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"
47 #include "libavutil/mem_internal.h"
52 #include "bytestream.h"
60 /* the different Cook versions */
61 #define MONO 0x1000001
62 #define STEREO 0x1000002
63 #define JOINT_STEREO 0x1000003
64 #define MC_COOK 0x2000000
66 #define SUBBAND_SIZE 20
67 #define MAX_SUBPACKETS 5
69 #define QUANT_VLC_BITS 9
70 #define COUPLING_VLC_BITS 6
72 typedef struct cook_gains {
77 typedef struct COOKSubpacket {
85 int samples_per_channel;
86 int log2_numvector_size;
87 unsigned int channel_mask;
90 int bits_per_subpacket;
93 int numvector_size; // 1 << log2_numvector_size;
95 float mono_previous_buffer1[1024];
96 float mono_previous_buffer2[1024];
106 typedef struct cook {
108 * The following 5 functions provide the lowlevel arithmetic on
109 * the internal audio buffers.
111 void (*scalar_dequant)(struct cook *q, int index, int quant_index,
112 int *subband_coef_index, int *subband_coef_sign,
115 void (*decouple)(struct cook *q,
119 float *decode_buffer,
120 float *mlt_buffer1, float *mlt_buffer2);
122 void (*imlt_window)(struct cook *q, float *buffer1,
123 cook_gains *gains_ptr, float *previous_buffer);
125 void (*interpolate)(struct cook *q, float *buffer,
126 int gain_index, int gain_index_next);
128 void (*saturate_output)(struct cook *q, float *out);
130 AVCodecContext* avctx;
131 AudioDSPContext adsp;
135 int samples_per_channel;
138 int discarded_packets;
145 VLC envelope_quant_index[13];
146 VLC sqvh[7]; // scalar quantization
148 /* generate tables and related variables */
149 int gain_size_factor;
150 float gain_table[31];
154 uint8_t* decoded_bytes_buffer;
155 DECLARE_ALIGNED(32, float, mono_mdct_output)[2048];
156 float decode_buffer_1[1024];
157 float decode_buffer_2[1024];
158 float decode_buffer_0[1060]; /* static allocation for joint decode */
160 const float *cplscales[5];
162 COOKSubpacket subpacket[MAX_SUBPACKETS];
165 static float pow2tab[127];
166 static float rootpow2tab[127];
168 /*************** init functions ***************/
170 /* table generator */
171 static av_cold void init_pow2table(void)
173 /* fast way of computing 2^i and 2^(0.5*i) for -63 <= i < 64 */
175 static const float exp2_tab[2] = {1, M_SQRT2};
176 float exp2_val = powf(2, -63);
177 float root_val = powf(2, -32);
178 for (i = -63; i < 64; i++) {
181 pow2tab[63 + i] = exp2_val;
182 rootpow2tab[63 + i] = root_val * exp2_tab[i & 1];
187 /* table generator */
188 static av_cold void init_gain_table(COOKContext *q)
191 q->gain_size_factor = q->samples_per_channel / 8;
192 for (i = 0; i < 31; i++)
193 q->gain_table[i] = pow(pow2tab[i + 48],
194 (1.0 / (double) q->gain_size_factor));
197 static av_cold int build_vlc(VLC *vlc, int nb_bits, const uint8_t counts[16],
198 const void *syms, int symbol_size, int offset,
201 uint8_t lens[MAX_COOK_VLC_ENTRIES];
204 for (int i = 0; i < 16; i++)
205 for (unsigned count = num + counts[i]; num < count; num++)
208 return ff_init_vlc_from_lengths(vlc, nb_bits, num, lens, 1,
209 syms, symbol_size, symbol_size,
213 static av_cold int init_cook_vlc_tables(COOKContext *q)
218 for (i = 0; i < 13; i++) {
219 result |= build_vlc(&q->envelope_quant_index[i], QUANT_VLC_BITS,
220 envelope_quant_index_huffcounts[i],
221 envelope_quant_index_huffsyms[i], 1, -12, q->avctx);
223 av_log(q->avctx, AV_LOG_DEBUG, "sqvh VLC init\n");
224 for (i = 0; i < 7; i++) {
225 int sym_size = 1 + (i == 3);
226 result |= build_vlc(&q->sqvh[i], vhvlcsize_tab[i],
228 cvh_huffsyms[i], sym_size, 0, q->avctx);
231 for (i = 0; i < q->num_subpackets; i++) {
232 if (q->subpacket[i].joint_stereo == 1) {
233 result |= build_vlc(&q->subpacket[i].channel_coupling, COUPLING_VLC_BITS,
234 ccpl_huffcounts[q->subpacket[i].js_vlc_bits - 2],
235 ccpl_huffsyms[q->subpacket[i].js_vlc_bits - 2], 1,
237 av_log(q->avctx, AV_LOG_DEBUG, "subpacket %i Joint-stereo VLC used.\n", i);
241 av_log(q->avctx, AV_LOG_DEBUG, "VLC tables initialized.\n");
245 static av_cold int init_cook_mlt(COOKContext *q)
248 int mlt_size = q->samples_per_channel;
250 if ((q->mlt_window = av_malloc_array(mlt_size, sizeof(*q->mlt_window))) == 0)
251 return AVERROR(ENOMEM);
253 /* Initialize the MLT window: simple sine window. */
254 ff_sine_window_init(q->mlt_window, mlt_size);
255 for (j = 0; j < mlt_size; j++)
256 q->mlt_window[j] *= sqrt(2.0 / q->samples_per_channel);
258 /* Initialize the MDCT. */
259 if ((ret = ff_mdct_init(&q->mdct_ctx, av_log2(mlt_size) + 1, 1, 1.0 / 32768.0))) {
260 av_freep(&q->mlt_window);
263 av_log(q->avctx, AV_LOG_DEBUG, "MDCT initialized, order = %d.\n",
264 av_log2(mlt_size) + 1);
269 static av_cold void init_cplscales_table(COOKContext *q)
272 for (i = 0; i < 5; i++)
273 q->cplscales[i] = cplscales[i];
276 /*************** init functions end ***********/
278 #define DECODE_BYTES_PAD1(bytes) (3 - ((bytes) + 3) % 4)
279 #define DECODE_BYTES_PAD2(bytes) ((bytes) % 4 + DECODE_BYTES_PAD1(2 * (bytes)))
282 * Cook indata decoding, every 32 bits are XORed with 0x37c511f2.
