2 * COOK compatible decoder
3 * Copyright (c) 2003 Sascha Sommer
4 * Copyright (c) 2005 Benjamin Larsson
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * Cook compatible decoder. Bastardization of the G.722.1 standard.
26 * This decoder handles RealNetworks, RealAudio G2 data.
27 * Cook is identified by the codec name cook in RM files.
29 * To use this decoder, a calling application must supply the extradata
30 * bytes provided from the RM container; 8+ bytes for mono streams and
31 * 16+ for stereo streams (maybe more).
33 * Codec technicalities (all this assume a buffer length of 1024):
34 * Cook works with several different techniques to achieve its compression.
35 * In the timedomain the buffer is divided into 8 pieces and quantized. If
36 * two neighboring pieces have different quantization index a smooth
37 * quantization curve is used to get a smooth overlap between the different
39 * To get to the transformdomain Cook uses a modulated lapped transform.
40 * The transform domain has 50 subbands with 20 elements each. This
41 * means only a maximum of 50*20=1000 coefficients are used out of the 1024
45 #include "libavutil/lfg.h"
49 #include "bytestream.h"
51 #include "libavutil/audioconvert.h"
56 /* the different Cook versions */
57 #define MONO 0x1000001
58 #define STEREO 0x1000002
59 #define JOINT_STEREO 0x1000003
60 #define MC_COOK 0x2000000 // multichannel Cook, not supported
62 #define SUBBAND_SIZE 20
63 #define MAX_SUBPACKETS 5
75 int samples_per_frame;
79 int samples_per_channel;
80 int log2_numvector_size;
81 unsigned int channel_mask;
82 VLC ccpl; ///< channel coupling
84 int bits_per_subpacket;
87 int numvector_size; ///< 1 << log2_numvector_size;
89 float mono_previous_buffer1[1024];
90 float mono_previous_buffer2[1024];
100 typedef struct cook {
102 * The following 5 functions provide the lowlevel arithmetic on
103 * the internal audio buffers.
105 void (*scalar_dequant)(struct cook *q, int index, int quant_index,
106 int *subband_coef_index, int *subband_coef_sign,
109 void (*decouple)(struct cook *q,
113 float *decode_buffer,
114 float *mlt_buffer1, float *mlt_buffer2);
116 void (*imlt_window)(struct cook *q, float *buffer1,
117 cook_gains *gains_ptr, float *previous_buffer);
119 void (*interpolate)(struct cook *q, float *buffer,
120 int gain_index, int gain_index_next);
122 void (*saturate_output)(struct cook *q, 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].ccpl, 6, (1 << q->subpacket[i].js_vlc_bits) - 1,
209 ccpl_huffbits[q->subpacket[i].js_vlc_bits - 2], 1, 1,
210 ccpl_huffcodes[q->subpacket[i].js_vlc_bits - 2], 2, 2, 0);
211 av_log(q->avctx, AV_LOG_DEBUG, "subpacket %i Joint-stereo VLC used.\n", i);
215 av_log(q->avctx, AV_LOG_DEBUG, "VLC tables initialized.\n");
219 static av_cold int init_cook_mlt(COOKContext *q)
222 int mlt_size = q->samples_per_channel;
224 if ((q->mlt_window = av_malloc(mlt_size * sizeof(*q->mlt_window))) == 0)
225 return AVERROR(ENOMEM);
227 /* Initialize the MLT window: simple sine window. */
228 ff_sine_window_init(q->mlt_window, mlt_size);
229 for (j = 0; j < mlt_size; j++)
230 q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
232 /* Initialize the MDCT. */
233 if ((ret = ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size) + 1, 1, 1.0 / 32768.0))) {
234 av_free(q->mlt_window);
237 av_log(q->avctx, AV_LOG_DEBUG, "MDCT initialized, order = %d.\n",
238 av_log2(mlt_size) + 1);
243 static const float *maybe_reformat_buffer32(COOKContext *q, const float *ptr, int n)
249 static av_cold void init_cplscales_table(COOKContext *q)
252 for (i = 0; i < 5; i++)
253 q->cplscales[i] = maybe_reformat_buffer32(q, cplscales[i], (1 << (i + 2)) - 1);
256 /*************** init functions end ***********/
258 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes) + 3) % 4)
259 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
262 * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
263 * Why? No idea, some checksum/error detection method maybe.
