpacket[0].samples_per_frame);
1047 PRINT("subbands", q->subpacket[0].subbands);
1048 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1049 PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1050 PRINT("numvector_size", q->subpacket[0].numvector_size);
1051 PRINT("total_subbands", q->subpacket[0].total_subbands);
1055 static av_cold int cook_count_channels(unsigned int mask)
1059 for (i = 0; i < 32; i++)
1060 if (mask & (1 << i))
1066 * Cook initialization
1068 * @param avctx pointer to the AVCodecContext
1070 static av_cold int cook_decode_init(AVCodecContext *avctx)
1072 COOKContext *q = avctx->priv_data;
1073 const uint8_t *edata_ptr = avctx->extradata;
1074 const uint8_t *edata_ptr_end = edata_ptr + avctx->extradata_size;
1075 int extradata_size = avctx->extradata_size;
1077 unsigned int channel_mask = 0;
1081 /* Take care of the codec specific extradata. */
1082 if (extradata_size <= 0) {
1083 av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1084 return AVERROR_INVALIDDATA;
1086 av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1088 /* Take data from the AVCodecContext (RM container). */
1089 q->sample_rate = avctx->sample_rate;
1090 q->nb_channels = avctx->channels;
1091 q->bit_rate = avctx->bit_rate;
1092 if (!q->nb_channels) {
1093 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1094 return AVERROR_INVALIDDATA;
1097 /* Initialize RNG. */
1098 av_lfg_init(&q->random_state, 0);
1100 ff_dsputil_init(&q->dsp, avctx);
1102 while (edata_ptr < edata_ptr_end) {
1103 /* 8 for mono, 16 for stereo, ? for multichannel
1104 Swap to right endianness so we don't need to care later on. */
1105 if (extradata_size >= 8) {
1106 q->subpacket[s].cookversion = bytestream_get_be32(&edata_ptr);
1107 q->subpacket[s].samples_per_frame = bytestream_get_be16(&edata_ptr);
1108 q->subpacket[s].subbands = bytestream_get_be16(&edata_ptr);
1109 extradata_size -= 8;
1111 if (extradata_size >= 8) {
1112 bytestream_get_be32(&edata_ptr); // Unknown unused
1113 q->subpacket[s].js_subband_start = bytestream_get_be16(&edata_ptr);
1114 q->subpacket[s].js_vlc_bits = bytestream_get_be16(&edata_ptr);
1115 extradata_size -= 8;
1118 /* Initialize extradata related variables. */
1119 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame / q->nb_channels;
1120 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1122 /* Initialize default data states. */
1123 q->subpacket[s].log2_numvector_size = 5;
1124 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1125 q->subpacket[s].num_channels = 1;
1127 /* Initialize version-dependent variables */
1129 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1130 q->subpacket[s].cookversion);
1131 q->subpacket[s].joint_stereo = 0;
1132 switch (q->subpacket[s].cookversion) {
1134 if (q->nb_channels != 1) {
1135 av_log_ask_for_sample(avctx, "Container channels != 1.\n");
1136 return AVERROR_PATCHWELCOME;
1138 av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1141 if (q->nb_channels != 1) {
1142 q->subpacket[s].bits_per_subpdiv = 1;
1143 q->subpacket[s].num_channels = 2;
1145 av_log(avctx, AV_LOG_DEBUG, "STEREO\n");
1148 if (q->nb_channels != 2) {
1149 av_log_ask_for_sample(avctx, "Container channels != 2.\n");
1150 return AVERROR_PATCHWELCOME;
1152 av_log(avctx, AV_LOG_DEBUG, "JOINT_STEREO\n");
1153 if (avctx->extradata_size >= 16) {
1154 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1155 q->subpacket[s].js_subband_start;
1156 q->subpacket[s].joint_stereo = 1;
1157 q->subpacket[s].num_channels = 2;
1159 if (q->subpacket[s].samples_per_channel > 256) {
1160 q->subpacket[s].log2_numvector_size = 6;
1162 if (q->subpacket[s].samples_per_channel > 512) {
1163 q->subpacket[s].log2_numvector_size = 7;
1167 av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1168 if (extradata_size >= 4)
1169 channel_mask |= q->subpacket[s].channel_mask = bytestream_get_be32(&edata_ptr);