283 * Why? No idea, some checksum/error detection method maybe.
285 * Out buffer size: extra bytes are needed to cope with
286 * padding/misalignment.
287 * Subpackets passed to the decoder can contain two, consecutive
288 * half-subpackets, of identical but arbitrary size.
289 * 1234 1234 1234 1234 extraA extraB
290 * Case 1: AAAA BBBB 0 0
291 * Case 2: AAAA ABBB BB-- 3 3
292 * Case 3: AAAA AABB BBBB 2 2
293 * Case 4: AAAA AAAB BBBB BB-- 1 5
295 * Nice way to waste CPU cycles.
297 * @param inbuffer pointer to byte array of indata
298 * @param out pointer to byte array of outdata
299 * @param bytes number of bytes
301 static inline int decode_bytes(const uint8_t *inbuffer, uint8_t *out, int bytes)
303 static const uint32_t tab[4] = {
304 AV_BE2NE32C(0x37c511f2u), AV_BE2NE32C(0xf237c511u),
305 AV_BE2NE32C(0x11f237c5u), AV_BE2NE32C(0xc511f237u),
310 uint32_t *obuf = (uint32_t *) out;
311 /* FIXME: 64 bit platforms would be able to do 64 bits at a time.
312 * I'm too lazy though, should be something like
313 * for (i = 0; i < bitamount / 64; i++)
314 * (int64_t) out[i] = 0x37c511f237c511f2 ^ av_be2ne64(int64_t) in[i]);
315 * Buffer alignment needs to be checked. */
317 off = (intptr_t) inbuffer & 3;
318 buf = (const uint32_t *) (inbuffer - off);
321 for (i = 0; i < bytes / 4; i++)
322 obuf[i] = c ^ buf[i];
327 static av_cold int cook_decode_close(AVCodecContext *avctx)
330 COOKContext *q = avctx->priv_data;
331 av_log(avctx, AV_LOG_DEBUG, "Deallocating memory.\n");
333 /* Free allocated memory buffers. */
334 av_freep(&q->mlt_window);
335 av_freep(&q->decoded_bytes_buffer);
337 /* Free the transform. */
338 ff_mdct_end(&q->mdct_ctx);
340 /* Free the VLC tables. */
341 for (i = 0; i < 13; i++)
342 ff_free_vlc(&q->envelope_quant_index[i]);
343 for (i = 0; i < 7; i++)
344 ff_free_vlc(&q->sqvh[i]);
345 for (i = 0; i < q->num_subpackets; i++)
346 ff_free_vlc(&q->subpacket[i].channel_coupling);
348 av_log(avctx, AV_LOG_DEBUG, "Memory deallocated.\n");
354 * Fill the gain array for the timedomain quantization.
356 * @param gb pointer to the GetBitContext
357 * @param gaininfo array[9] of gain indexes
359 static void decode_gain_info(GetBitContext *gb, int *gaininfo)
363 n = get_unary(gb, 0, get_bits_left(gb)); // amount of elements*2 to update
367 int index = get_bits(gb, 3);
368 int gain = get_bits1(gb) ? get_bits(gb, 4) - 7 : -1;
371 gaininfo[i++] = gain;
378 * Create the quant index table needed for the envelope.
380 * @param q pointer to the COOKContext
381 * @param quant_index_table pointer to the array
383 static int decode_envelope(COOKContext *q, COOKSubpacket *p,
384 int *quant_index_table)
388 quant_index_table[0] = get_bits(&q->gb, 6) - 6; // This is used later in categorize
390 for (i = 1; i < p->total_subbands; i++) {
392 if (i >= p->js_subband_start * 2) {
393 vlc_index -= p->js_subband_start;
400 vlc_index = 13; // the VLC tables >13 are identical to No. 13
402 j = get_vlc2(&q->gb, q->envelope_quant_index[vlc_index - 1].table,
404 quant_index_table[i] = quant_index_table[i - 1] + j; // differential encoding
405 if (quant_index_table[i] > 63 || quant_index_table[i] < -63) {
406 av_log(q->avctx, AV_LOG_ERROR,
407 "Invalid quantizer %d at position %d, outside [-63, 63] range\n",
408 quant_index_table[i], i);
409 return AVERROR_INVALIDDATA;
417 * Calculate the category and category_index vector.