265 * Out buffer size: extra bytes are needed to cope with
266 * padding/misalignment.
267 * Subpackets passed to the decoder can contain two, consecutive
268 * half-subpackets, of identical but arbitrary size.
269 * 1234 1234 1234 1234 extraA extraB
270 * Case 1: AAAA BBBB 0 0
271 * Case 2: AAAA ABBB BB-- 3 3
272 * Case 3: AAAA AABB BBBB 2 2
273 * Case 4: AAAA AAAB BBBB BB-- 1 5
275 * Nice way to waste CPU cycles.
277 * @param inbuffer pointer to byte array of indata
278 * @param out pointer to byte array of outdata
279 * @param bytes number of bytes
281 static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
283 static const uint32_t tab[4] = {
284 AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
285 AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
290 uint32_t *obuf = (uint32_t *) out;
291 /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
292 * I'm too lazy though, should be something like
293 * for (i = 0; i < bitamount / 64; i++)
294 * (int64_t) out[i] = 0x37c511f237c511f2 ^ av_be2ne64(int64_t) in[i]);
295 * Buffer alignment needs to be checked. */
297 off = (intptr_t) inbuffer & 3;
298 buf = (const uint32_t *) (inbuffer - off);
301 for (i = 0; i < bytes / 4; i++)
302 obuf[i] = c ^ buf[i];
310 static av_cold int cook_decode_close(AVCodecContext *avctx)
313 COOKContext *q = avctx->priv_data;
314 av_log(avctx, AV_LOG_DEBUG, "Deallocating memory.\n");
316 /* Free allocated memory buffers. */
317 av_free(q->mlt_window);
318 av_free(q->decoded_bytes_buffer);
320 /* Free the transform. */
321 ff_mdct_end(&q->mdct_ctx);
323 /* Free the VLC tables. */
324 for (i = 0; i < 13; i++)
325 ff_free_vlc(&q->envelope_quant_index[i]);
326 for (i = 0; i < 7; i++)
327 ff_free_vlc(&q->sqvh[i]);
328 for (i = 0; i < q->num_subpackets; i++)
329 ff_free_vlc(&q->subpacket[i].ccpl);
331 av_log(avctx, AV_LOG_DEBUG, "Memory deallocated.\n");
337 * Fill the gain array for the timedomain quantization.
339 * @param gb pointer to the GetBitContext
340 * @param gaininfo array[9] of gain indexes
342 static void decode_gain_info(GetBitContext *gb, int *gaininfo)
346 while (get_bits1(gb)) {
350 n = get_bits_count(gb) - 1; // amount of elements*2 to update
354 int index = get_bits(gb, 3);
355 int gain = get_bits1(gb) ? get_bits(gb, 4) - 7 : -1;
358 gaininfo[i++] = gain;
365 * Create the quant index table needed for the envelope.
367 * @param q pointer to the COOKContext
368 * @param quant_index_table pointer to the array
370 static int decode_envelope(COOKContext *q, COOKSubpacket *p,
371 int *quant_index_table)
375 quant_index_table[0] = get_bits(&q->gb, 6) - 6; // This is used later in categorize
377 for (i = 1; i < p->total_subbands; i++) {
379 if (i >= p->js_subband_start * 2) {
380 vlc_index -= p->js_subband_start;
387 vlc_index = 13; // the VLC tables >13 are identical to No. 13
389 j = get_vlc2(&q->gb, q->envelope_quant_index[vlc_index - 1].table,
390 q->envelope_quant_index[vlc_index - 1].bits, 2);
391 quant_index_table[i] = quant_index_table[i - 1] + j - 12; // differential encoding
392 if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
393 av_log(q->avctx, AV_LOG_ERROR,
394 "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
395 quant_index_table[i], i);
396 return AVERROR_INVALIDDATA;
404 * Calculate the category and category_index vector.