1171 if (cook_count_channels(q->subpacket[s].channel_mask) > 1) {
1172 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1173 q->subpacket[s].js_subband_start;
1174 q->subpacket[s].joint_stereo = 1;
1175 q->subpacket[s].num_channels = 2;
1176 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame >> 1;
1178 if (q->subpacket[s].samples_per_channel > 256) {
1179 q->subpacket[s].log2_numvector_size = 6;
1181 if (q->subpacket[s].samples_per_channel > 512) {
1182 q->subpacket[s].log2_numvector_size = 7;
1185 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame;
1189 av_log_ask_for_sample(avctx, "Unknown Cook version.\n");
1190 return AVERROR_PATCHWELCOME;
1193 if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1194 av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1195 return AVERROR_INVALIDDATA;
1197 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1200 /* Initialize variable relations */
1201 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1203 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1204 if (q->subpacket[s].total_subbands > 53) {
1205 av_log_ask_for_sample(avctx, "total_subbands > 53\n");
1206 return AVERROR_PATCHWELCOME;
1209 if ((q->subpacket[s].js_vlc_bits > 6) ||
1210 (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1211 av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1212 q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1213 return AVERROR_INVALIDDATA;
1216 if (q->subpacket[s].subbands > 50) {
1217 av_log_ask_for_sample(avctx, "subbands > 50\n");
1218 return AVERROR_PATCHWELCOME;
1220 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1221 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1222 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1223 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1225 q->num_subpackets++;
1227 if (s > MAX_SUBPACKETS) {
1228 av_log_ask_for_sample(avctx, "Too many subpackets > 5\n");
1229 return AVERROR_PATCHWELCOME;
1232 /* Generate tables */
1235 init_cplscales_table(q);
1237 if ((ret = init_cook_vlc_tables(q)))
1241 if (avctx->block_align >= UINT_MAX / 2)
1242 return AVERROR(EINVAL);
1244 /* Pad the databuffer with:
1245 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1246 FF_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1247 q->decoded_bytes_buffer =
1248 av_mallocz(avctx->block_align
1249 + DECODE_BYTES_PAD1(avctx->block_align)
1250 + FF_INPUT_BUFFER_PADDING_SIZE);
1251 if (q->decoded_bytes_buffer == NULL)
1252 return AVERROR(ENOMEM);
1254 /* Initialize transform. */
1255 if ((ret = init_cook_mlt(q)))
1258 /* Initialize COOK signal arithmetic handling */
1260 q->scalar_dequant = scalar_dequant_float;
1261 q->decouple = decouple_float;
1262 q->imlt_window = imlt_window_float;
1263 q->interpolate = interpolate_float;
1264 q->saturate_output = saturate_output_float;
1267 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1268 if ((q->samples_per_channel == 256) || (q->samples_per_channel == 512)
1269 || (q->samples_per_channel == 1024)) {
1271 av_log_ask_for_sample(avctx,
1272 "unknown amount of samples_per_channel = %d\n",
1273 q->samples_per_channel);
1274 return AVERROR_PATCHWELCOME;
1277 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1279 avctx->channel_layout = channel_mask;
1281 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1283 avcodec_get_frame_defaults(&q->frame);
1284 avctx->coded_frame = &q->frame;
1287 dump_cook_context(q);
1292 AVCodec ff_cook_decoder = {
1294 .type = AVMEDIA_TYPE_AUDIO,
1295 .id = AV_CODEC_ID_COOK,
1296 .priv_data_size = sizeof(COOKContext),
1297 .init = cook_decode_init,
1298 .close = cook_decode_close,
1299 .decode = cook_decode_frame,
1300 .capabilities = CODEC_CAP_DR1,
1301 .long_name = NULL_IF_CONFIG_SMALL("Cook / Cooker / Gecko (RealAudio G2)"),
1302 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1303 AV_SAMPLE_FMT_NONE },