419 * @param q pointer to the COOKContext
420 * @param quant_index_table pointer to the array
421 * @param category pointer to the category array
422 * @param category_index pointer to the category_index array
424 static void categorize(COOKContext *q, COOKSubpacket *p, const int *quant_index_table,
425 int *category, int *category_index)
427 int exp_idx, bias, tmpbias1, tmpbias2, bits_left, num_bits, index, v, i, j;
428 int exp_index2[102] = { 0 };
429 int exp_index1[102] = { 0 };
431 int tmp_categorize_array[128 * 2] = { 0 };
432 int tmp_categorize_array1_idx = p->numvector_size;
433 int tmp_categorize_array2_idx = p->numvector_size;
435 bits_left = p->bits_per_subpacket - get_bits_count(&q->gb);
437 if (bits_left > q->samples_per_channel)
438 bits_left = q->samples_per_channel +
439 ((bits_left - q->samples_per_channel) * 5) / 8;
444 for (i = 32; i > 0; i = i / 2) {
447 for (j = p->total_subbands; j > 0; j--) {
448 exp_idx = av_clip_uintp2((i - quant_index_table[index] + bias) / 2, 3);
450 num_bits += expbits_tab[exp_idx];
452 if (num_bits >= bits_left - 32)
456 /* Calculate total number of bits. */
458 for (i = 0; i < p->total_subbands; i++) {
459 exp_idx = av_clip_uintp2((bias - quant_index_table[i]) / 2, 3);
460 num_bits += expbits_tab[exp_idx];
461 exp_index1[i] = exp_idx;
462 exp_index2[i] = exp_idx;
464 tmpbias1 = tmpbias2 = num_bits;
466 for (j = 1; j < p->numvector_size; j++) {
467 if (tmpbias1 + tmpbias2 > 2 * bits_left) { /* ---> */
470 for (i = 0; i < p->total_subbands; i++) {
471 if (exp_index1[i] < 7) {
472 v = (-2 * exp_index1[i]) - quant_index_table[i] + bias;
481 tmp_categorize_array[tmp_categorize_array1_idx++] = index;
482 tmpbias1 -= expbits_tab[exp_index1[index]] -
483 expbits_tab[exp_index1[index] + 1];
488 for (i = 0; i < p->total_subbands; i++) {
489 if (exp_index2[i] > 0) {
490 v = (-2 * exp_index2[i]) - quant_index_table[i] + bias;
499 tmp_categorize_array[--tmp_categorize_array2_idx] = index;
500 tmpbias2 -= expbits_tab[exp_index2[index]] -
501 expbits_tab[exp_index2[index] - 1];
506 for (i = 0; i < p->total_subbands; i++)
507 category[i] = exp_index2[i];
509 for (i = 0; i < p->numvector_size - 1; i++)
510 category_index[i] = tmp_categorize_array[tmp_categorize_array2_idx++];
515 * Expand the category vector.
517 * @param q pointer to the COOKContext
518 * @param category pointer to the category array
519 * @param category_index pointer to the category_index array
521 static inline void expand_category(COOKContext *q, int *category,
525 for (i = 0; i < q->num_vectors; i++)
527 int idx = category_index[i];
528 if (++category[idx] >= FF_ARRAY_ELEMS(dither_tab))
534 * The real requantization of the mltcoefs
536 * @param q pointer to the COOKContext
538 * @param quant_index quantisation index
539 * @param subband_coef_index array of indexes to quant_centroid_tab
540 * @param subband_coef_sign signs of coefficients
541 * @param mlt_p pointer into the mlt buffer
543 static void scalar_dequant_float(COOKContext *q, int index, int quant_index,
544 int *subband_coef_index, int *subband_coef_sign,
550 for (i = 0; i < SUBBAND_SIZE; i++) {
551 if (subband_coef_index[i]) {
552 f1 = quant_centroid_tab[index][subband_coef_index[i]];
553 if (subband_coef_sign[i])
556 /* noise coding if subband_coef_index[i] == 0 */
557 f1 = dither_tab[index];
558 if (av_lfg_get(&q->random_state) < 0x80000000)
561 mlt_p[i] = f1 * rootpow2tab[quant_index + 63];
565 * Unpack the subband_coef_index and subband_coef_sign vectors.