406 * @param q pointer to the COOKContext
407 * @param quant_index_table pointer to the array
408 * @param category pointer to the category array
409 * @param category_index pointer to the category_index array
411 static void categorize(COOKContext *q, COOKSubpacket *p, const int *quant_index_table,
412 int *category, int *category_index)
414 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
415 int exp_index2[102] = { 0 };
416 int exp_index1[102] = { 0 };
418 int tmp_categorize_array[128 * 2] = { 0 };
419 int tmp_categorize_array1_idx = p->numvector_size;
420 int tmp_categorize_array2_idx = p->numvector_size;
422 bits_left = p->bits_per_subpacket - get_bits_count(&q->gb);
424 if (bits_left > q->samples_per_channel)
425 bits_left = q->samples_per_channel +
426 ((bits_left - q->samples_per_channel) * 5) / 8;
431 for (i = 32; i > 0; i = i / 2) {
434 for (j = p->total_subbands; j > 0; j--) {
435 exp_idx = av_clip((i - quant_index_table[index] + bias) / 2, 0, 7);
437 num_bits += expbits_tab[exp_idx];
439 if (num_bits >= bits_left - 32)
443 /* Calculate total number of bits. */
445 for (i = 0; i < p->total_subbands; i++) {
446 exp_idx = av_clip((bias - quant_index_table[i]) / 2, 0, 7);
447 num_bits += expbits_tab[exp_idx];
448 exp_index1[i] = exp_idx;
449 exp_index2[i] = exp_idx;
451 tmpbias1 = tmpbias2 = num_bits;
453 for (j = 1; j < p->numvector_size; j++) {
454 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
457 for (i = 0; i < p->total_subbands; i++) {
458 if (exp_index1[i] < 7) {
459 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
468 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
469 tmpbias1 -= expbits_tab[exp_index1[index]] -
470 expbits_tab[exp_index1[index] + 1];
475 for (i = 0; i < p->total_subbands; i++) {
476 if (exp_index2[i] > 0) {
477 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
486 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
487 tmpbias2 -= expbits_tab[exp_index2[index]] -
488 expbits_tab[exp_index2[index] - 1];
493 for (i = 0; i < p->total_subbands; i++)
494 category[i] = exp_index2[i];
496 for (i = 0; i < p->numvector_size - 1; i++)
497 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
502 * Expand the category vector.
504 * @param q pointer to the COOKContext
505 * @param category pointer to the category array
506 * @param category_index pointer to the category_index array
508 static inline void expand_category(COOKContext *q, int *category,
512 for (i = 0; i < q->num_vectors; i++)
514 int idx = category_index[i];
515 if (++category[idx] >= FF_ARRAY_ELEMS(dither_tab))
521 * The real requantization of the mltcoefs
523 * @param q pointer to the COOKContext
525 * @param quant_index quantisation index
526 * @param subband_coef_index array of indexes to quant_centroid_tab
527 * @param subband_coef_sign signs of coefficients
528 * @param mlt_p pointer into the mlt buffer
530 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
531 int *subband_coef_index, int *subband_coef_sign,
537 for (i = 0; i < SUBBAND_SIZE; i++) {
538 if (subband_coef_index[i]) {
539 f1 = quant_centroid_tab[index][subband_coef_index[i]];
540 if (subband_coef_sign[i])
543 /* noise coding if subband_coef_index[i] == 0 */
544 f1 = dither_tab[index];
545 if (av_lfg_get(&q->random_state) < 0x80000000)
548 mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
552 * Unpack the subband_coef_index and subband_coef_sign vectors.
554 * @param q pointer to the COOKContext
555 * @param category pointer to the category array
556 * @param subband_coef_index array of indexes to quant_centroid_tab
557 * @param subband_coef_sign signs of coefficients
559 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
560 int *subband_coef_index, int *subband_coef_sign)
563 int vlc, vd, tmp, result;
565 vd = vd_tab[category];
567 for (i = 0; i < vpr_tab[category]; i++) {
568 vlc = get_vlc2(&q->gb, q->sqvh[category].table, q->sqvh[category].bits, 3);
569 if (p->bits_per_subpacket < get_bits_count(&q->gb)) {
573 for (j = vd - 1; j >= 0; j--) {
574 tmp = (vlc * invradix_tab[category]) / 0x100000;
575 subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
578 for (j = 0; j < vd; j++) {
579 if (subband_coef_index[i * vd + j]) {
580 if (get_bits_count(&q->gb) < p->bits_per_subpacket) {
581 subband_coef_sign[i * vd + j] = get_bits1(&q->gb);
584 subband_coef_sign[i * vd + j] = 0;
587 subband_coef_sign[i * vd + j] = 0;
596 * Fill the mlt_buffer with mlt coefficients.