567 * @param q pointer to the COOKContext
568 * @param category pointer to the category array
569 * @param subband_coef_index array of indexes to quant_centroid_tab
570 * @param subband_coef_sign signs of coefficients
572 static int unpack_SQVH(COOKContext *q, COOKSubpacket *p, int category,
573 int *subband_coef_index, int *subband_coef_sign)
576 int vlc, vd, tmp, result;
578 vd = vd_tab[category];
580 for (i = 0; i < vpr_tab[category]; i++) {
581 vlc = get_vlc2(&q->gb, q->sqvh[category].table, q->sqvh[category].bits, 3);
582 if (p->bits_per_subpacket < get_bits_count(&q->gb)) {
586 for (j = vd - 1; j >= 0; j--) {
587 tmp = (vlc * invradix_tab[category]) / 0x100000;
588 subband_coef_index[vd * i + j] = vlc - tmp * (kmax_tab[category] + 1);
591 for (j = 0; j < vd; j++) {
592 if (subband_coef_index[i * vd + j]) {
593 if (get_bits_count(&q->gb) < p->bits_per_subpacket) {
594 subband_coef_sign[i * vd + j] = get_bits1(&q->gb);
597 subband_coef_sign[i * vd + j] = 0;
600 subband_coef_sign[i * vd + j] = 0;
609 * Fill the mlt_buffer with mlt coefficients.
611 * @param q pointer to the COOKContext
612 * @param category pointer to the category array
613 * @param quant_index_table pointer to the array
614 * @param mlt_buffer pointer to mlt coefficients
616 static void decode_vectors(COOKContext *q, COOKSubpacket *p, int *category,
617 int *quant_index_table, float *mlt_buffer)
619 /* A zero in this table means that the subband coefficient is
620 random noise coded. */
621 int subband_coef_index[SUBBAND_SIZE];
622 /* A zero in this table means that the subband coefficient is a
623 positive multiplicator. */
624 int subband_coef_sign[SUBBAND_SIZE];
628 for (band = 0; band < p->total_subbands; band++) {
629 index = category[band];
630 if (category[band] < 7) {
631 if (unpack_SQVH(q, p, category[band], subband_coef_index, subband_coef_sign)) {
633 for (j = 0; j < p->total_subbands; j++)
634 category[band + j] = 7;
638 memset(subband_coef_index, 0, sizeof(subband_coef_index));
639 memset(subband_coef_sign, 0, sizeof(subband_coef_sign));
641 q->scalar_dequant(q, index, quant_index_table[band],
642 subband_coef_index, subband_coef_sign,
643 &mlt_buffer[band * SUBBAND_SIZE]);
646 /* FIXME: should this be removed, or moved into loop above? */
647 if (p->total_subbands * SUBBAND_SIZE >= q->samples_per_channel)
652 static int mono_decode(COOKContext *q, COOKSubpacket *p, float *mlt_buffer)
654 int category_index[128] = { 0 };
655 int category[128] = { 0 };
656 int quant_index_table[102];
659 if ((res = decode_envelope(q, p, quant_index_table)) < 0)
661 q->num_vectors = get_bits(&q->gb, p->log2_numvector_size);
662 categorize(q, p, quant_index_table, category, category_index);
663 expand_category(q, category, category_index);
664 for (i=0; i<p->total_subbands; i++) {
666 return AVERROR_INVALIDDATA;
668 decode_vectors(q, p, category, quant_index_table, mlt_buffer);
675 * the actual requantization of the timedomain samples
677 * @param q pointer to the COOKContext
678 * @param buffer pointer to the timedomain buffer
679 * @param gain_index index for the block multiplier
680 * @param gain_index_next index for the next block multiplier
682 static void interpolate_float(COOKContext *q, float *buffer,
683 int gain_index, int gain_index_next)
687 fc1 = pow2tab[gain_index + 63];
689 if (gain_index == gain_index_next) { // static gain
690 for (i = 0; i < q->gain_size_factor; i++)
692 } else { // smooth gain
693 fc2 = q->gain_table[15 + (gain_index_next - gain_index)];
694 for (i = 0; i < q->gain_size_factor; i++) {
702 * Apply transform window, overlap buffers.
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_window_float(COOKContext *q, float *inbuffer,
710 cook_gains *gains_ptr, float *previous_buffer)
712 const float fc = pow2tab[gains_ptr->previous[0] + 63];
714 /* The weird thing here, is that the two halves of the time domain
715 * buffer are swapped. Also, the newest data, that we save away for
716 * next frame, has the wrong sign. Hence the subtraction below.
717 * Almost sounds like a complex conjugate/reverse data/FFT effect.
720 /* Apply window and overlap */
721 for (i = 0; i < q->samples_per_channel; i++)
722 inbuffer[i] = inbuffer[i] * fc * q->mlt_window[i] -
723 previous_buffer[i] * q->mlt_window[q->samples_per_channel - 1 - i];
727 * The modulated lapped transform, this takes transform coefficients
728 * and transforms them into timedomain samples.
729 * Apply transform window, overlap buffers, apply gain profile
730 * and buffer management.