598 * @param q pointer to the COOKContext
599 * @param category pointer to the category array
600 * @param quant_index_table pointer to the array
601 * @param mlt_buffer pointer to mlt coefficients
603 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
604 int *quant_index_table, float *mlt_buffer)
606 /* A zero in this table means that the subband coefficient is
607 random noise coded. */
608 int subband_coef_index[SUBBAND_SIZE];
609 /* A zero in this table means that the subband coefficient is a
610 positive multiplicator. */
611 int subband_coef_sign[SUBBAND_SIZE];
615 for (band = 0; band < p->total_subbands; band++) {
616 index = category[band];
617 if (category[band] < 7) {
618 if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
620 for (j = 0; j < p->total_subbands; j++)
621 category[band + j] = 7;
625 memset(subband_coef_index, 0, sizeof(subband_coef_index));
626 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
628 q->scalar_dequant(q, index, quant_index_table[band],
629 subband_coef_index, subband_coef_sign,
630 &mlt_buffer[band * SUBBAND_SIZE]);
633 /* FIXME: should this be removed, or moved into loop above? */
634 if (p->total_subbands * SUBBAND_SIZE >= q->samples_per_channel)
640 * function for decoding mono data
642 * @param q pointer to the COOKContext
643 * @param mlt_buffer pointer to mlt coefficients
645 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
647 int category_index[128] = { 0 };
648 int category[128] = { 0 };
649 int quant_index_table[102];
652 if ((res = decode_envelope(q, p, quant_index_table)) < 0)
654 q->num_vectors = get_bits(&q->gb, p->log2_numvector_size);
655 categorize(q, p, quant_index_table, category, category_index);
656 expand_category(q, category, category_index);
657 for (i=0; i<p->total_subbands; i++) {
659 return AVERROR_INVALIDDATA;
661 decode_vectors(q, p, category, quant_index_table, mlt_buffer);
668 * the actual requantization of the timedomain samples
670 * @param q pointer to the COOKContext
671 * @param buffer pointer to the timedomain buffer
672 * @param gain_index index for the block multiplier
673 * @param gain_index_next index for the next block multiplier
675 static void interpolate_float(COOKContext *q, float *buffer,
676 int gain_index, int gain_index_next)
680 fc1 = pow2tab[gain_index + 63];
682 if (gain_index == gain_index_next) { // static gain
683 for (i = 0; i < q->gain_size_factor; i++)
685 } else { // smooth gain
686 fc2 = q->gain_table[11 + (gain_index_next - gain_index)];
687 for (i = 0; i < q->gain_size_factor; i++) {
695 * Apply transform window, overlap buffers.
697 * @param q pointer to the COOKContext
698 * @param inbuffer pointer to the mltcoefficients
699 * @param gains_ptr current and previous gains
700 * @param previous_buffer pointer to the previous buffer to be used for overlapping
702 static void imlt_window_float(COOKContext *q, float *inbuffer,
703 cook_gains *gains_ptr, float *previous_buffer)
705 const float fc = pow2tab[gains_ptr->previous[0] + 63];
707 /* The weird thing here, is that the two halves of the time domain
708 * buffer are swapped. Also, the newest data, that we save away for
709 * next frame, has the wrong sign. Hence the subtraction below.
710 * Almost sounds like a complex conjugate/reverse data/FFT effect.
713 /* Apply window and overlap */
714 for (i = 0; i < q->samples_per_channel; i++)
715 inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
716 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
720 * The modulated lapped transform, this takes transform coefficients
721 * and transforms them into timedomain samples.
722 * Apply transform window, overlap buffers, apply gain profile
723 * and buffer management.
725 * @param q pointer to the COOKContext
726 * @param inbuffer pointer to the mltcoefficients
727 * @param gains_ptr current and previous gains
728 * @param previous_buffer pointer to the previous buffer to be used for overlapping
730 static void imlt_gain(COOKContext *q, float *inbuffer,
731 cook_gains *gains_ptr, float *previous_buffer)
733 float *buffer0 = q->mono_mdct_output;
734 float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
737 /* Inverse modified discrete cosine transform */
738 q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
740 q->imlt_window(q, buffer1, gains_ptr, previous_buffer);
742 /* Apply gain profile */
743 for (i = 0; i < 8; i++)
744 if (gains_ptr->now[i] || gains_ptr->now[i + 1])
745 q->interpolate(q, &buffer1[q->gain_size_factor * i],
746 gains_ptr->now[i], gains_ptr->now[i + 1]);
748 /* Save away the current to be previous block. */
749 memcpy(previous_buffer, buffer0,
750 q->samples_per_channel * sizeof(*previous_buffer));
755 * function for getting the jointstereo coupling information
757 * @param q pointer to the COOKContext
758 * @param decouple_tab decoupling array
761 static int decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
764 int vlc = get_bits1(&q->gb);
765 int start = cplband[p->js_subband_start];
766 int end = cplband[p->subbands - 1];
767 int length = end - start + 1;
773 for (i = 0; i < length; i++)
774 decouple_tab[start + i] = get_vlc2(&q->gb, p->ccpl.table, p->ccpl.bits, 2);
776 for (i = 0; i < length; i++) {
777 int v = get_bits(&q->gb, p->js_vlc_bits);
778 if (v == (1<<p->js_vlc_bits)-1) {
779 av_log(q->avctx, AV_LOG_ERROR, "decouple value too large\n");
780 return AVERROR_INVALIDDATA;
782 decouple_tab[start + i] = v;
788 * function decouples a pair of signals from a single signal via multiplication.
790 * @param q pointer to the COOKContext
791 * @param subband index of the current subband
792 * @param f1 multiplier for channel 1 extraction
793 * @param f2 multiplier for channel 2 extraction
794 * @param decode_buffer input buffer
795 * @param mlt_buffer1 pointer to left channel mlt coefficients
796 * @param mlt_buffer2 pointer to right channel mlt coefficients
798 static void decouple_float(COOKContext *q,
802 float *decode_buffer,
803 float *mlt_buffer1, float *mlt_buffer2)
806 for (j = 0; j < SUBBAND_SIZE; j++) {
807 tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
808 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
809 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
814 * function for decoding joint stereo data
816 * @param q pointer to the COOKContext
817 * @param mlt_buffer1 pointer to left channel mlt coefficients
818 * @param mlt_buffer2 pointer to right channel mlt coefficients
820 static int joint_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer1,
824 int decouple_tab[SUBBAND_SIZE] = { 0 };
825 float *decode_buffer = q->decode_buffer_0;
828 const float *cplscale;
830 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
832 /* Make sure the buffers are zeroed out. */
833 memset(mlt_buffer1, 0, 1024 * sizeof(*mlt_buffer1));
834 memset(mlt_buffer2, 0, 1024 * sizeof(*mlt_buffer2));
835 if ((res = decouple_info(q, p, decouple_tab)) < 0)
837 if ((res = mono_decode(q, p, decode_buffer)) < 0)
839 /* The two channels are stored interleaved in decode_buffer. */
840 for (i = 0; i < p->js_subband_start; i++) {
841 for (j = 0; j < SUBBAND_SIZE; j++) {
842 mlt_buffer1[i * 20 + j] = decode_buffer[i * 40 + j];
843 mlt_buffer2[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
847 /* When we reach js_subband_start (the higher frequencies)
848 the coefficients are stored in a coupling scheme. */
849 idx = (1 << p->js_vlc_bits) - 1;
850 for (i = p->js_subband_start; i < p->subbands; i++) {
851 cpl_tmp = cplband[i];
852 idx -= decouple_tab[cpl_tmp];
853 cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
854 f1 = cplscale[decouple_tab[cpl_tmp] + 1];
856 q->decouple(q, p, i, f1, f2, decode_buffer, mlt_buffer1, mlt_buffer2);
857 idx = (1 << p->js_vlc_bits) - 1;
864 * First part of subpacket decoding:
865 * decode raw stream bytes and read gain info.
867 * @param q pointer to the COOKContext
868 * @param inbuffer pointer to raw stream data
869 * @param gains_ptr array of current/prev gain pointers
871 static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
872 const uint8_t *inbuffer,
873 cook_gains *gains_ptr)
877 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
878 p->bits_per_subpacket / 8);
879 init_get_bits(&q->gb, q->decoded_bytes_buffer + offset,
880 p->bits_per_subpacket);
881 decode_gain_info(&q->gb, gains_ptr->now);
883 /* Swap current and previous gains */
884 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
888 * Saturate the output signal and interleave.