732 * @param q pointer to the COOKContext
733 * @param inbuffer pointer to the mltcoefficients
734 * @param gains_ptr current and previous gains
735 * @param previous_buffer pointer to the previous buffer to be used for overlapping
737 static void imlt_gain(COOKContext *q, float *inbuffer,
738 cook_gains *gains_ptr, float *previous_buffer)
740 float *buffer0 = q->mono_mdct_output;
741 float *buffer1 = q->mono_mdct_output + q->samples_per_channel;
744 /* Inverse modified discrete cosine transform */
745 q->mdct_ctx.imdct_calc(&q->mdct_ctx, q->mono_mdct_output, inbuffer);
747 q->imlt_window(q, buffer1, gains_ptr, previous_buffer);
749 /* Apply gain profile */
750 for (i = 0; i < 8; i++)
751 if (gains_ptr->now[i] || gains_ptr->now[i + 1])
752 q->interpolate(q, &buffer1[q->gain_size_factor * i],
753 gains_ptr->now[i], gains_ptr->now[i + 1]);
755 /* Save away the current to be previous block. */
756 memcpy(previous_buffer, buffer0,
757 q->samples_per_channel * sizeof(*previous_buffer));
762 * function for getting the jointstereo coupling information
764 * @param q pointer to the COOKContext
765 * @param decouple_tab decoupling array
767 static int decouple_info(COOKContext *q, COOKSubpacket *p, int *decouple_tab)
770 int vlc = get_bits1(&q->gb);
771 int start = cplband[p->js_subband_start];
772 int end = cplband[p->subbands - 1];
773 int length = end - start + 1;
779 for (i = 0; i < length; i++)
780 decouple_tab[start + i] = get_vlc2(&q->gb,
781 p->channel_coupling.table,
782 COUPLING_VLC_BITS, 3);
784 for (i = 0; i < length; i++) {
785 int v = get_bits(&q->gb, p->js_vlc_bits);
786 if (v == (1<<p->js_vlc_bits)-1) {
787 av_log(q->avctx, AV_LOG_ERROR, "decouple value too large\n");
788 return AVERROR_INVALIDDATA;
790 decouple_tab[start + i] = v;
796 * function decouples a pair of signals from a single signal via multiplication.
798 * @param q pointer to the COOKContext
799 * @param subband index of the current subband
800 * @param f1 multiplier for channel 1 extraction
801 * @param f2 multiplier for channel 2 extraction
802 * @param decode_buffer input buffer
803 * @param mlt_buffer1 pointer to left channel mlt coefficients
804 * @param mlt_buffer2 pointer to right channel mlt coefficients
806 static void decouple_float(COOKContext *q,
810 float *decode_buffer,
811 float *mlt_buffer1, float *mlt_buffer2)
814 for (j = 0; j < SUBBAND_SIZE; j++) {
815 tmp_idx = ((p->js_subband_start + subband) * SUBBAND_SIZE) + j;
816 mlt_buffer1[SUBBAND_SIZE * subband + j] = f1 * decode_buffer[tmp_idx];
817 mlt_buffer2[SUBBAND_SIZE * subband + j] = f2 * decode_buffer[tmp_idx];
822 * function for decoding joint stereo data
824 * @param q pointer to the COOKContext
825 * @param mlt_buffer1 pointer to left channel mlt coefficients
826 * @param mlt_buffer2 pointer to right channel mlt coefficients
828 static int joint_decode(COOKContext *q, COOKSubpacket *p,
829 float *mlt_buffer_left, float *mlt_buffer_right)
832 int decouple_tab[SUBBAND_SIZE] = { 0 };
833 float *decode_buffer = q->decode_buffer_0;
836 const float *cplscale;
838 memset(decode_buffer, 0, sizeof(q->decode_buffer_0));
840 /* Make sure the buffers are zeroed out. */
841 memset(mlt_buffer_left, 0, 1024 * sizeof(*mlt_buffer_left));
842 memset(mlt_buffer_right, 0, 1024 * sizeof(*mlt_buffer_right));
843 if ((res = decouple_info(q, p, decouple_tab)) < 0)
845 if ((res = mono_decode(q, p, decode_buffer)) < 0)
847 /* The two channels are stored interleaved in decode_buffer. */
848 for (i = 0; i < p->js_subband_start; i++) {
849 for (j = 0; j < SUBBAND_SIZE; j++) {
850 mlt_buffer_left[i * 20 + j] = decode_buffer[i * 40 + j];
851 mlt_buffer_right[i * 20 + j] = decode_buffer[i * 40 + 20 + j];
855 /* When we reach js_subband_start (the higher frequencies)
856 the coefficients are stored in a coupling scheme. */
857 idx = (1 << p->js_vlc_bits) - 1;
858 for (i = p->js_subband_start; i < p->subbands; i++) {
859 cpl_tmp = cplband[i];
860 idx -= decouple_tab[cpl_tmp];
861 cplscale = q->cplscales[p->js_vlc_bits - 2]; // choose decoupler table
862 f1 = cplscale[decouple_tab[cpl_tmp] + 1];
864 q->decouple(q, p, i, f1, f2, decode_buffer,
865 mlt_buffer_left, mlt_buffer_right);
866 idx = (1 << p->js_vlc_bits) - 1;
873 * First part of subpacket decoding:
874 * decode raw stream bytes and read gain info.
876 * @param q pointer to the COOKContext
877 * @param inbuffer pointer to raw stream data
878 * @param gains_ptr array of current/prev gain pointers
880 static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
881 const uint8_t *inbuffer,
882 cook_gains *gains_ptr)
886 offset = decode_bytes(inbuffer, q->decoded_bytes_buffer,
887 p->bits_per_subpacket / 8);
888 init_get_bits(&q->gb, q->decoded_bytes_buffer + offset,
889 p->bits_per_subpacket);
890 decode_gain_info(&q->gb, gains_ptr->now);
892 /* Swap current and previous gains */
893 FFSWAP(int *, gains_ptr->now, gains_ptr->previous);
897 * Saturate the output signal and interleave.