890 * @param q pointer to the COOKContext
891 * @param out pointer to the output vector
893 static void saturate_output_float(COOKContext *q, float *out)
895 q->dsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
896 -1.0f, 1.0f, FFALIGN(q->samples_per_channel, 8));
901 * Final part of subpacket decoding:
902 * Apply modulated lapped transform, gain compensation,
903 * clip and convert to integer.
905 * @param q pointer to the COOKContext
906 * @param decode_buffer pointer to the mlt coefficients
907 * @param gains_ptr array of current/prev gain pointers
908 * @param previous_buffer pointer to the previous buffer to be used for overlapping
909 * @param out pointer to the output buffer
911 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
912 cook_gains *gains_ptr, float *previous_buffer,
915 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
917 q->saturate_output(q, out);
922 * Cook subpacket decoding. This function returns one decoded subpacket,
923 * usually 1024 samples per channel.
925 * @param q pointer to the COOKContext
926 * @param inbuffer pointer to the inbuffer
927 * @param outbuffer pointer to the outbuffer
929 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
930 const uint8_t *inbuffer, float **outbuffer)
932 int sub_packet_size = p->size;
935 memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
936 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
938 if (p->joint_stereo) {
939 if ((res = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
942 if ((res = mono_decode(q, p, q->decode_buffer_1)) < 0)
945 if (p->num_channels == 2) {
946 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
947 if ((res = mono_decode(q, p, q->decode_buffer_2)) < 0)
952 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
953 p->mono_previous_buffer1,
954 outbuffer ? outbuffer[p->ch_idx] : NULL);
956 if (p->num_channels == 2)
958 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
959 p->mono_previous_buffer2,
960 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
962 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
963 p->mono_previous_buffer2,
964 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
971 * Cook frame decoding
973 * @param avctx pointer to the AVCodecContext
975 static int cook_decode_frame(AVCodecContext *avctx, void *data,
976 int *got_frame_ptr, AVPacket *avpkt)
978 const uint8_t *buf = avpkt->data;
979 int buf_size = avpkt->size;
980 COOKContext *q = avctx->priv_data;
981 float **samples = NULL;
986 if (buf_size < avctx->block_align)
989 /* get output buffer */
990 if (q->discarded_packets >= 2) {
991 q->frame.nb_samples = q->samples_per_channel;
992 if ((ret = avctx->get_buffer(avctx, &q->frame)) < 0) {
993 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
996 samples = (float **)q->frame.extended_data;
999 /* estimate subpacket sizes */
1000 q->subpacket[0].size = avctx->block_align;
1002 for (i = 1; i < q->num_subpackets; i++) {
1003 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
1004 q->subpacket[0].size -= q->subpacket[i].size + 1;
1005 if (q->subpacket[0].size < 0) {
1006 av_log(avctx, AV_LOG_DEBUG,
1007 "frame subpacket size total > avctx->block_align!\n");
1008 return AVERROR_INVALIDDATA;
1012 /* decode supbackets */
1013 for (i = 0; i < q->num_subpackets; i++) {
1014 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
1015 q->subpacket[i].bits_per_subpdiv;
1016 q->subpacket[i].ch_idx = chidx;
1017 av_log(avctx, AV_LOG_DEBUG,
1018 "subpacket[%i] size %i js %i %i block_align %i\n",
1019 i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
1020 avctx->block_align);
1022 if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
1024 offset += q->subpacket[i].size;
1025 chidx += q->subpacket[i].num_channels;
1026 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
1027 i, q->subpacket[i].size * 8, get_bits_count(&q->gb));
1030 /* Discard the first two frames: no valid audio. */
1031 if (q->discarded_packets < 2) {
1032 q->discarded_packets++;
1034 return avctx->block_align;
1038 *(AVFrame *) data = q->frame;
1040 return avctx->block_align;
1044 static void dump_cook_context(COOKContext *q)
1047 #define PRINT(a, b) av_log(q->avctx, AV_LOG_ERROR, " %s = %d\n", a, b);
1048 av_log(q->avctx, AV_LOG_ERROR, "COOKextradata\n");