899 * @param q pointer to the COOKContext
900 * @param out pointer to the output vector
902 static void saturate_output_float(COOKContext *q, float *out)
904 q->adsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
905 FFALIGN(q->samples_per_channel, 8), -1.0f, 1.0f);
910 * Final part of subpacket decoding:
911 * Apply modulated lapped transform, gain compensation,
912 * clip and convert to integer.
914 * @param q pointer to the COOKContext
915 * @param decode_buffer pointer to the mlt coefficients
916 * @param gains_ptr array of current/prev gain pointers
917 * @param previous_buffer pointer to the previous buffer to be used for overlapping
918 * @param out pointer to the output buffer
920 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
921 cook_gains *gains_ptr, float *previous_buffer,
924 imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
926 q->saturate_output(q, out);
931 * Cook subpacket decoding. This function returns one decoded subpacket,
932 * usually 1024 samples per channel.
934 * @param q pointer to the COOKContext
935 * @param inbuffer pointer to the inbuffer
936 * @param outbuffer pointer to the outbuffer
938 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
939 const uint8_t *inbuffer, float **outbuffer)
941 int sub_packet_size = p->size;
944 memset(q->decode_buffer_1, 0, sizeof(q->decode_buffer_1));
945 decode_bytes_and_gain(q, p, inbuffer, &p->gains1);
947 if (p->joint_stereo) {
948 if ((res = joint_decode(q, p, q->decode_buffer_1, q->decode_buffer_2)) < 0)
951 if ((res = mono_decode(q, p, q->decode_buffer_1)) < 0)
954 if (p->num_channels == 2) {
955 decode_bytes_and_gain(q, p, inbuffer + sub_packet_size / 2, &p->gains2);
956 if ((res = mono_decode(q, p, q->decode_buffer_2)) < 0)
961 mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
962 p->mono_previous_buffer1,
963 outbuffer ? outbuffer[p->ch_idx] : NULL);
965 if (p->num_channels == 2) {
967 mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
968 p->mono_previous_buffer2,
969 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
971 mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
972 p->mono_previous_buffer2,
973 outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
980 static int cook_decode_frame(AVCodecContext *avctx, void *data,
981 int *got_frame_ptr, AVPacket *avpkt)
983 AVFrame *frame = data;
984 const uint8_t *buf = avpkt->data;
985 int buf_size = avpkt->size;
986 COOKContext *q = avctx->priv_data;
987 float **samples = NULL;
992 if (buf_size < avctx->block_align)
995 /* get output buffer */
996 if (q->discarded_packets >= 2) {
997 frame->nb_samples = q->samples_per_channel;
998 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1000 samples = (float **)frame->extended_data;
1003 /* estimate subpacket sizes */
1004 q->subpacket[0].size = avctx->block_align;
1006 for (i = 1; i < q->num_subpackets; i++) {
1007 q->subpacket[i].size = 2 * buf[avctx->block_align - q->num_subpackets + i];
1008 q->subpacket[0].size -= q->subpacket[i].size + 1;
1009 if (q->subpacket[0].size < 0) {
1010 av_log(avctx, AV_LOG_DEBUG,
1011 "frame subpacket size total > avctx->block_align!\n");
1012 return AVERROR_INVALIDDATA;
1016 /* decode supbackets */
1017 for (i = 0; i < q->num_subpackets; i++) {
1018 q->subpacket[i].bits_per_subpacket = (q->subpacket[i].size * 8) >>
1019 q->subpacket[i].bits_per_subpdiv;
1020 q->subpacket[i].ch_idx = chidx;
1021 av_log(avctx, AV_LOG_DEBUG,
1022 "subpacket[%i] size %i js %i %i block_align %i\n",
1023 i, q->subpacket[i].size, q->subpacket[i].joint_stereo, offset,
1024 avctx->block_align);
1026 if ((ret = decode_subpacket(q, &q->subpacket[i], buf + offset, samples)) < 0)
1028 offset += q->subpacket[i].size;
1029 chidx += q->subpacket[i].num_channels;
1030 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i] %i %i\n",
1031 i, q->subpacket[i].size * 8, get_bits_count(&q->gb));
1034 /* Discard the first two frames: no valid audio. */
1035 if (q->discarded_packets < 2) {
1036 q->discarded_packets++;
1038 return avctx->block_align;
1043 return avctx->block_align;
1046 static void dump_cook_context(COOKContext *q)
1049 #define PRINT(a, b) ff_dlog(q->avctx, " %s = %d\n", a, b);
1050 ff_dlog(q->avctx, "COOKextradata\n");
1051 ff_dlog(q->avctx, "cookversion=%x\n", q->subpacket[0].cookversion);
1052 if (q->subpacket[0].cookversion > STEREO) {
1053 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1054 PRINT("js_vlc_bits", q->subpacket[0].js_vlc_bits);
1056 ff_dlog(q->avctx, "COOKContext\n");
1057 PRINT("nb_channels", q->avctx->channels);
1058 PRINT("bit_rate", (int)q->avctx->bit_rate);
1059 PRINT("sample_rate", q->avctx->sample_rate);
1060 PRINT("samples_per_channel", q->subpacket[0].samples_per_channel);
1061 PRINT("subbands", q->subpacket[0].subbands);
1062 PRINT("js_subband_start", q->subpacket[0].js_subband_start);
1063 PRINT("log2_numvector_size", q->subpacket[0].log2_numvector_size);
1064 PRINT("numvector_size", q->subpacket[0].numvector_size);
1065 PRINT("total_subbands", q->subpacket[0].total_subbands);
1069 * Cook initialization
1071 * @param avctx pointer to the AVCodecContext
1073 static av_cold int cook_decode_init(AVCodecContext *avctx)
1075 COOKContext *q = avctx->priv_data;
1078 unsigned int channel_mask = 0;
1079 int samples_per_frame = 0;
1083 /* Take care of the codec specific extradata. */
1084 if (avctx->extradata_size < 8) {
1085 av_log(avctx, AV_LOG_ERROR, "Necessary extradata missing!\n");
1086 return AVERROR_INVALIDDATA;
1088 av_log(avctx, AV_LOG_DEBUG, "codecdata_length=%d\n", avctx->extradata_size);
1090 bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
1092 /* Take data from the AVCodecContext (RM container). */
1093 if (!avctx->channels) {
1094 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1095 return AVERROR_INVALIDDATA;
1098 if (avctx->block_align >= INT_MAX / 8)
1099 return AVERROR(EINVAL);
1101 /* Initialize RNG. */
1102 av_lfg_init(&q->random_state, 0);
1104 ff_audiodsp_init(&q->adsp);
1106 while (bytestream2_get_bytes_left(&gb)) {
1107 if (s >= FFMIN(MAX_SUBPACKETS, avctx->block_align)) {
1108 avpriv_request_sample(avctx, "subpackets > %d", FFMIN(MAX_SUBPACKETS, avctx->block_align));
1109 return AVERROR_PATCHWELCOME;
1111 /* 8 for mono, 16 for stereo, ? for multichannel
1112 Swap to right endianness so we don't need to care later on. */
1113 q->subpacket[s].cookversion = bytestream2_get_be32(&gb);
1114 samples_per_frame = bytestream2_get_be16(&gb);
1115 q->subpacket[s].subbands = bytestream2_get_be16(&gb);
1116 bytestream2_get_be32(&gb); // Unknown unused
1117 q->subpacket[s].js_subband_start = bytestream2_get_be16(&gb);
1118 if (q->subpacket[s].js_subband_start >= 51) {
1119 av_log(avctx, AV_LOG_ERROR, "js_subband_start %d is too large\n", q->subpacket[s].js_subband_start);
1120 return AVERROR_INVALIDDATA;
1122 q->subpacket[s].js_vlc_bits = bytestream2_get_be16(&gb);
1124 /* Initialize extradata related variables. */
1125 q->subpacket[s].samples_per_channel = samples_per_frame / avctx->channels;
1126 q->subpacket[s].bits_per_subpacket = avctx->block_align * 8;
1128 /* Initialize default data states. */
1129 q->subpacket[s].log2_numvector_size = 5;
1130 q->subpacket[s].total_subbands = q->subpacket[s].subbands;
1131 q->subpacket[s].num_channels = 1;
1133 /* Initialize version-dependent variables */
1135 av_log(avctx, AV_LOG_DEBUG, "subpacket[%i].cookversion=%x\n", s,
1136 q->subpacket[s].cookversion);
1137 q->subpacket[s].joint_stereo = 0;
1138 switch (q->subpacket[s].cookversion) {
1140 if (avctx->channels != 1) {
1141 avpriv_request_sample(avctx, "Container channels != 1");
1142 return AVERROR_PATCHWELCOME;
1144 av_log(avctx, AV_LOG_DEBUG, "MONO\n");
1147 if (avctx->channels != 1) {
1148 q->subpacket[s].bits_per_subpdiv = 1;
1149 q->subpacket[s].num_channels = 2;
1151 av_log(avctx, AV_LOG_DEBUG, "STEREO\n");
1154 if (avctx->channels != 2) {
1155 avpriv_request_sample(avctx, "Container channels != 2");
1156 return AVERROR_PATCHWELCOME;
1158 av_log(avctx, AV_LOG_DEBUG, "JOINT_STEREO\n");
1159 if (avctx->extradata_size >= 16) {
1160 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1161 q->subpacket[s].js_subband_start;
1162 q->subpacket[s].joint_stereo = 1;
1163 q->subpacket[s].num_channels = 2;
1165 if (q->subpacket[s].samples_per_channel > 256) {
1166 q->subpacket[s].log2_numvector_size = 6;
1168 if (q->subpacket[s].samples_per_channel > 512) {
1169 q->subpacket[s].log2_numvector_size = 7;
1173 av_log(avctx, AV_LOG_DEBUG, "MULTI_CHANNEL\n");