1049 av_log(q->avctx, AV_LOG_ERROR, "cookversion=%x\n", q->subpacket[0].cookversion);
1050 if (q->subpacket[0].cookversion > STEREO) {
1051 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1052 PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1054 av_log(q->avctx, AV_LOG_ERROR, "COOKContext\n");
1055 PRINT("nb_channels", q->nb_channels);
1056 PRINT("bit_rate", q->bit_rate);
1057 PRINT("sample_rate", q->sample_rate);
1058 PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1059 PRINT("samples_per_frame", q->subpacket[0].samples_per_frame);
1060 PRINT("subbands", q->subpacket[0].subbands);
1061 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1062 PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1063 PRINT("numvector_size", q->subpacket[0].numvector_size);
1064 PRINT("total_subbands", q->subpacket[0].total_subbands);
1068 static av_cold int cook_count_channels(unsigned int mask)
1072 for (i = 0; i < 32; i++)
1073 if (mask & (1 << i))
1079 * Cook initialization
1081 * @param avctx pointer to the AVCodecContext
1083 static av_cold int cook_decode_init(AVCodecContext *avctx)
1085 COOKContext *q = avctx->priv_data;
1086 const uint8_t *edata_ptr = avctx->extradata;
1087 const uint8_t *edata_ptr_end = edata_ptr + avctx->extradata_size;
1088 int extradata_size = avctx->extradata_size;
1090 unsigned int channel_mask = 0;
1094 /* Take care of the codec specific extradata. */
1095 if (extradata_size <= 0) {
1096 av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1097 return AVERROR_INVALIDDATA;
1099 av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1101 /* Take data from the AVCodecContext (RM container). */
1102 q->sample_rate = avctx->sample_rate;
1103 q->nb_channels = avctx->channels;
1104 q->bit_rate = avctx->bit_rate;
1105 if (!q->nb_channels) {
1106 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1107 return AVERROR_INVALIDDATA;
1110 /* Initialize RNG. */
1111 av_lfg_init(&q->random_state, 0);
1113 ff_dsputil_init(&q->dsp, avctx);
1115 while (edata_ptr < edata_ptr_end) {
1116 /* 8 for mono, 16 for stereo, ? for multichannel
1117 Swap to right endianness so we don't need to care later on. */
1118 if (extradata_size >= 8) {
1119 q->subpacket[s].cookversion = bytestream_get_be32(&edata_ptr);
1120 q->subpacket[s].samples_per_frame = bytestream_get_be16(&edata_ptr);
1121 q->subpacket[s].subbands = bytestream_get_be16(&edata_ptr);
1122 extradata_size -= 8;
1124 if (extradata_size >= 8) {
1125 bytestream_get_be32(&edata_ptr); // Unknown unused
1126 q->subpacket[s].js_subband_start = bytestream_get_be16(&edata_ptr);
1127 q->subpacket[s].js_vlc_bits = bytestream_get_be16(&edata_ptr);
1128 extradata_size -= 8;
1131 /* Initialize extradata related variables. */
1132 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame / q->nb_channels;
1133 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1135 /* Initialize default data states. */
1136 q->subpacket[s].log2_numvector_size = 5;
1137 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1138 q->subpacket[s].num_channels = 1;
1140 /* Initialize version-dependent variables */
1142 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1143 q->subpacket[s].cookversion);
1144 q->subpacket[s].joint_stereo = 0;
1145 switch (q->subpacket[s].cookversion) {
1147 if (q->nb_channels != 1) {
1148 av_log_ask_for_sample(avctx, "Container channels != 1.\n");
1149 return AVERROR_PATCHWELCOME;
1151 av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1154 if (q->nb_channels != 1) {
1155 q->subpacket[s].bits_per_subpdiv = 1;
1156 q->subpacket[s].num_channels = 2;
1158 av_log(avctx, AV_LOG_DEBUG, "STEREO\n");
1161 if (q->nb_channels != 2) {
1162 av_log_ask_for_sample(avctx, "Container channels != 2.\n");
1163 return AVERROR_PATCHWELCOME;
1165 av_log(avctx, AV_LOG_DEBUG, "JOINT_STEREO\n");
1166 if (avctx->extradata_size >= 16) {
1167 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1168 q->subpacket[s].js_subband_start;
1169 q->subpacket[s].joint_stereo = 1;
1170 q->subpacket[s].num_channels = 2;
1172 if (q->subpacket[s].samples_per_channel > 256) {
1173 q->subpacket[s].log2_numvector_size = 6;
1175 if (q->subpacket[s].samples_per_channel > 512) {
1176 q->subpacket[s].log2_numvector_size = 7;