1174 channel_mask |= q->subpacket[s].channel_mask = bytestream2_get_be32(&gb);
1176 if (av_get_channel_layout_nb_channels(q->subpacket[s].channel_mask) > 1) {
1177 q->subpacket[s].total_subbands = q->subpacket[s].subbands +
1178 q->subpacket[s].js_subband_start;
1179 q->subpacket[s].joint_stereo = 1;
1180 q->subpacket[s].num_channels = 2;
1181 q->subpacket[s].samples_per_channel = samples_per_frame >> 1;
1183 if (q->subpacket[s].samples_per_channel > 256) {
1184 q->subpacket[s].log2_numvector_size = 6;
1186 if (q->subpacket[s].samples_per_channel > 512) {
1187 q->subpacket[s].log2_numvector_size = 7;
1190 q->subpacket[s].samples_per_channel = samples_per_frame;
1194 avpriv_request_sample(avctx, "Cook version %d",
1195 q->subpacket[s].cookversion);
1196 return AVERROR_PATCHWELCOME;
1199 if (s > 1 && q->subpacket[s].samples_per_channel != q->samples_per_channel) {
1200 av_log(avctx, AV_LOG_ERROR, "different number of samples per channel!\n");
1201 return AVERROR_INVALIDDATA;
1203 q->samples_per_channel = q->subpacket[0].samples_per_channel;
1206 /* Initialize variable relations */
1207 q->subpacket[s].numvector_size = (1 << q->subpacket[s].log2_numvector_size);
1209 /* Try to catch some obviously faulty streams, otherwise it might be exploitable */
1210 if (q->subpacket[s].total_subbands > 53) {
1211 avpriv_request_sample(avctx, "total_subbands > 53");
1212 return AVERROR_PATCHWELCOME;
1215 if ((q->subpacket[s].js_vlc_bits > 6) ||
1216 (q->subpacket[s].js_vlc_bits < 2 * q->subpacket[s].joint_stereo)) {
1217 av_log(avctx, AV_LOG_ERROR, "js_vlc_bits = %d, only >= %d and <= 6 allowed!\n",
1218 q->subpacket[s].js_vlc_bits, 2 * q->subpacket[s].joint_stereo);
1219 return AVERROR_INVALIDDATA;
1222 if (q->subpacket[s].subbands > 50) {
1223 avpriv_request_sample(avctx, "subbands > 50");
1224 return AVERROR_PATCHWELCOME;
1226 if (q->subpacket[s].subbands == 0) {
1227 avpriv_request_sample(avctx, "subbands = 0");
1228 return AVERROR_PATCHWELCOME;
1230 q->subpacket[s].gains1.now = q->subpacket[s].gain_1;
1231 q->subpacket[s].gains1.previous = q->subpacket[s].gain_2;
1232 q->subpacket[s].gains2.now = q->subpacket[s].gain_3;
1233 q->subpacket[s].gains2.previous = q->subpacket[s].gain_4;
1235 if (q->num_subpackets + q->subpacket[s].num_channels > q->avctx->channels) {
1236 av_log(avctx, AV_LOG_ERROR, "Too many subpackets %d for channels %d\n", q->num_subpackets, q->avctx->channels);
1237 return AVERROR_INVALIDDATA;
1240 q->num_subpackets++;
1244 /* Try to catch some obviously faulty streams, otherwise it might be exploitable */
1245 if (q->samples_per_channel != 256 && q->samples_per_channel != 512 &&
1246 q->samples_per_channel != 1024) {
1247 avpriv_request_sample(avctx, "samples_per_channel = %d",
1248 q->samples_per_channel);
1249 return AVERROR_PATCHWELCOME;
1252 /* Generate tables */
1255 init_cplscales_table(q);
1257 if ((ret = init_cook_vlc_tables(q)))
1260 /* Pad the databuffer with:
1261 DECODE_BYTES_PAD1 or DECODE_BYTES_PAD2 for decode_bytes(),
1262 AV_INPUT_BUFFER_PADDING_SIZE, for the bitstreamreader. */
1263 q->decoded_bytes_buffer =
1264 av_mallocz(avctx->block_align
1265 + DECODE_BYTES_PAD1(avctx->block_align)
1266 + AV_INPUT_BUFFER_PADDING_SIZE);
1267 if (!q->decoded_bytes_buffer)
1268 return AVERROR(ENOMEM);
1270 /* Initialize transform. */
1271 if ((ret = init_cook_mlt(q)))
1274 /* Initialize COOK signal arithmetic handling */
1276 q->scalar_dequant = scalar_dequant_float;
1277 q->decouple = decouple_float;
1278 q->imlt_window = imlt_window_float;
1279 q->interpolate = interpolate_float;
1280 q->saturate_output = saturate_output_float;
1283 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1285 avctx->channel_layout = channel_mask;
1287 avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
1290 dump_cook_context(q);
1295 AVCodec ff_cook_decoder = {
1297 .long_name = NULL_IF_CONFIG_SMALL("Cook / Cooker / Gecko (RealAudio G2)"),
1298 .type = AVMEDIA_TYPE_AUDIO,
1299 .id = AV_CODEC_ID_COOK,
1300 .priv_data_size = sizeof(COOKContext),
1301 .init = cook_decode_init,
1302 .close = cook_decode_close,
1303 .decode = cook_decode_frame,
1304 .capabilities = AV_CODEC_CAP_DR1,
1305 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
1306 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1307 AV_SAMPLE_FMT_NONE },