1180 av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1181 if (extradata_size >= 4)
1182 channel_mask |= q->subpacket[s].channel_mask = bytestream_get_be32(&edata_ptr);
1184 if (cook_count_channels(q->subpacket[s].channel_mask) > 1) {
1185 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1186 q->subpacket[s].js_subband_start;
1187 q->subpacket[s].joint_stereo = 1;
1188 q->subpacket[s].num_channels = 2;
1189 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame >> 1;
1191 if (q->subpacket[s].samples_per_channel > 256) {
1192 q->subpacket[s].log2_numvector_size = 6;
1194 if (q->subpacket[s].samples_per_channel > 512) {
1195 q->subpacket[s].log2_numvector_size = 7;
1198 q->subpacket[s].samples_per_channel = q->subpacket[s].samples_per_frame;
1202 av_log_ask_for_sample(avctx, "Unknown Cook version.\n");
1203 return AVERROR_PATCHWELCOME;
1206 if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1207 av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1208 return AVERROR_INVALIDDATA;
1210 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1213 /* Initialize variable relations */
1214 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1216 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1217 if (q->subpacket[s].total_subbands > 53) {
1218 av_log_ask_for_sample(avctx, "total_subbands > 53\n");
1219 return AVERROR_PATCHWELCOME;
1222 if ((q->subpacket[s].js_vlc_bits > 6) ||
1223 (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1224 av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1225 q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1226 return AVERROR_INVALIDDATA;
1229 if (q->subpacket[s].subbands > 50) {
1230 av_log_ask_for_sample(avctx, "subbands > 50\n");
1231 return AVERROR_PATCHWELCOME;
1233 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1234 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1235 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1236 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1238 if (q->num_subpackets + q->subpacket[s].num_channels > q->nb_channels) {
1239 av_log(avctx, AV_LOG_ERROR, "Too many subpackets %d for channels %d\n", q->num_subpackets, q->nb_channels);
1240 return AVERROR_INVALIDDATA;
1243 q->num_subpackets++;
1245 if (s > MAX_SUBPACKETS) {
1246 av_log_ask_for_sample(avctx, "Too many subpackets > 5\n");
1247 return AVERROR_PATCHWELCOME;
1250 /* Generate tables */
1253 init_cplscales_table(q);
1255 if ((ret = init_cook_vlc_tables(q)))
1259 if (avctx->block_align >= UINT_MAX / 2)
1260 return AVERROR(EINVAL);
1262 /* Pad the databuffer with:
1263 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1264 FF_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1265 q->decoded_bytes_buffer =
1266 av_mallocz(avctx->block_align
1267 + DECODE_BYTES_PAD1(avctx->block_align)
1268 + FF_INPUT_BUFFER_PADDING_SIZE);
1269 if (q->decoded_bytes_buffer == NULL)
1270 return AVERROR(ENOMEM);
1272 /* Initialize transform. */
1273 if ((ret = init_cook_mlt(q)))
1276 /* Initialize COOK signal arithmetic handling */
1278 q->scalar_dequant = scalar_dequant_float;
1279 q->decouple = decouple_float;
1280 q->imlt_window = imlt_window_float;
1281 q->interpolate = interpolate_float;
1282 q->saturate_output = saturate_output_float;
1285 /* Try to catch some obviously faulty streams, othervise it might be exploitable */
1286 if ((q->samples_per_channel == 256) || (q->samples_per_channel == 512)
1287 || (q->samples_per_channel == 1024)) {
1289 av_log_ask_for_sample(avctx,
1290 "unknown amount of samples_per_channel = %d\n",
1291 q->samples_per_channel);
1292 return AVERROR_PATCHWELCOME;
1295 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1297 avctx->channel_layout = channel_mask;
1299 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1301 avcodec_get_frame_defaults(&q->frame);
1302 avctx->coded_frame = &q->frame;
1305 dump_cook_context(q);
1310 AVCodec ff_cook_decoder = {
1312 .type = AVMEDIA_TYPE_AUDIO,
1313 .id = AV_CODEC_ID_COOK,
1314 .priv_data_size = sizeof(COOKContext),
1315 .init = cook_decode_init,
1316 .close = cook_decode_close,
1317 .decode = cook_decode_frame,
1318 .capabilities = CODEC_CAP_DR1,
1319 .long_name = NULL_IF_CONFIG_SMALL("Cook / Cooker / Gecko (RealAudio G2)"),
1320 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1321 AV_SAMPLE_FMT_NONE },